1 /*
   2  * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "jvm.h"
  27 #include "classfile/classLoader.hpp"
  28 #include "classfile/systemDictionary.hpp"
  29 #include "classfile/vmSymbols.hpp"
  30 #include "code/icBuffer.hpp"
  31 #include "code/vtableStubs.hpp"
  32 #include "compiler/compileBroker.hpp"
  33 #include "compiler/disassembler.hpp"
  34 #include "interpreter/interpreter.hpp"
  35 #include "logging/log.hpp"
  36 #include "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_misc.hpp"
  43 #include "runtime/arguments.hpp"
  44 #include "runtime/atomic.hpp"
  45 #include "runtime/extendedPC.hpp"
  46 #include "runtime/globals.hpp"
  47 #include "runtime/interfaceSupport.inline.hpp"
  48 #include "runtime/java.hpp"
  49 #include "runtime/javaCalls.hpp"
  50 #include "runtime/mutexLocker.hpp"
  51 #include "runtime/objectMonitor.hpp"
  52 #include "runtime/orderAccess.inline.hpp"
  53 #include "runtime/osThread.hpp"
  54 #include "runtime/perfMemory.hpp"
  55 #include "runtime/sharedRuntime.hpp"
  56 #include "runtime/statSampler.hpp"
  57 #include "runtime/stubRoutines.hpp"
  58 #include "runtime/thread.inline.hpp"
  59 #include "runtime/threadCritical.hpp"
  60 #include "runtime/timer.hpp"
  61 #include "runtime/vm_version.hpp"
  62 #include "semaphore_posix.hpp"
  63 #include "services/attachListener.hpp"
  64 #include "services/memTracker.hpp"
  65 #include "services/runtimeService.hpp"
  66 #include "utilities/align.hpp"
  67 #include "utilities/decoder.hpp"
  68 #include "utilities/defaultStream.hpp"
  69 #include "utilities/events.hpp"
  70 #include "utilities/growableArray.hpp"
  71 #include "utilities/macros.hpp"
  72 #include "utilities/vmError.hpp"
  73 
  74 // put OS-includes here
  75 # include <dlfcn.h>
  76 # include <errno.h>
  77 # include <exception>
  78 # include <link.h>
  79 # include <poll.h>
  80 # include <pthread.h>
  81 # include <schedctl.h>
  82 # include <setjmp.h>
  83 # include <signal.h>
  84 # include <stdio.h>
  85 # include <alloca.h>
  86 # include <sys/filio.h>
  87 # include <sys/ipc.h>
  88 # include <sys/lwp.h>
  89 # include <sys/machelf.h>     // for elf Sym structure used by dladdr1
  90 # include <sys/mman.h>
  91 # include <sys/processor.h>
  92 # include <sys/procset.h>
  93 # include <sys/pset.h>
  94 # include <sys/resource.h>
  95 # include <sys/shm.h>
  96 # include <sys/socket.h>
  97 # include <sys/stat.h>
  98 # include <sys/systeminfo.h>
  99 # include <sys/time.h>
 100 # include <sys/times.h>
 101 # include <sys/types.h>
 102 # include <sys/wait.h>
 103 # include <sys/utsname.h>
 104 # include <thread.h>
 105 # include <unistd.h>
 106 # include <sys/priocntl.h>
 107 # include <sys/rtpriocntl.h>
 108 # include <sys/tspriocntl.h>
 109 # include <sys/iapriocntl.h>
 110 # include <sys/fxpriocntl.h>
 111 # include <sys/loadavg.h>
 112 # include <string.h>
 113 # include <stdio.h>
 114 
 115 # define _STRUCTURED_PROC 1  //  this gets us the new structured proc interfaces of 5.6 & later
 116 # include <sys/procfs.h>     //  see comment in <sys/procfs.h>
 117 
 118 #define MAX_PATH (2 * K)
 119 
 120 // for timer info max values which include all bits
 121 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 122 
 123 
 124 // Here are some liblgrp types from sys/lgrp_user.h to be able to
 125 // compile on older systems without this header file.
 126 
 127 #ifndef MADV_ACCESS_LWP
 128   #define  MADV_ACCESS_LWP   7       /* next LWP to access heavily */
 129 #endif
 130 #ifndef MADV_ACCESS_MANY
 131   #define  MADV_ACCESS_MANY  8       /* many processes to access heavily */
 132 #endif
 133 
 134 #ifndef LGRP_RSRC_CPU
 135   #define LGRP_RSRC_CPU      0       /* CPU resources */
 136 #endif
 137 #ifndef LGRP_RSRC_MEM
 138   #define LGRP_RSRC_MEM      1       /* memory resources */
 139 #endif
 140 
 141 // Values for ThreadPriorityPolicy == 1
 142 int prio_policy1[CriticalPriority+1] = {
 143   -99999,  0, 16,  32,  48,  64,
 144           80, 96, 112, 124, 127, 127 };
 145 
 146 // System parameters used internally
 147 static clock_t clock_tics_per_sec = 100;
 148 
 149 // Track if we have called enable_extended_FILE_stdio (on Solaris 10u4+)
 150 static bool enabled_extended_FILE_stdio = false;
 151 
 152 // For diagnostics to print a message once. see run_periodic_checks
 153 static bool check_addr0_done = false;
 154 static sigset_t check_signal_done;
 155 static bool check_signals = true;
 156 
 157 address os::Solaris::handler_start;  // start pc of thr_sighndlrinfo
 158 address os::Solaris::handler_end;    // end pc of thr_sighndlrinfo
 159 
 160 address os::Solaris::_main_stack_base = NULL;  // 4352906 workaround
 161 
 162 os::Solaris::pthread_setname_np_func_t os::Solaris::_pthread_setname_np = NULL;
 163 
 164 // "default" initializers for missing libc APIs
 165 extern "C" {
 166   static int lwp_mutex_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
 167   static int lwp_mutex_destroy(mutex_t *mx)                 { return 0; }
 168 
 169   static int lwp_cond_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
 170   static int lwp_cond_destroy(cond_t *cv)                   { return 0; }
 171 }
 172 
 173 // "default" initializers for pthread-based synchronization
 174 extern "C" {
 175   static int pthread_mutex_default_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
 176   static int pthread_cond_default_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
 177 }
 178 
 179 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 180 
 181 static inline size_t adjust_stack_size(address base, size_t size) {
 182   if ((ssize_t)size < 0) {
 183     // 4759953: Compensate for ridiculous stack size.
 184     size = max_intx;
 185   }
 186   if (size > (size_t)base) {
 187     // 4812466: Make sure size doesn't allow the stack to wrap the address space.
 188     size = (size_t)base;
 189   }
 190   return size;
 191 }
 192 
 193 static inline stack_t get_stack_info() {
 194   stack_t st;
 195   int retval = thr_stksegment(&st);
 196   st.ss_size = adjust_stack_size((address)st.ss_sp, st.ss_size);
 197   assert(retval == 0, "incorrect return value from thr_stksegment");
 198   assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
 199   assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
 200   return st;
 201 }
 202 
 203 bool os::is_primordial_thread(void) {
 204   int r = thr_main();
 205   guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
 206   return r == 1;
 207 }
 208 
 209 address os::current_stack_base() {
 210   bool _is_primordial_thread = is_primordial_thread();
 211 
 212   // Workaround 4352906, avoid calls to thr_stksegment by
 213   // thr_main after the first one (it looks like we trash
 214   // some data, causing the value for ss_sp to be incorrect).
 215   if (!_is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
 216     stack_t st = get_stack_info();
 217     if (_is_primordial_thread) {
 218       // cache initial value of stack base
 219       os::Solaris::_main_stack_base = (address)st.ss_sp;
 220     }
 221     return (address)st.ss_sp;
 222   } else {
 223     guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
 224     return os::Solaris::_main_stack_base;
 225   }
 226 }
 227 
 228 size_t os::current_stack_size() {
 229   size_t size;
 230 
 231   if (!is_primordial_thread()) {
 232     size = get_stack_info().ss_size;
 233   } else {
 234     struct rlimit limits;
 235     getrlimit(RLIMIT_STACK, &limits);
 236     size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
 237   }
 238   // base may not be page aligned
 239   address base = current_stack_base();
 240   address bottom = align_up(base - size, os::vm_page_size());;
 241   return (size_t)(base - bottom);
 242 }
 243 
 244 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
 245   return localtime_r(clock, res);
 246 }
 247 
 248 void os::Solaris::try_enable_extended_io() {
 249   typedef int (*enable_extended_FILE_stdio_t)(int, int);
 250 
 251   if (!UseExtendedFileIO) {
 252     return;
 253   }
 254 
 255   enable_extended_FILE_stdio_t enabler =
 256     (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
 257                                          "enable_extended_FILE_stdio");
 258   if (enabler) {
 259     enabler(-1, -1);
 260   }
 261 }
 262 
 263 static int _processors_online = 0;
 264 
 265 jint os::Solaris::_os_thread_limit = 0;
 266 volatile jint os::Solaris::_os_thread_count = 0;
 267 
 268 julong os::available_memory() {
 269   return Solaris::available_memory();
 270 }
 271 
 272 julong os::Solaris::available_memory() {
 273   return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
 274 }
 275 
 276 julong os::Solaris::_physical_memory = 0;
 277 
 278 julong os::physical_memory() {
 279   return Solaris::physical_memory();
 280 }
 281 
 282 static hrtime_t first_hrtime = 0;
 283 static const hrtime_t hrtime_hz = 1000*1000*1000;
 284 static volatile hrtime_t max_hrtime = 0;
 285 
 286 
 287 void os::Solaris::initialize_system_info() {
 288   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 289   _processors_online = sysconf(_SC_NPROCESSORS_ONLN);
 290   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) *
 291                                      (julong)sysconf(_SC_PAGESIZE);
 292 }
 293 
 294 int os::active_processor_count() {
 295   // User has overridden the number of active processors
 296   if (ActiveProcessorCount > 0) {
 297     log_trace(os)("active_processor_count: "
 298                   "active processor count set by user : %d",
 299                   ActiveProcessorCount);
 300     return ActiveProcessorCount;
 301   }
 302 
 303   int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 304   pid_t pid = getpid();
 305   psetid_t pset = PS_NONE;
 306   // Are we running in a processor set or is there any processor set around?
 307   if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
 308     uint_t pset_cpus;
 309     // Query the number of cpus available to us.
 310     if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
 311       assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
 312       _processors_online = pset_cpus;
 313       return pset_cpus;
 314     }
 315   }
 316   // Otherwise return number of online cpus
 317   return online_cpus;
 318 }
 319 
 320 static bool find_processors_in_pset(psetid_t        pset,
 321                                     processorid_t** id_array,
 322                                     uint_t*         id_length) {
 323   bool result = false;
 324   // Find the number of processors in the processor set.
 325   if (pset_info(pset, NULL, id_length, NULL) == 0) {
 326     // Make up an array to hold their ids.
 327     *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
 328     // Fill in the array with their processor ids.
 329     if (pset_info(pset, NULL, id_length, *id_array) == 0) {
 330       result = true;
 331     }
 332   }
 333   return result;
 334 }
 335 
 336 // Callers of find_processors_online() must tolerate imprecise results --
 337 // the system configuration can change asynchronously because of DR
 338 // or explicit psradm operations.
 339 //
 340 // We also need to take care that the loop (below) terminates as the
 341 // number of processors online can change between the _SC_NPROCESSORS_ONLN
 342 // request and the loop that builds the list of processor ids.   Unfortunately
 343 // there's no reliable way to determine the maximum valid processor id,
 344 // so we use a manifest constant, MAX_PROCESSOR_ID, instead.  See p_online
 345 // man pages, which claim the processor id set is "sparse, but
 346 // not too sparse".  MAX_PROCESSOR_ID is used to ensure that we eventually
 347 // exit the loop.
 348 //
 349 // In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
 350 // not available on S8.0.
 351 
 352 static bool find_processors_online(processorid_t** id_array,
 353                                    uint*           id_length) {
 354   const processorid_t MAX_PROCESSOR_ID = 100000;
 355   // Find the number of processors online.
 356   *id_length = sysconf(_SC_NPROCESSORS_ONLN);
 357   // Make up an array to hold their ids.
 358   *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
 359   // Processors need not be numbered consecutively.
 360   long found = 0;
 361   processorid_t next = 0;
 362   while (found < *id_length && next < MAX_PROCESSOR_ID) {
 363     processor_info_t info;
 364     if (processor_info(next, &info) == 0) {
 365       // NB, PI_NOINTR processors are effectively online ...
 366       if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
 367         (*id_array)[found] = next;
 368         found += 1;
 369       }
 370     }
 371     next += 1;
 372   }
 373   if (found < *id_length) {
 374     // The loop above didn't identify the expected number of processors.
 375     // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
 376     // and re-running the loop, above, but there's no guarantee of progress
 377     // if the system configuration is in flux.  Instead, we just return what
 378     // we've got.  Note that in the worst case find_processors_online() could
 379     // return an empty set.  (As a fall-back in the case of the empty set we
 380     // could just return the ID of the current processor).
 381     *id_length = found;
 382   }
 383 
 384   return true;
 385 }
 386 
 387 static bool assign_distribution(processorid_t* id_array,
 388                                 uint           id_length,
 389                                 uint*          distribution,
 390                                 uint           distribution_length) {
 391   // We assume we can assign processorid_t's to uint's.
 392   assert(sizeof(processorid_t) == sizeof(uint),
 393          "can't convert processorid_t to uint");
 394   // Quick check to see if we won't succeed.
 395   if (id_length < distribution_length) {
 396     return false;
 397   }
 398   // Assign processor ids to the distribution.
 399   // Try to shuffle processors to distribute work across boards,
 400   // assuming 4 processors per board.
 401   const uint processors_per_board = ProcessDistributionStride;
 402   // Find the maximum processor id.
 403   processorid_t max_id = 0;
 404   for (uint m = 0; m < id_length; m += 1) {
 405     max_id = MAX2(max_id, id_array[m]);
 406   }
 407   // The next id, to limit loops.
 408   const processorid_t limit_id = max_id + 1;
 409   // Make up markers for available processors.
 410   bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id, mtInternal);
 411   for (uint c = 0; c < limit_id; c += 1) {
 412     available_id[c] = false;
 413   }
 414   for (uint a = 0; a < id_length; a += 1) {
 415     available_id[id_array[a]] = true;
 416   }
 417   // Step by "boards", then by "slot", copying to "assigned".
 418   // NEEDS_CLEANUP: The assignment of processors should be stateful,
 419   //                remembering which processors have been assigned by
 420   //                previous calls, etc., so as to distribute several
 421   //                independent calls of this method.  What we'd like is
 422   //                It would be nice to have an API that let us ask
 423   //                how many processes are bound to a processor,
 424   //                but we don't have that, either.
 425   //                In the short term, "board" is static so that
 426   //                subsequent distributions don't all start at board 0.
 427   static uint board = 0;
 428   uint assigned = 0;
 429   // Until we've found enough processors ....
 430   while (assigned < distribution_length) {
 431     // ... find the next available processor in the board.
 432     for (uint slot = 0; slot < processors_per_board; slot += 1) {
 433       uint try_id = board * processors_per_board + slot;
 434       if ((try_id < limit_id) && (available_id[try_id] == true)) {
 435         distribution[assigned] = try_id;
 436         available_id[try_id] = false;
 437         assigned += 1;
 438         break;
 439       }
 440     }
 441     board += 1;
 442     if (board * processors_per_board + 0 >= limit_id) {
 443       board = 0;
 444     }
 445   }
 446   if (available_id != NULL) {
 447     FREE_C_HEAP_ARRAY(bool, available_id);
 448   }
 449   return true;
 450 }
 451 
 452 void os::set_native_thread_name(const char *name) {
 453   if (Solaris::_pthread_setname_np != NULL) {
 454     // Only the first 31 bytes of 'name' are processed by pthread_setname_np
 455     // but we explicitly copy into a size-limited buffer to avoid any
 456     // possible overflow.
 457     char buf[32];
 458     snprintf(buf, sizeof(buf), "%s", name);
 459     buf[sizeof(buf) - 1] = '\0';
 460     Solaris::_pthread_setname_np(pthread_self(), buf);
 461   }
 462 }
 463 
 464 bool os::distribute_processes(uint length, uint* distribution) {
 465   bool result = false;
 466   // Find the processor id's of all the available CPUs.
 467   processorid_t* id_array  = NULL;
 468   uint           id_length = 0;
 469   // There are some races between querying information and using it,
 470   // since processor sets can change dynamically.
 471   psetid_t pset = PS_NONE;
 472   // Are we running in a processor set?
 473   if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
 474     result = find_processors_in_pset(pset, &id_array, &id_length);
 475   } else {
 476     result = find_processors_online(&id_array, &id_length);
 477   }
 478   if (result == true) {
 479     if (id_length >= length) {
 480       result = assign_distribution(id_array, id_length, distribution, length);
 481     } else {
 482       result = false;
 483     }
 484   }
 485   if (id_array != NULL) {
 486     FREE_C_HEAP_ARRAY(processorid_t, id_array);
 487   }
 488   return result;
 489 }
 490 
 491 bool os::bind_to_processor(uint processor_id) {
 492   // We assume that a processorid_t can be stored in a uint.
 493   assert(sizeof(uint) == sizeof(processorid_t),
 494          "can't convert uint to processorid_t");
 495   int bind_result =
 496     processor_bind(P_LWPID,                       // bind LWP.
 497                    P_MYID,                        // bind current LWP.
 498                    (processorid_t) processor_id,  // id.
 499                    NULL);                         // don't return old binding.
 500   return (bind_result == 0);
 501 }
 502 
 503 // Return true if user is running as root.
 504 
 505 bool os::have_special_privileges() {
 506   static bool init = false;
 507   static bool privileges = false;
 508   if (!init) {
 509     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 510     init = true;
 511   }
 512   return privileges;
 513 }
 514 
 515 
 516 void os::init_system_properties_values() {
 517   // The next steps are taken in the product version:
 518   //
 519   // Obtain the JAVA_HOME value from the location of libjvm.so.
 520   // This library should be located at:
 521   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 522   //
 523   // If "/jre/lib/" appears at the right place in the path, then we
 524   // assume libjvm.so is installed in a JDK and we use this path.
 525   //
 526   // Otherwise exit with message: "Could not create the Java virtual machine."
 527   //
 528   // The following extra steps are taken in the debugging version:
 529   //
 530   // If "/jre/lib/" does NOT appear at the right place in the path
 531   // instead of exit check for $JAVA_HOME environment variable.
 532   //
 533   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 534   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 535   // it looks like libjvm.so is installed there
 536   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 537   //
 538   // Otherwise exit.
 539   //
 540   // Important note: if the location of libjvm.so changes this
 541   // code needs to be changed accordingly.
 542 
 543 // Base path of extensions installed on the system.
 544 #define SYS_EXT_DIR     "/usr/jdk/packages"
 545 #define EXTENSIONS_DIR  "/lib/ext"
 546 
 547   // Buffer that fits several sprintfs.
 548   // Note that the space for the colon and the trailing null are provided
 549   // by the nulls included by the sizeof operator.
 550   const size_t bufsize =
 551     MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
 552          sizeof(SYS_EXT_DIR) + sizeof("/lib/"), // invariant ld_library_path
 553          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 554   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 555 
 556   // sysclasspath, java_home, dll_dir
 557   {
 558     char *pslash;
 559     os::jvm_path(buf, bufsize);
 560 
 561     // Found the full path to libjvm.so.
 562     // Now cut the path to <java_home>/jre if we can.
 563     *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
 564     pslash = strrchr(buf, '/');
 565     if (pslash != NULL) {
 566       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 567     }
 568     Arguments::set_dll_dir(buf);
 569 
 570     if (pslash != NULL) {
 571       pslash = strrchr(buf, '/');
 572       if (pslash != NULL) {
 573         *pslash = '\0';        // Get rid of /lib.
 574       }
 575     }
 576     Arguments::set_java_home(buf);
 577     set_boot_path('/', ':');
 578   }
 579 
 580   // Where to look for native libraries.
 581   {
 582     // Use dlinfo() to determine the correct java.library.path.
 583     //
 584     // If we're launched by the Java launcher, and the user
 585     // does not set java.library.path explicitly on the commandline,
 586     // the Java launcher sets LD_LIBRARY_PATH for us and unsets
 587     // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64.  In this case
 588     // dlinfo returns LD_LIBRARY_PATH + crle settings (including
 589     // /usr/lib), which is exactly what we want.
 590     //
 591     // If the user does set java.library.path, it completely
 592     // overwrites this setting, and always has.
 593     //
 594     // If we're not launched by the Java launcher, we may
 595     // get here with any/all of the LD_LIBRARY_PATH[_32|64]
 596     // settings.  Again, dlinfo does exactly what we want.
 597 
 598     Dl_serinfo     info_sz, *info = &info_sz;
 599     Dl_serpath     *path;
 600     char           *library_path;
 601     char           *common_path = buf;
 602 
 603     // Determine search path count and required buffer size.
 604     if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
 605       FREE_C_HEAP_ARRAY(char, buf);
 606       vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
 607     }
 608 
 609     // Allocate new buffer and initialize.
 610     info = (Dl_serinfo*)NEW_C_HEAP_ARRAY(char, info_sz.dls_size, mtInternal);
 611     info->dls_size = info_sz.dls_size;
 612     info->dls_cnt = info_sz.dls_cnt;
 613 
 614     // Obtain search path information.
 615     if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
 616       FREE_C_HEAP_ARRAY(char, buf);
 617       FREE_C_HEAP_ARRAY(char, info);
 618       vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
 619     }
 620 
 621     path = &info->dls_serpath[0];
 622 
 623     // Note: Due to a legacy implementation, most of the library path
 624     // is set in the launcher. This was to accomodate linking restrictions
 625     // on legacy Solaris implementations (which are no longer supported).
 626     // Eventually, all the library path setting will be done here.
 627     //
 628     // However, to prevent the proliferation of improperly built native
 629     // libraries, the new path component /usr/jdk/packages is added here.
 630 
 631     // Construct the invariant part of ld_library_path.
 632     sprintf(common_path, SYS_EXT_DIR "/lib");
 633 
 634     // Struct size is more than sufficient for the path components obtained
 635     // through the dlinfo() call, so only add additional space for the path
 636     // components explicitly added here.
 637     size_t library_path_size = info->dls_size + strlen(common_path);
 638     library_path = (char *)NEW_C_HEAP_ARRAY(char, library_path_size, mtInternal);
 639     library_path[0] = '\0';
 640 
 641     // Construct the desired Java library path from the linker's library
 642     // search path.
 643     //
 644     // For compatibility, it is optimal that we insert the additional path
 645     // components specific to the Java VM after those components specified
 646     // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
 647     // infrastructure.
 648     if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it.
 649       strcpy(library_path, common_path);
 650     } else {
 651       int inserted = 0;
 652       int i;
 653       for (i = 0; i < info->dls_cnt; i++, path++) {
 654         uint_t flags = path->dls_flags & LA_SER_MASK;
 655         if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
 656           strcat(library_path, common_path);
 657           strcat(library_path, os::path_separator());
 658           inserted = 1;
 659         }
 660         strcat(library_path, path->dls_name);
 661         strcat(library_path, os::path_separator());
 662       }
 663       // Eliminate trailing path separator.
 664       library_path[strlen(library_path)-1] = '\0';
 665     }
 666 
 667     // happens before argument parsing - can't use a trace flag
 668     // tty->print_raw("init_system_properties_values: native lib path: ");
 669     // tty->print_raw_cr(library_path);
 670 
 671     // Callee copies into its own buffer.
 672     Arguments::set_library_path(library_path);
 673 
 674     FREE_C_HEAP_ARRAY(char, library_path);
 675     FREE_C_HEAP_ARRAY(char, info);
 676   }
 677 
 678   // Extensions directories.
 679   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 680   Arguments::set_ext_dirs(buf);
 681 
 682   FREE_C_HEAP_ARRAY(char, buf);
 683 
 684 #undef SYS_EXT_DIR
 685 #undef EXTENSIONS_DIR
 686 }
 687 
 688 void os::breakpoint() {
 689   BREAKPOINT;
 690 }
 691 
 692 bool os::obsolete_option(const JavaVMOption *option) {
 693   if (!strncmp(option->optionString, "-Xt", 3)) {
 694     return true;
 695   } else if (!strncmp(option->optionString, "-Xtm", 4)) {
 696     return true;
 697   } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
 698     return true;
 699   } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
 700     return true;
 701   }
 702   return false;
 703 }
 704 
 705 bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
 706   address  stackStart  = (address)thread->stack_base();
 707   address  stackEnd    = (address)(stackStart - (address)thread->stack_size());
 708   if (sp < stackStart && sp >= stackEnd) return true;
 709   return false;
 710 }
 711 
 712 extern "C" void breakpoint() {
 713   // use debugger to set breakpoint here
 714 }
 715 
 716 static thread_t main_thread;
 717 
 718 // Thread start routine for all newly created threads
 719 extern "C" void* thread_native_entry(void* thread_addr) {
 720   // Try to randomize the cache line index of hot stack frames.
 721   // This helps when threads of the same stack traces evict each other's
 722   // cache lines. The threads can be either from the same JVM instance, or
 723   // from different JVM instances. The benefit is especially true for
 724   // processors with hyperthreading technology.
 725   static int counter = 0;
 726   int pid = os::current_process_id();
 727   alloca(((pid ^ counter++) & 7) * 128);
 728 
 729   int prio;
 730   Thread* thread = (Thread*)thread_addr;
 731 
 732   thread->initialize_thread_current();
 733 
 734   OSThread* osthr = thread->osthread();
 735 
 736   osthr->set_lwp_id(_lwp_self());  // Store lwp in case we are bound
 737   thread->_schedctl = (void *) schedctl_init();
 738 
 739   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ").",
 740     os::current_thread_id());
 741 
 742   if (UseNUMA) {
 743     int lgrp_id = os::numa_get_group_id();
 744     if (lgrp_id != -1) {
 745       thread->set_lgrp_id(lgrp_id);
 746     }
 747   }
 748 
 749   // Our priority was set when we were created, and stored in the
 750   // osthread, but couldn't be passed through to our LWP until now.
 751   // So read back the priority and set it again.
 752 
 753   if (osthr->thread_id() != -1) {
 754     if (UseThreadPriorities) {
 755       int prio = osthr->native_priority();
 756       if (ThreadPriorityVerbose) {
 757         tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is "
 758                       INTPTR_FORMAT ", setting priority: %d\n",
 759                       osthr->thread_id(), osthr->lwp_id(), prio);
 760       }
 761       os::set_native_priority(thread, prio);
 762     }
 763   } else if (ThreadPriorityVerbose) {
 764     warning("Can't set priority in _start routine, thread id hasn't been set\n");
 765   }
 766 
 767   assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
 768 
 769   // initialize signal mask for this thread
 770   os::Solaris::hotspot_sigmask(thread);
 771 
 772   thread->run();
 773 
 774   // One less thread is executing
 775   // When the VMThread gets here, the main thread may have already exited
 776   // which frees the CodeHeap containing the Atomic::dec code
 777   if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
 778     Atomic::dec(&os::Solaris::_os_thread_count);
 779   }
 780 
 781   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ").", os::current_thread_id());
 782 
 783   // If a thread has not deleted itself ("delete this") as part of its
 784   // termination sequence, we have to ensure thread-local-storage is
 785   // cleared before we actually terminate. No threads should ever be
 786   // deleted asynchronously with respect to their termination.
 787   if (Thread::current_or_null_safe() != NULL) {
 788     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 789     thread->clear_thread_current();
 790   }
 791 
 792   if (UseDetachedThreads) {
 793     thr_exit(NULL);
 794     ShouldNotReachHere();
 795   }
 796   return NULL;
 797 }
 798 
 799 static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
 800   // Allocate the OSThread object
 801   OSThread* osthread = new OSThread(NULL, NULL);
 802   if (osthread == NULL) return NULL;
 803 
 804   // Store info on the Solaris thread into the OSThread
 805   osthread->set_thread_id(thread_id);
 806   osthread->set_lwp_id(_lwp_self());
 807   thread->_schedctl = (void *) schedctl_init();
 808 
 809   if (UseNUMA) {
 810     int lgrp_id = os::numa_get_group_id();
 811     if (lgrp_id != -1) {
 812       thread->set_lgrp_id(lgrp_id);
 813     }
 814   }
 815 
 816   if (ThreadPriorityVerbose) {
 817     tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
 818                   osthread->thread_id(), osthread->lwp_id());
 819   }
 820 
 821   // Initial thread state is INITIALIZED, not SUSPENDED
 822   osthread->set_state(INITIALIZED);
 823 
 824   return osthread;
 825 }
 826 
 827 void os::Solaris::hotspot_sigmask(Thread* thread) {
 828   //Save caller's signal mask
 829   sigset_t sigmask;
 830   pthread_sigmask(SIG_SETMASK, NULL, &sigmask);
 831   OSThread *osthread = thread->osthread();
 832   osthread->set_caller_sigmask(sigmask);
 833 
 834   pthread_sigmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
 835   if (!ReduceSignalUsage) {
 836     if (thread->is_VM_thread()) {
 837       // Only the VM thread handles BREAK_SIGNAL ...
 838       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 839     } else {
 840       // ... all other threads block BREAK_SIGNAL
 841       assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
 842       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 843     }
 844   }
 845 }
 846 
 847 bool os::create_attached_thread(JavaThread* thread) {
 848 #ifdef ASSERT
 849   thread->verify_not_published();
 850 #endif
 851   OSThread* osthread = create_os_thread(thread, thr_self());
 852   if (osthread == NULL) {
 853     return false;
 854   }
 855 
 856   // Initial thread state is RUNNABLE
 857   osthread->set_state(RUNNABLE);
 858   thread->set_osthread(osthread);
 859 
 860   // initialize signal mask for this thread
 861   // and save the caller's signal mask
 862   os::Solaris::hotspot_sigmask(thread);
 863 
 864   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ").",
 865     os::current_thread_id());
 866 
 867   return true;
 868 }
 869 
 870 bool os::create_main_thread(JavaThread* thread) {
 871 #ifdef ASSERT
 872   thread->verify_not_published();
 873 #endif
 874   if (_starting_thread == NULL) {
 875     _starting_thread = create_os_thread(thread, main_thread);
 876     if (_starting_thread == NULL) {
 877       return false;
 878     }
 879   }
 880 
 881   // The primodial thread is runnable from the start
 882   _starting_thread->set_state(RUNNABLE);
 883 
 884   thread->set_osthread(_starting_thread);
 885 
 886   // initialize signal mask for this thread
 887   // and save the caller's signal mask
 888   os::Solaris::hotspot_sigmask(thread);
 889 
 890   return true;
 891 }
 892 
 893 // Helper function to trace thread attributes, similar to os::Posix::describe_pthread_attr()
 894 static char* describe_thr_create_attributes(char* buf, size_t buflen,
 895                                             size_t stacksize, long flags) {
 896   stringStream ss(buf, buflen);
 897   ss.print("stacksize: " SIZE_FORMAT "k, ", stacksize / 1024);
 898   ss.print("flags: ");
 899   #define PRINT_FLAG(f) if (flags & f) ss.print( #f " ");
 900   #define ALL(X) \
 901     X(THR_SUSPENDED) \
 902     X(THR_DETACHED) \
 903     X(THR_BOUND) \
 904     X(THR_NEW_LWP) \
 905     X(THR_DAEMON)
 906   ALL(PRINT_FLAG)
 907   #undef ALL
 908   #undef PRINT_FLAG
 909   return buf;
 910 }
 911 
 912 // return default stack size for thr_type
 913 size_t os::Posix::default_stack_size(os::ThreadType thr_type) {
 914   // default stack size when not specified by caller is 1M (2M for LP64)
 915   size_t s = (BytesPerWord >> 2) * K * K;
 916   return s;
 917 }
 918 
 919 bool os::create_thread(Thread* thread, ThreadType thr_type,
 920                        size_t req_stack_size) {
 921   // Allocate the OSThread object
 922   OSThread* osthread = new OSThread(NULL, NULL);
 923   if (osthread == NULL) {
 924     return false;
 925   }
 926 
 927   if (ThreadPriorityVerbose) {
 928     char *thrtyp;
 929     switch (thr_type) {
 930     case vm_thread:
 931       thrtyp = (char *)"vm";
 932       break;
 933     case cgc_thread:
 934       thrtyp = (char *)"cgc";
 935       break;
 936     case pgc_thread:
 937       thrtyp = (char *)"pgc";
 938       break;
 939     case java_thread:
 940       thrtyp = (char *)"java";
 941       break;
 942     case compiler_thread:
 943       thrtyp = (char *)"compiler";
 944       break;
 945     case watcher_thread:
 946       thrtyp = (char *)"watcher";
 947       break;
 948     default:
 949       thrtyp = (char *)"unknown";
 950       break;
 951     }
 952     tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
 953   }
 954 
 955   // calculate stack size if it's not specified by caller
 956   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 957 
 958   // Initial state is ALLOCATED but not INITIALIZED
 959   osthread->set_state(ALLOCATED);
 960 
 961   if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
 962     // We got lots of threads. Check if we still have some address space left.
 963     // Need to be at least 5Mb of unreserved address space. We do check by
 964     // trying to reserve some.
 965     const size_t VirtualMemoryBangSize = 20*K*K;
 966     char* mem = os::reserve_memory(VirtualMemoryBangSize);
 967     if (mem == NULL) {
 968       delete osthread;
 969       return false;
 970     } else {
 971       // Release the memory again
 972       os::release_memory(mem, VirtualMemoryBangSize);
 973     }
 974   }
 975 
 976   // Setup osthread because the child thread may need it.
 977   thread->set_osthread(osthread);
 978 
 979   // Create the Solaris thread
 980   thread_t tid = 0;
 981   long     flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED;
 982   int      status;
 983 
 984   // Mark that we don't have an lwp or thread id yet.
 985   // In case we attempt to set the priority before the thread starts.
 986   osthread->set_lwp_id(-1);
 987   osthread->set_thread_id(-1);
 988 
 989   status = thr_create(NULL, stack_size, thread_native_entry, thread, flags, &tid);
 990 
 991   char buf[64];
 992   if (status == 0) {
 993     log_info(os, thread)("Thread started (tid: " UINTX_FORMAT ", attributes: %s). ",
 994       (uintx) tid, describe_thr_create_attributes(buf, sizeof(buf), stack_size, flags));
 995   } else {
 996     log_warning(os, thread)("Failed to start thread - thr_create failed (%s) for attributes: %s.",
 997       os::errno_name(status), describe_thr_create_attributes(buf, sizeof(buf), stack_size, flags));
 998   }
 999 
1000   if (status != 0) {
1001     thread->set_osthread(NULL);
1002     // Need to clean up stuff we've allocated so far
1003     delete osthread;
1004     return false;
1005   }
1006 
1007   Atomic::inc(&os::Solaris::_os_thread_count);
1008 
1009   // Store info on the Solaris thread into the OSThread
1010   osthread->set_thread_id(tid);
1011 
1012   // Remember that we created this thread so we can set priority on it
1013   osthread->set_vm_created();
1014 
1015   // Most thread types will set an explicit priority before starting the thread,
1016   // but for those that don't we need a valid value to read back in thread_native_entry.
1017   osthread->set_native_priority(NormPriority);
1018 
1019   // Initial thread state is INITIALIZED, not SUSPENDED
1020   osthread->set_state(INITIALIZED);
1021 
1022   // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
1023   return true;
1024 }
1025 
1026 debug_only(static bool signal_sets_initialized = false);
1027 static sigset_t unblocked_sigs, vm_sigs;
1028 
1029 bool os::Solaris::is_sig_ignored(int sig) {
1030   struct sigaction oact;
1031   sigaction(sig, (struct sigaction*)NULL, &oact);
1032   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
1033                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
1034   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
1035     return true;
1036   } else {
1037     return false;
1038   }
1039 }
1040 
1041 void os::Solaris::signal_sets_init() {
1042   // Should also have an assertion stating we are still single-threaded.
1043   assert(!signal_sets_initialized, "Already initialized");
1044   // Fill in signals that are necessarily unblocked for all threads in
1045   // the VM. Currently, we unblock the following signals:
1046   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1047   //                         by -Xrs (=ReduceSignalUsage));
1048   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1049   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1050   // the dispositions or masks wrt these signals.
1051   // Programs embedding the VM that want to use the above signals for their
1052   // own purposes must, at this time, use the "-Xrs" option to prevent
1053   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1054   // (See bug 4345157, and other related bugs).
1055   // In reality, though, unblocking these signals is really a nop, since
1056   // these signals are not blocked by default.
1057   sigemptyset(&unblocked_sigs);
1058   sigaddset(&unblocked_sigs, SIGILL);
1059   sigaddset(&unblocked_sigs, SIGSEGV);
1060   sigaddset(&unblocked_sigs, SIGBUS);
1061   sigaddset(&unblocked_sigs, SIGFPE);
1062   sigaddset(&unblocked_sigs, ASYNC_SIGNAL);
1063 
1064   if (!ReduceSignalUsage) {
1065     if (!os::Solaris::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1066       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1067     }
1068     if (!os::Solaris::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1069       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1070     }
1071     if (!os::Solaris::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1072       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1073     }
1074   }
1075   // Fill in signals that are blocked by all but the VM thread.
1076   sigemptyset(&vm_sigs);
1077   if (!ReduceSignalUsage) {
1078     sigaddset(&vm_sigs, BREAK_SIGNAL);
1079   }
1080   debug_only(signal_sets_initialized = true);
1081 
1082   // For diagnostics only used in run_periodic_checks
1083   sigemptyset(&check_signal_done);
1084 }
1085 
1086 // These are signals that are unblocked while a thread is running Java.
1087 // (For some reason, they get blocked by default.)
1088 sigset_t* os::Solaris::unblocked_signals() {
1089   assert(signal_sets_initialized, "Not initialized");
1090   return &unblocked_sigs;
1091 }
1092 
1093 // These are the signals that are blocked while a (non-VM) thread is
1094 // running Java. Only the VM thread handles these signals.
1095 sigset_t* os::Solaris::vm_signals() {
1096   assert(signal_sets_initialized, "Not initialized");
1097   return &vm_sigs;
1098 }
1099 
1100 void _handle_uncaught_cxx_exception() {
1101   VMError::report_and_die("An uncaught C++ exception");
1102 }
1103 
1104 
1105 // First crack at OS-specific initialization, from inside the new thread.
1106 void os::initialize_thread(Thread* thr) {
1107   if (is_primordial_thread()) {
1108     JavaThread* jt = (JavaThread *)thr;
1109     assert(jt != NULL, "Sanity check");
1110     size_t stack_size;
1111     address base = jt->stack_base();
1112     if (Arguments::created_by_java_launcher()) {
1113       // Use 2MB to allow for Solaris 7 64 bit mode.
1114       stack_size = JavaThread::stack_size_at_create() == 0
1115         ? 2048*K : JavaThread::stack_size_at_create();
1116 
1117       // There are rare cases when we may have already used more than
1118       // the basic stack size allotment before this method is invoked.
1119       // Attempt to allow for a normally sized java_stack.
1120       size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1121       stack_size += ReservedSpace::page_align_size_down(current_stack_offset);
1122     } else {
1123       // 6269555: If we were not created by a Java launcher, i.e. if we are
1124       // running embedded in a native application, treat the primordial thread
1125       // as much like a native attached thread as possible.  This means using
1126       // the current stack size from thr_stksegment(), unless it is too large
1127       // to reliably setup guard pages.  A reasonable max size is 8MB.
1128       size_t current_size = current_stack_size();
1129       // This should never happen, but just in case....
1130       if (current_size == 0) current_size = 2 * K * K;
1131       stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1132     }
1133     address bottom = align_up(base - stack_size, os::vm_page_size());;
1134     stack_size = (size_t)(base - bottom);
1135 
1136     assert(stack_size > 0, "Stack size calculation problem");
1137 
1138     if (stack_size > jt->stack_size()) {
1139 #ifndef PRODUCT
1140       struct rlimit limits;
1141       getrlimit(RLIMIT_STACK, &limits);
1142       size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1143       assert(size >= jt->stack_size(), "Stack size problem in main thread");
1144 #endif
1145       tty->print_cr("Stack size of %d Kb exceeds current limit of %d Kb.\n"
1146                     "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1147                     "See limit(1) to increase the stack size limit.",
1148                     stack_size / K, jt->stack_size() / K);
1149       vm_exit(1);
1150     }
1151     assert(jt->stack_size() >= stack_size,
1152            "Attempt to map more stack than was allocated");
1153     jt->set_stack_size(stack_size);
1154   }
1155 
1156   // With the T2 libthread (T1 is no longer supported) threads are always bound
1157   // and we use stackbanging in all cases.
1158 
1159   os::Solaris::init_thread_fpu_state();
1160   std::set_terminate(_handle_uncaught_cxx_exception);
1161 }
1162 
1163 
1164 
1165 // Free Solaris resources related to the OSThread
1166 void os::free_thread(OSThread* osthread) {
1167   assert(osthread != NULL, "os::free_thread but osthread not set");
1168 
1169   // We are told to free resources of the argument thread,
1170   // but we can only really operate on the current thread.
1171   assert(Thread::current()->osthread() == osthread,
1172          "os::free_thread but not current thread");
1173 
1174   // Restore caller's signal mask
1175   sigset_t sigmask = osthread->caller_sigmask();
1176   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1177 
1178   delete osthread;
1179 }
1180 
1181 void os::pd_start_thread(Thread* thread) {
1182   int status = thr_continue(thread->osthread()->thread_id());
1183   assert_status(status == 0, status, "thr_continue failed");
1184 }
1185 
1186 
1187 intx os::current_thread_id() {
1188   return (intx)thr_self();
1189 }
1190 
1191 static pid_t _initial_pid = 0;
1192 
1193 int os::current_process_id() {
1194   return (int)(_initial_pid ? _initial_pid : getpid());
1195 }
1196 
1197 // gethrtime() should be monotonic according to the documentation,
1198 // but some virtualized platforms are known to break this guarantee.
1199 // getTimeNanos() must be guaranteed not to move backwards, so we
1200 // are forced to add a check here.
1201 inline hrtime_t getTimeNanos() {
1202   const hrtime_t now = gethrtime();
1203   const hrtime_t prev = max_hrtime;
1204   if (now <= prev) {
1205     return prev;   // same or retrograde time;
1206   }
1207   const hrtime_t obsv = Atomic::cmpxchg(now, &max_hrtime, prev);
1208   assert(obsv >= prev, "invariant");   // Monotonicity
1209   // If the CAS succeeded then we're done and return "now".
1210   // If the CAS failed and the observed value "obsv" is >= now then
1211   // we should return "obsv".  If the CAS failed and now > obsv > prv then
1212   // some other thread raced this thread and installed a new value, in which case
1213   // we could either (a) retry the entire operation, (b) retry trying to install now
1214   // or (c) just return obsv.  We use (c).   No loop is required although in some cases
1215   // we might discard a higher "now" value in deference to a slightly lower but freshly
1216   // installed obsv value.   That's entirely benign -- it admits no new orderings compared
1217   // to (a) or (b) -- and greatly reduces coherence traffic.
1218   // We might also condition (c) on the magnitude of the delta between obsv and now.
1219   // Avoiding excessive CAS operations to hot RW locations is critical.
1220   // See https://blogs.oracle.com/dave/entry/cas_and_cache_trivia_invalidate
1221   return (prev == obsv) ? now : obsv;
1222 }
1223 
1224 // Time since start-up in seconds to a fine granularity.
1225 // Used by VMSelfDestructTimer and the MemProfiler.
1226 double os::elapsedTime() {
1227   return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1228 }
1229 
1230 jlong os::elapsed_counter() {
1231   return (jlong)(getTimeNanos() - first_hrtime);
1232 }
1233 
1234 jlong os::elapsed_frequency() {
1235   return hrtime_hz;
1236 }
1237 
1238 // Return the real, user, and system times in seconds from an
1239 // arbitrary fixed point in the past.
1240 bool os::getTimesSecs(double* process_real_time,
1241                       double* process_user_time,
1242                       double* process_system_time) {
1243   struct tms ticks;
1244   clock_t real_ticks = times(&ticks);
1245 
1246   if (real_ticks == (clock_t) (-1)) {
1247     return false;
1248   } else {
1249     double ticks_per_second = (double) clock_tics_per_sec;
1250     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1251     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1252     // For consistency return the real time from getTimeNanos()
1253     // converted to seconds.
1254     *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1255 
1256     return true;
1257   }
1258 }
1259 
1260 bool os::supports_vtime() { return true; }
1261 bool os::enable_vtime() { return false; }
1262 bool os::vtime_enabled() { return false; }
1263 
1264 double os::elapsedVTime() {
1265   return (double)gethrvtime() / (double)hrtime_hz;
1266 }
1267 
1268 // Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1269 jlong os::javaTimeMillis() {
1270   timeval t;
1271   if (gettimeofday(&t, NULL) == -1) {
1272     fatal("os::javaTimeMillis: gettimeofday (%s)", os::strerror(errno));
1273   }
1274   return jlong(t.tv_sec) * 1000  +  jlong(t.tv_usec) / 1000;
1275 }
1276 
1277 // Must return seconds+nanos since Jan 1 1970. This must use the same
1278 // time source as javaTimeMillis and can't use get_nsec_fromepoch as
1279 // we need better than 1ms accuracy
1280 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1281   timeval t;
1282   if (gettimeofday(&t, NULL) == -1) {
1283     fatal("os::javaTimeSystemUTC: gettimeofday (%s)", os::strerror(errno));
1284   }
1285   seconds = jlong(t.tv_sec);
1286   nanos = jlong(t.tv_usec) * 1000;
1287 }
1288 
1289 
1290 jlong os::javaTimeNanos() {
1291   return (jlong)getTimeNanos();
1292 }
1293 
1294 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1295   info_ptr->max_value = ALL_64_BITS;      // gethrtime() uses all 64 bits
1296   info_ptr->may_skip_backward = false;    // not subject to resetting or drifting
1297   info_ptr->may_skip_forward = false;     // not subject to resetting or drifting
1298   info_ptr->kind = JVMTI_TIMER_ELAPSED;   // elapsed not CPU time
1299 }
1300 
1301 char * os::local_time_string(char *buf, size_t buflen) {
1302   struct tm t;
1303   time_t long_time;
1304   time(&long_time);
1305   localtime_r(&long_time, &t);
1306   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1307                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1308                t.tm_hour, t.tm_min, t.tm_sec);
1309   return buf;
1310 }
1311 
1312 // Note: os::shutdown() might be called very early during initialization, or
1313 // called from signal handler. Before adding something to os::shutdown(), make
1314 // sure it is async-safe and can handle partially initialized VM.
1315 void os::shutdown() {
1316 
1317   // allow PerfMemory to attempt cleanup of any persistent resources
1318   perfMemory_exit();
1319 
1320   // needs to remove object in file system
1321   AttachListener::abort();
1322 
1323   // flush buffered output, finish log files
1324   ostream_abort();
1325 
1326   // Check for abort hook
1327   abort_hook_t abort_hook = Arguments::abort_hook();
1328   if (abort_hook != NULL) {
1329     abort_hook();
1330   }
1331 }
1332 
1333 // Note: os::abort() might be called very early during initialization, or
1334 // called from signal handler. Before adding something to os::abort(), make
1335 // sure it is async-safe and can handle partially initialized VM.
1336 void os::abort(bool dump_core, void* siginfo, const void* context) {
1337   os::shutdown();
1338   if (dump_core) {
1339 #ifndef PRODUCT
1340     fdStream out(defaultStream::output_fd());
1341     out.print_raw("Current thread is ");
1342     char buf[16];
1343     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1344     out.print_raw_cr(buf);
1345     out.print_raw_cr("Dumping core ...");
1346 #endif
1347     ::abort(); // dump core (for debugging)
1348   }
1349 
1350   ::exit(1);
1351 }
1352 
1353 // Die immediately, no exit hook, no abort hook, no cleanup.
1354 void os::die() {
1355   ::abort(); // dump core (for debugging)
1356 }
1357 
1358 // DLL functions
1359 
1360 const char* os::dll_file_extension() { return ".so"; }
1361 
1362 // This must be hard coded because it's the system's temporary
1363 // directory not the java application's temp directory, ala java.io.tmpdir.
1364 const char* os::get_temp_directory() { return "/tmp"; }
1365 
1366 // check if addr is inside libjvm.so
1367 bool os::address_is_in_vm(address addr) {
1368   static address libjvm_base_addr;
1369   Dl_info dlinfo;
1370 
1371   if (libjvm_base_addr == NULL) {
1372     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1373       libjvm_base_addr = (address)dlinfo.dli_fbase;
1374     }
1375     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1376   }
1377 
1378   if (dladdr((void *)addr, &dlinfo) != 0) {
1379     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1380   }
1381 
1382   return false;
1383 }
1384 
1385 typedef int (*dladdr1_func_type)(void *, Dl_info *, void **, int);
1386 static dladdr1_func_type dladdr1_func = NULL;
1387 
1388 bool os::dll_address_to_function_name(address addr, char *buf,
1389                                       int buflen, int * offset,
1390                                       bool demangle) {
1391   // buf is not optional, but offset is optional
1392   assert(buf != NULL, "sanity check");
1393 
1394   Dl_info dlinfo;
1395 
1396   // dladdr1_func was initialized in os::init()
1397   if (dladdr1_func != NULL) {
1398     // yes, we have dladdr1
1399 
1400     // Support for dladdr1 is checked at runtime; it may be
1401     // available even if the vm is built on a machine that does
1402     // not have dladdr1 support.  Make sure there is a value for
1403     // RTLD_DL_SYMENT.
1404 #ifndef RTLD_DL_SYMENT
1405   #define RTLD_DL_SYMENT 1
1406 #endif
1407 #ifdef _LP64
1408     Elf64_Sym * info;
1409 #else
1410     Elf32_Sym * info;
1411 #endif
1412     if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1413                      RTLD_DL_SYMENT) != 0) {
1414       // see if we have a matching symbol that covers our address
1415       if (dlinfo.dli_saddr != NULL &&
1416           (char *)dlinfo.dli_saddr + info->st_size > (char *)addr) {
1417         if (dlinfo.dli_sname != NULL) {
1418           if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1419             jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1420           }
1421           if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1422           return true;
1423         }
1424       }
1425       // no matching symbol so try for just file info
1426       if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1427         if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1428                             buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1429           return true;
1430         }
1431       }
1432     }
1433     buf[0] = '\0';
1434     if (offset != NULL) *offset  = -1;
1435     return false;
1436   }
1437 
1438   // no, only dladdr is available
1439   if (dladdr((void *)addr, &dlinfo) != 0) {
1440     // see if we have a matching symbol
1441     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1442       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1443         jio_snprintf(buf, buflen, dlinfo.dli_sname);
1444       }
1445       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1446       return true;
1447     }
1448     // no matching symbol so try for just file info
1449     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1450       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1451                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1452         return true;
1453       }
1454     }
1455   }
1456   buf[0] = '\0';
1457   if (offset != NULL) *offset  = -1;
1458   return false;
1459 }
1460 
1461 bool os::dll_address_to_library_name(address addr, char* buf,
1462                                      int buflen, int* offset) {
1463   // buf is not optional, but offset is optional
1464   assert(buf != NULL, "sanity check");
1465 
1466   Dl_info dlinfo;
1467 
1468   if (dladdr((void*)addr, &dlinfo) != 0) {
1469     if (dlinfo.dli_fname != NULL) {
1470       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1471     }
1472     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1473       *offset = addr - (address)dlinfo.dli_fbase;
1474     }
1475     return true;
1476   }
1477 
1478   buf[0] = '\0';
1479   if (offset) *offset = -1;
1480   return false;
1481 }
1482 
1483 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1484   Dl_info dli;
1485   // Sanity check?
1486   if (dladdr(CAST_FROM_FN_PTR(void *, os::get_loaded_modules_info), &dli) == 0 ||
1487       dli.dli_fname == NULL) {
1488     return 1;
1489   }
1490 
1491   void * handle = dlopen(dli.dli_fname, RTLD_LAZY);
1492   if (handle == NULL) {
1493     return 1;
1494   }
1495 
1496   Link_map *map;
1497   dlinfo(handle, RTLD_DI_LINKMAP, &map);
1498   if (map == NULL) {
1499     dlclose(handle);
1500     return 1;
1501   }
1502 
1503   while (map->l_prev != NULL) {
1504     map = map->l_prev;
1505   }
1506 
1507   while (map != NULL) {
1508     // Iterate through all map entries and call callback with fields of interest
1509     if(callback(map->l_name, (address)map->l_addr, (address)0, param)) {
1510       dlclose(handle);
1511       return 1;
1512     }
1513     map = map->l_next;
1514   }
1515 
1516   dlclose(handle);
1517   return 0;
1518 }
1519 
1520 int _print_dll_info_cb(const char * name, address base_address, address top_address, void * param) {
1521   outputStream * out = (outputStream *) param;
1522   out->print_cr(PTR_FORMAT " \t%s", base_address, name);
1523   return 0;
1524 }
1525 
1526 void os::print_dll_info(outputStream * st) {
1527   st->print_cr("Dynamic libraries:"); st->flush();
1528   if (get_loaded_modules_info(_print_dll_info_cb, (void *)st)) {
1529     st->print_cr("Error: Cannot print dynamic libraries.");
1530   }
1531 }
1532 
1533 // Loads .dll/.so and
1534 // in case of error it checks if .dll/.so was built for the
1535 // same architecture as Hotspot is running on
1536 
1537 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1538   void * result= ::dlopen(filename, RTLD_LAZY);
1539   if (result != NULL) {
1540     // Successful loading
1541     return result;
1542   }
1543 
1544   Elf32_Ehdr elf_head;
1545 
1546   // Read system error message into ebuf
1547   // It may or may not be overwritten below
1548   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1549   ebuf[ebuflen-1]='\0';
1550   int diag_msg_max_length=ebuflen-strlen(ebuf);
1551   char* diag_msg_buf=ebuf+strlen(ebuf);
1552 
1553   if (diag_msg_max_length==0) {
1554     // No more space in ebuf for additional diagnostics message
1555     return NULL;
1556   }
1557 
1558 
1559   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1560 
1561   if (file_descriptor < 0) {
1562     // Can't open library, report dlerror() message
1563     return NULL;
1564   }
1565 
1566   bool failed_to_read_elf_head=
1567     (sizeof(elf_head)!=
1568      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1569 
1570   ::close(file_descriptor);
1571   if (failed_to_read_elf_head) {
1572     // file i/o error - report dlerror() msg
1573     return NULL;
1574   }
1575 
1576   typedef struct {
1577     Elf32_Half  code;         // Actual value as defined in elf.h
1578     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1579     char        elf_class;    // 32 or 64 bit
1580     char        endianess;    // MSB or LSB
1581     char*       name;         // String representation
1582   } arch_t;
1583 
1584   static const arch_t arch_array[]={
1585     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1586     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1587     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1588     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1589     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1590     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1591     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1592     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1593     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1594     {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM 32"}
1595   };
1596 
1597 #if  (defined IA32)
1598   static  Elf32_Half running_arch_code=EM_386;
1599 #elif   (defined AMD64)
1600   static  Elf32_Half running_arch_code=EM_X86_64;
1601 #elif  (defined IA64)
1602   static  Elf32_Half running_arch_code=EM_IA_64;
1603 #elif  (defined __sparc) && (defined _LP64)
1604   static  Elf32_Half running_arch_code=EM_SPARCV9;
1605 #elif  (defined __sparc) && (!defined _LP64)
1606   static  Elf32_Half running_arch_code=EM_SPARC;
1607 #elif  (defined __powerpc64__)
1608   static  Elf32_Half running_arch_code=EM_PPC64;
1609 #elif  (defined __powerpc__)
1610   static  Elf32_Half running_arch_code=EM_PPC;
1611 #elif (defined ARM)
1612   static  Elf32_Half running_arch_code=EM_ARM;
1613 #else
1614   #error Method os::dll_load requires that one of following is defined:\
1615        IA32, AMD64, IA64, __sparc, __powerpc__, ARM, ARM
1616 #endif
1617 
1618   // Identify compatability class for VM's architecture and library's architecture
1619   // Obtain string descriptions for architectures
1620 
1621   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1622   int running_arch_index=-1;
1623 
1624   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1625     if (running_arch_code == arch_array[i].code) {
1626       running_arch_index    = i;
1627     }
1628     if (lib_arch.code == arch_array[i].code) {
1629       lib_arch.compat_class = arch_array[i].compat_class;
1630       lib_arch.name         = arch_array[i].name;
1631     }
1632   }
1633 
1634   assert(running_arch_index != -1,
1635          "Didn't find running architecture code (running_arch_code) in arch_array");
1636   if (running_arch_index == -1) {
1637     // Even though running architecture detection failed
1638     // we may still continue with reporting dlerror() message
1639     return NULL;
1640   }
1641 
1642   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1643     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1644     return NULL;
1645   }
1646 
1647   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1648     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1649     return NULL;
1650   }
1651 
1652   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1653     if (lib_arch.name!=NULL) {
1654       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1655                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1656                  lib_arch.name, arch_array[running_arch_index].name);
1657     } else {
1658       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1659                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1660                  lib_arch.code,
1661                  arch_array[running_arch_index].name);
1662     }
1663   }
1664 
1665   return NULL;
1666 }
1667 
1668 void* os::dll_lookup(void* handle, const char* name) {
1669   return dlsym(handle, name);
1670 }
1671 
1672 void* os::get_default_process_handle() {
1673   return (void*)::dlopen(NULL, RTLD_LAZY);
1674 }
1675 
1676 int os::stat(const char *path, struct stat *sbuf) {
1677   char pathbuf[MAX_PATH];
1678   if (strlen(path) > MAX_PATH - 1) {
1679     errno = ENAMETOOLONG;
1680     return -1;
1681   }
1682   os::native_path(strcpy(pathbuf, path));
1683   return ::stat(pathbuf, sbuf);
1684 }
1685 
1686 static inline time_t get_mtime(const char* filename) {
1687   struct stat st;
1688   int ret = os::stat(filename, &st);
1689   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
1690   return st.st_mtime;
1691 }
1692 
1693 int os::compare_file_modified_times(const char* file1, const char* file2) {
1694   time_t t1 = get_mtime(file1);
1695   time_t t2 = get_mtime(file2);
1696   return t1 - t2;
1697 }
1698 
1699 static bool _print_ascii_file(const char* filename, outputStream* st) {
1700   int fd = ::open(filename, O_RDONLY);
1701   if (fd == -1) {
1702     return false;
1703   }
1704 
1705   char buf[32];
1706   int bytes;
1707   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1708     st->print_raw(buf, bytes);
1709   }
1710 
1711   ::close(fd);
1712 
1713   return true;
1714 }
1715 
1716 void os::print_os_info_brief(outputStream* st) {
1717   os::Solaris::print_distro_info(st);
1718 
1719   os::Posix::print_uname_info(st);
1720 
1721   os::Solaris::print_libversion_info(st);
1722 }
1723 
1724 void os::print_os_info(outputStream* st) {
1725   st->print("OS:");
1726 
1727   os::Solaris::print_distro_info(st);
1728 
1729   os::Posix::print_uname_info(st);
1730 
1731   os::Solaris::print_libversion_info(st);
1732 
1733   os::Posix::print_rlimit_info(st);
1734 
1735   os::Posix::print_load_average(st);
1736 }
1737 
1738 void os::Solaris::print_distro_info(outputStream* st) {
1739   if (!_print_ascii_file("/etc/release", st)) {
1740     st->print("Solaris");
1741   }
1742   st->cr();
1743 }
1744 
1745 void os::get_summary_os_info(char* buf, size_t buflen) {
1746   strncpy(buf, "Solaris", buflen);  // default to plain solaris
1747   FILE* fp = fopen("/etc/release", "r");
1748   if (fp != NULL) {
1749     char tmp[256];
1750     // Only get the first line and chop out everything but the os name.
1751     if (fgets(tmp, sizeof(tmp), fp)) {
1752       char* ptr = tmp;
1753       // skip past whitespace characters
1754       while (*ptr != '\0' && (*ptr == ' ' || *ptr == '\t' || *ptr == '\n')) ptr++;
1755       if (*ptr != '\0') {
1756         char* nl = strchr(ptr, '\n');
1757         if (nl != NULL) *nl = '\0';
1758         strncpy(buf, ptr, buflen);
1759       }
1760     }
1761     fclose(fp);
1762   }
1763 }
1764 
1765 void os::Solaris::print_libversion_info(outputStream* st) {
1766   st->print("  (T2 libthread)");
1767   st->cr();
1768 }
1769 
1770 static bool check_addr0(outputStream* st) {
1771   jboolean status = false;
1772   const int read_chunk = 200;
1773   int ret = 0;
1774   int nmap = 0;
1775   int fd = ::open("/proc/self/map",O_RDONLY);
1776   if (fd >= 0) {
1777     prmap_t *p = NULL;
1778     char *mbuff = (char *) calloc(read_chunk, sizeof(prmap_t));
1779     if (NULL == mbuff) {
1780       ::close(fd);
1781       return status;
1782     }
1783     while ((ret = ::read(fd, mbuff, read_chunk*sizeof(prmap_t))) > 0) {
1784       //check if read() has not read partial data
1785       if( 0 != ret % sizeof(prmap_t)){
1786         break;
1787       }
1788       nmap = ret / sizeof(prmap_t);
1789       p = (prmap_t *)mbuff;
1790       for(int i = 0; i < nmap; i++){
1791         if (p->pr_vaddr == 0x0) {
1792           st->print("Warning: Address: " PTR_FORMAT ", Size: " SIZE_FORMAT "K, ",p->pr_vaddr, p->pr_size/1024);
1793           st->print("Mapped file: %s, ", p->pr_mapname[0] == '\0' ? "None" : p->pr_mapname);
1794           st->print("Access: ");
1795           st->print("%s",(p->pr_mflags & MA_READ)  ? "r" : "-");
1796           st->print("%s",(p->pr_mflags & MA_WRITE) ? "w" : "-");
1797           st->print("%s",(p->pr_mflags & MA_EXEC)  ? "x" : "-");
1798           st->cr();
1799           status = true;
1800         }
1801         p++;
1802       }
1803     }
1804     free(mbuff);
1805     ::close(fd);
1806   }
1807   return status;
1808 }
1809 
1810 void os::get_summary_cpu_info(char* buf, size_t buflen) {
1811   // Get MHz with system call. We don't seem to already have this.
1812   processor_info_t stats;
1813   processorid_t id = getcpuid();
1814   int clock = 0;
1815   if (processor_info(id, &stats) != -1) {
1816     clock = stats.pi_clock;  // pi_processor_type isn't more informative than below
1817   }
1818 #ifdef AMD64
1819   snprintf(buf, buflen, "x86 64 bit %d MHz", clock);
1820 #else
1821   // must be sparc
1822   snprintf(buf, buflen, "Sparcv9 64 bit %d MHz", clock);
1823 #endif
1824 }
1825 
1826 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
1827   // Nothing to do for now.
1828 }
1829 
1830 void os::print_memory_info(outputStream* st) {
1831   st->print("Memory:");
1832   st->print(" %dk page", os::vm_page_size()>>10);
1833   st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
1834   st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
1835   st->cr();
1836   (void) check_addr0(st);
1837 }
1838 
1839 // Moved from whole group, because we need them here for diagnostic
1840 // prints.
1841 static int Maxsignum = 0;
1842 static int *ourSigFlags = NULL;
1843 
1844 int os::Solaris::get_our_sigflags(int sig) {
1845   assert(ourSigFlags!=NULL, "signal data structure not initialized");
1846   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
1847   return ourSigFlags[sig];
1848 }
1849 
1850 void os::Solaris::set_our_sigflags(int sig, int flags) {
1851   assert(ourSigFlags!=NULL, "signal data structure not initialized");
1852   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
1853   ourSigFlags[sig] = flags;
1854 }
1855 
1856 
1857 static const char* get_signal_handler_name(address handler,
1858                                            char* buf, int buflen) {
1859   int offset;
1860   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
1861   if (found) {
1862     // skip directory names
1863     const char *p1, *p2;
1864     p1 = buf;
1865     size_t len = strlen(os::file_separator());
1866     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
1867     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
1868   } else {
1869     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
1870   }
1871   return buf;
1872 }
1873 
1874 static void print_signal_handler(outputStream* st, int sig,
1875                                  char* buf, size_t buflen) {
1876   struct sigaction sa;
1877 
1878   sigaction(sig, NULL, &sa);
1879 
1880   st->print("%s: ", os::exception_name(sig, buf, buflen));
1881 
1882   address handler = (sa.sa_flags & SA_SIGINFO)
1883                   ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
1884                   : CAST_FROM_FN_PTR(address, sa.sa_handler);
1885 
1886   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
1887     st->print("SIG_DFL");
1888   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
1889     st->print("SIG_IGN");
1890   } else {
1891     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
1892   }
1893 
1894   st->print(", sa_mask[0]=");
1895   os::Posix::print_signal_set_short(st, &sa.sa_mask);
1896 
1897   address rh = VMError::get_resetted_sighandler(sig);
1898   // May be, handler was resetted by VMError?
1899   if (rh != NULL) {
1900     handler = rh;
1901     sa.sa_flags = VMError::get_resetted_sigflags(sig);
1902   }
1903 
1904   st->print(", sa_flags=");
1905   os::Posix::print_sa_flags(st, sa.sa_flags);
1906 
1907   // Check: is it our handler?
1908   if (handler == CAST_FROM_FN_PTR(address, signalHandler)) {
1909     // It is our signal handler
1910     // check for flags
1911     if (sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
1912       st->print(
1913                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
1914                 os::Solaris::get_our_sigflags(sig));
1915     }
1916   }
1917   st->cr();
1918 }
1919 
1920 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
1921   st->print_cr("Signal Handlers:");
1922   print_signal_handler(st, SIGSEGV, buf, buflen);
1923   print_signal_handler(st, SIGBUS , buf, buflen);
1924   print_signal_handler(st, SIGFPE , buf, buflen);
1925   print_signal_handler(st, SIGPIPE, buf, buflen);
1926   print_signal_handler(st, SIGXFSZ, buf, buflen);
1927   print_signal_handler(st, SIGILL , buf, buflen);
1928   print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
1929   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
1930   print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
1931   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
1932   print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
1933 }
1934 
1935 static char saved_jvm_path[MAXPATHLEN] = { 0 };
1936 
1937 // Find the full path to the current module, libjvm.so
1938 void os::jvm_path(char *buf, jint buflen) {
1939   // Error checking.
1940   if (buflen < MAXPATHLEN) {
1941     assert(false, "must use a large-enough buffer");
1942     buf[0] = '\0';
1943     return;
1944   }
1945   // Lazy resolve the path to current module.
1946   if (saved_jvm_path[0] != 0) {
1947     strcpy(buf, saved_jvm_path);
1948     return;
1949   }
1950 
1951   Dl_info dlinfo;
1952   int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
1953   assert(ret != 0, "cannot locate libjvm");
1954   if (ret != 0 && dlinfo.dli_fname != NULL) {
1955     if (os::Posix::realpath((char *)dlinfo.dli_fname, buf, buflen) == NULL) {
1956       return;
1957     }
1958   } else {
1959     buf[0] = '\0';
1960     return;
1961   }
1962 
1963   if (Arguments::sun_java_launcher_is_altjvm()) {
1964     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
1965     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
1966     // If "/jre/lib/" appears at the right place in the string, then
1967     // assume we are installed in a JDK and we're done.  Otherwise, check
1968     // for a JAVA_HOME environment variable and fix up the path so it
1969     // looks like libjvm.so is installed there (append a fake suffix
1970     // hotspot/libjvm.so).
1971     const char *p = buf + strlen(buf) - 1;
1972     for (int count = 0; p > buf && count < 5; ++count) {
1973       for (--p; p > buf && *p != '/'; --p)
1974         /* empty */ ;
1975     }
1976 
1977     if (strncmp(p, "/jre/lib/", 9) != 0) {
1978       // Look for JAVA_HOME in the environment.
1979       char* java_home_var = ::getenv("JAVA_HOME");
1980       if (java_home_var != NULL && java_home_var[0] != 0) {
1981         char* jrelib_p;
1982         int   len;
1983 
1984         // Check the current module name "libjvm.so".
1985         p = strrchr(buf, '/');
1986         assert(strstr(p, "/libjvm") == p, "invalid library name");
1987 
1988         if (os::Posix::realpath(java_home_var, buf, buflen) == NULL) {
1989           return;
1990         }
1991         // determine if this is a legacy image or modules image
1992         // modules image doesn't have "jre" subdirectory
1993         len = strlen(buf);
1994         assert(len < buflen, "Ran out of buffer space");
1995         jrelib_p = buf + len;
1996         snprintf(jrelib_p, buflen-len, "/jre/lib");
1997         if (0 != access(buf, F_OK)) {
1998           snprintf(jrelib_p, buflen-len, "/lib");
1999         }
2000 
2001         if (0 == access(buf, F_OK)) {
2002           // Use current module name "libjvm.so"
2003           len = strlen(buf);
2004           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2005         } else {
2006           // Go back to path of .so
2007           if (os::Posix::realpath((char *)dlinfo.dli_fname, buf, buflen) == NULL) {
2008             return;
2009           }
2010         }
2011       }
2012     }
2013   }
2014 
2015   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2016   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2017 }
2018 
2019 
2020 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2021   // no prefix required, not even "_"
2022 }
2023 
2024 
2025 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2026   // no suffix required
2027 }
2028 
2029 // This method is a copy of JDK's sysGetLastErrorString
2030 // from src/solaris/hpi/src/system_md.c
2031 
2032 size_t os::lasterror(char *buf, size_t len) {
2033   if (errno == 0)  return 0;
2034 
2035   const char *s = os::strerror(errno);
2036   size_t n = ::strlen(s);
2037   if (n >= len) {
2038     n = len - 1;
2039   }
2040   ::strncpy(buf, s, n);
2041   buf[n] = '\0';
2042   return n;
2043 }
2044 
2045 
2046 // sun.misc.Signal
2047 
2048 extern "C" {
2049   static void UserHandler(int sig, void *siginfo, void *context) {
2050     // Ctrl-C is pressed during error reporting, likely because the error
2051     // handler fails to abort. Let VM die immediately.
2052     if (sig == SIGINT && VMError::is_error_reported()) {
2053       os::die();
2054     }
2055 
2056     os::signal_notify(sig);
2057     // We do not need to reinstate the signal handler each time...
2058   }
2059 }
2060 
2061 void* os::user_handler() {
2062   return CAST_FROM_FN_PTR(void*, UserHandler);
2063 }
2064 
2065 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2066   struct timespec ts;
2067   unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2068 
2069   return ts;
2070 }
2071 
2072 extern "C" {
2073   typedef void (*sa_handler_t)(int);
2074   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2075 }
2076 
2077 void* os::signal(int signal_number, void* handler) {
2078   struct sigaction sigAct, oldSigAct;
2079   sigfillset(&(sigAct.sa_mask));
2080   sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2081   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2082 
2083   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2084     // -1 means registration failed
2085     return (void *)-1;
2086   }
2087 
2088   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2089 }
2090 
2091 void os::signal_raise(int signal_number) {
2092   raise(signal_number);
2093 }
2094 
2095 // The following code is moved from os.cpp for making this
2096 // code platform specific, which it is by its very nature.
2097 
2098 // a counter for each possible signal value
2099 static int Sigexit = 0;
2100 static jint *pending_signals = NULL;
2101 static int *preinstalled_sigs = NULL;
2102 static struct sigaction *chainedsigactions = NULL;
2103 static Semaphore* sig_sem = NULL;
2104 typedef int (*version_getting_t)();
2105 version_getting_t os::Solaris::get_libjsig_version = NULL;
2106 
2107 int os::sigexitnum_pd() {
2108   assert(Sigexit > 0, "signal memory not yet initialized");
2109   return Sigexit;
2110 }
2111 
2112 void os::Solaris::init_signal_mem() {
2113   // Initialize signal structures
2114   Maxsignum = SIGRTMAX;
2115   Sigexit = Maxsignum+1;
2116   assert(Maxsignum >0, "Unable to obtain max signal number");
2117 
2118   // Initialize signal structures
2119   // pending_signals has one int per signal
2120   // The additional signal is for SIGEXIT - exit signal to signal_thread
2121   pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1), mtInternal);
2122   memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2123 
2124   if (UseSignalChaining) {
2125     chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2126                                                    * (Maxsignum + 1), mtInternal);
2127     memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2128     preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2129     memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2130   }
2131   ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2132   memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2133 }
2134 
2135 void os::signal_init_pd() {
2136   // Initialize signal semaphore
2137   sig_sem = new Semaphore();
2138 }
2139 
2140 void os::signal_notify(int sig) {
2141   if (sig_sem != NULL) {
2142     Atomic::inc(&pending_signals[sig]);
2143     sig_sem->signal();
2144   } else {
2145     // Signal thread is not created with ReduceSignalUsage and signal_init_pd
2146     // initialization isn't called.
2147     assert(ReduceSignalUsage, "signal semaphore should be created");
2148   }
2149 }
2150 
2151 static int check_pending_signals() {
2152   int ret;
2153   while (true) {
2154     for (int i = 0; i < Sigexit + 1; i++) {
2155       jint n = pending_signals[i];
2156       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2157         return i;
2158       }
2159     }
2160     JavaThread *thread = JavaThread::current();
2161     ThreadBlockInVM tbivm(thread);
2162 
2163     bool threadIsSuspended;
2164     do {
2165       thread->set_suspend_equivalent();
2166       sig_sem->wait();
2167 
2168       // were we externally suspended while we were waiting?
2169       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2170       if (threadIsSuspended) {
2171         // The semaphore has been incremented, but while we were waiting
2172         // another thread suspended us. We don't want to continue running
2173         // while suspended because that would surprise the thread that
2174         // suspended us.
2175         sig_sem->signal();
2176 
2177         thread->java_suspend_self();
2178       }
2179     } while (threadIsSuspended);
2180   }
2181 }
2182 
2183 int os::signal_wait() {
2184   return check_pending_signals();
2185 }
2186 
2187 ////////////////////////////////////////////////////////////////////////////////
2188 // Virtual Memory
2189 
2190 static int page_size = -1;
2191 
2192 int os::vm_page_size() {
2193   assert(page_size != -1, "must call os::init");
2194   return page_size;
2195 }
2196 
2197 // Solaris allocates memory by pages.
2198 int os::vm_allocation_granularity() {
2199   assert(page_size != -1, "must call os::init");
2200   return page_size;
2201 }
2202 
2203 static bool recoverable_mmap_error(int err) {
2204   // See if the error is one we can let the caller handle. This
2205   // list of errno values comes from the Solaris mmap(2) man page.
2206   switch (err) {
2207   case EBADF:
2208   case EINVAL:
2209   case ENOTSUP:
2210     // let the caller deal with these errors
2211     return true;
2212 
2213   default:
2214     // Any remaining errors on this OS can cause our reserved mapping
2215     // to be lost. That can cause confusion where different data
2216     // structures think they have the same memory mapped. The worst
2217     // scenario is if both the VM and a library think they have the
2218     // same memory mapped.
2219     return false;
2220   }
2221 }
2222 
2223 static void warn_fail_commit_memory(char* addr, size_t bytes, bool exec,
2224                                     int err) {
2225   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2226           ", %d) failed; error='%s' (errno=%d)", addr, bytes, exec,
2227           os::strerror(err), err);
2228 }
2229 
2230 static void warn_fail_commit_memory(char* addr, size_t bytes,
2231                                     size_t alignment_hint, bool exec,
2232                                     int err) {
2233   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2234           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, bytes,
2235           alignment_hint, exec, os::strerror(err), err);
2236 }
2237 
2238 int os::Solaris::commit_memory_impl(char* addr, size_t bytes, bool exec) {
2239   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2240   size_t size = bytes;
2241   char *res = Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2242   if (res != NULL) {
2243     if (UseNUMAInterleaving) {
2244       numa_make_global(addr, bytes);
2245     }
2246     return 0;
2247   }
2248 
2249   int err = errno;  // save errno from mmap() call in mmap_chunk()
2250 
2251   if (!recoverable_mmap_error(err)) {
2252     warn_fail_commit_memory(addr, bytes, exec, err);
2253     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "committing reserved memory.");
2254   }
2255 
2256   return err;
2257 }
2258 
2259 bool os::pd_commit_memory(char* addr, size_t bytes, bool exec) {
2260   return Solaris::commit_memory_impl(addr, bytes, exec) == 0;
2261 }
2262 
2263 void os::pd_commit_memory_or_exit(char* addr, size_t bytes, bool exec,
2264                                   const char* mesg) {
2265   assert(mesg != NULL, "mesg must be specified");
2266   int err = os::Solaris::commit_memory_impl(addr, bytes, exec);
2267   if (err != 0) {
2268     // the caller wants all commit errors to exit with the specified mesg:
2269     warn_fail_commit_memory(addr, bytes, exec, err);
2270     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2271   }
2272 }
2273 
2274 size_t os::Solaris::page_size_for_alignment(size_t alignment) {
2275   assert(is_aligned(alignment, (size_t) vm_page_size()),
2276          SIZE_FORMAT " is not aligned to " SIZE_FORMAT,
2277          alignment, (size_t) vm_page_size());
2278 
2279   for (int i = 0; _page_sizes[i] != 0; i++) {
2280     if (is_aligned(alignment, _page_sizes[i])) {
2281       return _page_sizes[i];
2282     }
2283   }
2284 
2285   return (size_t) vm_page_size();
2286 }
2287 
2288 int os::Solaris::commit_memory_impl(char* addr, size_t bytes,
2289                                     size_t alignment_hint, bool exec) {
2290   int err = Solaris::commit_memory_impl(addr, bytes, exec);
2291   if (err == 0 && UseLargePages && alignment_hint > 0) {
2292     assert(is_aligned(bytes, alignment_hint),
2293            SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, alignment_hint);
2294 
2295     // The syscall memcntl requires an exact page size (see man memcntl for details).
2296     size_t page_size = page_size_for_alignment(alignment_hint);
2297     if (page_size > (size_t) vm_page_size()) {
2298       (void)Solaris::setup_large_pages(addr, bytes, page_size);
2299     }
2300   }
2301   return err;
2302 }
2303 
2304 bool os::pd_commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2305                           bool exec) {
2306   return Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec) == 0;
2307 }
2308 
2309 void os::pd_commit_memory_or_exit(char* addr, size_t bytes,
2310                                   size_t alignment_hint, bool exec,
2311                                   const char* mesg) {
2312   assert(mesg != NULL, "mesg must be specified");
2313   int err = os::Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec);
2314   if (err != 0) {
2315     // the caller wants all commit errors to exit with the specified mesg:
2316     warn_fail_commit_memory(addr, bytes, alignment_hint, exec, err);
2317     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2318   }
2319 }
2320 
2321 // Uncommit the pages in a specified region.
2322 void os::pd_free_memory(char* addr, size_t bytes, size_t alignment_hint) {
2323   if (madvise(addr, bytes, MADV_FREE) < 0) {
2324     debug_only(warning("MADV_FREE failed."));
2325     return;
2326   }
2327 }
2328 
2329 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2330   return os::commit_memory(addr, size, !ExecMem);
2331 }
2332 
2333 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2334   return os::uncommit_memory(addr, size);
2335 }
2336 
2337 // Change the page size in a given range.
2338 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2339   assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2340   assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2341   if (UseLargePages) {
2342     size_t page_size = Solaris::page_size_for_alignment(alignment_hint);
2343     if (page_size > (size_t) vm_page_size()) {
2344       Solaris::setup_large_pages(addr, bytes, page_size);
2345     }
2346   }
2347 }
2348 
2349 // Tell the OS to make the range local to the first-touching LWP
2350 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2351   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2352   if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2353     debug_only(warning("MADV_ACCESS_LWP failed."));
2354   }
2355 }
2356 
2357 // Tell the OS that this range would be accessed from different LWPs.
2358 void os::numa_make_global(char *addr, size_t bytes) {
2359   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2360   if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2361     debug_only(warning("MADV_ACCESS_MANY failed."));
2362   }
2363 }
2364 
2365 // Get the number of the locality groups.
2366 size_t os::numa_get_groups_num() {
2367   size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2368   return n != -1 ? n : 1;
2369 }
2370 
2371 // Get a list of leaf locality groups. A leaf lgroup is group that
2372 // doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2373 // board. An LWP is assigned to one of these groups upon creation.
2374 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2375   if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2376     ids[0] = 0;
2377     return 1;
2378   }
2379   int result_size = 0, top = 1, bottom = 0, cur = 0;
2380   for (int k = 0; k < size; k++) {
2381     int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2382                                    (Solaris::lgrp_id_t*)&ids[top], size - top);
2383     if (r == -1) {
2384       ids[0] = 0;
2385       return 1;
2386     }
2387     if (!r) {
2388       // That's a leaf node.
2389       assert(bottom <= cur, "Sanity check");
2390       // Check if the node has memory
2391       if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2392                                   NULL, 0, LGRP_RSRC_MEM) > 0) {
2393         ids[bottom++] = ids[cur];
2394       }
2395     }
2396     top += r;
2397     cur++;
2398   }
2399   if (bottom == 0) {
2400     // Handle a situation, when the OS reports no memory available.
2401     // Assume UMA architecture.
2402     ids[0] = 0;
2403     return 1;
2404   }
2405   return bottom;
2406 }
2407 
2408 // Detect the topology change. Typically happens during CPU plugging-unplugging.
2409 bool os::numa_topology_changed() {
2410   int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2411   if (is_stale != -1 && is_stale) {
2412     Solaris::lgrp_fini(Solaris::lgrp_cookie());
2413     Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2414     assert(c != 0, "Failure to initialize LGRP API");
2415     Solaris::set_lgrp_cookie(c);
2416     return true;
2417   }
2418   return false;
2419 }
2420 
2421 // Get the group id of the current LWP.
2422 int os::numa_get_group_id() {
2423   int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2424   if (lgrp_id == -1) {
2425     return 0;
2426   }
2427   const int size = os::numa_get_groups_num();
2428   int *ids = (int*)alloca(size * sizeof(int));
2429 
2430   // Get the ids of all lgroups with memory; r is the count.
2431   int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2432                                   (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2433   if (r <= 0) {
2434     return 0;
2435   }
2436   return ids[os::random() % r];
2437 }
2438 
2439 // Request information about the page.
2440 bool os::get_page_info(char *start, page_info* info) {
2441   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2442   uint64_t addr = (uintptr_t)start;
2443   uint64_t outdata[2];
2444   uint_t validity = 0;
2445 
2446   if (meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2447     return false;
2448   }
2449 
2450   info->size = 0;
2451   info->lgrp_id = -1;
2452 
2453   if ((validity & 1) != 0) {
2454     if ((validity & 2) != 0) {
2455       info->lgrp_id = outdata[0];
2456     }
2457     if ((validity & 4) != 0) {
2458       info->size = outdata[1];
2459     }
2460     return true;
2461   }
2462   return false;
2463 }
2464 
2465 // Scan the pages from start to end until a page different than
2466 // the one described in the info parameter is encountered.
2467 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2468                      page_info* page_found) {
2469   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2470   const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2471   uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT + 1];
2472   uint_t validity[MAX_MEMINFO_CNT];
2473 
2474   size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2475   uint64_t p = (uint64_t)start;
2476   while (p < (uint64_t)end) {
2477     addrs[0] = p;
2478     size_t addrs_count = 1;
2479     while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] + page_size < (uint64_t)end) {
2480       addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2481       addrs_count++;
2482     }
2483 
2484     if (meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2485       return NULL;
2486     }
2487 
2488     size_t i = 0;
2489     for (; i < addrs_count; i++) {
2490       if ((validity[i] & 1) != 0) {
2491         if ((validity[i] & 4) != 0) {
2492           if (outdata[types * i + 1] != page_expected->size) {
2493             break;
2494           }
2495         } else if (page_expected->size != 0) {
2496           break;
2497         }
2498 
2499         if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2500           if (outdata[types * i] != page_expected->lgrp_id) {
2501             break;
2502           }
2503         }
2504       } else {
2505         return NULL;
2506       }
2507     }
2508 
2509     if (i < addrs_count) {
2510       if ((validity[i] & 2) != 0) {
2511         page_found->lgrp_id = outdata[types * i];
2512       } else {
2513         page_found->lgrp_id = -1;
2514       }
2515       if ((validity[i] & 4) != 0) {
2516         page_found->size = outdata[types * i + 1];
2517       } else {
2518         page_found->size = 0;
2519       }
2520       return (char*)addrs[i];
2521     }
2522 
2523     p = addrs[addrs_count - 1] + page_size;
2524   }
2525   return end;
2526 }
2527 
2528 bool os::pd_uncommit_memory(char* addr, size_t bytes) {
2529   size_t size = bytes;
2530   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2531   // uncommitted page. Otherwise, the read/write might succeed if we
2532   // have enough swap space to back the physical page.
2533   return
2534     NULL != Solaris::mmap_chunk(addr, size,
2535                                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
2536                                 PROT_NONE);
2537 }
2538 
2539 char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
2540   char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
2541 
2542   if (b == MAP_FAILED) {
2543     return NULL;
2544   }
2545   return b;
2546 }
2547 
2548 char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes,
2549                              size_t alignment_hint, bool fixed) {
2550   char* addr = requested_addr;
2551   int flags = MAP_PRIVATE | MAP_NORESERVE;
2552 
2553   assert(!(fixed && (alignment_hint > 0)),
2554          "alignment hint meaningless with fixed mmap");
2555 
2556   if (fixed) {
2557     flags |= MAP_FIXED;
2558   } else if (alignment_hint > (size_t) vm_page_size()) {
2559     flags |= MAP_ALIGN;
2560     addr = (char*) alignment_hint;
2561   }
2562 
2563   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2564   // uncommitted page. Otherwise, the read/write might succeed if we
2565   // have enough swap space to back the physical page.
2566   return mmap_chunk(addr, bytes, flags, PROT_NONE);
2567 }
2568 
2569 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2570                             size_t alignment_hint) {
2571   char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint,
2572                                   (requested_addr != NULL));
2573 
2574   guarantee(requested_addr == NULL || requested_addr == addr,
2575             "OS failed to return requested mmap address.");
2576   return addr;
2577 }
2578 
2579 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
2580   assert(file_desc >= 0, "file_desc is not valid");
2581   char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
2582   if (result != NULL) {
2583     if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
2584       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
2585     }
2586   }
2587   return result;
2588 }
2589 
2590 // Reserve memory at an arbitrary address, only if that area is
2591 // available (and not reserved for something else).
2592 
2593 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2594   const int max_tries = 10;
2595   char* base[max_tries];
2596   size_t size[max_tries];
2597 
2598   // Solaris adds a gap between mmap'ed regions.  The size of the gap
2599   // is dependent on the requested size and the MMU.  Our initial gap
2600   // value here is just a guess and will be corrected later.
2601   bool had_top_overlap = false;
2602   bool have_adjusted_gap = false;
2603   size_t gap = 0x400000;
2604 
2605   // Assert only that the size is a multiple of the page size, since
2606   // that's all that mmap requires, and since that's all we really know
2607   // about at this low abstraction level.  If we need higher alignment,
2608   // we can either pass an alignment to this method or verify alignment
2609   // in one of the methods further up the call chain.  See bug 5044738.
2610   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2611 
2612   // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
2613   // Give it a try, if the kernel honors the hint we can return immediately.
2614   char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
2615 
2616   volatile int err = errno;
2617   if (addr == requested_addr) {
2618     return addr;
2619   } else if (addr != NULL) {
2620     pd_unmap_memory(addr, bytes);
2621   }
2622 
2623   if (log_is_enabled(Warning, os)) {
2624     char buf[256];
2625     buf[0] = '\0';
2626     if (addr == NULL) {
2627       jio_snprintf(buf, sizeof(buf), ": %s", os::strerror(err));
2628     }
2629     log_info(os)("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
2630             PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
2631             "%s", bytes, requested_addr, addr, buf);
2632   }
2633 
2634   // Address hint method didn't work.  Fall back to the old method.
2635   // In theory, once SNV becomes our oldest supported platform, this
2636   // code will no longer be needed.
2637   //
2638   // Repeatedly allocate blocks until the block is allocated at the
2639   // right spot. Give up after max_tries.
2640   int i;
2641   for (i = 0; i < max_tries; ++i) {
2642     base[i] = reserve_memory(bytes);
2643 
2644     if (base[i] != NULL) {
2645       // Is this the block we wanted?
2646       if (base[i] == requested_addr) {
2647         size[i] = bytes;
2648         break;
2649       }
2650 
2651       // check that the gap value is right
2652       if (had_top_overlap && !have_adjusted_gap) {
2653         size_t actual_gap = base[i-1] - base[i] - bytes;
2654         if (gap != actual_gap) {
2655           // adjust the gap value and retry the last 2 allocations
2656           assert(i > 0, "gap adjustment code problem");
2657           have_adjusted_gap = true;  // adjust the gap only once, just in case
2658           gap = actual_gap;
2659           log_info(os)("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
2660           unmap_memory(base[i], bytes);
2661           unmap_memory(base[i-1], size[i-1]);
2662           i-=2;
2663           continue;
2664         }
2665       }
2666 
2667       // Does this overlap the block we wanted? Give back the overlapped
2668       // parts and try again.
2669       //
2670       // There is still a bug in this code: if top_overlap == bytes,
2671       // the overlap is offset from requested region by the value of gap.
2672       // In this case giving back the overlapped part will not work,
2673       // because we'll give back the entire block at base[i] and
2674       // therefore the subsequent allocation will not generate a new gap.
2675       // This could be fixed with a new algorithm that used larger
2676       // or variable size chunks to find the requested region -
2677       // but such a change would introduce additional complications.
2678       // It's rare enough that the planets align for this bug,
2679       // so we'll just wait for a fix for 6204603/5003415 which
2680       // will provide a mmap flag to allow us to avoid this business.
2681 
2682       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2683       if (top_overlap >= 0 && top_overlap < bytes) {
2684         had_top_overlap = true;
2685         unmap_memory(base[i], top_overlap);
2686         base[i] += top_overlap;
2687         size[i] = bytes - top_overlap;
2688       } else {
2689         size_t bottom_overlap = base[i] + bytes - requested_addr;
2690         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2691           if (bottom_overlap == 0) {
2692             log_info(os)("attempt_reserve_memory_at: possible alignment bug");
2693           }
2694           unmap_memory(requested_addr, bottom_overlap);
2695           size[i] = bytes - bottom_overlap;
2696         } else {
2697           size[i] = bytes;
2698         }
2699       }
2700     }
2701   }
2702 
2703   // Give back the unused reserved pieces.
2704 
2705   for (int j = 0; j < i; ++j) {
2706     if (base[j] != NULL) {
2707       unmap_memory(base[j], size[j]);
2708     }
2709   }
2710 
2711   return (i < max_tries) ? requested_addr : NULL;
2712 }
2713 
2714 bool os::pd_release_memory(char* addr, size_t bytes) {
2715   size_t size = bytes;
2716   return munmap(addr, size) == 0;
2717 }
2718 
2719 static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
2720   assert(addr == (char*)align_down((uintptr_t)addr, os::vm_page_size()),
2721          "addr must be page aligned");
2722   int retVal = mprotect(addr, bytes, prot);
2723   return retVal == 0;
2724 }
2725 
2726 // Protect memory (Used to pass readonly pages through
2727 // JNI GetArray<type>Elements with empty arrays.)
2728 // Also, used for serialization page and for compressed oops null pointer
2729 // checking.
2730 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2731                         bool is_committed) {
2732   unsigned int p = 0;
2733   switch (prot) {
2734   case MEM_PROT_NONE: p = PROT_NONE; break;
2735   case MEM_PROT_READ: p = PROT_READ; break;
2736   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2737   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2738   default:
2739     ShouldNotReachHere();
2740   }
2741   // is_committed is unused.
2742   return solaris_mprotect(addr, bytes, p);
2743 }
2744 
2745 // guard_memory and unguard_memory only happens within stack guard pages.
2746 // Since ISM pertains only to the heap, guard and unguard memory should not
2747 /// happen with an ISM region.
2748 bool os::guard_memory(char* addr, size_t bytes) {
2749   return solaris_mprotect(addr, bytes, PROT_NONE);
2750 }
2751 
2752 bool os::unguard_memory(char* addr, size_t bytes) {
2753   return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
2754 }
2755 
2756 // Large page support
2757 static size_t _large_page_size = 0;
2758 
2759 // Insertion sort for small arrays (descending order).
2760 static void insertion_sort_descending(size_t* array, int len) {
2761   for (int i = 0; i < len; i++) {
2762     size_t val = array[i];
2763     for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
2764       size_t tmp = array[key];
2765       array[key] = array[key - 1];
2766       array[key - 1] = tmp;
2767     }
2768   }
2769 }
2770 
2771 bool os::Solaris::mpss_sanity_check(bool warn, size_t* page_size) {
2772   const unsigned int usable_count = VM_Version::page_size_count();
2773   if (usable_count == 1) {
2774     return false;
2775   }
2776 
2777   // Find the right getpagesizes interface.  When solaris 11 is the minimum
2778   // build platform, getpagesizes() (without the '2') can be called directly.
2779   typedef int (*gps_t)(size_t[], int);
2780   gps_t gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes2"));
2781   if (gps_func == NULL) {
2782     gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes"));
2783     if (gps_func == NULL) {
2784       if (warn) {
2785         warning("MPSS is not supported by the operating system.");
2786       }
2787       return false;
2788     }
2789   }
2790 
2791   // Fill the array of page sizes.
2792   int n = (*gps_func)(_page_sizes, page_sizes_max);
2793   assert(n > 0, "Solaris bug?");
2794 
2795   if (n == page_sizes_max) {
2796     // Add a sentinel value (necessary only if the array was completely filled
2797     // since it is static (zeroed at initialization)).
2798     _page_sizes[--n] = 0;
2799     DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
2800   }
2801   assert(_page_sizes[n] == 0, "missing sentinel");
2802   trace_page_sizes("available page sizes", _page_sizes, n);
2803 
2804   if (n == 1) return false;     // Only one page size available.
2805 
2806   // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
2807   // select up to usable_count elements.  First sort the array, find the first
2808   // acceptable value, then copy the usable sizes to the top of the array and
2809   // trim the rest.  Make sure to include the default page size :-).
2810   //
2811   // A better policy could get rid of the 4M limit by taking the sizes of the
2812   // important VM memory regions (java heap and possibly the code cache) into
2813   // account.
2814   insertion_sort_descending(_page_sizes, n);
2815   const size_t size_limit =
2816     FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
2817   int beg;
2818   for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */;
2819   const int end = MIN2((int)usable_count, n) - 1;
2820   for (int cur = 0; cur < end; ++cur, ++beg) {
2821     _page_sizes[cur] = _page_sizes[beg];
2822   }
2823   _page_sizes[end] = vm_page_size();
2824   _page_sizes[end + 1] = 0;
2825 
2826   if (_page_sizes[end] > _page_sizes[end - 1]) {
2827     // Default page size is not the smallest; sort again.
2828     insertion_sort_descending(_page_sizes, end + 1);
2829   }
2830   *page_size = _page_sizes[0];
2831 
2832   trace_page_sizes("usable page sizes", _page_sizes, end + 1);
2833   return true;
2834 }
2835 
2836 void os::large_page_init() {
2837   if (UseLargePages) {
2838     // print a warning if any large page related flag is specified on command line
2839     bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
2840                            !FLAG_IS_DEFAULT(LargePageSizeInBytes);
2841 
2842     UseLargePages = Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
2843   }
2844 }
2845 
2846 bool os::Solaris::is_valid_page_size(size_t bytes) {
2847   for (int i = 0; _page_sizes[i] != 0; i++) {
2848     if (_page_sizes[i] == bytes) {
2849       return true;
2850     }
2851   }
2852   return false;
2853 }
2854 
2855 bool os::Solaris::setup_large_pages(caddr_t start, size_t bytes, size_t align) {
2856   assert(is_valid_page_size(align), SIZE_FORMAT " is not a valid page size", align);
2857   assert(is_aligned((void*) start, align),
2858          PTR_FORMAT " is not aligned to " SIZE_FORMAT, p2i((void*) start), align);
2859   assert(is_aligned(bytes, align),
2860          SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, align);
2861 
2862   // Signal to OS that we want large pages for addresses
2863   // from addr, addr + bytes
2864   struct memcntl_mha mpss_struct;
2865   mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
2866   mpss_struct.mha_pagesize = align;
2867   mpss_struct.mha_flags = 0;
2868   // Upon successful completion, memcntl() returns 0
2869   if (memcntl(start, bytes, MC_HAT_ADVISE, (caddr_t) &mpss_struct, 0, 0)) {
2870     debug_only(warning("Attempt to use MPSS failed."));
2871     return false;
2872   }
2873   return true;
2874 }
2875 
2876 char* os::reserve_memory_special(size_t size, size_t alignment, char* addr, bool exec) {
2877   fatal("os::reserve_memory_special should not be called on Solaris.");
2878   return NULL;
2879 }
2880 
2881 bool os::release_memory_special(char* base, size_t bytes) {
2882   fatal("os::release_memory_special should not be called on Solaris.");
2883   return false;
2884 }
2885 
2886 size_t os::large_page_size() {
2887   return _large_page_size;
2888 }
2889 
2890 // MPSS allows application to commit large page memory on demand; with ISM
2891 // the entire memory region must be allocated as shared memory.
2892 bool os::can_commit_large_page_memory() {
2893   return true;
2894 }
2895 
2896 bool os::can_execute_large_page_memory() {
2897   return true;
2898 }
2899 
2900 // Read calls from inside the vm need to perform state transitions
2901 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2902   size_t res;
2903   JavaThread* thread = (JavaThread*)Thread::current();
2904   assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
2905   ThreadBlockInVM tbiv(thread);
2906   RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
2907   return res;
2908 }
2909 
2910 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
2911   size_t res;
2912   JavaThread* thread = (JavaThread*)Thread::current();
2913   assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
2914   ThreadBlockInVM tbiv(thread);
2915   RESTARTABLE(::pread(fd, buf, (size_t) nBytes, offset), res);
2916   return res;
2917 }
2918 
2919 size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
2920   size_t res;
2921   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
2922          "Assumed _thread_in_native");
2923   RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
2924   return res;
2925 }
2926 
2927 void os::naked_short_sleep(jlong ms) {
2928   assert(ms < 1000, "Un-interruptable sleep, short time use only");
2929 
2930   // usleep is deprecated and removed from POSIX, in favour of nanosleep, but
2931   // Solaris requires -lrt for this.
2932   usleep((ms * 1000));
2933 
2934   return;
2935 }
2936 
2937 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2938 void os::infinite_sleep() {
2939   while (true) {    // sleep forever ...
2940     ::sleep(100);   // ... 100 seconds at a time
2941   }
2942 }
2943 
2944 // Used to convert frequent JVM_Yield() to nops
2945 bool os::dont_yield() {
2946   if (DontYieldALot) {
2947     static hrtime_t last_time = 0;
2948     hrtime_t diff = getTimeNanos() - last_time;
2949 
2950     if (diff < DontYieldALotInterval * 1000000) {
2951       return true;
2952     }
2953 
2954     last_time += diff;
2955 
2956     return false;
2957   } else {
2958     return false;
2959   }
2960 }
2961 
2962 // Note that yield semantics are defined by the scheduling class to which
2963 // the thread currently belongs.  Typically, yield will _not yield to
2964 // other equal or higher priority threads that reside on the dispatch queues
2965 // of other CPUs.
2966 
2967 void os::naked_yield() {
2968   thr_yield();
2969 }
2970 
2971 // Interface for setting lwp priorities.  We are using T2 libthread,
2972 // which forces the use of bound threads, so all of our threads will
2973 // be assigned to real lwp's.  Using the thr_setprio function is
2974 // meaningless in this mode so we must adjust the real lwp's priority.
2975 // The routines below implement the getting and setting of lwp priorities.
2976 //
2977 // Note: There are three priority scales used on Solaris.  Java priotities
2978 //       which range from 1 to 10, libthread "thr_setprio" scale which range
2979 //       from 0 to 127, and the current scheduling class of the process we
2980 //       are running in.  This is typically from -60 to +60.
2981 //       The setting of the lwp priorities in done after a call to thr_setprio
2982 //       so Java priorities are mapped to libthread priorities and we map from
2983 //       the latter to lwp priorities.  We don't keep priorities stored in
2984 //       Java priorities since some of our worker threads want to set priorities
2985 //       higher than all Java threads.
2986 //
2987 // For related information:
2988 // (1)  man -s 2 priocntl
2989 // (2)  man -s 4 priocntl
2990 // (3)  man dispadmin
2991 // =    librt.so
2992 // =    libthread/common/rtsched.c - thrp_setlwpprio().
2993 // =    ps -cL <pid> ... to validate priority.
2994 // =    sched_get_priority_min and _max
2995 //              pthread_create
2996 //              sched_setparam
2997 //              pthread_setschedparam
2998 //
2999 // Assumptions:
3000 // +    We assume that all threads in the process belong to the same
3001 //              scheduling class.   IE. an homogenous process.
3002 // +    Must be root or in IA group to change change "interactive" attribute.
3003 //              Priocntl() will fail silently.  The only indication of failure is when
3004 //              we read-back the value and notice that it hasn't changed.
3005 // +    Interactive threads enter the runq at the head, non-interactive at the tail.
3006 // +    For RT, change timeslice as well.  Invariant:
3007 //              constant "priority integral"
3008 //              Konst == TimeSlice * (60-Priority)
3009 //              Given a priority, compute appropriate timeslice.
3010 // +    Higher numerical values have higher priority.
3011 
3012 // sched class attributes
3013 typedef struct {
3014   int   schedPolicy;              // classID
3015   int   maxPrio;
3016   int   minPrio;
3017 } SchedInfo;
3018 
3019 
3020 static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
3021 
3022 #ifdef ASSERT
3023 static int  ReadBackValidate = 1;
3024 #endif
3025 static int  myClass     = 0;
3026 static int  myMin       = 0;
3027 static int  myMax       = 0;
3028 static int  myCur       = 0;
3029 static bool priocntl_enable = false;
3030 
3031 static const int criticalPrio = FXCriticalPriority;
3032 static int java_MaxPriority_to_os_priority = 0; // Saved mapping
3033 
3034 
3035 // lwp_priocntl_init
3036 //
3037 // Try to determine the priority scale for our process.
3038 //
3039 // Return errno or 0 if OK.
3040 //
3041 static int lwp_priocntl_init() {
3042   int rslt;
3043   pcinfo_t ClassInfo;
3044   pcparms_t ParmInfo;
3045   int i;
3046 
3047   if (!UseThreadPriorities) return 0;
3048 
3049   // If ThreadPriorityPolicy is 1, switch tables
3050   if (ThreadPriorityPolicy == 1) {
3051     for (i = 0; i < CriticalPriority+1; i++)
3052       os::java_to_os_priority[i] = prio_policy1[i];
3053   }
3054   if (UseCriticalJavaThreadPriority) {
3055     // MaxPriority always maps to the FX scheduling class and criticalPrio.
3056     // See set_native_priority() and set_lwp_class_and_priority().
3057     // Save original MaxPriority mapping in case attempt to
3058     // use critical priority fails.
3059     java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
3060     // Set negative to distinguish from other priorities
3061     os::java_to_os_priority[MaxPriority] = -criticalPrio;
3062   }
3063 
3064   // Get IDs for a set of well-known scheduling classes.
3065   // TODO-FIXME: GETCLINFO returns the current # of classes in the
3066   // the system.  We should have a loop that iterates over the
3067   // classID values, which are known to be "small" integers.
3068 
3069   strcpy(ClassInfo.pc_clname, "TS");
3070   ClassInfo.pc_cid = -1;
3071   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3072   if (rslt < 0) return errno;
3073   assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3074   tsLimits.schedPolicy = ClassInfo.pc_cid;
3075   tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3076   tsLimits.minPrio = -tsLimits.maxPrio;
3077 
3078   strcpy(ClassInfo.pc_clname, "IA");
3079   ClassInfo.pc_cid = -1;
3080   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3081   if (rslt < 0) return errno;
3082   assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3083   iaLimits.schedPolicy = ClassInfo.pc_cid;
3084   iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3085   iaLimits.minPrio = -iaLimits.maxPrio;
3086 
3087   strcpy(ClassInfo.pc_clname, "RT");
3088   ClassInfo.pc_cid = -1;
3089   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3090   if (rslt < 0) return errno;
3091   assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3092   rtLimits.schedPolicy = ClassInfo.pc_cid;
3093   rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3094   rtLimits.minPrio = 0;
3095 
3096   strcpy(ClassInfo.pc_clname, "FX");
3097   ClassInfo.pc_cid = -1;
3098   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3099   if (rslt < 0) return errno;
3100   assert(ClassInfo.pc_cid != -1, "cid for FX class is -1");
3101   fxLimits.schedPolicy = ClassInfo.pc_cid;
3102   fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
3103   fxLimits.minPrio = 0;
3104 
3105   // Query our "current" scheduling class.
3106   // This will normally be IA, TS or, rarely, FX or RT.
3107   memset(&ParmInfo, 0, sizeof(ParmInfo));
3108   ParmInfo.pc_cid = PC_CLNULL;
3109   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3110   if (rslt < 0) return errno;
3111   myClass = ParmInfo.pc_cid;
3112 
3113   // We now know our scheduling classId, get specific information
3114   // about the class.
3115   ClassInfo.pc_cid = myClass;
3116   ClassInfo.pc_clname[0] = 0;
3117   rslt = priocntl((idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
3118   if (rslt < 0) return errno;
3119 
3120   if (ThreadPriorityVerbose) {
3121     tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3122   }
3123 
3124   memset(&ParmInfo, 0, sizeof(pcparms_t));
3125   ParmInfo.pc_cid = PC_CLNULL;
3126   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3127   if (rslt < 0) return errno;
3128 
3129   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3130     myMin = rtLimits.minPrio;
3131     myMax = rtLimits.maxPrio;
3132   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3133     iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3134     myMin = iaLimits.minPrio;
3135     myMax = iaLimits.maxPrio;
3136     myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3137   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3138     tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3139     myMin = tsLimits.minPrio;
3140     myMax = tsLimits.maxPrio;
3141     myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3142   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3143     fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
3144     myMin = fxLimits.minPrio;
3145     myMax = fxLimits.maxPrio;
3146     myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
3147   } else {
3148     // No clue - punt
3149     if (ThreadPriorityVerbose) {
3150       tty->print_cr("Unknown scheduling class: %s ... \n",
3151                     ClassInfo.pc_clname);
3152     }
3153     return EINVAL;      // no clue, punt
3154   }
3155 
3156   if (ThreadPriorityVerbose) {
3157     tty->print_cr("Thread priority Range: [%d..%d]\n", myMin, myMax);
3158   }
3159 
3160   priocntl_enable = true;  // Enable changing priorities
3161   return 0;
3162 }
3163 
3164 #define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3165 #define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3166 #define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3167 #define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
3168 
3169 
3170 // scale_to_lwp_priority
3171 //
3172 // Convert from the libthread "thr_setprio" scale to our current
3173 // lwp scheduling class scale.
3174 //
3175 static int scale_to_lwp_priority(int rMin, int rMax, int x) {
3176   int v;
3177 
3178   if (x == 127) return rMax;            // avoid round-down
3179   v = (((x*(rMax-rMin)))/128)+rMin;
3180   return v;
3181 }
3182 
3183 
3184 // set_lwp_class_and_priority
3185 int set_lwp_class_and_priority(int ThreadID, int lwpid,
3186                                int newPrio, int new_class, bool scale) {
3187   int rslt;
3188   int Actual, Expected, prv;
3189   pcparms_t ParmInfo;                   // for GET-SET
3190 #ifdef ASSERT
3191   pcparms_t ReadBack;                   // for readback
3192 #endif
3193 
3194   // Set priority via PC_GETPARMS, update, PC_SETPARMS
3195   // Query current values.
3196   // TODO: accelerate this by eliminating the PC_GETPARMS call.
3197   // Cache "pcparms_t" in global ParmCache.
3198   // TODO: elide set-to-same-value
3199 
3200   // If something went wrong on init, don't change priorities.
3201   if (!priocntl_enable) {
3202     if (ThreadPriorityVerbose) {
3203       tty->print_cr("Trying to set priority but init failed, ignoring");
3204     }
3205     return EINVAL;
3206   }
3207 
3208   // If lwp hasn't started yet, just return
3209   // the _start routine will call us again.
3210   if (lwpid <= 0) {
3211     if (ThreadPriorityVerbose) {
3212       tty->print_cr("deferring the set_lwp_class_and_priority of thread "
3213                     INTPTR_FORMAT " to %d, lwpid not set",
3214                     ThreadID, newPrio);
3215     }
3216     return 0;
3217   }
3218 
3219   if (ThreadPriorityVerbose) {
3220     tty->print_cr ("set_lwp_class_and_priority("
3221                    INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3222                    ThreadID, lwpid, newPrio);
3223   }
3224 
3225   memset(&ParmInfo, 0, sizeof(pcparms_t));
3226   ParmInfo.pc_cid = PC_CLNULL;
3227   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3228   if (rslt < 0) return errno;
3229 
3230   int cur_class = ParmInfo.pc_cid;
3231   ParmInfo.pc_cid = (id_t)new_class;
3232 
3233   if (new_class == rtLimits.schedPolicy) {
3234     rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3235     rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
3236                                                        rtLimits.maxPrio, newPrio)
3237                                : newPrio;
3238     rtInfo->rt_tqsecs  = RT_NOCHANGE;
3239     rtInfo->rt_tqnsecs = RT_NOCHANGE;
3240     if (ThreadPriorityVerbose) {
3241       tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3242     }
3243   } else if (new_class == iaLimits.schedPolicy) {
3244     iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3245     int maxClamped     = MIN2(iaLimits.maxPrio,
3246                               cur_class == new_class
3247                               ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
3248     iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
3249                                                        maxClamped, newPrio)
3250                                : newPrio;
3251     iaInfo->ia_uprilim = cur_class == new_class
3252                            ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
3253     iaInfo->ia_mode    = IA_NOCHANGE;
3254     if (ThreadPriorityVerbose) {
3255       tty->print_cr("IA: [%d...%d] %d->%d\n",
3256                     iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3257     }
3258   } else if (new_class == tsLimits.schedPolicy) {
3259     tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3260     int maxClamped     = MIN2(tsLimits.maxPrio,
3261                               cur_class == new_class
3262                               ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
3263     tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
3264                                                        maxClamped, newPrio)
3265                                : newPrio;
3266     tsInfo->ts_uprilim = cur_class == new_class
3267                            ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
3268     if (ThreadPriorityVerbose) {
3269       tty->print_cr("TS: [%d...%d] %d->%d\n",
3270                     tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3271     }
3272   } else if (new_class == fxLimits.schedPolicy) {
3273     fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
3274     int maxClamped     = MIN2(fxLimits.maxPrio,
3275                               cur_class == new_class
3276                               ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
3277     fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
3278                                                        maxClamped, newPrio)
3279                                : newPrio;
3280     fxInfo->fx_uprilim = cur_class == new_class
3281                            ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
3282     fxInfo->fx_tqsecs  = FX_NOCHANGE;
3283     fxInfo->fx_tqnsecs = FX_NOCHANGE;
3284     if (ThreadPriorityVerbose) {
3285       tty->print_cr("FX: [%d...%d] %d->%d\n",
3286                     fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
3287     }
3288   } else {
3289     if (ThreadPriorityVerbose) {
3290       tty->print_cr("Unknown new scheduling class %d\n", new_class);
3291     }
3292     return EINVAL;    // no clue, punt
3293   }
3294 
3295   rslt = priocntl(P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3296   if (ThreadPriorityVerbose && rslt) {
3297     tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3298   }
3299   if (rslt < 0) return errno;
3300 
3301 #ifdef ASSERT
3302   // Sanity check: read back what we just attempted to set.
3303   // In theory it could have changed in the interim ...
3304   //
3305   // The priocntl system call is tricky.
3306   // Sometimes it'll validate the priority value argument and
3307   // return EINVAL if unhappy.  At other times it fails silently.
3308   // Readbacks are prudent.
3309 
3310   if (!ReadBackValidate) return 0;
3311 
3312   memset(&ReadBack, 0, sizeof(pcparms_t));
3313   ReadBack.pc_cid = PC_CLNULL;
3314   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3315   assert(rslt >= 0, "priocntl failed");
3316   Actual = Expected = 0xBAD;
3317   assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3318   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3319     Actual   = RTPRI(ReadBack)->rt_pri;
3320     Expected = RTPRI(ParmInfo)->rt_pri;
3321   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3322     Actual   = IAPRI(ReadBack)->ia_upri;
3323     Expected = IAPRI(ParmInfo)->ia_upri;
3324   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3325     Actual   = TSPRI(ReadBack)->ts_upri;
3326     Expected = TSPRI(ParmInfo)->ts_upri;
3327   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3328     Actual   = FXPRI(ReadBack)->fx_upri;
3329     Expected = FXPRI(ParmInfo)->fx_upri;
3330   } else {
3331     if (ThreadPriorityVerbose) {
3332       tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
3333                     ParmInfo.pc_cid);
3334     }
3335   }
3336 
3337   if (Actual != Expected) {
3338     if (ThreadPriorityVerbose) {
3339       tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
3340                      lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
3341     }
3342   }
3343 #endif
3344 
3345   return 0;
3346 }
3347 
3348 // Solaris only gives access to 128 real priorities at a time,
3349 // so we expand Java's ten to fill this range.  This would be better
3350 // if we dynamically adjusted relative priorities.
3351 //
3352 // The ThreadPriorityPolicy option allows us to select 2 different
3353 // priority scales.
3354 //
3355 // ThreadPriorityPolicy=0
3356 // Since the Solaris' default priority is MaximumPriority, we do not
3357 // set a priority lower than Max unless a priority lower than
3358 // NormPriority is requested.
3359 //
3360 // ThreadPriorityPolicy=1
3361 // This mode causes the priority table to get filled with
3362 // linear values.  NormPriority get's mapped to 50% of the
3363 // Maximum priority an so on.  This will cause VM threads
3364 // to get unfair treatment against other Solaris processes
3365 // which do not explicitly alter their thread priorities.
3366 
3367 int os::java_to_os_priority[CriticalPriority + 1] = {
3368   -99999,         // 0 Entry should never be used
3369 
3370   0,              // 1 MinPriority
3371   32,             // 2
3372   64,             // 3
3373 
3374   96,             // 4
3375   127,            // 5 NormPriority
3376   127,            // 6
3377 
3378   127,            // 7
3379   127,            // 8
3380   127,            // 9 NearMaxPriority
3381 
3382   127,            // 10 MaxPriority
3383 
3384   -criticalPrio   // 11 CriticalPriority
3385 };
3386 
3387 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3388   OSThread* osthread = thread->osthread();
3389 
3390   // Save requested priority in case the thread hasn't been started
3391   osthread->set_native_priority(newpri);
3392 
3393   // Check for critical priority request
3394   bool fxcritical = false;
3395   if (newpri == -criticalPrio) {
3396     fxcritical = true;
3397     newpri = criticalPrio;
3398   }
3399 
3400   assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
3401   if (!UseThreadPriorities) return OS_OK;
3402 
3403   int status = 0;
3404 
3405   if (!fxcritical) {
3406     // Use thr_setprio only if we have a priority that thr_setprio understands
3407     status = thr_setprio(thread->osthread()->thread_id(), newpri);
3408   }
3409 
3410   int lwp_status =
3411           set_lwp_class_and_priority(osthread->thread_id(),
3412                                      osthread->lwp_id(),
3413                                      newpri,
3414                                      fxcritical ? fxLimits.schedPolicy : myClass,
3415                                      !fxcritical);
3416   if (lwp_status != 0 && fxcritical) {
3417     // Try again, this time without changing the scheduling class
3418     newpri = java_MaxPriority_to_os_priority;
3419     lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
3420                                             osthread->lwp_id(),
3421                                             newpri, myClass, false);
3422   }
3423   status |= lwp_status;
3424   return (status == 0) ? OS_OK : OS_ERR;
3425 }
3426 
3427 
3428 OSReturn os::get_native_priority(const Thread* const thread,
3429                                  int *priority_ptr) {
3430   int p;
3431   if (!UseThreadPriorities) {
3432     *priority_ptr = NormalPriority;
3433     return OS_OK;
3434   }
3435   int status = thr_getprio(thread->osthread()->thread_id(), &p);
3436   if (status != 0) {
3437     return OS_ERR;
3438   }
3439   *priority_ptr = p;
3440   return OS_OK;
3441 }
3442 
3443 
3444 // Hint to the underlying OS that a task switch would not be good.
3445 // Void return because it's a hint and can fail.
3446 void os::hint_no_preempt() {
3447   schedctl_start(schedctl_init());
3448 }
3449 
3450 ////////////////////////////////////////////////////////////////////////////////
3451 // suspend/resume support
3452 
3453 //  The low-level signal-based suspend/resume support is a remnant from the
3454 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3455 //  within hotspot. Currently used by JFR's OSThreadSampler
3456 //
3457 //  The remaining code is greatly simplified from the more general suspension
3458 //  code that used to be used.
3459 //
3460 //  The protocol is quite simple:
3461 //  - suspend:
3462 //      - sends a signal to the target thread
3463 //      - polls the suspend state of the osthread using a yield loop
3464 //      - target thread signal handler (SR_handler) sets suspend state
3465 //        and blocks in sigsuspend until continued
3466 //  - resume:
3467 //      - sets target osthread state to continue
3468 //      - sends signal to end the sigsuspend loop in the SR_handler
3469 //
3470 //  Note that the SR_lock plays no role in this suspend/resume protocol,
3471 //  but is checked for NULL in SR_handler as a thread termination indicator.
3472 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
3473 //
3474 //  Note that resume_clear_context() and suspend_save_context() are needed
3475 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
3476 //  which in part is used by:
3477 //    - Forte Analyzer: AsyncGetCallTrace()
3478 //    - StackBanging: get_frame_at_stack_banging_point()
3479 //    - JFR: get_topframe()-->....-->get_valid_uc_in_signal_handler()
3480 
3481 static void resume_clear_context(OSThread *osthread) {
3482   osthread->set_ucontext(NULL);
3483 }
3484 
3485 static void suspend_save_context(OSThread *osthread, ucontext_t* context) {
3486   osthread->set_ucontext(context);
3487 }
3488 
3489 static PosixSemaphore sr_semaphore;
3490 
3491 void os::Solaris::SR_handler(Thread* thread, ucontext_t* context) {
3492   // Save and restore errno to avoid confusing native code with EINTR
3493   // after sigsuspend.
3494   int old_errno = errno;
3495 
3496   OSThread* osthread = thread->osthread();
3497   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3498 
3499   os::SuspendResume::State current = osthread->sr.state();
3500   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3501     suspend_save_context(osthread, context);
3502 
3503     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3504     os::SuspendResume::State state = osthread->sr.suspended();
3505     if (state == os::SuspendResume::SR_SUSPENDED) {
3506       sigset_t suspend_set;  // signals for sigsuspend()
3507 
3508       // get current set of blocked signals and unblock resume signal
3509       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3510       sigdelset(&suspend_set, ASYNC_SIGNAL);
3511 
3512       sr_semaphore.signal();
3513       // wait here until we are resumed
3514       while (1) {
3515         sigsuspend(&suspend_set);
3516 
3517         os::SuspendResume::State result = osthread->sr.running();
3518         if (result == os::SuspendResume::SR_RUNNING) {
3519           sr_semaphore.signal();
3520           break;
3521         }
3522       }
3523 
3524     } else if (state == os::SuspendResume::SR_RUNNING) {
3525       // request was cancelled, continue
3526     } else {
3527       ShouldNotReachHere();
3528     }
3529 
3530     resume_clear_context(osthread);
3531   } else if (current == os::SuspendResume::SR_RUNNING) {
3532     // request was cancelled, continue
3533   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3534     // ignore
3535   } else {
3536     // ignore
3537   }
3538 
3539   errno = old_errno;
3540 }
3541 
3542 void os::print_statistics() {
3543 }
3544 
3545 bool os::message_box(const char* title, const char* message) {
3546   int i;
3547   fdStream err(defaultStream::error_fd());
3548   for (i = 0; i < 78; i++) err.print_raw("=");
3549   err.cr();
3550   err.print_raw_cr(title);
3551   for (i = 0; i < 78; i++) err.print_raw("-");
3552   err.cr();
3553   err.print_raw_cr(message);
3554   for (i = 0; i < 78; i++) err.print_raw("=");
3555   err.cr();
3556 
3557   char buf[16];
3558   // Prevent process from exiting upon "read error" without consuming all CPU
3559   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3560 
3561   return buf[0] == 'y' || buf[0] == 'Y';
3562 }
3563 
3564 static int sr_notify(OSThread* osthread) {
3565   int status = thr_kill(osthread->thread_id(), ASYNC_SIGNAL);
3566   assert_status(status == 0, status, "thr_kill");
3567   return status;
3568 }
3569 
3570 // "Randomly" selected value for how long we want to spin
3571 // before bailing out on suspending a thread, also how often
3572 // we send a signal to a thread we want to resume
3573 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3574 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3575 
3576 static bool do_suspend(OSThread* osthread) {
3577   assert(osthread->sr.is_running(), "thread should be running");
3578   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3579 
3580   // mark as suspended and send signal
3581   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3582     // failed to switch, state wasn't running?
3583     ShouldNotReachHere();
3584     return false;
3585   }
3586 
3587   if (sr_notify(osthread) != 0) {
3588     ShouldNotReachHere();
3589   }
3590 
3591   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3592   while (true) {
3593     if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2000 * NANOSECS_PER_MILLISEC))) {
3594       break;
3595     } else {
3596       // timeout
3597       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3598       if (cancelled == os::SuspendResume::SR_RUNNING) {
3599         return false;
3600       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3601         // make sure that we consume the signal on the semaphore as well
3602         sr_semaphore.wait();
3603         break;
3604       } else {
3605         ShouldNotReachHere();
3606         return false;
3607       }
3608     }
3609   }
3610 
3611   guarantee(osthread->sr.is_suspended(), "Must be suspended");
3612   return true;
3613 }
3614 
3615 static void do_resume(OSThread* osthread) {
3616   assert(osthread->sr.is_suspended(), "thread should be suspended");
3617   assert(!sr_semaphore.trywait(), "invalid semaphore state");
3618 
3619   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3620     // failed to switch to WAKEUP_REQUEST
3621     ShouldNotReachHere();
3622     return;
3623   }
3624 
3625   while (true) {
3626     if (sr_notify(osthread) == 0) {
3627       if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
3628         if (osthread->sr.is_running()) {
3629           return;
3630         }
3631       }
3632     } else {
3633       ShouldNotReachHere();
3634     }
3635   }
3636 
3637   guarantee(osthread->sr.is_running(), "Must be running!");
3638 }
3639 
3640 void os::SuspendedThreadTask::internal_do_task() {
3641   if (do_suspend(_thread->osthread())) {
3642     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
3643     do_task(context);
3644     do_resume(_thread->osthread());
3645   }
3646 }
3647 
3648 // This does not do anything on Solaris. This is basically a hook for being
3649 // able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
3650 void os::os_exception_wrapper(java_call_t f, JavaValue* value,
3651                               const methodHandle& method, JavaCallArguments* args,
3652                               Thread* thread) {
3653   f(value, method, args, thread);
3654 }
3655 
3656 // This routine may be used by user applications as a "hook" to catch signals.
3657 // The user-defined signal handler must pass unrecognized signals to this
3658 // routine, and if it returns true (non-zero), then the signal handler must
3659 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3660 // routine will never retun false (zero), but instead will execute a VM panic
3661 // routine kill the process.
3662 //
3663 // If this routine returns false, it is OK to call it again.  This allows
3664 // the user-defined signal handler to perform checks either before or after
3665 // the VM performs its own checks.  Naturally, the user code would be making
3666 // a serious error if it tried to handle an exception (such as a null check
3667 // or breakpoint) that the VM was generating for its own correct operation.
3668 //
3669 // This routine may recognize any of the following kinds of signals:
3670 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
3671 // ASYNC_SIGNAL.
3672 // It should be consulted by handlers for any of those signals.
3673 //
3674 // The caller of this routine must pass in the three arguments supplied
3675 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3676 // field of the structure passed to sigaction().  This routine assumes that
3677 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3678 //
3679 // Note that the VM will print warnings if it detects conflicting signal
3680 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3681 //
3682 extern "C" JNIEXPORT int JVM_handle_solaris_signal(int signo,
3683                                                    siginfo_t* siginfo,
3684                                                    void* ucontext,
3685                                                    int abort_if_unrecognized);
3686 
3687 
3688 void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
3689   int orig_errno = errno;  // Preserve errno value over signal handler.
3690   JVM_handle_solaris_signal(sig, info, ucVoid, true);
3691   errno = orig_errno;
3692 }
3693 
3694 // This boolean allows users to forward their own non-matching signals
3695 // to JVM_handle_solaris_signal, harmlessly.
3696 bool os::Solaris::signal_handlers_are_installed = false;
3697 
3698 // For signal-chaining
3699 bool os::Solaris::libjsig_is_loaded = false;
3700 typedef struct sigaction *(*get_signal_t)(int);
3701 get_signal_t os::Solaris::get_signal_action = NULL;
3702 
3703 struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
3704   struct sigaction *actp = NULL;
3705 
3706   if ((libjsig_is_loaded)  && (sig <= Maxsignum)) {
3707     // Retrieve the old signal handler from libjsig
3708     actp = (*get_signal_action)(sig);
3709   }
3710   if (actp == NULL) {
3711     // Retrieve the preinstalled signal handler from jvm
3712     actp = get_preinstalled_handler(sig);
3713   }
3714 
3715   return actp;
3716 }
3717 
3718 static bool call_chained_handler(struct sigaction *actp, int sig,
3719                                  siginfo_t *siginfo, void *context) {
3720   // Call the old signal handler
3721   if (actp->sa_handler == SIG_DFL) {
3722     // It's more reasonable to let jvm treat it as an unexpected exception
3723     // instead of taking the default action.
3724     return false;
3725   } else if (actp->sa_handler != SIG_IGN) {
3726     if ((actp->sa_flags & SA_NODEFER) == 0) {
3727       // automaticlly block the signal
3728       sigaddset(&(actp->sa_mask), sig);
3729     }
3730 
3731     sa_handler_t hand;
3732     sa_sigaction_t sa;
3733     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3734     // retrieve the chained handler
3735     if (siginfo_flag_set) {
3736       sa = actp->sa_sigaction;
3737     } else {
3738       hand = actp->sa_handler;
3739     }
3740 
3741     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3742       actp->sa_handler = SIG_DFL;
3743     }
3744 
3745     // try to honor the signal mask
3746     sigset_t oset;
3747     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3748 
3749     // call into the chained handler
3750     if (siginfo_flag_set) {
3751       (*sa)(sig, siginfo, context);
3752     } else {
3753       (*hand)(sig);
3754     }
3755 
3756     // restore the signal mask
3757     pthread_sigmask(SIG_SETMASK, &oset, 0);
3758   }
3759   // Tell jvm's signal handler the signal is taken care of.
3760   return true;
3761 }
3762 
3763 bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3764   bool chained = false;
3765   // signal-chaining
3766   if (UseSignalChaining) {
3767     struct sigaction *actp = get_chained_signal_action(sig);
3768     if (actp != NULL) {
3769       chained = call_chained_handler(actp, sig, siginfo, context);
3770     }
3771   }
3772   return chained;
3773 }
3774 
3775 struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
3776   assert((chainedsigactions != (struct sigaction *)NULL) &&
3777          (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3778   if (preinstalled_sigs[sig] != 0) {
3779     return &chainedsigactions[sig];
3780   }
3781   return NULL;
3782 }
3783 
3784 void os::Solaris::save_preinstalled_handler(int sig,
3785                                             struct sigaction& oldAct) {
3786   assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
3787   assert((chainedsigactions != (struct sigaction *)NULL) &&
3788          (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3789   chainedsigactions[sig] = oldAct;
3790   preinstalled_sigs[sig] = 1;
3791 }
3792 
3793 void os::Solaris::set_signal_handler(int sig, bool set_installed,
3794                                      bool oktochain) {
3795   // Check for overwrite.
3796   struct sigaction oldAct;
3797   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3798   void* oldhand =
3799       oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3800                           : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3801   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3802       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3803       oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
3804     if (AllowUserSignalHandlers || !set_installed) {
3805       // Do not overwrite; user takes responsibility to forward to us.
3806       return;
3807     } else if (UseSignalChaining) {
3808       if (oktochain) {
3809         // save the old handler in jvm
3810         save_preinstalled_handler(sig, oldAct);
3811       } else {
3812         vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal.");
3813       }
3814       // libjsig also interposes the sigaction() call below and saves the
3815       // old sigaction on it own.
3816     } else {
3817       fatal("Encountered unexpected pre-existing sigaction handler "
3818             "%#lx for signal %d.", (long)oldhand, sig);
3819     }
3820   }
3821 
3822   struct sigaction sigAct;
3823   sigfillset(&(sigAct.sa_mask));
3824   sigAct.sa_handler = SIG_DFL;
3825 
3826   sigAct.sa_sigaction = signalHandler;
3827   // Handle SIGSEGV on alternate signal stack if
3828   // not using stack banging
3829   if (!UseStackBanging && sig == SIGSEGV) {
3830     sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
3831   } else {
3832     sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
3833   }
3834   os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
3835 
3836   sigaction(sig, &sigAct, &oldAct);
3837 
3838   void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3839                                        : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3840   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3841 }
3842 
3843 
3844 #define DO_SIGNAL_CHECK(sig)                      \
3845   do {                                            \
3846     if (!sigismember(&check_signal_done, sig)) {  \
3847       os::Solaris::check_signal_handler(sig);     \
3848     }                                             \
3849   } while (0)
3850 
3851 // This method is a periodic task to check for misbehaving JNI applications
3852 // under CheckJNI, we can add any periodic checks here
3853 
3854 void os::run_periodic_checks() {
3855   // A big source of grief is hijacking virt. addr 0x0 on Solaris,
3856   // thereby preventing a NULL checks.
3857   if (!check_addr0_done) check_addr0_done = check_addr0(tty);
3858 
3859   if (check_signals == false) return;
3860 
3861   // SEGV and BUS if overridden could potentially prevent
3862   // generation of hs*.log in the event of a crash, debugging
3863   // such a case can be very challenging, so we absolutely
3864   // check for the following for a good measure:
3865   DO_SIGNAL_CHECK(SIGSEGV);
3866   DO_SIGNAL_CHECK(SIGILL);
3867   DO_SIGNAL_CHECK(SIGFPE);
3868   DO_SIGNAL_CHECK(SIGBUS);
3869   DO_SIGNAL_CHECK(SIGPIPE);
3870   DO_SIGNAL_CHECK(SIGXFSZ);
3871   DO_SIGNAL_CHECK(ASYNC_SIGNAL);
3872 
3873   // ReduceSignalUsage allows the user to override these handlers
3874   // see comments at the very top and jvm_solaris.h
3875   if (!ReduceSignalUsage) {
3876     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3877     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3878     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3879     DO_SIGNAL_CHECK(BREAK_SIGNAL);
3880   }
3881 }
3882 
3883 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3884 
3885 static os_sigaction_t os_sigaction = NULL;
3886 
3887 void os::Solaris::check_signal_handler(int sig) {
3888   char buf[O_BUFLEN];
3889   address jvmHandler = NULL;
3890 
3891   struct sigaction act;
3892   if (os_sigaction == NULL) {
3893     // only trust the default sigaction, in case it has been interposed
3894     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3895     if (os_sigaction == NULL) return;
3896   }
3897 
3898   os_sigaction(sig, (struct sigaction*)NULL, &act);
3899 
3900   address thisHandler = (act.sa_flags & SA_SIGINFO)
3901     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3902     : CAST_FROM_FN_PTR(address, act.sa_handler);
3903 
3904 
3905   switch (sig) {
3906   case SIGSEGV:
3907   case SIGBUS:
3908   case SIGFPE:
3909   case SIGPIPE:
3910   case SIGXFSZ:
3911   case SIGILL:
3912   case ASYNC_SIGNAL:
3913     jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
3914     break;
3915 
3916   case SHUTDOWN1_SIGNAL:
3917   case SHUTDOWN2_SIGNAL:
3918   case SHUTDOWN3_SIGNAL:
3919   case BREAK_SIGNAL:
3920     jvmHandler = (address)user_handler();
3921     break;
3922 
3923   default:
3924       return;
3925   }
3926 
3927   if (thisHandler != jvmHandler) {
3928     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3929     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3930     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3931     // No need to check this sig any longer
3932     sigaddset(&check_signal_done, sig);
3933     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
3934     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
3935       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
3936                     exception_name(sig, buf, O_BUFLEN));
3937     }
3938   } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
3939     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3940     tty->print("expected:");
3941     os::Posix::print_sa_flags(tty, os::Solaris::get_our_sigflags(sig));
3942     tty->cr();
3943     tty->print("  found:");
3944     os::Posix::print_sa_flags(tty, act.sa_flags);
3945     tty->cr();
3946     // No need to check this sig any longer
3947     sigaddset(&check_signal_done, sig);
3948   }
3949 
3950   // Print all the signal handler state
3951   if (sigismember(&check_signal_done, sig)) {
3952     print_signal_handlers(tty, buf, O_BUFLEN);
3953   }
3954 
3955 }
3956 
3957 void os::Solaris::install_signal_handlers() {
3958   signal_handlers_are_installed = true;
3959 
3960   // signal-chaining
3961   typedef void (*signal_setting_t)();
3962   signal_setting_t begin_signal_setting = NULL;
3963   signal_setting_t end_signal_setting = NULL;
3964   begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3965                                         dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3966   if (begin_signal_setting != NULL) {
3967     end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3968                                         dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3969     get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3970                                        dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3971     get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
3972                                          dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
3973     libjsig_is_loaded = true;
3974     if (os::Solaris::get_libjsig_version != NULL) {
3975       int libjsigversion =  (*os::Solaris::get_libjsig_version)();
3976       assert(libjsigversion == JSIG_VERSION_1_4_1, "libjsig version mismatch");
3977     }
3978     assert(UseSignalChaining, "should enable signal-chaining");
3979   }
3980   if (libjsig_is_loaded) {
3981     // Tell libjsig jvm is setting signal handlers
3982     (*begin_signal_setting)();
3983   }
3984 
3985   set_signal_handler(SIGSEGV, true, true);
3986   set_signal_handler(SIGPIPE, true, true);
3987   set_signal_handler(SIGXFSZ, true, true);
3988   set_signal_handler(SIGBUS, true, true);
3989   set_signal_handler(SIGILL, true, true);
3990   set_signal_handler(SIGFPE, true, true);
3991   set_signal_handler(ASYNC_SIGNAL, true, true);
3992 
3993   if (libjsig_is_loaded) {
3994     // Tell libjsig jvm finishes setting signal handlers
3995     (*end_signal_setting)();
3996   }
3997 
3998   // We don't activate signal checker if libjsig is in place, we trust ourselves
3999   // and if UserSignalHandler is installed all bets are off.
4000   // Log that signal checking is off only if -verbose:jni is specified.
4001   if (CheckJNICalls) {
4002     if (libjsig_is_loaded) {
4003       if (PrintJNIResolving) {
4004         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4005       }
4006       check_signals = false;
4007     }
4008     if (AllowUserSignalHandlers) {
4009       if (PrintJNIResolving) {
4010         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4011       }
4012       check_signals = false;
4013     }
4014   }
4015 }
4016 
4017 
4018 void report_error(const char* file_name, int line_no, const char* title,
4019                   const char* format, ...);
4020 
4021 // (Static) wrappers for the liblgrp API
4022 os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4023 os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4024 os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4025 os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4026 os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4027 os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4028 os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4029 os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4030 os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4031 
4032 static address resolve_symbol_lazy(const char* name) {
4033   address addr = (address) dlsym(RTLD_DEFAULT, name);
4034   if (addr == NULL) {
4035     // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4036     addr = (address) dlsym(RTLD_NEXT, name);
4037   }
4038   return addr;
4039 }
4040 
4041 static address resolve_symbol(const char* name) {
4042   address addr = resolve_symbol_lazy(name);
4043   if (addr == NULL) {
4044     fatal(dlerror());
4045   }
4046   return addr;
4047 }
4048 
4049 void os::Solaris::libthread_init() {
4050   address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4051 
4052   lwp_priocntl_init();
4053 
4054   // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4055   if (func == NULL) {
4056     func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4057     // Guarantee that this VM is running on an new enough OS (5.6 or
4058     // later) that it will have a new enough libthread.so.
4059     guarantee(func != NULL, "libthread.so is too old.");
4060   }
4061 
4062   int size;
4063   void (*handler_info_func)(address *, int *);
4064   handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4065   handler_info_func(&handler_start, &size);
4066   handler_end = handler_start + size;
4067 }
4068 
4069 
4070 int_fnP_mutex_tP os::Solaris::_mutex_lock;
4071 int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4072 int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4073 int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4074 int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4075 int os::Solaris::_mutex_scope = USYNC_THREAD;
4076 
4077 int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4078 int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4079 int_fnP_cond_tP os::Solaris::_cond_signal;
4080 int_fnP_cond_tP os::Solaris::_cond_broadcast;
4081 int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4082 int_fnP_cond_tP os::Solaris::_cond_destroy;
4083 int os::Solaris::_cond_scope = USYNC_THREAD;
4084 bool os::Solaris::_synchronization_initialized;
4085 
4086 void os::Solaris::synchronization_init() {
4087   if (UseLWPSynchronization) {
4088     os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4089     os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4090     os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4091     os::Solaris::set_mutex_init(lwp_mutex_init);
4092     os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4093     os::Solaris::set_mutex_scope(USYNC_THREAD);
4094 
4095     os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4096     os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4097     os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4098     os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4099     os::Solaris::set_cond_init(lwp_cond_init);
4100     os::Solaris::set_cond_destroy(lwp_cond_destroy);
4101     os::Solaris::set_cond_scope(USYNC_THREAD);
4102   } else {
4103     os::Solaris::set_mutex_scope(USYNC_THREAD);
4104     os::Solaris::set_cond_scope(USYNC_THREAD);
4105 
4106     if (UsePthreads) {
4107       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4108       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4109       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4110       os::Solaris::set_mutex_init(pthread_mutex_default_init);
4111       os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4112 
4113       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4114       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4115       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4116       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4117       os::Solaris::set_cond_init(pthread_cond_default_init);
4118       os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4119     } else {
4120       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4121       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4122       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4123       os::Solaris::set_mutex_init(::mutex_init);
4124       os::Solaris::set_mutex_destroy(::mutex_destroy);
4125 
4126       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4127       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4128       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4129       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4130       os::Solaris::set_cond_init(::cond_init);
4131       os::Solaris::set_cond_destroy(::cond_destroy);
4132     }
4133   }
4134   _synchronization_initialized = true;
4135 }
4136 
4137 bool os::Solaris::liblgrp_init() {
4138   void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4139   if (handle != NULL) {
4140     os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4141     os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4142     os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4143     os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4144     os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4145     os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4146     os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4147     os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4148                                                       dlsym(handle, "lgrp_cookie_stale")));
4149 
4150     lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4151     set_lgrp_cookie(c);
4152     return true;
4153   }
4154   return false;
4155 }
4156 
4157 // int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4158 typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4159 static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4160 
4161 void init_pset_getloadavg_ptr(void) {
4162   pset_getloadavg_ptr =
4163     (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
4164   if (pset_getloadavg_ptr == NULL) {
4165     log_warning(os)("pset_getloadavg function not found");
4166   }
4167 }
4168 
4169 int os::Solaris::_dev_zero_fd = -1;
4170 
4171 // this is called _before_ the global arguments have been parsed
4172 void os::init(void) {
4173   _initial_pid = getpid();
4174 
4175   max_hrtime = first_hrtime = gethrtime();
4176 
4177   init_random(1234567);
4178 
4179   page_size = sysconf(_SC_PAGESIZE);
4180   if (page_size == -1) {
4181     fatal("os_solaris.cpp: os::init: sysconf failed (%s)", os::strerror(errno));
4182   }
4183   init_page_sizes((size_t) page_size);
4184 
4185   Solaris::initialize_system_info();
4186 
4187   int fd = ::open("/dev/zero", O_RDWR);
4188   if (fd < 0) {
4189     fatal("os::init: cannot open /dev/zero (%s)", os::strerror(errno));
4190   } else {
4191     Solaris::set_dev_zero_fd(fd);
4192 
4193     // Close on exec, child won't inherit.
4194     fcntl(fd, F_SETFD, FD_CLOEXEC);
4195   }
4196 
4197   clock_tics_per_sec = CLK_TCK;
4198 
4199   // check if dladdr1() exists; dladdr1 can provide more information than
4200   // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
4201   // and is available on linker patches for 5.7 and 5.8.
4202   // libdl.so must have been loaded, this call is just an entry lookup
4203   void * hdl = dlopen("libdl.so", RTLD_NOW);
4204   if (hdl) {
4205     dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
4206   }
4207 
4208   // main_thread points to the thread that created/loaded the JVM.
4209   main_thread = thr_self();
4210 
4211   // dynamic lookup of functions that may not be available in our lowest
4212   // supported Solaris release
4213   void * handle = dlopen("libc.so.1", RTLD_LAZY);
4214   if (handle != NULL) {
4215     Solaris::_pthread_setname_np =  // from 11.3
4216         (Solaris::pthread_setname_np_func_t)dlsym(handle, "pthread_setname_np");
4217   }
4218 }
4219 
4220 // To install functions for atexit system call
4221 extern "C" {
4222   static void perfMemory_exit_helper() {
4223     perfMemory_exit();
4224   }
4225 }
4226 
4227 // this is called _after_ the global arguments have been parsed
4228 jint os::init_2(void) {
4229   // try to enable extended file IO ASAP, see 6431278
4230   os::Solaris::try_enable_extended_io();
4231 
4232   // Check and sets minimum stack sizes against command line options
4233   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4234     return JNI_ERR;
4235   }
4236 
4237   Solaris::libthread_init();
4238 
4239   if (UseNUMA) {
4240     if (!Solaris::liblgrp_init()) {
4241       UseNUMA = false;
4242     } else {
4243       size_t lgrp_limit = os::numa_get_groups_num();
4244       int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit, mtInternal);
4245       size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
4246       FREE_C_HEAP_ARRAY(int, lgrp_ids);
4247       if (lgrp_num < 2) {
4248         // There's only one locality group, disable NUMA.
4249         UseNUMA = false;
4250       }
4251     }
4252     if (!UseNUMA && ForceNUMA) {
4253       UseNUMA = true;
4254     }
4255   }
4256 
4257   Solaris::signal_sets_init();
4258   Solaris::init_signal_mem();
4259   Solaris::install_signal_handlers();
4260 
4261   // initialize synchronization primitives to use either thread or
4262   // lwp synchronization (controlled by UseLWPSynchronization)
4263   Solaris::synchronization_init();
4264 
4265   if (MaxFDLimit) {
4266     // set the number of file descriptors to max. print out error
4267     // if getrlimit/setrlimit fails but continue regardless.
4268     struct rlimit nbr_files;
4269     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4270     if (status != 0) {
4271       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4272     } else {
4273       nbr_files.rlim_cur = nbr_files.rlim_max;
4274       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4275       if (status != 0) {
4276         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4277       }
4278     }
4279   }
4280 
4281   // Calculate theoretical max. size of Threads to guard gainst
4282   // artifical out-of-memory situations, where all available address-
4283   // space has been reserved by thread stacks. Default stack size is 1Mb.
4284   size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
4285     JavaThread::stack_size_at_create() : (1*K*K);
4286   assert(pre_thread_stack_size != 0, "Must have a stack");
4287   // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
4288   // we should start doing Virtual Memory banging. Currently when the threads will
4289   // have used all but 200Mb of space.
4290   size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
4291   Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
4292 
4293   // at-exit methods are called in the reverse order of their registration.
4294   // In Solaris 7 and earlier, atexit functions are called on return from
4295   // main or as a result of a call to exit(3C). There can be only 32 of
4296   // these functions registered and atexit() does not set errno. In Solaris
4297   // 8 and later, there is no limit to the number of functions registered
4298   // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
4299   // functions are called upon dlclose(3DL) in addition to return from main
4300   // and exit(3C).
4301 
4302   if (PerfAllowAtExitRegistration) {
4303     // only register atexit functions if PerfAllowAtExitRegistration is set.
4304     // atexit functions can be delayed until process exit time, which
4305     // can be problematic for embedded VM situations. Embedded VMs should
4306     // call DestroyJavaVM() to assure that VM resources are released.
4307 
4308     // note: perfMemory_exit_helper atexit function may be removed in
4309     // the future if the appropriate cleanup code can be added to the
4310     // VM_Exit VMOperation's doit method.
4311     if (atexit(perfMemory_exit_helper) != 0) {
4312       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4313     }
4314   }
4315 
4316   // Init pset_loadavg function pointer
4317   init_pset_getloadavg_ptr();
4318 
4319   return JNI_OK;
4320 }
4321 
4322 // Mark the polling page as unreadable
4323 void os::make_polling_page_unreadable(void) {
4324   if (mprotect((char *)_polling_page, page_size, PROT_NONE) != 0) {
4325     fatal("Could not disable polling page");
4326   }
4327 }
4328 
4329 // Mark the polling page as readable
4330 void os::make_polling_page_readable(void) {
4331   if (mprotect((char *)_polling_page, page_size, PROT_READ) != 0) {
4332     fatal("Could not enable polling page");
4333   }
4334 }
4335 
4336 // Is a (classpath) directory empty?
4337 bool os::dir_is_empty(const char* path) {
4338   DIR *dir = NULL;
4339   struct dirent *ptr;
4340 
4341   dir = opendir(path);
4342   if (dir == NULL) return true;
4343 
4344   // Scan the directory
4345   bool result = true;
4346   char buf[sizeof(struct dirent) + MAX_PATH];
4347   struct dirent *dbuf = (struct dirent *) buf;
4348   while (result && (ptr = readdir(dir, dbuf)) != NULL) {
4349     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4350       result = false;
4351     }
4352   }
4353   closedir(dir);
4354   return result;
4355 }
4356 
4357 // This code originates from JDK's sysOpen and open64_w
4358 // from src/solaris/hpi/src/system_md.c
4359 
4360 int os::open(const char *path, int oflag, int mode) {
4361   if (strlen(path) > MAX_PATH - 1) {
4362     errno = ENAMETOOLONG;
4363     return -1;
4364   }
4365   int fd;
4366 
4367   fd = ::open64(path, oflag, mode);
4368   if (fd == -1) return -1;
4369 
4370   // If the open succeeded, the file might still be a directory
4371   {
4372     struct stat64 buf64;
4373     int ret = ::fstat64(fd, &buf64);
4374     int st_mode = buf64.st_mode;
4375 
4376     if (ret != -1) {
4377       if ((st_mode & S_IFMT) == S_IFDIR) {
4378         errno = EISDIR;
4379         ::close(fd);
4380         return -1;
4381       }
4382     } else {
4383       ::close(fd);
4384       return -1;
4385     }
4386   }
4387 
4388   // 32-bit Solaris systems suffer from:
4389   //
4390   // - an historical default soft limit of 256 per-process file
4391   //   descriptors that is too low for many Java programs.
4392   //
4393   // - a design flaw where file descriptors created using stdio
4394   //   fopen must be less than 256, _even_ when the first limit above
4395   //   has been raised.  This can cause calls to fopen (but not calls to
4396   //   open, for example) to fail mysteriously, perhaps in 3rd party
4397   //   native code (although the JDK itself uses fopen).  One can hardly
4398   //   criticize them for using this most standard of all functions.
4399   //
4400   // We attempt to make everything work anyways by:
4401   //
4402   // - raising the soft limit on per-process file descriptors beyond
4403   //   256
4404   //
4405   // - As of Solaris 10u4, we can request that Solaris raise the 256
4406   //   stdio fopen limit by calling function enable_extended_FILE_stdio.
4407   //   This is done in init_2 and recorded in enabled_extended_FILE_stdio
4408   //
4409   // - If we are stuck on an old (pre 10u4) Solaris system, we can
4410   //   workaround the bug by remapping non-stdio file descriptors below
4411   //   256 to ones beyond 256, which is done below.
4412   //
4413   // See:
4414   // 1085341: 32-bit stdio routines should support file descriptors >255
4415   // 6533291: Work around 32-bit Solaris stdio limit of 256 open files
4416   // 6431278: Netbeans crash on 32 bit Solaris: need to call
4417   //          enable_extended_FILE_stdio() in VM initialisation
4418   // Giri Mandalika's blog
4419   // http://technopark02.blogspot.com/2005_05_01_archive.html
4420   //
4421 #ifndef  _LP64
4422   if ((!enabled_extended_FILE_stdio) && fd < 256) {
4423     int newfd = ::fcntl(fd, F_DUPFD, 256);
4424     if (newfd != -1) {
4425       ::close(fd);
4426       fd = newfd;
4427     }
4428   }
4429 #endif // 32-bit Solaris
4430 
4431   // All file descriptors that are opened in the JVM and not
4432   // specifically destined for a subprocess should have the
4433   // close-on-exec flag set.  If we don't set it, then careless 3rd
4434   // party native code might fork and exec without closing all
4435   // appropriate file descriptors (e.g. as we do in closeDescriptors in
4436   // UNIXProcess.c), and this in turn might:
4437   //
4438   // - cause end-of-file to fail to be detected on some file
4439   //   descriptors, resulting in mysterious hangs, or
4440   //
4441   // - might cause an fopen in the subprocess to fail on a system
4442   //   suffering from bug 1085341.
4443   //
4444   // (Yes, the default setting of the close-on-exec flag is a Unix
4445   // design flaw)
4446   //
4447   // See:
4448   // 1085341: 32-bit stdio routines should support file descriptors >255
4449   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4450   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4451   //
4452 #ifdef FD_CLOEXEC
4453   {
4454     int flags = ::fcntl(fd, F_GETFD);
4455     if (flags != -1) {
4456       ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4457     }
4458   }
4459 #endif
4460 
4461   return fd;
4462 }
4463 
4464 // create binary file, rewriting existing file if required
4465 int os::create_binary_file(const char* path, bool rewrite_existing) {
4466   int oflags = O_WRONLY | O_CREAT;
4467   if (!rewrite_existing) {
4468     oflags |= O_EXCL;
4469   }
4470   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4471 }
4472 
4473 // return current position of file pointer
4474 jlong os::current_file_offset(int fd) {
4475   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4476 }
4477 
4478 // move file pointer to the specified offset
4479 jlong os::seek_to_file_offset(int fd, jlong offset) {
4480   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4481 }
4482 
4483 jlong os::lseek(int fd, jlong offset, int whence) {
4484   return (jlong) ::lseek64(fd, offset, whence);
4485 }
4486 
4487 char * os::native_path(char *path) {
4488   return path;
4489 }
4490 
4491 int os::ftruncate(int fd, jlong length) {
4492   return ::ftruncate64(fd, length);
4493 }
4494 
4495 int os::fsync(int fd)  {
4496   RESTARTABLE_RETURN_INT(::fsync(fd));
4497 }
4498 
4499 int os::available(int fd, jlong *bytes) {
4500   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
4501          "Assumed _thread_in_native");
4502   jlong cur, end;
4503   int mode;
4504   struct stat64 buf64;
4505 
4506   if (::fstat64(fd, &buf64) >= 0) {
4507     mode = buf64.st_mode;
4508     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4509       int n,ioctl_return;
4510 
4511       RESTARTABLE(::ioctl(fd, FIONREAD, &n), ioctl_return);
4512       if (ioctl_return>= 0) {
4513         *bytes = n;
4514         return 1;
4515       }
4516     }
4517   }
4518   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4519     return 0;
4520   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4521     return 0;
4522   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4523     return 0;
4524   }
4525   *bytes = end - cur;
4526   return 1;
4527 }
4528 
4529 // Map a block of memory.
4530 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4531                         char *addr, size_t bytes, bool read_only,
4532                         bool allow_exec) {
4533   int prot;
4534   int flags;
4535 
4536   if (read_only) {
4537     prot = PROT_READ;
4538     flags = MAP_SHARED;
4539   } else {
4540     prot = PROT_READ | PROT_WRITE;
4541     flags = MAP_PRIVATE;
4542   }
4543 
4544   if (allow_exec) {
4545     prot |= PROT_EXEC;
4546   }
4547 
4548   if (addr != NULL) {
4549     flags |= MAP_FIXED;
4550   }
4551 
4552   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4553                                      fd, file_offset);
4554   if (mapped_address == MAP_FAILED) {
4555     return NULL;
4556   }
4557   return mapped_address;
4558 }
4559 
4560 
4561 // Remap a block of memory.
4562 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4563                           char *addr, size_t bytes, bool read_only,
4564                           bool allow_exec) {
4565   // same as map_memory() on this OS
4566   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4567                         allow_exec);
4568 }
4569 
4570 
4571 // Unmap a block of memory.
4572 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4573   return munmap(addr, bytes) == 0;
4574 }
4575 
4576 void os::pause() {
4577   char filename[MAX_PATH];
4578   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4579     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4580   } else {
4581     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4582   }
4583 
4584   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4585   if (fd != -1) {
4586     struct stat buf;
4587     ::close(fd);
4588     while (::stat(filename, &buf) == 0) {
4589       (void)::poll(NULL, 0, 100);
4590     }
4591   } else {
4592     jio_fprintf(stderr,
4593                 "Could not open pause file '%s', continuing immediately.\n", filename);
4594   }
4595 }
4596 
4597 #ifndef PRODUCT
4598 #ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4599 // Turn this on if you need to trace synch operations.
4600 // Set RECORD_SYNCH_LIMIT to a large-enough value,
4601 // and call record_synch_enable and record_synch_disable
4602 // around the computation of interest.
4603 
4604 void record_synch(char* name, bool returning);  // defined below
4605 
4606 class RecordSynch {
4607   char* _name;
4608  public:
4609   RecordSynch(char* name) :_name(name) { record_synch(_name, false); }
4610   ~RecordSynch()                       { record_synch(_name, true); }
4611 };
4612 
4613 #define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
4614 extern "C" ret name params {                                    \
4615   typedef ret name##_t params;                                  \
4616   static name##_t* implem = NULL;                               \
4617   static int callcount = 0;                                     \
4618   if (implem == NULL) {                                         \
4619     implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
4620     if (implem == NULL)  fatal(dlerror());                      \
4621   }                                                             \
4622   ++callcount;                                                  \
4623   RecordSynch _rs(#name);                                       \
4624   inner;                                                        \
4625   return implem args;                                           \
4626 }
4627 // in dbx, examine callcounts this way:
4628 // for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
4629 
4630 #define CHECK_POINTER_OK(p) \
4631   (!Universe::is_fully_initialized() || !Universe::is_reserved_heap((oop)(p)))
4632 #define CHECK_MU \
4633   if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
4634 #define CHECK_CV \
4635   if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
4636 #define CHECK_P(p) \
4637   if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
4638 
4639 #define CHECK_MUTEX(mutex_op) \
4640   CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
4641 
4642 CHECK_MUTEX(   mutex_lock)
4643 CHECK_MUTEX(  _mutex_lock)
4644 CHECK_MUTEX( mutex_unlock)
4645 CHECK_MUTEX(_mutex_unlock)
4646 CHECK_MUTEX( mutex_trylock)
4647 CHECK_MUTEX(_mutex_trylock)
4648 
4649 #define CHECK_COND(cond_op) \
4650   CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU; CHECK_CV);
4651 
4652 CHECK_COND( cond_wait);
4653 CHECK_COND(_cond_wait);
4654 CHECK_COND(_cond_wait_cancel);
4655 
4656 #define CHECK_COND2(cond_op) \
4657   CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU; CHECK_CV);
4658 
4659 CHECK_COND2( cond_timedwait);
4660 CHECK_COND2(_cond_timedwait);
4661 CHECK_COND2(_cond_timedwait_cancel);
4662 
4663 // do the _lwp_* versions too
4664 #define mutex_t lwp_mutex_t
4665 #define cond_t  lwp_cond_t
4666 CHECK_MUTEX(  _lwp_mutex_lock)
4667 CHECK_MUTEX(  _lwp_mutex_unlock)
4668 CHECK_MUTEX(  _lwp_mutex_trylock)
4669 CHECK_MUTEX( __lwp_mutex_lock)
4670 CHECK_MUTEX( __lwp_mutex_unlock)
4671 CHECK_MUTEX( __lwp_mutex_trylock)
4672 CHECK_MUTEX(___lwp_mutex_lock)
4673 CHECK_MUTEX(___lwp_mutex_unlock)
4674 
4675 CHECK_COND(  _lwp_cond_wait);
4676 CHECK_COND( __lwp_cond_wait);
4677 CHECK_COND(___lwp_cond_wait);
4678 
4679 CHECK_COND2(  _lwp_cond_timedwait);
4680 CHECK_COND2( __lwp_cond_timedwait);
4681 #undef mutex_t
4682 #undef cond_t
4683 
4684 CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4685 CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4686 CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4687 CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4688 CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4689 CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4690 CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4691 CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4692 
4693 
4694 // recording machinery:
4695 
4696 enum { RECORD_SYNCH_LIMIT = 200 };
4697 char* record_synch_name[RECORD_SYNCH_LIMIT];
4698 void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
4699 bool record_synch_returning[RECORD_SYNCH_LIMIT];
4700 thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
4701 int record_synch_count = 0;
4702 bool record_synch_enabled = false;
4703 
4704 // in dbx, examine recorded data this way:
4705 // for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
4706 
4707 void record_synch(char* name, bool returning) {
4708   if (record_synch_enabled) {
4709     if (record_synch_count < RECORD_SYNCH_LIMIT) {
4710       record_synch_name[record_synch_count] = name;
4711       record_synch_returning[record_synch_count] = returning;
4712       record_synch_thread[record_synch_count] = thr_self();
4713       record_synch_arg0ptr[record_synch_count] = &name;
4714       record_synch_count++;
4715     }
4716     // put more checking code here:
4717     // ...
4718   }
4719 }
4720 
4721 void record_synch_enable() {
4722   // start collecting trace data, if not already doing so
4723   if (!record_synch_enabled)  record_synch_count = 0;
4724   record_synch_enabled = true;
4725 }
4726 
4727 void record_synch_disable() {
4728   // stop collecting trace data
4729   record_synch_enabled = false;
4730 }
4731 
4732 #endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4733 #endif // PRODUCT
4734 
4735 const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
4736 const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
4737                                (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
4738 
4739 
4740 // JVMTI & JVM monitoring and management support
4741 // The thread_cpu_time() and current_thread_cpu_time() are only
4742 // supported if is_thread_cpu_time_supported() returns true.
4743 // They are not supported on Solaris T1.
4744 
4745 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4746 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4747 // of a thread.
4748 //
4749 // current_thread_cpu_time() and thread_cpu_time(Thread *)
4750 // returns the fast estimate available on the platform.
4751 
4752 // hrtime_t gethrvtime() return value includes
4753 // user time but does not include system time
4754 jlong os::current_thread_cpu_time() {
4755   return (jlong) gethrvtime();
4756 }
4757 
4758 jlong os::thread_cpu_time(Thread *thread) {
4759   // return user level CPU time only to be consistent with
4760   // what current_thread_cpu_time returns.
4761   // thread_cpu_time_info() must be changed if this changes
4762   return os::thread_cpu_time(thread, false /* user time only */);
4763 }
4764 
4765 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4766   if (user_sys_cpu_time) {
4767     return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
4768   } else {
4769     return os::current_thread_cpu_time();
4770   }
4771 }
4772 
4773 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4774   char proc_name[64];
4775   int count;
4776   prusage_t prusage;
4777   jlong lwp_time;
4778   int fd;
4779 
4780   sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
4781           getpid(),
4782           thread->osthread()->lwp_id());
4783   fd = ::open(proc_name, O_RDONLY);
4784   if (fd == -1) return -1;
4785 
4786   do {
4787     count = ::pread(fd,
4788                     (void *)&prusage.pr_utime,
4789                     thr_time_size,
4790                     thr_time_off);
4791   } while (count < 0 && errno == EINTR);
4792   ::close(fd);
4793   if (count < 0) return -1;
4794 
4795   if (user_sys_cpu_time) {
4796     // user + system CPU time
4797     lwp_time = (((jlong)prusage.pr_stime.tv_sec +
4798                  (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
4799                  (jlong)prusage.pr_stime.tv_nsec +
4800                  (jlong)prusage.pr_utime.tv_nsec;
4801   } else {
4802     // user level CPU time only
4803     lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
4804                 (jlong)prusage.pr_utime.tv_nsec;
4805   }
4806 
4807   return (lwp_time);
4808 }
4809 
4810 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4811   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
4812   info_ptr->may_skip_backward = false;    // elapsed time not wall time
4813   info_ptr->may_skip_forward = false;     // elapsed time not wall time
4814   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
4815 }
4816 
4817 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4818   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
4819   info_ptr->may_skip_backward = false;    // elapsed time not wall time
4820   info_ptr->may_skip_forward = false;     // elapsed time not wall time
4821   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
4822 }
4823 
4824 bool os::is_thread_cpu_time_supported() {
4825   return true;
4826 }
4827 
4828 // System loadavg support.  Returns -1 if load average cannot be obtained.
4829 // Return the load average for our processor set if the primitive exists
4830 // (Solaris 9 and later).  Otherwise just return system wide loadavg.
4831 int os::loadavg(double loadavg[], int nelem) {
4832   if (pset_getloadavg_ptr != NULL) {
4833     return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
4834   } else {
4835     return ::getloadavg(loadavg, nelem);
4836   }
4837 }
4838 
4839 //---------------------------------------------------------------------------------
4840 
4841 bool os::find(address addr, outputStream* st) {
4842   Dl_info dlinfo;
4843   memset(&dlinfo, 0, sizeof(dlinfo));
4844   if (dladdr(addr, &dlinfo) != 0) {
4845     st->print(PTR_FORMAT ": ", addr);
4846     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4847       st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
4848     } else if (dlinfo.dli_fbase != NULL) {
4849       st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
4850     } else {
4851       st->print("<absolute address>");
4852     }
4853     if (dlinfo.dli_fname != NULL) {
4854       st->print(" in %s", dlinfo.dli_fname);
4855     }
4856     if (dlinfo.dli_fbase != NULL) {
4857       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4858     }
4859     st->cr();
4860 
4861     if (Verbose) {
4862       // decode some bytes around the PC
4863       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4864       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4865       address       lowest = (address) dlinfo.dli_sname;
4866       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4867       if (begin < lowest)  begin = lowest;
4868       Dl_info dlinfo2;
4869       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4870           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4871         end = (address) dlinfo2.dli_saddr;
4872       }
4873       Disassembler::decode(begin, end, st);
4874     }
4875     return true;
4876   }
4877   return false;
4878 }
4879 
4880 // Following function has been added to support HotSparc's libjvm.so running
4881 // under Solaris production JDK 1.2.2 / 1.3.0.  These came from
4882 // src/solaris/hpi/native_threads in the EVM codebase.
4883 //
4884 // NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
4885 // libraries and should thus be removed. We will leave it behind for a while
4886 // until we no longer want to able to run on top of 1.3.0 Solaris production
4887 // JDK. See 4341971.
4888 
4889 #define STACK_SLACK 0x800
4890 
4891 extern "C" {
4892   intptr_t sysThreadAvailableStackWithSlack() {
4893     stack_t st;
4894     intptr_t retval, stack_top;
4895     retval = thr_stksegment(&st);
4896     assert(retval == 0, "incorrect return value from thr_stksegment");
4897     assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
4898     assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
4899     stack_top=(intptr_t)st.ss_sp-st.ss_size;
4900     return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
4901   }
4902 }
4903 
4904 // ObjectMonitor park-unpark infrastructure ...
4905 //
4906 // We implement Solaris and Linux PlatformEvents with the
4907 // obvious condvar-mutex-flag triple.
4908 // Another alternative that works quite well is pipes:
4909 // Each PlatformEvent consists of a pipe-pair.
4910 // The thread associated with the PlatformEvent
4911 // calls park(), which reads from the input end of the pipe.
4912 // Unpark() writes into the other end of the pipe.
4913 // The write-side of the pipe must be set NDELAY.
4914 // Unfortunately pipes consume a large # of handles.
4915 // Native solaris lwp_park() and lwp_unpark() work nicely, too.
4916 // Using pipes for the 1st few threads might be workable, however.
4917 //
4918 // park() is permitted to return spuriously.
4919 // Callers of park() should wrap the call to park() in
4920 // an appropriate loop.  A litmus test for the correct
4921 // usage of park is the following: if park() were modified
4922 // to immediately return 0 your code should still work,
4923 // albeit degenerating to a spin loop.
4924 //
4925 // In a sense, park()-unpark() just provides more polite spinning
4926 // and polling with the key difference over naive spinning being
4927 // that a parked thread needs to be explicitly unparked() in order
4928 // to wake up and to poll the underlying condition.
4929 //
4930 // Assumption:
4931 //    Only one parker can exist on an event, which is why we allocate
4932 //    them per-thread. Multiple unparkers can coexist.
4933 //
4934 // _Event transitions in park()
4935 //   -1 => -1 : illegal
4936 //    1 =>  0 : pass - return immediately
4937 //    0 => -1 : block; then set _Event to 0 before returning
4938 //
4939 // _Event transitions in unpark()
4940 //    0 => 1 : just return
4941 //    1 => 1 : just return
4942 //   -1 => either 0 or 1; must signal target thread
4943 //         That is, we can safely transition _Event from -1 to either
4944 //         0 or 1.
4945 //
4946 // _Event serves as a restricted-range semaphore.
4947 //   -1 : thread is blocked, i.e. there is a waiter
4948 //    0 : neutral: thread is running or ready,
4949 //        could have been signaled after a wait started
4950 //    1 : signaled - thread is running or ready
4951 //
4952 // Another possible encoding of _Event would be with
4953 // explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
4954 //
4955 // TODO-FIXME: add DTRACE probes for:
4956 // 1.   Tx parks
4957 // 2.   Ty unparks Tx
4958 // 3.   Tx resumes from park
4959 
4960 
4961 // value determined through experimentation
4962 #define ROUNDINGFIX 11
4963 
4964 // utility to compute the abstime argument to timedwait.
4965 // TODO-FIXME: switch from compute_abstime() to unpackTime().
4966 
4967 static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
4968   // millis is the relative timeout time
4969   // abstime will be the absolute timeout time
4970   if (millis < 0)  millis = 0;
4971   struct timeval now;
4972   int status = gettimeofday(&now, NULL);
4973   assert(status == 0, "gettimeofday");
4974   jlong seconds = millis / 1000;
4975   jlong max_wait_period;
4976 
4977   if (UseLWPSynchronization) {
4978     // forward port of fix for 4275818 (not sleeping long enough)
4979     // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
4980     // _lwp_cond_timedwait() used a round_down algorithm rather
4981     // than a round_up. For millis less than our roundfactor
4982     // it rounded down to 0 which doesn't meet the spec.
4983     // For millis > roundfactor we may return a bit sooner, but
4984     // since we can not accurately identify the patch level and
4985     // this has already been fixed in Solaris 9 and 8 we will
4986     // leave it alone rather than always rounding down.
4987 
4988     if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
4989     // It appears that when we go directly through Solaris _lwp_cond_timedwait()
4990     // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
4991     max_wait_period = 21000000;
4992   } else {
4993     max_wait_period = 50000000;
4994   }
4995   millis %= 1000;
4996   if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
4997     seconds = max_wait_period;
4998   }
4999   abstime->tv_sec = now.tv_sec  + seconds;
5000   long       usec = now.tv_usec + millis * 1000;
5001   if (usec >= 1000000) {
5002     abstime->tv_sec += 1;
5003     usec -= 1000000;
5004   }
5005   abstime->tv_nsec = usec * 1000;
5006   return abstime;
5007 }
5008 
5009 void os::PlatformEvent::park() {           // AKA: down()
5010   // Transitions for _Event:
5011   //   -1 => -1 : illegal
5012   //    1 =>  0 : pass - return immediately
5013   //    0 => -1 : block; then set _Event to 0 before returning
5014 
5015   // Invariant: Only the thread associated with the Event/PlatformEvent
5016   // may call park().
5017   assert(_nParked == 0, "invariant");
5018 
5019   int v;
5020   for (;;) {
5021     v = _Event;
5022     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5023   }
5024   guarantee(v >= 0, "invariant");
5025   if (v == 0) {
5026     // Do this the hard way by blocking ...
5027     // See http://monaco.sfbay/detail.jsf?cr=5094058.
5028     int status = os::Solaris::mutex_lock(_mutex);
5029     assert_status(status == 0, status, "mutex_lock");
5030     guarantee(_nParked == 0, "invariant");
5031     ++_nParked;
5032     while (_Event < 0) {
5033       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5034       // Treat this the same as if the wait was interrupted
5035       // With usr/lib/lwp going to kernel, always handle ETIME
5036       status = os::Solaris::cond_wait(_cond, _mutex);
5037       if (status == ETIME) status = EINTR;
5038       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5039     }
5040     --_nParked;
5041     _Event = 0;
5042     status = os::Solaris::mutex_unlock(_mutex);
5043     assert_status(status == 0, status, "mutex_unlock");
5044     // Paranoia to ensure our locked and lock-free paths interact
5045     // correctly with each other.
5046     OrderAccess::fence();
5047   }
5048 }
5049 
5050 int os::PlatformEvent::park(jlong millis) {
5051   // Transitions for _Event:
5052   //   -1 => -1 : illegal
5053   //    1 =>  0 : pass - return immediately
5054   //    0 => -1 : block; then set _Event to 0 before returning
5055 
5056   guarantee(_nParked == 0, "invariant");
5057   int v;
5058   for (;;) {
5059     v = _Event;
5060     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5061   }
5062   guarantee(v >= 0, "invariant");
5063   if (v != 0) return OS_OK;
5064 
5065   int ret = OS_TIMEOUT;
5066   timestruc_t abst;
5067   compute_abstime(&abst, millis);
5068 
5069   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5070   int status = os::Solaris::mutex_lock(_mutex);
5071   assert_status(status == 0, status, "mutex_lock");
5072   guarantee(_nParked == 0, "invariant");
5073   ++_nParked;
5074   while (_Event < 0) {
5075     int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
5076     assert_status(status == 0 || status == EINTR ||
5077                   status == ETIME || status == ETIMEDOUT,
5078                   status, "cond_timedwait");
5079     if (!FilterSpuriousWakeups) break;                // previous semantics
5080     if (status == ETIME || status == ETIMEDOUT) break;
5081     // We consume and ignore EINTR and spurious wakeups.
5082   }
5083   --_nParked;
5084   if (_Event >= 0) ret = OS_OK;
5085   _Event = 0;
5086   status = os::Solaris::mutex_unlock(_mutex);
5087   assert_status(status == 0, status, "mutex_unlock");
5088   // Paranoia to ensure our locked and lock-free paths interact
5089   // correctly with each other.
5090   OrderAccess::fence();
5091   return ret;
5092 }
5093 
5094 void os::PlatformEvent::unpark() {
5095   // Transitions for _Event:
5096   //    0 => 1 : just return
5097   //    1 => 1 : just return
5098   //   -1 => either 0 or 1; must signal target thread
5099   //         That is, we can safely transition _Event from -1 to either
5100   //         0 or 1.
5101   // See also: "Semaphores in Plan 9" by Mullender & Cox
5102   //
5103   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5104   // that it will take two back-to-back park() calls for the owning
5105   // thread to block. This has the benefit of forcing a spurious return
5106   // from the first park() call after an unpark() call which will help
5107   // shake out uses of park() and unpark() without condition variables.
5108 
5109   if (Atomic::xchg(1, &_Event) >= 0) return;
5110 
5111   // If the thread associated with the event was parked, wake it.
5112   // Wait for the thread assoc with the PlatformEvent to vacate.
5113   int status = os::Solaris::mutex_lock(_mutex);
5114   assert_status(status == 0, status, "mutex_lock");
5115   int AnyWaiters = _nParked;
5116   status = os::Solaris::mutex_unlock(_mutex);
5117   assert_status(status == 0, status, "mutex_unlock");
5118   guarantee(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5119   if (AnyWaiters != 0) {
5120     // Note that we signal() *after* dropping the lock for "immortal" Events.
5121     // This is safe and avoids a common class of  futile wakeups.  In rare
5122     // circumstances this can cause a thread to return prematurely from
5123     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5124     // will simply re-test the condition and re-park itself.
5125     // This provides particular benefit if the underlying platform does not
5126     // provide wait morphing.
5127     status = os::Solaris::cond_signal(_cond);
5128     assert_status(status == 0, status, "cond_signal");
5129   }
5130 }
5131 
5132 // JSR166
5133 // -------------------------------------------------------
5134 
5135 // The solaris and linux implementations of park/unpark are fairly
5136 // conservative for now, but can be improved. They currently use a
5137 // mutex/condvar pair, plus _counter.
5138 // Park decrements _counter if > 0, else does a condvar wait.  Unpark
5139 // sets count to 1 and signals condvar.  Only one thread ever waits
5140 // on the condvar. Contention seen when trying to park implies that someone
5141 // is unparking you, so don't wait. And spurious returns are fine, so there
5142 // is no need to track notifications.
5143 
5144 #define MAX_SECS 100000000
5145 
5146 // This code is common to linux and solaris and will be moved to a
5147 // common place in dolphin.
5148 //
5149 // The passed in time value is either a relative time in nanoseconds
5150 // or an absolute time in milliseconds. Either way it has to be unpacked
5151 // into suitable seconds and nanoseconds components and stored in the
5152 // given timespec structure.
5153 // Given time is a 64-bit value and the time_t used in the timespec is only
5154 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5155 // overflow if times way in the future are given. Further on Solaris versions
5156 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5157 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5158 // As it will be 28 years before "now + 100000000" will overflow we can
5159 // ignore overflow and just impose a hard-limit on seconds using the value
5160 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5161 // years from "now".
5162 //
5163 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5164   assert(time > 0, "convertTime");
5165 
5166   struct timeval now;
5167   int status = gettimeofday(&now, NULL);
5168   assert(status == 0, "gettimeofday");
5169 
5170   time_t max_secs = now.tv_sec + MAX_SECS;
5171 
5172   if (isAbsolute) {
5173     jlong secs = time / 1000;
5174     if (secs > max_secs) {
5175       absTime->tv_sec = max_secs;
5176     } else {
5177       absTime->tv_sec = secs;
5178     }
5179     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5180   } else {
5181     jlong secs = time / NANOSECS_PER_SEC;
5182     if (secs >= MAX_SECS) {
5183       absTime->tv_sec = max_secs;
5184       absTime->tv_nsec = 0;
5185     } else {
5186       absTime->tv_sec = now.tv_sec + secs;
5187       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5188       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5189         absTime->tv_nsec -= NANOSECS_PER_SEC;
5190         ++absTime->tv_sec; // note: this must be <= max_secs
5191       }
5192     }
5193   }
5194   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5195   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5196   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5197   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5198 }
5199 
5200 void Parker::park(bool isAbsolute, jlong time) {
5201   // Ideally we'd do something useful while spinning, such
5202   // as calling unpackTime().
5203 
5204   // Optional fast-path check:
5205   // Return immediately if a permit is available.
5206   // We depend on Atomic::xchg() having full barrier semantics
5207   // since we are doing a lock-free update to _counter.
5208   if (Atomic::xchg(0, &_counter) > 0) return;
5209 
5210   // Optional fast-exit: Check interrupt before trying to wait
5211   Thread* thread = Thread::current();
5212   assert(thread->is_Java_thread(), "Must be JavaThread");
5213   JavaThread *jt = (JavaThread *)thread;
5214   if (Thread::is_interrupted(thread, false)) {
5215     return;
5216   }
5217 
5218   // First, demultiplex/decode time arguments
5219   timespec absTime;
5220   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5221     return;
5222   }
5223   if (time > 0) {
5224     // Warning: this code might be exposed to the old Solaris time
5225     // round-down bugs.  Grep "roundingFix" for details.
5226     unpackTime(&absTime, isAbsolute, time);
5227   }
5228 
5229   // Enter safepoint region
5230   // Beware of deadlocks such as 6317397.
5231   // The per-thread Parker:: _mutex is a classic leaf-lock.
5232   // In particular a thread must never block on the Threads_lock while
5233   // holding the Parker:: mutex.  If safepoints are pending both the
5234   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5235   ThreadBlockInVM tbivm(jt);
5236 
5237   // Don't wait if cannot get lock since interference arises from
5238   // unblocking.  Also. check interrupt before trying wait
5239   if (Thread::is_interrupted(thread, false) ||
5240       os::Solaris::mutex_trylock(_mutex) != 0) {
5241     return;
5242   }
5243 
5244   int status;
5245 
5246   if (_counter > 0)  { // no wait needed
5247     _counter = 0;
5248     status = os::Solaris::mutex_unlock(_mutex);
5249     assert(status == 0, "invariant");
5250     // Paranoia to ensure our locked and lock-free paths interact
5251     // correctly with each other and Java-level accesses.
5252     OrderAccess::fence();
5253     return;
5254   }
5255 
5256   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5257   jt->set_suspend_equivalent();
5258   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5259 
5260   // Do this the hard way by blocking ...
5261   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5262   if (time == 0) {
5263     status = os::Solaris::cond_wait(_cond, _mutex);
5264   } else {
5265     status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
5266   }
5267   // Note that an untimed cond_wait() can sometimes return ETIME on older
5268   // versions of the Solaris.
5269   assert_status(status == 0 || status == EINTR ||
5270                 status == ETIME || status == ETIMEDOUT,
5271                 status, "cond_timedwait");
5272 
5273   _counter = 0;
5274   status = os::Solaris::mutex_unlock(_mutex);
5275   assert_status(status == 0, status, "mutex_unlock");
5276   // Paranoia to ensure our locked and lock-free paths interact
5277   // correctly with each other and Java-level accesses.
5278   OrderAccess::fence();
5279 
5280   // If externally suspended while waiting, re-suspend
5281   if (jt->handle_special_suspend_equivalent_condition()) {
5282     jt->java_suspend_self();
5283   }
5284 }
5285 
5286 void Parker::unpark() {
5287   int status = os::Solaris::mutex_lock(_mutex);
5288   assert(status == 0, "invariant");
5289   const int s = _counter;
5290   _counter = 1;
5291   status = os::Solaris::mutex_unlock(_mutex);
5292   assert(status == 0, "invariant");
5293 
5294   if (s < 1) {
5295     status = os::Solaris::cond_signal(_cond);
5296     assert(status == 0, "invariant");
5297   }
5298 }
5299 
5300 extern char** environ;
5301 
5302 // Run the specified command in a separate process. Return its exit value,
5303 // or -1 on failure (e.g. can't fork a new process).
5304 // Unlike system(), this function can be called from signal handler. It
5305 // doesn't block SIGINT et al.
5306 int os::fork_and_exec(char* cmd) {
5307   char * argv[4];
5308   argv[0] = (char *)"sh";
5309   argv[1] = (char *)"-c";
5310   argv[2] = cmd;
5311   argv[3] = NULL;
5312 
5313   // fork is async-safe, fork1 is not so can't use in signal handler
5314   pid_t pid;
5315   Thread* t = Thread::current_or_null_safe();
5316   if (t != NULL && t->is_inside_signal_handler()) {
5317     pid = fork();
5318   } else {
5319     pid = fork1();
5320   }
5321 
5322   if (pid < 0) {
5323     // fork failed
5324     warning("fork failed: %s", os::strerror(errno));
5325     return -1;
5326 
5327   } else if (pid == 0) {
5328     // child process
5329 
5330     // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
5331     execve("/usr/bin/sh", argv, environ);
5332 
5333     // execve failed
5334     _exit(-1);
5335 
5336   } else  {
5337     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5338     // care about the actual exit code, for now.
5339 
5340     int status;
5341 
5342     // Wait for the child process to exit.  This returns immediately if
5343     // the child has already exited. */
5344     while (waitpid(pid, &status, 0) < 0) {
5345       switch (errno) {
5346       case ECHILD: return 0;
5347       case EINTR: break;
5348       default: return -1;
5349       }
5350     }
5351 
5352     if (WIFEXITED(status)) {
5353       // The child exited normally; get its exit code.
5354       return WEXITSTATUS(status);
5355     } else if (WIFSIGNALED(status)) {
5356       // The child exited because of a signal
5357       // The best value to return is 0x80 + signal number,
5358       // because that is what all Unix shells do, and because
5359       // it allows callers to distinguish between process exit and
5360       // process death by signal.
5361       return 0x80 + WTERMSIG(status);
5362     } else {
5363       // Unknown exit code; pass it through
5364       return status;
5365     }
5366   }
5367 }
5368 
5369 size_t os::write(int fd, const void *buf, unsigned int nBytes) {
5370   size_t res;
5371   RESTARTABLE((size_t) ::write(fd, buf, (size_t) nBytes), res);
5372   return res;
5373 }
5374 
5375 int os::close(int fd) {
5376   return ::close(fd);
5377 }
5378 
5379 int os::socket_close(int fd) {
5380   return ::close(fd);
5381 }
5382 
5383 int os::recv(int fd, char* buf, size_t nBytes, uint flags) {
5384   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5385          "Assumed _thread_in_native");
5386   RESTARTABLE_RETURN_INT((int)::recv(fd, buf, nBytes, flags));
5387 }
5388 
5389 int os::send(int fd, char* buf, size_t nBytes, uint flags) {
5390   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5391          "Assumed _thread_in_native");
5392   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5393 }
5394 
5395 int os::raw_send(int fd, char* buf, size_t nBytes, uint flags) {
5396   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5397 }
5398 
5399 // As both poll and select can be interrupted by signals, we have to be
5400 // prepared to restart the system call after updating the timeout, unless
5401 // a poll() is done with timeout == -1, in which case we repeat with this
5402 // "wait forever" value.
5403 
5404 int os::connect(int fd, struct sockaddr *him, socklen_t len) {
5405   int _result;
5406   _result = ::connect(fd, him, len);
5407 
5408   // On Solaris, when a connect() call is interrupted, the connection
5409   // can be established asynchronously (see 6343810). Subsequent calls
5410   // to connect() must check the errno value which has the semantic
5411   // described below (copied from the connect() man page). Handling
5412   // of asynchronously established connections is required for both
5413   // blocking and non-blocking sockets.
5414   //     EINTR            The  connection  attempt  was   interrupted
5415   //                      before  any data arrived by the delivery of
5416   //                      a signal. The connection, however, will  be
5417   //                      established asynchronously.
5418   //
5419   //     EINPROGRESS      The socket is non-blocking, and the connec-
5420   //                      tion  cannot  be completed immediately.
5421   //
5422   //     EALREADY         The socket is non-blocking,  and a previous
5423   //                      connection  attempt  has  not yet been com-
5424   //                      pleted.
5425   //
5426   //     EISCONN          The socket is already connected.
5427   if (_result == OS_ERR && errno == EINTR) {
5428     // restarting a connect() changes its errno semantics
5429     RESTARTABLE(::connect(fd, him, len), _result);
5430     // undo these changes
5431     if (_result == OS_ERR) {
5432       if (errno == EALREADY) {
5433         errno = EINPROGRESS; // fall through
5434       } else if (errno == EISCONN) {
5435         errno = 0;
5436         return OS_OK;
5437       }
5438     }
5439   }
5440   return _result;
5441 }
5442 
5443 // Get the default path to the core file
5444 // Returns the length of the string
5445 int os::get_core_path(char* buffer, size_t bufferSize) {
5446   const char* p = get_current_directory(buffer, bufferSize);
5447 
5448   if (p == NULL) {
5449     assert(p != NULL, "failed to get current directory");
5450     return 0;
5451   }
5452 
5453   jio_snprintf(buffer, bufferSize, "%s/core or core.%d",
5454                                               p, current_process_id());
5455 
5456   return strlen(buffer);
5457 }
5458 
5459 #ifndef PRODUCT
5460 void TestReserveMemorySpecial_test() {
5461   // No tests available for this platform
5462 }
5463 #endif
5464 
5465 bool os::start_debugging(char *buf, int buflen) {
5466   int len = (int)strlen(buf);
5467   char *p = &buf[len];
5468 
5469   jio_snprintf(p, buflen-len,
5470                "\n\n"
5471                "Do you want to debug the problem?\n\n"
5472                "To debug, run 'dbx - %d'; then switch to thread " INTX_FORMAT "\n"
5473                "Enter 'yes' to launch dbx automatically (PATH must include dbx)\n"
5474                "Otherwise, press RETURN to abort...",
5475                os::current_process_id(), os::current_thread_id());
5476 
5477   bool yes = os::message_box("Unexpected Error", buf);
5478 
5479   if (yes) {
5480     // yes, user asked VM to launch debugger
5481     jio_snprintf(buf, sizeof(buf), "dbx - %d", os::current_process_id());
5482 
5483     os::fork_and_exec(buf);
5484     yes = false;
5485   }
5486   return yes;
5487 }