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