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 static inline time_t get_mtime(const char* filename) {
1671   struct stat st;
1672   int ret = os::stat(filename, &st);
1673   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
1674   return st.st_mtime;
1675 }
1676 
1677 int os::compare_file_modified_times(const char* file1, const char* file2) {
1678   time_t t1 = get_mtime(file1);
1679   time_t t2 = get_mtime(file2);
1680   return t1 - t2;
1681 }
1682 
1683 static bool _print_ascii_file(const char* filename, outputStream* st) {
1684   int fd = ::open(filename, O_RDONLY);
1685   if (fd == -1) {
1686     return false;
1687   }
1688 
1689   char buf[32];
1690   int bytes;
1691   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1692     st->print_raw(buf, bytes);
1693   }
1694 
1695   ::close(fd);
1696 
1697   return true;
1698 }
1699 
1700 void os::print_os_info_brief(outputStream* st) {
1701   os::Solaris::print_distro_info(st);
1702 
1703   os::Posix::print_uname_info(st);
1704 
1705   os::Solaris::print_libversion_info(st);
1706 }
1707 
1708 void os::print_os_info(outputStream* st) {
1709   st->print("OS:");
1710 
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   os::Posix::print_rlimit_info(st);
1718 
1719   os::Posix::print_load_average(st);
1720 }
1721 
1722 void os::Solaris::print_distro_info(outputStream* st) {
1723   if (!_print_ascii_file("/etc/release", st)) {
1724     st->print("Solaris");
1725   }
1726   st->cr();
1727 }
1728 
1729 void os::get_summary_os_info(char* buf, size_t buflen) {
1730   strncpy(buf, "Solaris", buflen);  // default to plain solaris
1731   FILE* fp = fopen("/etc/release", "r");
1732   if (fp != NULL) {
1733     char tmp[256];
1734     // Only get the first line and chop out everything but the os name.
1735     if (fgets(tmp, sizeof(tmp), fp)) {
1736       char* ptr = tmp;
1737       // skip past whitespace characters
1738       while (*ptr != '\0' && (*ptr == ' ' || *ptr == '\t' || *ptr == '\n')) ptr++;
1739       if (*ptr != '\0') {
1740         char* nl = strchr(ptr, '\n');
1741         if (nl != NULL) *nl = '\0';
1742         strncpy(buf, ptr, buflen);
1743       }
1744     }
1745     fclose(fp);
1746   }
1747 }
1748 
1749 void os::Solaris::print_libversion_info(outputStream* st) {
1750   st->print("  (T2 libthread)");
1751   st->cr();
1752 }
1753 
1754 static bool check_addr0(outputStream* st) {
1755   jboolean status = false;
1756   const int read_chunk = 200;
1757   int ret = 0;
1758   int nmap = 0;
1759   int fd = ::open("/proc/self/map",O_RDONLY);
1760   if (fd >= 0) {
1761     prmap_t *p = NULL;
1762     char *mbuff = (char *) calloc(read_chunk, sizeof(prmap_t));
1763     if (NULL == mbuff) {
1764       ::close(fd);
1765       return status;
1766     }
1767     while ((ret = ::read(fd, mbuff, read_chunk*sizeof(prmap_t))) > 0) {
1768       //check if read() has not read partial data
1769       if( 0 != ret % sizeof(prmap_t)){
1770         break;
1771       }
1772       nmap = ret / sizeof(prmap_t);
1773       p = (prmap_t *)mbuff;
1774       for(int i = 0; i < nmap; i++){
1775         if (p->pr_vaddr == 0x0) {
1776           st->print("Warning: Address: " PTR_FORMAT ", Size: " SIZE_FORMAT "K, ",p->pr_vaddr, p->pr_size/1024);
1777           st->print("Mapped file: %s, ", p->pr_mapname[0] == '\0' ? "None" : p->pr_mapname);
1778           st->print("Access: ");
1779           st->print("%s",(p->pr_mflags & MA_READ)  ? "r" : "-");
1780           st->print("%s",(p->pr_mflags & MA_WRITE) ? "w" : "-");
1781           st->print("%s",(p->pr_mflags & MA_EXEC)  ? "x" : "-");
1782           st->cr();
1783           status = true;
1784         }
1785         p++;
1786       }
1787     }
1788     free(mbuff);
1789     ::close(fd);
1790   }
1791   return status;
1792 }
1793 
1794 void os::get_summary_cpu_info(char* buf, size_t buflen) {
1795   // Get MHz with system call. We don't seem to already have this.
1796   processor_info_t stats;
1797   processorid_t id = getcpuid();
1798   int clock = 0;
1799   if (processor_info(id, &stats) != -1) {
1800     clock = stats.pi_clock;  // pi_processor_type isn't more informative than below
1801   }
1802 #ifdef AMD64
1803   snprintf(buf, buflen, "x86 64 bit %d MHz", clock);
1804 #else
1805   // must be sparc
1806   snprintf(buf, buflen, "Sparcv9 64 bit %d MHz", clock);
1807 #endif
1808 }
1809 
1810 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
1811   // Nothing to do for now.
1812 }
1813 
1814 void os::print_memory_info(outputStream* st) {
1815   st->print("Memory:");
1816   st->print(" %dk page", os::vm_page_size()>>10);
1817   st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
1818   st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
1819   st->cr();
1820   (void) check_addr0(st);
1821 }
1822 
1823 // Moved from whole group, because we need them here for diagnostic
1824 // prints.
1825 static int Maxsignum = 0;
1826 static int *ourSigFlags = NULL;
1827 
1828 int os::Solaris::get_our_sigflags(int sig) {
1829   assert(ourSigFlags!=NULL, "signal data structure not initialized");
1830   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
1831   return ourSigFlags[sig];
1832 }
1833 
1834 void os::Solaris::set_our_sigflags(int sig, int flags) {
1835   assert(ourSigFlags!=NULL, "signal data structure not initialized");
1836   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
1837   ourSigFlags[sig] = flags;
1838 }
1839 
1840 
1841 static const char* get_signal_handler_name(address handler,
1842                                            char* buf, int buflen) {
1843   int offset;
1844   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
1845   if (found) {
1846     // skip directory names
1847     const char *p1, *p2;
1848     p1 = buf;
1849     size_t len = strlen(os::file_separator());
1850     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
1851     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
1852   } else {
1853     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
1854   }
1855   return buf;
1856 }
1857 
1858 static void print_signal_handler(outputStream* st, int sig,
1859                                  char* buf, size_t buflen) {
1860   struct sigaction sa;
1861 
1862   sigaction(sig, NULL, &sa);
1863 
1864   st->print("%s: ", os::exception_name(sig, buf, buflen));
1865 
1866   address handler = (sa.sa_flags & SA_SIGINFO)
1867                   ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
1868                   : CAST_FROM_FN_PTR(address, sa.sa_handler);
1869 
1870   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
1871     st->print("SIG_DFL");
1872   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
1873     st->print("SIG_IGN");
1874   } else {
1875     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
1876   }
1877 
1878   st->print(", sa_mask[0]=");
1879   os::Posix::print_signal_set_short(st, &sa.sa_mask);
1880 
1881   address rh = VMError::get_resetted_sighandler(sig);
1882   // May be, handler was resetted by VMError?
1883   if (rh != NULL) {
1884     handler = rh;
1885     sa.sa_flags = VMError::get_resetted_sigflags(sig);
1886   }
1887 
1888   st->print(", sa_flags=");
1889   os::Posix::print_sa_flags(st, sa.sa_flags);
1890 
1891   // Check: is it our handler?
1892   if (handler == CAST_FROM_FN_PTR(address, signalHandler)) {
1893     // It is our signal handler
1894     // check for flags
1895     if (sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
1896       st->print(
1897                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
1898                 os::Solaris::get_our_sigflags(sig));
1899     }
1900   }
1901   st->cr();
1902 }
1903 
1904 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
1905   st->print_cr("Signal Handlers:");
1906   print_signal_handler(st, SIGSEGV, buf, buflen);
1907   print_signal_handler(st, SIGBUS , buf, buflen);
1908   print_signal_handler(st, SIGFPE , buf, buflen);
1909   print_signal_handler(st, SIGPIPE, buf, buflen);
1910   print_signal_handler(st, SIGXFSZ, buf, buflen);
1911   print_signal_handler(st, SIGILL , buf, buflen);
1912   print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
1913   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
1914   print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
1915   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
1916   print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
1917 }
1918 
1919 static char saved_jvm_path[MAXPATHLEN] = { 0 };
1920 
1921 // Find the full path to the current module, libjvm.so
1922 void os::jvm_path(char *buf, jint buflen) {
1923   // Error checking.
1924   if (buflen < MAXPATHLEN) {
1925     assert(false, "must use a large-enough buffer");
1926     buf[0] = '\0';
1927     return;
1928   }
1929   // Lazy resolve the path to current module.
1930   if (saved_jvm_path[0] != 0) {
1931     strcpy(buf, saved_jvm_path);
1932     return;
1933   }
1934 
1935   Dl_info dlinfo;
1936   int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
1937   assert(ret != 0, "cannot locate libjvm");
1938   if (ret != 0 && dlinfo.dli_fname != NULL) {
1939     if (os::Posix::realpath((char *)dlinfo.dli_fname, buf, buflen) == NULL) {
1940       return;
1941     }
1942   } else {
1943     buf[0] = '\0';
1944     return;
1945   }
1946 
1947   if (Arguments::sun_java_launcher_is_altjvm()) {
1948     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
1949     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
1950     // If "/jre/lib/" appears at the right place in the string, then
1951     // assume we are installed in a JDK and we're done.  Otherwise, check
1952     // for a JAVA_HOME environment variable and fix up the path so it
1953     // looks like libjvm.so is installed there (append a fake suffix
1954     // hotspot/libjvm.so).
1955     const char *p = buf + strlen(buf) - 1;
1956     for (int count = 0; p > buf && count < 5; ++count) {
1957       for (--p; p > buf && *p != '/'; --p)
1958         /* empty */ ;
1959     }
1960 
1961     if (strncmp(p, "/jre/lib/", 9) != 0) {
1962       // Look for JAVA_HOME in the environment.
1963       char* java_home_var = ::getenv("JAVA_HOME");
1964       if (java_home_var != NULL && java_home_var[0] != 0) {
1965         char* jrelib_p;
1966         int   len;
1967 
1968         // Check the current module name "libjvm.so".
1969         p = strrchr(buf, '/');
1970         assert(strstr(p, "/libjvm") == p, "invalid library name");
1971 
1972         if (os::Posix::realpath(java_home_var, buf, buflen) == NULL) {
1973           return;
1974         }
1975         // determine if this is a legacy image or modules image
1976         // modules image doesn't have "jre" subdirectory
1977         len = strlen(buf);
1978         assert(len < buflen, "Ran out of buffer space");
1979         jrelib_p = buf + len;
1980         snprintf(jrelib_p, buflen-len, "/jre/lib");
1981         if (0 != access(buf, F_OK)) {
1982           snprintf(jrelib_p, buflen-len, "/lib");
1983         }
1984 
1985         if (0 == access(buf, F_OK)) {
1986           // Use current module name "libjvm.so"
1987           len = strlen(buf);
1988           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
1989         } else {
1990           // Go back to path of .so
1991           if (os::Posix::realpath((char *)dlinfo.dli_fname, buf, buflen) == NULL) {
1992             return;
1993           }
1994         }
1995       }
1996     }
1997   }
1998 
1999   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2000   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2001 }
2002 
2003 
2004 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2005   // no prefix required, not even "_"
2006 }
2007 
2008 
2009 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2010   // no suffix required
2011 }
2012 
2013 // sun.misc.Signal
2014 
2015 extern "C" {
2016   static void UserHandler(int sig, void *siginfo, void *context) {
2017     // Ctrl-C is pressed during error reporting, likely because the error
2018     // handler fails to abort. Let VM die immediately.
2019     if (sig == SIGINT && VMError::is_error_reported()) {
2020       os::die();
2021     }
2022 
2023     os::signal_notify(sig);
2024     // We do not need to reinstate the signal handler each time...
2025   }
2026 }
2027 
2028 void* os::user_handler() {
2029   return CAST_FROM_FN_PTR(void*, UserHandler);
2030 }
2031 
2032 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2033   struct timespec ts;
2034   unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2035 
2036   return ts;
2037 }
2038 
2039 extern "C" {
2040   typedef void (*sa_handler_t)(int);
2041   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2042 }
2043 
2044 void* os::signal(int signal_number, void* handler) {
2045   struct sigaction sigAct, oldSigAct;
2046   sigfillset(&(sigAct.sa_mask));
2047   sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2048   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2049 
2050   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2051     // -1 means registration failed
2052     return (void *)-1;
2053   }
2054 
2055   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2056 }
2057 
2058 void os::signal_raise(int signal_number) {
2059   raise(signal_number);
2060 }
2061 
2062 // The following code is moved from os.cpp for making this
2063 // code platform specific, which it is by its very nature.
2064 
2065 // a counter for each possible signal value
2066 static int Sigexit = 0;
2067 static jint *pending_signals = NULL;
2068 static int *preinstalled_sigs = NULL;
2069 static struct sigaction *chainedsigactions = NULL;
2070 static Semaphore* sig_sem = NULL;
2071 
2072 int os::sigexitnum_pd() {
2073   assert(Sigexit > 0, "signal memory not yet initialized");
2074   return Sigexit;
2075 }
2076 
2077 void os::Solaris::init_signal_mem() {
2078   // Initialize signal structures
2079   Maxsignum = SIGRTMAX;
2080   Sigexit = Maxsignum+1;
2081   assert(Maxsignum >0, "Unable to obtain max signal number");
2082 
2083   // Initialize signal structures
2084   // pending_signals has one int per signal
2085   // The additional signal is for SIGEXIT - exit signal to signal_thread
2086   pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1), mtInternal);
2087   memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2088 
2089   if (UseSignalChaining) {
2090     chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2091                                                    * (Maxsignum + 1), mtInternal);
2092     memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2093     preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2094     memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2095   }
2096   ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2097   memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2098 }
2099 
2100 static void jdk_misc_signal_init() {
2101   // Initialize signal semaphore
2102   sig_sem = new Semaphore();
2103 }
2104 
2105 void os::signal_notify(int sig) {
2106   if (sig_sem != NULL) {
2107     Atomic::inc(&pending_signals[sig]);
2108     sig_sem->signal();
2109   } else {
2110     // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2111     // initialization isn't called.
2112     assert(ReduceSignalUsage, "signal semaphore should be created");
2113   }
2114 }
2115 
2116 static int check_pending_signals() {
2117   int ret;
2118   while (true) {
2119     for (int i = 0; i < Sigexit + 1; i++) {
2120       jint n = pending_signals[i];
2121       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2122         return i;
2123       }
2124     }
2125     JavaThread *thread = JavaThread::current();
2126     ThreadBlockInVM tbivm(thread);
2127 
2128     bool threadIsSuspended;
2129     do {
2130       thread->set_suspend_equivalent();
2131       sig_sem->wait();
2132 
2133       // were we externally suspended while we were waiting?
2134       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2135       if (threadIsSuspended) {
2136         // The semaphore has been incremented, but while we were waiting
2137         // another thread suspended us. We don't want to continue running
2138         // while suspended because that would surprise the thread that
2139         // suspended us.
2140         sig_sem->signal();
2141 
2142         thread->java_suspend_self();
2143       }
2144     } while (threadIsSuspended);
2145   }
2146 }
2147 
2148 int os::signal_wait() {
2149   return check_pending_signals();
2150 }
2151 
2152 ////////////////////////////////////////////////////////////////////////////////
2153 // Virtual Memory
2154 
2155 static int page_size = -1;
2156 
2157 int os::vm_page_size() {
2158   assert(page_size != -1, "must call os::init");
2159   return page_size;
2160 }
2161 
2162 // Solaris allocates memory by pages.
2163 int os::vm_allocation_granularity() {
2164   assert(page_size != -1, "must call os::init");
2165   return page_size;
2166 }
2167 
2168 static bool recoverable_mmap_error(int err) {
2169   // See if the error is one we can let the caller handle. This
2170   // list of errno values comes from the Solaris mmap(2) man page.
2171   switch (err) {
2172   case EBADF:
2173   case EINVAL:
2174   case ENOTSUP:
2175     // let the caller deal with these errors
2176     return true;
2177 
2178   default:
2179     // Any remaining errors on this OS can cause our reserved mapping
2180     // to be lost. That can cause confusion where different data
2181     // structures think they have the same memory mapped. The worst
2182     // scenario is if both the VM and a library think they have the
2183     // same memory mapped.
2184     return false;
2185   }
2186 }
2187 
2188 static void warn_fail_commit_memory(char* addr, size_t bytes, bool exec,
2189                                     int err) {
2190   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2191           ", %d) failed; error='%s' (errno=%d)", addr, bytes, exec,
2192           os::strerror(err), err);
2193 }
2194 
2195 static void warn_fail_commit_memory(char* addr, size_t bytes,
2196                                     size_t alignment_hint, bool exec,
2197                                     int err) {
2198   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2199           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, bytes,
2200           alignment_hint, exec, os::strerror(err), err);
2201 }
2202 
2203 int os::Solaris::commit_memory_impl(char* addr, size_t bytes, bool exec) {
2204   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2205   size_t size = bytes;
2206   char *res = Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2207   if (res != NULL) {
2208     if (UseNUMAInterleaving) {
2209       numa_make_global(addr, bytes);
2210     }
2211     return 0;
2212   }
2213 
2214   int err = errno;  // save errno from mmap() call in mmap_chunk()
2215 
2216   if (!recoverable_mmap_error(err)) {
2217     warn_fail_commit_memory(addr, bytes, exec, err);
2218     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "committing reserved memory.");
2219   }
2220 
2221   return err;
2222 }
2223 
2224 bool os::pd_commit_memory(char* addr, size_t bytes, bool exec) {
2225   return Solaris::commit_memory_impl(addr, bytes, exec) == 0;
2226 }
2227 
2228 void os::pd_commit_memory_or_exit(char* addr, size_t bytes, bool exec,
2229                                   const char* mesg) {
2230   assert(mesg != NULL, "mesg must be specified");
2231   int err = os::Solaris::commit_memory_impl(addr, bytes, exec);
2232   if (err != 0) {
2233     // the caller wants all commit errors to exit with the specified mesg:
2234     warn_fail_commit_memory(addr, bytes, exec, err);
2235     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2236   }
2237 }
2238 
2239 size_t os::Solaris::page_size_for_alignment(size_t alignment) {
2240   assert(is_aligned(alignment, (size_t) vm_page_size()),
2241          SIZE_FORMAT " is not aligned to " SIZE_FORMAT,
2242          alignment, (size_t) vm_page_size());
2243 
2244   for (int i = 0; _page_sizes[i] != 0; i++) {
2245     if (is_aligned(alignment, _page_sizes[i])) {
2246       return _page_sizes[i];
2247     }
2248   }
2249 
2250   return (size_t) vm_page_size();
2251 }
2252 
2253 int os::Solaris::commit_memory_impl(char* addr, size_t bytes,
2254                                     size_t alignment_hint, bool exec) {
2255   int err = Solaris::commit_memory_impl(addr, bytes, exec);
2256   if (err == 0 && UseLargePages && alignment_hint > 0) {
2257     assert(is_aligned(bytes, alignment_hint),
2258            SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, alignment_hint);
2259 
2260     // The syscall memcntl requires an exact page size (see man memcntl for details).
2261     size_t page_size = page_size_for_alignment(alignment_hint);
2262     if (page_size > (size_t) vm_page_size()) {
2263       (void)Solaris::setup_large_pages(addr, bytes, page_size);
2264     }
2265   }
2266   return err;
2267 }
2268 
2269 bool os::pd_commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2270                           bool exec) {
2271   return Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec) == 0;
2272 }
2273 
2274 void os::pd_commit_memory_or_exit(char* addr, size_t bytes,
2275                                   size_t alignment_hint, bool exec,
2276                                   const char* mesg) {
2277   assert(mesg != NULL, "mesg must be specified");
2278   int err = os::Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec);
2279   if (err != 0) {
2280     // the caller wants all commit errors to exit with the specified mesg:
2281     warn_fail_commit_memory(addr, bytes, alignment_hint, exec, err);
2282     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2283   }
2284 }
2285 
2286 // Uncommit the pages in a specified region.
2287 void os::pd_free_memory(char* addr, size_t bytes, size_t alignment_hint) {
2288   if (madvise(addr, bytes, MADV_FREE) < 0) {
2289     debug_only(warning("MADV_FREE failed."));
2290     return;
2291   }
2292 }
2293 
2294 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2295   return os::commit_memory(addr, size, !ExecMem);
2296 }
2297 
2298 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2299   return os::uncommit_memory(addr, size);
2300 }
2301 
2302 // Change the page size in a given range.
2303 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2304   assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2305   assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2306   if (UseLargePages) {
2307     size_t page_size = Solaris::page_size_for_alignment(alignment_hint);
2308     if (page_size > (size_t) vm_page_size()) {
2309       Solaris::setup_large_pages(addr, bytes, page_size);
2310     }
2311   }
2312 }
2313 
2314 // Tell the OS to make the range local to the first-touching LWP
2315 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2316   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2317   if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2318     debug_only(warning("MADV_ACCESS_LWP failed."));
2319   }
2320 }
2321 
2322 // Tell the OS that this range would be accessed from different LWPs.
2323 void os::numa_make_global(char *addr, size_t bytes) {
2324   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2325   if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2326     debug_only(warning("MADV_ACCESS_MANY failed."));
2327   }
2328 }
2329 
2330 // Get the number of the locality groups.
2331 size_t os::numa_get_groups_num() {
2332   size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2333   return n != -1 ? n : 1;
2334 }
2335 
2336 // Get a list of leaf locality groups. A leaf lgroup is group that
2337 // doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2338 // board. An LWP is assigned to one of these groups upon creation.
2339 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2340   if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2341     ids[0] = 0;
2342     return 1;
2343   }
2344   int result_size = 0, top = 1, bottom = 0, cur = 0;
2345   for (int k = 0; k < size; k++) {
2346     int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2347                                    (Solaris::lgrp_id_t*)&ids[top], size - top);
2348     if (r == -1) {
2349       ids[0] = 0;
2350       return 1;
2351     }
2352     if (!r) {
2353       // That's a leaf node.
2354       assert(bottom <= cur, "Sanity check");
2355       // Check if the node has memory
2356       if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2357                                   NULL, 0, LGRP_RSRC_MEM) > 0) {
2358         ids[bottom++] = ids[cur];
2359       }
2360     }
2361     top += r;
2362     cur++;
2363   }
2364   if (bottom == 0) {
2365     // Handle a situation, when the OS reports no memory available.
2366     // Assume UMA architecture.
2367     ids[0] = 0;
2368     return 1;
2369   }
2370   return bottom;
2371 }
2372 
2373 // Detect the topology change. Typically happens during CPU plugging-unplugging.
2374 bool os::numa_topology_changed() {
2375   int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2376   if (is_stale != -1 && is_stale) {
2377     Solaris::lgrp_fini(Solaris::lgrp_cookie());
2378     Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2379     assert(c != 0, "Failure to initialize LGRP API");
2380     Solaris::set_lgrp_cookie(c);
2381     return true;
2382   }
2383   return false;
2384 }
2385 
2386 // Get the group id of the current LWP.
2387 int os::numa_get_group_id() {
2388   int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2389   if (lgrp_id == -1) {
2390     return 0;
2391   }
2392   const int size = os::numa_get_groups_num();
2393   int *ids = (int*)alloca(size * sizeof(int));
2394 
2395   // Get the ids of all lgroups with memory; r is the count.
2396   int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2397                                   (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2398   if (r <= 0) {
2399     return 0;
2400   }
2401   return ids[os::random() % r];
2402 }
2403 
2404 // Request information about the page.
2405 bool os::get_page_info(char *start, page_info* info) {
2406   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2407   uint64_t addr = (uintptr_t)start;
2408   uint64_t outdata[2];
2409   uint_t validity = 0;
2410 
2411   if (meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2412     return false;
2413   }
2414 
2415   info->size = 0;
2416   info->lgrp_id = -1;
2417 
2418   if ((validity & 1) != 0) {
2419     if ((validity & 2) != 0) {
2420       info->lgrp_id = outdata[0];
2421     }
2422     if ((validity & 4) != 0) {
2423       info->size = outdata[1];
2424     }
2425     return true;
2426   }
2427   return false;
2428 }
2429 
2430 // Scan the pages from start to end until a page different than
2431 // the one described in the info parameter is encountered.
2432 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2433                      page_info* page_found) {
2434   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2435   const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2436   uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT + 1];
2437   uint_t validity[MAX_MEMINFO_CNT];
2438 
2439   size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2440   uint64_t p = (uint64_t)start;
2441   while (p < (uint64_t)end) {
2442     addrs[0] = p;
2443     size_t addrs_count = 1;
2444     while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] + page_size < (uint64_t)end) {
2445       addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2446       addrs_count++;
2447     }
2448 
2449     if (meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2450       return NULL;
2451     }
2452 
2453     size_t i = 0;
2454     for (; i < addrs_count; i++) {
2455       if ((validity[i] & 1) != 0) {
2456         if ((validity[i] & 4) != 0) {
2457           if (outdata[types * i + 1] != page_expected->size) {
2458             break;
2459           }
2460         } else if (page_expected->size != 0) {
2461           break;
2462         }
2463 
2464         if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2465           if (outdata[types * i] != page_expected->lgrp_id) {
2466             break;
2467           }
2468         }
2469       } else {
2470         return NULL;
2471       }
2472     }
2473 
2474     if (i < addrs_count) {
2475       if ((validity[i] & 2) != 0) {
2476         page_found->lgrp_id = outdata[types * i];
2477       } else {
2478         page_found->lgrp_id = -1;
2479       }
2480       if ((validity[i] & 4) != 0) {
2481         page_found->size = outdata[types * i + 1];
2482       } else {
2483         page_found->size = 0;
2484       }
2485       return (char*)addrs[i];
2486     }
2487 
2488     p = addrs[addrs_count - 1] + page_size;
2489   }
2490   return end;
2491 }
2492 
2493 bool os::pd_uncommit_memory(char* addr, size_t bytes) {
2494   size_t size = bytes;
2495   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2496   // uncommitted page. Otherwise, the read/write might succeed if we
2497   // have enough swap space to back the physical page.
2498   return
2499     NULL != Solaris::mmap_chunk(addr, size,
2500                                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
2501                                 PROT_NONE);
2502 }
2503 
2504 char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
2505   char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
2506 
2507   if (b == MAP_FAILED) {
2508     return NULL;
2509   }
2510   return b;
2511 }
2512 
2513 char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes,
2514                              size_t alignment_hint, bool fixed) {
2515   char* addr = requested_addr;
2516   int flags = MAP_PRIVATE | MAP_NORESERVE;
2517 
2518   assert(!(fixed && (alignment_hint > 0)),
2519          "alignment hint meaningless with fixed mmap");
2520 
2521   if (fixed) {
2522     flags |= MAP_FIXED;
2523   } else if (alignment_hint > (size_t) vm_page_size()) {
2524     flags |= MAP_ALIGN;
2525     addr = (char*) alignment_hint;
2526   }
2527 
2528   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2529   // uncommitted page. Otherwise, the read/write might succeed if we
2530   // have enough swap space to back the physical page.
2531   return mmap_chunk(addr, bytes, flags, PROT_NONE);
2532 }
2533 
2534 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2535                             size_t alignment_hint) {
2536   char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint,
2537                                   (requested_addr != NULL));
2538 
2539   guarantee(requested_addr == NULL || requested_addr == addr,
2540             "OS failed to return requested mmap address.");
2541   return addr;
2542 }
2543 
2544 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
2545   assert(file_desc >= 0, "file_desc is not valid");
2546   char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
2547   if (result != NULL) {
2548     if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
2549       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
2550     }
2551   }
2552   return result;
2553 }
2554 
2555 // Reserve memory at an arbitrary address, only if that area is
2556 // available (and not reserved for something else).
2557 
2558 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2559   const int max_tries = 10;
2560   char* base[max_tries];
2561   size_t size[max_tries];
2562 
2563   // Solaris adds a gap between mmap'ed regions.  The size of the gap
2564   // is dependent on the requested size and the MMU.  Our initial gap
2565   // value here is just a guess and will be corrected later.
2566   bool had_top_overlap = false;
2567   bool have_adjusted_gap = false;
2568   size_t gap = 0x400000;
2569 
2570   // Assert only that the size is a multiple of the page size, since
2571   // that's all that mmap requires, and since that's all we really know
2572   // about at this low abstraction level.  If we need higher alignment,
2573   // we can either pass an alignment to this method or verify alignment
2574   // in one of the methods further up the call chain.  See bug 5044738.
2575   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2576 
2577   // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
2578   // Give it a try, if the kernel honors the hint we can return immediately.
2579   char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
2580 
2581   volatile int err = errno;
2582   if (addr == requested_addr) {
2583     return addr;
2584   } else if (addr != NULL) {
2585     pd_unmap_memory(addr, bytes);
2586   }
2587 
2588   if (log_is_enabled(Warning, os)) {
2589     char buf[256];
2590     buf[0] = '\0';
2591     if (addr == NULL) {
2592       jio_snprintf(buf, sizeof(buf), ": %s", os::strerror(err));
2593     }
2594     log_info(os)("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
2595             PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
2596             "%s", bytes, requested_addr, addr, buf);
2597   }
2598 
2599   // Address hint method didn't work.  Fall back to the old method.
2600   // In theory, once SNV becomes our oldest supported platform, this
2601   // code will no longer be needed.
2602   //
2603   // Repeatedly allocate blocks until the block is allocated at the
2604   // right spot. Give up after max_tries.
2605   int i;
2606   for (i = 0; i < max_tries; ++i) {
2607     base[i] = reserve_memory(bytes);
2608 
2609     if (base[i] != NULL) {
2610       // Is this the block we wanted?
2611       if (base[i] == requested_addr) {
2612         size[i] = bytes;
2613         break;
2614       }
2615 
2616       // check that the gap value is right
2617       if (had_top_overlap && !have_adjusted_gap) {
2618         size_t actual_gap = base[i-1] - base[i] - bytes;
2619         if (gap != actual_gap) {
2620           // adjust the gap value and retry the last 2 allocations
2621           assert(i > 0, "gap adjustment code problem");
2622           have_adjusted_gap = true;  // adjust the gap only once, just in case
2623           gap = actual_gap;
2624           log_info(os)("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
2625           unmap_memory(base[i], bytes);
2626           unmap_memory(base[i-1], size[i-1]);
2627           i-=2;
2628           continue;
2629         }
2630       }
2631 
2632       // Does this overlap the block we wanted? Give back the overlapped
2633       // parts and try again.
2634       //
2635       // There is still a bug in this code: if top_overlap == bytes,
2636       // the overlap is offset from requested region by the value of gap.
2637       // In this case giving back the overlapped part will not work,
2638       // because we'll give back the entire block at base[i] and
2639       // therefore the subsequent allocation will not generate a new gap.
2640       // This could be fixed with a new algorithm that used larger
2641       // or variable size chunks to find the requested region -
2642       // but such a change would introduce additional complications.
2643       // It's rare enough that the planets align for this bug,
2644       // so we'll just wait for a fix for 6204603/5003415 which
2645       // will provide a mmap flag to allow us to avoid this business.
2646 
2647       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2648       if (top_overlap >= 0 && top_overlap < bytes) {
2649         had_top_overlap = true;
2650         unmap_memory(base[i], top_overlap);
2651         base[i] += top_overlap;
2652         size[i] = bytes - top_overlap;
2653       } else {
2654         size_t bottom_overlap = base[i] + bytes - requested_addr;
2655         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2656           if (bottom_overlap == 0) {
2657             log_info(os)("attempt_reserve_memory_at: possible alignment bug");
2658           }
2659           unmap_memory(requested_addr, bottom_overlap);
2660           size[i] = bytes - bottom_overlap;
2661         } else {
2662           size[i] = bytes;
2663         }
2664       }
2665     }
2666   }
2667 
2668   // Give back the unused reserved pieces.
2669 
2670   for (int j = 0; j < i; ++j) {
2671     if (base[j] != NULL) {
2672       unmap_memory(base[j], size[j]);
2673     }
2674   }
2675 
2676   return (i < max_tries) ? requested_addr : NULL;
2677 }
2678 
2679 bool os::pd_release_memory(char* addr, size_t bytes) {
2680   size_t size = bytes;
2681   return munmap(addr, size) == 0;
2682 }
2683 
2684 static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
2685   assert(addr == (char*)align_down((uintptr_t)addr, os::vm_page_size()),
2686          "addr must be page aligned");
2687   int retVal = mprotect(addr, bytes, prot);
2688   return retVal == 0;
2689 }
2690 
2691 // Protect memory (Used to pass readonly pages through
2692 // JNI GetArray<type>Elements with empty arrays.)
2693 // Also, used for serialization page and for compressed oops null pointer
2694 // checking.
2695 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2696                         bool is_committed) {
2697   unsigned int p = 0;
2698   switch (prot) {
2699   case MEM_PROT_NONE: p = PROT_NONE; break;
2700   case MEM_PROT_READ: p = PROT_READ; break;
2701   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2702   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2703   default:
2704     ShouldNotReachHere();
2705   }
2706   // is_committed is unused.
2707   return solaris_mprotect(addr, bytes, p);
2708 }
2709 
2710 // guard_memory and unguard_memory only happens within stack guard pages.
2711 // Since ISM pertains only to the heap, guard and unguard memory should not
2712 /// happen with an ISM region.
2713 bool os::guard_memory(char* addr, size_t bytes) {
2714   return solaris_mprotect(addr, bytes, PROT_NONE);
2715 }
2716 
2717 bool os::unguard_memory(char* addr, size_t bytes) {
2718   return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
2719 }
2720 
2721 // Large page support
2722 static size_t _large_page_size = 0;
2723 
2724 // Insertion sort for small arrays (descending order).
2725 static void insertion_sort_descending(size_t* array, int len) {
2726   for (int i = 0; i < len; i++) {
2727     size_t val = array[i];
2728     for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
2729       size_t tmp = array[key];
2730       array[key] = array[key - 1];
2731       array[key - 1] = tmp;
2732     }
2733   }
2734 }
2735 
2736 bool os::Solaris::mpss_sanity_check(bool warn, size_t* page_size) {
2737   const unsigned int usable_count = VM_Version::page_size_count();
2738   if (usable_count == 1) {
2739     return false;
2740   }
2741 
2742   // Find the right getpagesizes interface.  When solaris 11 is the minimum
2743   // build platform, getpagesizes() (without the '2') can be called directly.
2744   typedef int (*gps_t)(size_t[], int);
2745   gps_t gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes2"));
2746   if (gps_func == NULL) {
2747     gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes"));
2748     if (gps_func == NULL) {
2749       if (warn) {
2750         warning("MPSS is not supported by the operating system.");
2751       }
2752       return false;
2753     }
2754   }
2755 
2756   // Fill the array of page sizes.
2757   int n = (*gps_func)(_page_sizes, page_sizes_max);
2758   assert(n > 0, "Solaris bug?");
2759 
2760   if (n == page_sizes_max) {
2761     // Add a sentinel value (necessary only if the array was completely filled
2762     // since it is static (zeroed at initialization)).
2763     _page_sizes[--n] = 0;
2764     DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
2765   }
2766   assert(_page_sizes[n] == 0, "missing sentinel");
2767   trace_page_sizes("available page sizes", _page_sizes, n);
2768 
2769   if (n == 1) return false;     // Only one page size available.
2770 
2771   // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
2772   // select up to usable_count elements.  First sort the array, find the first
2773   // acceptable value, then copy the usable sizes to the top of the array and
2774   // trim the rest.  Make sure to include the default page size :-).
2775   //
2776   // A better policy could get rid of the 4M limit by taking the sizes of the
2777   // important VM memory regions (java heap and possibly the code cache) into
2778   // account.
2779   insertion_sort_descending(_page_sizes, n);
2780   const size_t size_limit =
2781     FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
2782   int beg;
2783   for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */;
2784   const int end = MIN2((int)usable_count, n) - 1;
2785   for (int cur = 0; cur < end; ++cur, ++beg) {
2786     _page_sizes[cur] = _page_sizes[beg];
2787   }
2788   _page_sizes[end] = vm_page_size();
2789   _page_sizes[end + 1] = 0;
2790 
2791   if (_page_sizes[end] > _page_sizes[end - 1]) {
2792     // Default page size is not the smallest; sort again.
2793     insertion_sort_descending(_page_sizes, end + 1);
2794   }
2795   *page_size = _page_sizes[0];
2796 
2797   trace_page_sizes("usable page sizes", _page_sizes, end + 1);
2798   return true;
2799 }
2800 
2801 void os::large_page_init() {
2802   if (UseLargePages) {
2803     // print a warning if any large page related flag is specified on command line
2804     bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
2805                            !FLAG_IS_DEFAULT(LargePageSizeInBytes);
2806 
2807     UseLargePages = Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
2808   }
2809 }
2810 
2811 bool os::Solaris::is_valid_page_size(size_t bytes) {
2812   for (int i = 0; _page_sizes[i] != 0; i++) {
2813     if (_page_sizes[i] == bytes) {
2814       return true;
2815     }
2816   }
2817   return false;
2818 }
2819 
2820 bool os::Solaris::setup_large_pages(caddr_t start, size_t bytes, size_t align) {
2821   assert(is_valid_page_size(align), SIZE_FORMAT " is not a valid page size", align);
2822   assert(is_aligned((void*) start, align),
2823          PTR_FORMAT " is not aligned to " SIZE_FORMAT, p2i((void*) start), align);
2824   assert(is_aligned(bytes, align),
2825          SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, align);
2826 
2827   // Signal to OS that we want large pages for addresses
2828   // from addr, addr + bytes
2829   struct memcntl_mha mpss_struct;
2830   mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
2831   mpss_struct.mha_pagesize = align;
2832   mpss_struct.mha_flags = 0;
2833   // Upon successful completion, memcntl() returns 0
2834   if (memcntl(start, bytes, MC_HAT_ADVISE, (caddr_t) &mpss_struct, 0, 0)) {
2835     debug_only(warning("Attempt to use MPSS failed."));
2836     return false;
2837   }
2838   return true;
2839 }
2840 
2841 char* os::reserve_memory_special(size_t size, size_t alignment, char* addr, bool exec) {
2842   fatal("os::reserve_memory_special should not be called on Solaris.");
2843   return NULL;
2844 }
2845 
2846 bool os::release_memory_special(char* base, size_t bytes) {
2847   fatal("os::release_memory_special should not be called on Solaris.");
2848   return false;
2849 }
2850 
2851 size_t os::large_page_size() {
2852   return _large_page_size;
2853 }
2854 
2855 // MPSS allows application to commit large page memory on demand; with ISM
2856 // the entire memory region must be allocated as shared memory.
2857 bool os::can_commit_large_page_memory() {
2858   return true;
2859 }
2860 
2861 bool os::can_execute_large_page_memory() {
2862   return true;
2863 }
2864 
2865 // Read calls from inside the vm need to perform state transitions
2866 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2867   size_t res;
2868   JavaThread* thread = (JavaThread*)Thread::current();
2869   assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
2870   ThreadBlockInVM tbiv(thread);
2871   RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
2872   return res;
2873 }
2874 
2875 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
2876   size_t res;
2877   JavaThread* thread = (JavaThread*)Thread::current();
2878   assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
2879   ThreadBlockInVM tbiv(thread);
2880   RESTARTABLE(::pread(fd, buf, (size_t) nBytes, offset), res);
2881   return res;
2882 }
2883 
2884 size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
2885   size_t res;
2886   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
2887          "Assumed _thread_in_native");
2888   RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
2889   return res;
2890 }
2891 
2892 void os::naked_short_sleep(jlong ms) {
2893   assert(ms < 1000, "Un-interruptable sleep, short time use only");
2894 
2895   // usleep is deprecated and removed from POSIX, in favour of nanosleep, but
2896   // Solaris requires -lrt for this.
2897   usleep((ms * 1000));
2898 
2899   return;
2900 }
2901 
2902 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2903 void os::infinite_sleep() {
2904   while (true) {    // sleep forever ...
2905     ::sleep(100);   // ... 100 seconds at a time
2906   }
2907 }
2908 
2909 // Used to convert frequent JVM_Yield() to nops
2910 bool os::dont_yield() {
2911   if (DontYieldALot) {
2912     static hrtime_t last_time = 0;
2913     hrtime_t diff = getTimeNanos() - last_time;
2914 
2915     if (diff < DontYieldALotInterval * 1000000) {
2916       return true;
2917     }
2918 
2919     last_time += diff;
2920 
2921     return false;
2922   } else {
2923     return false;
2924   }
2925 }
2926 
2927 // Note that yield semantics are defined by the scheduling class to which
2928 // the thread currently belongs.  Typically, yield will _not yield to
2929 // other equal or higher priority threads that reside on the dispatch queues
2930 // of other CPUs.
2931 
2932 void os::naked_yield() {
2933   thr_yield();
2934 }
2935 
2936 // Interface for setting lwp priorities.  We are using T2 libthread,
2937 // which forces the use of bound threads, so all of our threads will
2938 // be assigned to real lwp's.  Using the thr_setprio function is
2939 // meaningless in this mode so we must adjust the real lwp's priority.
2940 // The routines below implement the getting and setting of lwp priorities.
2941 //
2942 // Note: There are three priority scales used on Solaris.  Java priotities
2943 //       which range from 1 to 10, libthread "thr_setprio" scale which range
2944 //       from 0 to 127, and the current scheduling class of the process we
2945 //       are running in.  This is typically from -60 to +60.
2946 //       The setting of the lwp priorities in done after a call to thr_setprio
2947 //       so Java priorities are mapped to libthread priorities and we map from
2948 //       the latter to lwp priorities.  We don't keep priorities stored in
2949 //       Java priorities since some of our worker threads want to set priorities
2950 //       higher than all Java threads.
2951 //
2952 // For related information:
2953 // (1)  man -s 2 priocntl
2954 // (2)  man -s 4 priocntl
2955 // (3)  man dispadmin
2956 // =    librt.so
2957 // =    libthread/common/rtsched.c - thrp_setlwpprio().
2958 // =    ps -cL <pid> ... to validate priority.
2959 // =    sched_get_priority_min and _max
2960 //              pthread_create
2961 //              sched_setparam
2962 //              pthread_setschedparam
2963 //
2964 // Assumptions:
2965 // +    We assume that all threads in the process belong to the same
2966 //              scheduling class.   IE. an homogenous process.
2967 // +    Must be root or in IA group to change change "interactive" attribute.
2968 //              Priocntl() will fail silently.  The only indication of failure is when
2969 //              we read-back the value and notice that it hasn't changed.
2970 // +    Interactive threads enter the runq at the head, non-interactive at the tail.
2971 // +    For RT, change timeslice as well.  Invariant:
2972 //              constant "priority integral"
2973 //              Konst == TimeSlice * (60-Priority)
2974 //              Given a priority, compute appropriate timeslice.
2975 // +    Higher numerical values have higher priority.
2976 
2977 // sched class attributes
2978 typedef struct {
2979   int   schedPolicy;              // classID
2980   int   maxPrio;
2981   int   minPrio;
2982 } SchedInfo;
2983 
2984 
2985 static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
2986 
2987 #ifdef ASSERT
2988 static int  ReadBackValidate = 1;
2989 #endif
2990 static int  myClass     = 0;
2991 static int  myMin       = 0;
2992 static int  myMax       = 0;
2993 static int  myCur       = 0;
2994 static bool priocntl_enable = false;
2995 
2996 static const int criticalPrio = FXCriticalPriority;
2997 static int java_MaxPriority_to_os_priority = 0; // Saved mapping
2998 
2999 
3000 // lwp_priocntl_init
3001 //
3002 // Try to determine the priority scale for our process.
3003 //
3004 // Return errno or 0 if OK.
3005 //
3006 static int lwp_priocntl_init() {
3007   int rslt;
3008   pcinfo_t ClassInfo;
3009   pcparms_t ParmInfo;
3010   int i;
3011 
3012   if (!UseThreadPriorities) return 0;
3013 
3014   // If ThreadPriorityPolicy is 1, switch tables
3015   if (ThreadPriorityPolicy == 1) {
3016     for (i = 0; i < CriticalPriority+1; i++)
3017       os::java_to_os_priority[i] = prio_policy1[i];
3018   }
3019   if (UseCriticalJavaThreadPriority) {
3020     // MaxPriority always maps to the FX scheduling class and criticalPrio.
3021     // See set_native_priority() and set_lwp_class_and_priority().
3022     // Save original MaxPriority mapping in case attempt to
3023     // use critical priority fails.
3024     java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
3025     // Set negative to distinguish from other priorities
3026     os::java_to_os_priority[MaxPriority] = -criticalPrio;
3027   }
3028 
3029   // Get IDs for a set of well-known scheduling classes.
3030   // TODO-FIXME: GETCLINFO returns the current # of classes in the
3031   // the system.  We should have a loop that iterates over the
3032   // classID values, which are known to be "small" integers.
3033 
3034   strcpy(ClassInfo.pc_clname, "TS");
3035   ClassInfo.pc_cid = -1;
3036   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3037   if (rslt < 0) return errno;
3038   assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3039   tsLimits.schedPolicy = ClassInfo.pc_cid;
3040   tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3041   tsLimits.minPrio = -tsLimits.maxPrio;
3042 
3043   strcpy(ClassInfo.pc_clname, "IA");
3044   ClassInfo.pc_cid = -1;
3045   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3046   if (rslt < 0) return errno;
3047   assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3048   iaLimits.schedPolicy = ClassInfo.pc_cid;
3049   iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3050   iaLimits.minPrio = -iaLimits.maxPrio;
3051 
3052   strcpy(ClassInfo.pc_clname, "RT");
3053   ClassInfo.pc_cid = -1;
3054   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3055   if (rslt < 0) return errno;
3056   assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3057   rtLimits.schedPolicy = ClassInfo.pc_cid;
3058   rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3059   rtLimits.minPrio = 0;
3060 
3061   strcpy(ClassInfo.pc_clname, "FX");
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 FX class is -1");
3066   fxLimits.schedPolicy = ClassInfo.pc_cid;
3067   fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
3068   fxLimits.minPrio = 0;
3069 
3070   // Query our "current" scheduling class.
3071   // This will normally be IA, TS or, rarely, FX or RT.
3072   memset(&ParmInfo, 0, sizeof(ParmInfo));
3073   ParmInfo.pc_cid = PC_CLNULL;
3074   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3075   if (rslt < 0) return errno;
3076   myClass = ParmInfo.pc_cid;
3077 
3078   // We now know our scheduling classId, get specific information
3079   // about the class.
3080   ClassInfo.pc_cid = myClass;
3081   ClassInfo.pc_clname[0] = 0;
3082   rslt = priocntl((idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
3083   if (rslt < 0) return errno;
3084 
3085   if (ThreadPriorityVerbose) {
3086     tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3087   }
3088 
3089   memset(&ParmInfo, 0, sizeof(pcparms_t));
3090   ParmInfo.pc_cid = PC_CLNULL;
3091   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3092   if (rslt < 0) return errno;
3093 
3094   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3095     myMin = rtLimits.minPrio;
3096     myMax = rtLimits.maxPrio;
3097   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3098     iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3099     myMin = iaLimits.minPrio;
3100     myMax = iaLimits.maxPrio;
3101     myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3102   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3103     tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3104     myMin = tsLimits.minPrio;
3105     myMax = tsLimits.maxPrio;
3106     myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3107   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3108     fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
3109     myMin = fxLimits.minPrio;
3110     myMax = fxLimits.maxPrio;
3111     myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
3112   } else {
3113     // No clue - punt
3114     if (ThreadPriorityVerbose) {
3115       tty->print_cr("Unknown scheduling class: %s ... \n",
3116                     ClassInfo.pc_clname);
3117     }
3118     return EINVAL;      // no clue, punt
3119   }
3120 
3121   if (ThreadPriorityVerbose) {
3122     tty->print_cr("Thread priority Range: [%d..%d]\n", myMin, myMax);
3123   }
3124 
3125   priocntl_enable = true;  // Enable changing priorities
3126   return 0;
3127 }
3128 
3129 #define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3130 #define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3131 #define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3132 #define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
3133 
3134 
3135 // scale_to_lwp_priority
3136 //
3137 // Convert from the libthread "thr_setprio" scale to our current
3138 // lwp scheduling class scale.
3139 //
3140 static int scale_to_lwp_priority(int rMin, int rMax, int x) {
3141   int v;
3142 
3143   if (x == 127) return rMax;            // avoid round-down
3144   v = (((x*(rMax-rMin)))/128)+rMin;
3145   return v;
3146 }
3147 
3148 
3149 // set_lwp_class_and_priority
3150 int set_lwp_class_and_priority(int ThreadID, int lwpid,
3151                                int newPrio, int new_class, bool scale) {
3152   int rslt;
3153   int Actual, Expected, prv;
3154   pcparms_t ParmInfo;                   // for GET-SET
3155 #ifdef ASSERT
3156   pcparms_t ReadBack;                   // for readback
3157 #endif
3158 
3159   // Set priority via PC_GETPARMS, update, PC_SETPARMS
3160   // Query current values.
3161   // TODO: accelerate this by eliminating the PC_GETPARMS call.
3162   // Cache "pcparms_t" in global ParmCache.
3163   // TODO: elide set-to-same-value
3164 
3165   // If something went wrong on init, don't change priorities.
3166   if (!priocntl_enable) {
3167     if (ThreadPriorityVerbose) {
3168       tty->print_cr("Trying to set priority but init failed, ignoring");
3169     }
3170     return EINVAL;
3171   }
3172 
3173   // If lwp hasn't started yet, just return
3174   // the _start routine will call us again.
3175   if (lwpid <= 0) {
3176     if (ThreadPriorityVerbose) {
3177       tty->print_cr("deferring the set_lwp_class_and_priority of thread "
3178                     INTPTR_FORMAT " to %d, lwpid not set",
3179                     ThreadID, newPrio);
3180     }
3181     return 0;
3182   }
3183 
3184   if (ThreadPriorityVerbose) {
3185     tty->print_cr ("set_lwp_class_and_priority("
3186                    INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3187                    ThreadID, lwpid, newPrio);
3188   }
3189 
3190   memset(&ParmInfo, 0, sizeof(pcparms_t));
3191   ParmInfo.pc_cid = PC_CLNULL;
3192   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3193   if (rslt < 0) return errno;
3194 
3195   int cur_class = ParmInfo.pc_cid;
3196   ParmInfo.pc_cid = (id_t)new_class;
3197 
3198   if (new_class == rtLimits.schedPolicy) {
3199     rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3200     rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
3201                                                        rtLimits.maxPrio, newPrio)
3202                                : newPrio;
3203     rtInfo->rt_tqsecs  = RT_NOCHANGE;
3204     rtInfo->rt_tqnsecs = RT_NOCHANGE;
3205     if (ThreadPriorityVerbose) {
3206       tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3207     }
3208   } else if (new_class == iaLimits.schedPolicy) {
3209     iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3210     int maxClamped     = MIN2(iaLimits.maxPrio,
3211                               cur_class == new_class
3212                               ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
3213     iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
3214                                                        maxClamped, newPrio)
3215                                : newPrio;
3216     iaInfo->ia_uprilim = cur_class == new_class
3217                            ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
3218     iaInfo->ia_mode    = IA_NOCHANGE;
3219     if (ThreadPriorityVerbose) {
3220       tty->print_cr("IA: [%d...%d] %d->%d\n",
3221                     iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3222     }
3223   } else if (new_class == tsLimits.schedPolicy) {
3224     tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3225     int maxClamped     = MIN2(tsLimits.maxPrio,
3226                               cur_class == new_class
3227                               ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
3228     tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
3229                                                        maxClamped, newPrio)
3230                                : newPrio;
3231     tsInfo->ts_uprilim = cur_class == new_class
3232                            ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
3233     if (ThreadPriorityVerbose) {
3234       tty->print_cr("TS: [%d...%d] %d->%d\n",
3235                     tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3236     }
3237   } else if (new_class == fxLimits.schedPolicy) {
3238     fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
3239     int maxClamped     = MIN2(fxLimits.maxPrio,
3240                               cur_class == new_class
3241                               ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
3242     fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
3243                                                        maxClamped, newPrio)
3244                                : newPrio;
3245     fxInfo->fx_uprilim = cur_class == new_class
3246                            ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
3247     fxInfo->fx_tqsecs  = FX_NOCHANGE;
3248     fxInfo->fx_tqnsecs = FX_NOCHANGE;
3249     if (ThreadPriorityVerbose) {
3250       tty->print_cr("FX: [%d...%d] %d->%d\n",
3251                     fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
3252     }
3253   } else {
3254     if (ThreadPriorityVerbose) {
3255       tty->print_cr("Unknown new scheduling class %d\n", new_class);
3256     }
3257     return EINVAL;    // no clue, punt
3258   }
3259 
3260   rslt = priocntl(P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3261   if (ThreadPriorityVerbose && rslt) {
3262     tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3263   }
3264   if (rslt < 0) return errno;
3265 
3266 #ifdef ASSERT
3267   // Sanity check: read back what we just attempted to set.
3268   // In theory it could have changed in the interim ...
3269   //
3270   // The priocntl system call is tricky.
3271   // Sometimes it'll validate the priority value argument and
3272   // return EINVAL if unhappy.  At other times it fails silently.
3273   // Readbacks are prudent.
3274 
3275   if (!ReadBackValidate) return 0;
3276 
3277   memset(&ReadBack, 0, sizeof(pcparms_t));
3278   ReadBack.pc_cid = PC_CLNULL;
3279   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3280   assert(rslt >= 0, "priocntl failed");
3281   Actual = Expected = 0xBAD;
3282   assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3283   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3284     Actual   = RTPRI(ReadBack)->rt_pri;
3285     Expected = RTPRI(ParmInfo)->rt_pri;
3286   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3287     Actual   = IAPRI(ReadBack)->ia_upri;
3288     Expected = IAPRI(ParmInfo)->ia_upri;
3289   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3290     Actual   = TSPRI(ReadBack)->ts_upri;
3291     Expected = TSPRI(ParmInfo)->ts_upri;
3292   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3293     Actual   = FXPRI(ReadBack)->fx_upri;
3294     Expected = FXPRI(ParmInfo)->fx_upri;
3295   } else {
3296     if (ThreadPriorityVerbose) {
3297       tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
3298                     ParmInfo.pc_cid);
3299     }
3300   }
3301 
3302   if (Actual != Expected) {
3303     if (ThreadPriorityVerbose) {
3304       tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
3305                      lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
3306     }
3307   }
3308 #endif
3309 
3310   return 0;
3311 }
3312 
3313 // Solaris only gives access to 128 real priorities at a time,
3314 // so we expand Java's ten to fill this range.  This would be better
3315 // if we dynamically adjusted relative priorities.
3316 //
3317 // The ThreadPriorityPolicy option allows us to select 2 different
3318 // priority scales.
3319 //
3320 // ThreadPriorityPolicy=0
3321 // Since the Solaris' default priority is MaximumPriority, we do not
3322 // set a priority lower than Max unless a priority lower than
3323 // NormPriority is requested.
3324 //
3325 // ThreadPriorityPolicy=1
3326 // This mode causes the priority table to get filled with
3327 // linear values.  NormPriority get's mapped to 50% of the
3328 // Maximum priority an so on.  This will cause VM threads
3329 // to get unfair treatment against other Solaris processes
3330 // which do not explicitly alter their thread priorities.
3331 
3332 int os::java_to_os_priority[CriticalPriority + 1] = {
3333   -99999,         // 0 Entry should never be used
3334 
3335   0,              // 1 MinPriority
3336   32,             // 2
3337   64,             // 3
3338 
3339   96,             // 4
3340   127,            // 5 NormPriority
3341   127,            // 6
3342 
3343   127,            // 7
3344   127,            // 8
3345   127,            // 9 NearMaxPriority
3346 
3347   127,            // 10 MaxPriority
3348 
3349   -criticalPrio   // 11 CriticalPriority
3350 };
3351 
3352 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3353   OSThread* osthread = thread->osthread();
3354 
3355   // Save requested priority in case the thread hasn't been started
3356   osthread->set_native_priority(newpri);
3357 
3358   // Check for critical priority request
3359   bool fxcritical = false;
3360   if (newpri == -criticalPrio) {
3361     fxcritical = true;
3362     newpri = criticalPrio;
3363   }
3364 
3365   assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
3366   if (!UseThreadPriorities) return OS_OK;
3367 
3368   int status = 0;
3369 
3370   if (!fxcritical) {
3371     // Use thr_setprio only if we have a priority that thr_setprio understands
3372     status = thr_setprio(thread->osthread()->thread_id(), newpri);
3373   }
3374 
3375   int lwp_status =
3376           set_lwp_class_and_priority(osthread->thread_id(),
3377                                      osthread->lwp_id(),
3378                                      newpri,
3379                                      fxcritical ? fxLimits.schedPolicy : myClass,
3380                                      !fxcritical);
3381   if (lwp_status != 0 && fxcritical) {
3382     // Try again, this time without changing the scheduling class
3383     newpri = java_MaxPriority_to_os_priority;
3384     lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
3385                                             osthread->lwp_id(),
3386                                             newpri, myClass, false);
3387   }
3388   status |= lwp_status;
3389   return (status == 0) ? OS_OK : OS_ERR;
3390 }
3391 
3392 
3393 OSReturn os::get_native_priority(const Thread* const thread,
3394                                  int *priority_ptr) {
3395   int p;
3396   if (!UseThreadPriorities) {
3397     *priority_ptr = NormalPriority;
3398     return OS_OK;
3399   }
3400   int status = thr_getprio(thread->osthread()->thread_id(), &p);
3401   if (status != 0) {
3402     return OS_ERR;
3403   }
3404   *priority_ptr = p;
3405   return OS_OK;
3406 }
3407 
3408 
3409 // Hint to the underlying OS that a task switch would not be good.
3410 // Void return because it's a hint and can fail.
3411 void os::hint_no_preempt() {
3412   schedctl_start(schedctl_init());
3413 }
3414 
3415 ////////////////////////////////////////////////////////////////////////////////
3416 // suspend/resume support
3417 
3418 //  The low-level signal-based suspend/resume support is a remnant from the
3419 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3420 //  within hotspot. Currently used by JFR's OSThreadSampler
3421 //
3422 //  The remaining code is greatly simplified from the more general suspension
3423 //  code that used to be used.
3424 //
3425 //  The protocol is quite simple:
3426 //  - suspend:
3427 //      - sends a signal to the target thread
3428 //      - polls the suspend state of the osthread using a yield loop
3429 //      - target thread signal handler (SR_handler) sets suspend state
3430 //        and blocks in sigsuspend until continued
3431 //  - resume:
3432 //      - sets target osthread state to continue
3433 //      - sends signal to end the sigsuspend loop in the SR_handler
3434 //
3435 //  Note that the SR_lock plays no role in this suspend/resume protocol,
3436 //  but is checked for NULL in SR_handler as a thread termination indicator.
3437 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
3438 //
3439 //  Note that resume_clear_context() and suspend_save_context() are needed
3440 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
3441 //  which in part is used by:
3442 //    - Forte Analyzer: AsyncGetCallTrace()
3443 //    - StackBanging: get_frame_at_stack_banging_point()
3444 //    - JFR: get_topframe()-->....-->get_valid_uc_in_signal_handler()
3445 
3446 static void resume_clear_context(OSThread *osthread) {
3447   osthread->set_ucontext(NULL);
3448 }
3449 
3450 static void suspend_save_context(OSThread *osthread, ucontext_t* context) {
3451   osthread->set_ucontext(context);
3452 }
3453 
3454 static PosixSemaphore sr_semaphore;
3455 
3456 void os::Solaris::SR_handler(Thread* thread, ucontext_t* context) {
3457   // Save and restore errno to avoid confusing native code with EINTR
3458   // after sigsuspend.
3459   int old_errno = errno;
3460 
3461   OSThread* osthread = thread->osthread();
3462   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3463 
3464   os::SuspendResume::State current = osthread->sr.state();
3465   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3466     suspend_save_context(osthread, context);
3467 
3468     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3469     os::SuspendResume::State state = osthread->sr.suspended();
3470     if (state == os::SuspendResume::SR_SUSPENDED) {
3471       sigset_t suspend_set;  // signals for sigsuspend()
3472 
3473       // get current set of blocked signals and unblock resume signal
3474       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3475       sigdelset(&suspend_set, ASYNC_SIGNAL);
3476 
3477       sr_semaphore.signal();
3478       // wait here until we are resumed
3479       while (1) {
3480         sigsuspend(&suspend_set);
3481 
3482         os::SuspendResume::State result = osthread->sr.running();
3483         if (result == os::SuspendResume::SR_RUNNING) {
3484           sr_semaphore.signal();
3485           break;
3486         }
3487       }
3488 
3489     } else if (state == os::SuspendResume::SR_RUNNING) {
3490       // request was cancelled, continue
3491     } else {
3492       ShouldNotReachHere();
3493     }
3494 
3495     resume_clear_context(osthread);
3496   } else if (current == os::SuspendResume::SR_RUNNING) {
3497     // request was cancelled, continue
3498   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3499     // ignore
3500   } else {
3501     // ignore
3502   }
3503 
3504   errno = old_errno;
3505 }
3506 
3507 void os::print_statistics() {
3508 }
3509 
3510 bool os::message_box(const char* title, const char* message) {
3511   int i;
3512   fdStream err(defaultStream::error_fd());
3513   for (i = 0; i < 78; i++) err.print_raw("=");
3514   err.cr();
3515   err.print_raw_cr(title);
3516   for (i = 0; i < 78; i++) err.print_raw("-");
3517   err.cr();
3518   err.print_raw_cr(message);
3519   for (i = 0; i < 78; i++) err.print_raw("=");
3520   err.cr();
3521 
3522   char buf[16];
3523   // Prevent process from exiting upon "read error" without consuming all CPU
3524   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3525 
3526   return buf[0] == 'y' || buf[0] == 'Y';
3527 }
3528 
3529 static int sr_notify(OSThread* osthread) {
3530   int status = thr_kill(osthread->thread_id(), ASYNC_SIGNAL);
3531   assert_status(status == 0, status, "thr_kill");
3532   return status;
3533 }
3534 
3535 // "Randomly" selected value for how long we want to spin
3536 // before bailing out on suspending a thread, also how often
3537 // we send a signal to a thread we want to resume
3538 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3539 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3540 
3541 static bool do_suspend(OSThread* osthread) {
3542   assert(osthread->sr.is_running(), "thread should be running");
3543   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3544 
3545   // mark as suspended and send signal
3546   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3547     // failed to switch, state wasn't running?
3548     ShouldNotReachHere();
3549     return false;
3550   }
3551 
3552   if (sr_notify(osthread) != 0) {
3553     ShouldNotReachHere();
3554   }
3555 
3556   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3557   while (true) {
3558     if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2000 * NANOSECS_PER_MILLISEC))) {
3559       break;
3560     } else {
3561       // timeout
3562       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3563       if (cancelled == os::SuspendResume::SR_RUNNING) {
3564         return false;
3565       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3566         // make sure that we consume the signal on the semaphore as well
3567         sr_semaphore.wait();
3568         break;
3569       } else {
3570         ShouldNotReachHere();
3571         return false;
3572       }
3573     }
3574   }
3575 
3576   guarantee(osthread->sr.is_suspended(), "Must be suspended");
3577   return true;
3578 }
3579 
3580 static void do_resume(OSThread* osthread) {
3581   assert(osthread->sr.is_suspended(), "thread should be suspended");
3582   assert(!sr_semaphore.trywait(), "invalid semaphore state");
3583 
3584   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3585     // failed to switch to WAKEUP_REQUEST
3586     ShouldNotReachHere();
3587     return;
3588   }
3589 
3590   while (true) {
3591     if (sr_notify(osthread) == 0) {
3592       if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
3593         if (osthread->sr.is_running()) {
3594           return;
3595         }
3596       }
3597     } else {
3598       ShouldNotReachHere();
3599     }
3600   }
3601 
3602   guarantee(osthread->sr.is_running(), "Must be running!");
3603 }
3604 
3605 void os::SuspendedThreadTask::internal_do_task() {
3606   if (do_suspend(_thread->osthread())) {
3607     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
3608     do_task(context);
3609     do_resume(_thread->osthread());
3610   }
3611 }
3612 
3613 // This does not do anything on Solaris. This is basically a hook for being
3614 // able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
3615 void os::os_exception_wrapper(java_call_t f, JavaValue* value,
3616                               const methodHandle& method, JavaCallArguments* args,
3617                               Thread* thread) {
3618   f(value, method, args, thread);
3619 }
3620 
3621 // This routine may be used by user applications as a "hook" to catch signals.
3622 // The user-defined signal handler must pass unrecognized signals to this
3623 // routine, and if it returns true (non-zero), then the signal handler must
3624 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3625 // routine will never retun false (zero), but instead will execute a VM panic
3626 // routine kill the process.
3627 //
3628 // If this routine returns false, it is OK to call it again.  This allows
3629 // the user-defined signal handler to perform checks either before or after
3630 // the VM performs its own checks.  Naturally, the user code would be making
3631 // a serious error if it tried to handle an exception (such as a null check
3632 // or breakpoint) that the VM was generating for its own correct operation.
3633 //
3634 // This routine may recognize any of the following kinds of signals:
3635 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
3636 // ASYNC_SIGNAL.
3637 // It should be consulted by handlers for any of those signals.
3638 //
3639 // The caller of this routine must pass in the three arguments supplied
3640 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3641 // field of the structure passed to sigaction().  This routine assumes that
3642 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3643 //
3644 // Note that the VM will print warnings if it detects conflicting signal
3645 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3646 //
3647 extern "C" JNIEXPORT int JVM_handle_solaris_signal(int signo,
3648                                                    siginfo_t* siginfo,
3649                                                    void* ucontext,
3650                                                    int abort_if_unrecognized);
3651 
3652 
3653 void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
3654   int orig_errno = errno;  // Preserve errno value over signal handler.
3655   JVM_handle_solaris_signal(sig, info, ucVoid, true);
3656   errno = orig_errno;
3657 }
3658 
3659 // This boolean allows users to forward their own non-matching signals
3660 // to JVM_handle_solaris_signal, harmlessly.
3661 bool os::Solaris::signal_handlers_are_installed = false;
3662 
3663 // For signal-chaining
3664 bool os::Solaris::libjsig_is_loaded = false;
3665 typedef struct sigaction *(*get_signal_t)(int);
3666 get_signal_t os::Solaris::get_signal_action = NULL;
3667 
3668 struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
3669   struct sigaction *actp = NULL;
3670 
3671   if ((libjsig_is_loaded)  && (sig <= Maxsignum)) {
3672     // Retrieve the old signal handler from libjsig
3673     actp = (*get_signal_action)(sig);
3674   }
3675   if (actp == NULL) {
3676     // Retrieve the preinstalled signal handler from jvm
3677     actp = get_preinstalled_handler(sig);
3678   }
3679 
3680   return actp;
3681 }
3682 
3683 static bool call_chained_handler(struct sigaction *actp, int sig,
3684                                  siginfo_t *siginfo, void *context) {
3685   // Call the old signal handler
3686   if (actp->sa_handler == SIG_DFL) {
3687     // It's more reasonable to let jvm treat it as an unexpected exception
3688     // instead of taking the default action.
3689     return false;
3690   } else if (actp->sa_handler != SIG_IGN) {
3691     if ((actp->sa_flags & SA_NODEFER) == 0) {
3692       // automaticlly block the signal
3693       sigaddset(&(actp->sa_mask), sig);
3694     }
3695 
3696     sa_handler_t hand;
3697     sa_sigaction_t sa;
3698     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3699     // retrieve the chained handler
3700     if (siginfo_flag_set) {
3701       sa = actp->sa_sigaction;
3702     } else {
3703       hand = actp->sa_handler;
3704     }
3705 
3706     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3707       actp->sa_handler = SIG_DFL;
3708     }
3709 
3710     // try to honor the signal mask
3711     sigset_t oset;
3712     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3713 
3714     // call into the chained handler
3715     if (siginfo_flag_set) {
3716       (*sa)(sig, siginfo, context);
3717     } else {
3718       (*hand)(sig);
3719     }
3720 
3721     // restore the signal mask
3722     pthread_sigmask(SIG_SETMASK, &oset, 0);
3723   }
3724   // Tell jvm's signal handler the signal is taken care of.
3725   return true;
3726 }
3727 
3728 bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3729   bool chained = false;
3730   // signal-chaining
3731   if (UseSignalChaining) {
3732     struct sigaction *actp = get_chained_signal_action(sig);
3733     if (actp != NULL) {
3734       chained = call_chained_handler(actp, sig, siginfo, context);
3735     }
3736   }
3737   return chained;
3738 }
3739 
3740 struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
3741   assert((chainedsigactions != (struct sigaction *)NULL) &&
3742          (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3743   if (preinstalled_sigs[sig] != 0) {
3744     return &chainedsigactions[sig];
3745   }
3746   return NULL;
3747 }
3748 
3749 void os::Solaris::save_preinstalled_handler(int sig,
3750                                             struct sigaction& oldAct) {
3751   assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
3752   assert((chainedsigactions != (struct sigaction *)NULL) &&
3753          (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3754   chainedsigactions[sig] = oldAct;
3755   preinstalled_sigs[sig] = 1;
3756 }
3757 
3758 void os::Solaris::set_signal_handler(int sig, bool set_installed,
3759                                      bool oktochain) {
3760   // Check for overwrite.
3761   struct sigaction oldAct;
3762   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3763   void* oldhand =
3764       oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3765                           : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3766   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3767       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3768       oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
3769     if (AllowUserSignalHandlers || !set_installed) {
3770       // Do not overwrite; user takes responsibility to forward to us.
3771       return;
3772     } else if (UseSignalChaining) {
3773       if (oktochain) {
3774         // save the old handler in jvm
3775         save_preinstalled_handler(sig, oldAct);
3776       } else {
3777         vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal.");
3778       }
3779       // libjsig also interposes the sigaction() call below and saves the
3780       // old sigaction on it own.
3781     } else {
3782       fatal("Encountered unexpected pre-existing sigaction handler "
3783             "%#lx for signal %d.", (long)oldhand, sig);
3784     }
3785   }
3786 
3787   struct sigaction sigAct;
3788   sigfillset(&(sigAct.sa_mask));
3789   sigAct.sa_handler = SIG_DFL;
3790 
3791   sigAct.sa_sigaction = signalHandler;
3792   // Handle SIGSEGV on alternate signal stack if
3793   // not using stack banging
3794   if (!UseStackBanging && sig == SIGSEGV) {
3795     sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
3796   } else {
3797     sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
3798   }
3799   os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
3800 
3801   sigaction(sig, &sigAct, &oldAct);
3802 
3803   void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3804                                        : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3805   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3806 }
3807 
3808 
3809 #define DO_SIGNAL_CHECK(sig)                      \
3810   do {                                            \
3811     if (!sigismember(&check_signal_done, sig)) {  \
3812       os::Solaris::check_signal_handler(sig);     \
3813     }                                             \
3814   } while (0)
3815 
3816 // This method is a periodic task to check for misbehaving JNI applications
3817 // under CheckJNI, we can add any periodic checks here
3818 
3819 void os::run_periodic_checks() {
3820   // A big source of grief is hijacking virt. addr 0x0 on Solaris,
3821   // thereby preventing a NULL checks.
3822   if (!check_addr0_done) check_addr0_done = check_addr0(tty);
3823 
3824   if (check_signals == false) return;
3825 
3826   // SEGV and BUS if overridden could potentially prevent
3827   // generation of hs*.log in the event of a crash, debugging
3828   // such a case can be very challenging, so we absolutely
3829   // check for the following for a good measure:
3830   DO_SIGNAL_CHECK(SIGSEGV);
3831   DO_SIGNAL_CHECK(SIGILL);
3832   DO_SIGNAL_CHECK(SIGFPE);
3833   DO_SIGNAL_CHECK(SIGBUS);
3834   DO_SIGNAL_CHECK(SIGPIPE);
3835   DO_SIGNAL_CHECK(SIGXFSZ);
3836   DO_SIGNAL_CHECK(ASYNC_SIGNAL);
3837 
3838   // ReduceSignalUsage allows the user to override these handlers
3839   // see comments at the very top and jvm_solaris.h
3840   if (!ReduceSignalUsage) {
3841     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3842     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3843     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3844     DO_SIGNAL_CHECK(BREAK_SIGNAL);
3845   }
3846 }
3847 
3848 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3849 
3850 static os_sigaction_t os_sigaction = NULL;
3851 
3852 void os::Solaris::check_signal_handler(int sig) {
3853   char buf[O_BUFLEN];
3854   address jvmHandler = NULL;
3855 
3856   struct sigaction act;
3857   if (os_sigaction == NULL) {
3858     // only trust the default sigaction, in case it has been interposed
3859     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3860     if (os_sigaction == NULL) return;
3861   }
3862 
3863   os_sigaction(sig, (struct sigaction*)NULL, &act);
3864 
3865   address thisHandler = (act.sa_flags & SA_SIGINFO)
3866     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3867     : CAST_FROM_FN_PTR(address, act.sa_handler);
3868 
3869 
3870   switch (sig) {
3871   case SIGSEGV:
3872   case SIGBUS:
3873   case SIGFPE:
3874   case SIGPIPE:
3875   case SIGXFSZ:
3876   case SIGILL:
3877   case ASYNC_SIGNAL:
3878     jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
3879     break;
3880 
3881   case SHUTDOWN1_SIGNAL:
3882   case SHUTDOWN2_SIGNAL:
3883   case SHUTDOWN3_SIGNAL:
3884   case BREAK_SIGNAL:
3885     jvmHandler = (address)user_handler();
3886     break;
3887 
3888   default:
3889       return;
3890   }
3891 
3892   if (thisHandler != jvmHandler) {
3893     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3894     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3895     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3896     // No need to check this sig any longer
3897     sigaddset(&check_signal_done, sig);
3898     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
3899     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
3900       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
3901                     exception_name(sig, buf, O_BUFLEN));
3902     }
3903   } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
3904     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3905     tty->print("expected:");
3906     os::Posix::print_sa_flags(tty, os::Solaris::get_our_sigflags(sig));
3907     tty->cr();
3908     tty->print("  found:");
3909     os::Posix::print_sa_flags(tty, act.sa_flags);
3910     tty->cr();
3911     // No need to check this sig any longer
3912     sigaddset(&check_signal_done, sig);
3913   }
3914 
3915   // Print all the signal handler state
3916   if (sigismember(&check_signal_done, sig)) {
3917     print_signal_handlers(tty, buf, O_BUFLEN);
3918   }
3919 
3920 }
3921 
3922 void os::Solaris::install_signal_handlers() {
3923   signal_handlers_are_installed = true;
3924 
3925   // signal-chaining
3926   typedef void (*signal_setting_t)();
3927   signal_setting_t begin_signal_setting = NULL;
3928   signal_setting_t end_signal_setting = NULL;
3929   begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3930                                         dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3931   if (begin_signal_setting != NULL) {
3932     end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3933                                         dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3934     get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3935                                        dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3936     libjsig_is_loaded = true;
3937     assert(UseSignalChaining, "should enable signal-chaining");
3938   }
3939   if (libjsig_is_loaded) {
3940     // Tell libjsig jvm is setting signal handlers
3941     (*begin_signal_setting)();
3942   }
3943 
3944   set_signal_handler(SIGSEGV, true, true);
3945   set_signal_handler(SIGPIPE, true, true);
3946   set_signal_handler(SIGXFSZ, true, true);
3947   set_signal_handler(SIGBUS, true, true);
3948   set_signal_handler(SIGILL, true, true);
3949   set_signal_handler(SIGFPE, true, true);
3950   set_signal_handler(ASYNC_SIGNAL, true, true);
3951 
3952   if (libjsig_is_loaded) {
3953     // Tell libjsig jvm finishes setting signal handlers
3954     (*end_signal_setting)();
3955   }
3956 
3957   // We don't activate signal checker if libjsig is in place, we trust ourselves
3958   // and if UserSignalHandler is installed all bets are off.
3959   // Log that signal checking is off only if -verbose:jni is specified.
3960   if (CheckJNICalls) {
3961     if (libjsig_is_loaded) {
3962       if (PrintJNIResolving) {
3963         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3964       }
3965       check_signals = false;
3966     }
3967     if (AllowUserSignalHandlers) {
3968       if (PrintJNIResolving) {
3969         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3970       }
3971       check_signals = false;
3972     }
3973   }
3974 }
3975 
3976 
3977 void report_error(const char* file_name, int line_no, const char* title,
3978                   const char* format, ...);
3979 
3980 // (Static) wrappers for the liblgrp API
3981 os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
3982 os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
3983 os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
3984 os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
3985 os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
3986 os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
3987 os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
3988 os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
3989 os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
3990 
3991 static address resolve_symbol_lazy(const char* name) {
3992   address addr = (address) dlsym(RTLD_DEFAULT, name);
3993   if (addr == NULL) {
3994     // RTLD_DEFAULT was not defined on some early versions of 2.5.1
3995     addr = (address) dlsym(RTLD_NEXT, name);
3996   }
3997   return addr;
3998 }
3999 
4000 static address resolve_symbol(const char* name) {
4001   address addr = resolve_symbol_lazy(name);
4002   if (addr == NULL) {
4003     fatal(dlerror());
4004   }
4005   return addr;
4006 }
4007 
4008 void os::Solaris::libthread_init() {
4009   address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4010 
4011   lwp_priocntl_init();
4012 
4013   // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4014   if (func == NULL) {
4015     func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4016     // Guarantee that this VM is running on an new enough OS (5.6 or
4017     // later) that it will have a new enough libthread.so.
4018     guarantee(func != NULL, "libthread.so is too old.");
4019   }
4020 
4021   int size;
4022   void (*handler_info_func)(address *, int *);
4023   handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4024   handler_info_func(&handler_start, &size);
4025   handler_end = handler_start + size;
4026 }
4027 
4028 
4029 int_fnP_mutex_tP os::Solaris::_mutex_lock;
4030 int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4031 int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4032 int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4033 int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4034 int os::Solaris::_mutex_scope = USYNC_THREAD;
4035 
4036 int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4037 int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4038 int_fnP_cond_tP os::Solaris::_cond_signal;
4039 int_fnP_cond_tP os::Solaris::_cond_broadcast;
4040 int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4041 int_fnP_cond_tP os::Solaris::_cond_destroy;
4042 int os::Solaris::_cond_scope = USYNC_THREAD;
4043 bool os::Solaris::_synchronization_initialized;
4044 
4045 void os::Solaris::synchronization_init() {
4046   if (UseLWPSynchronization) {
4047     os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4048     os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4049     os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4050     os::Solaris::set_mutex_init(lwp_mutex_init);
4051     os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4052     os::Solaris::set_mutex_scope(USYNC_THREAD);
4053 
4054     os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4055     os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4056     os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4057     os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4058     os::Solaris::set_cond_init(lwp_cond_init);
4059     os::Solaris::set_cond_destroy(lwp_cond_destroy);
4060     os::Solaris::set_cond_scope(USYNC_THREAD);
4061   } else {
4062     os::Solaris::set_mutex_scope(USYNC_THREAD);
4063     os::Solaris::set_cond_scope(USYNC_THREAD);
4064 
4065     if (UsePthreads) {
4066       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4067       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4068       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4069       os::Solaris::set_mutex_init(pthread_mutex_default_init);
4070       os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4071 
4072       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4073       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4074       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4075       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4076       os::Solaris::set_cond_init(pthread_cond_default_init);
4077       os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4078     } else {
4079       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4080       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4081       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4082       os::Solaris::set_mutex_init(::mutex_init);
4083       os::Solaris::set_mutex_destroy(::mutex_destroy);
4084 
4085       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4086       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4087       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4088       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4089       os::Solaris::set_cond_init(::cond_init);
4090       os::Solaris::set_cond_destroy(::cond_destroy);
4091     }
4092   }
4093   _synchronization_initialized = true;
4094 }
4095 
4096 bool os::Solaris::liblgrp_init() {
4097   void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4098   if (handle != NULL) {
4099     os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4100     os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4101     os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4102     os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4103     os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4104     os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4105     os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4106     os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4107                                                       dlsym(handle, "lgrp_cookie_stale")));
4108 
4109     lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4110     set_lgrp_cookie(c);
4111     return true;
4112   }
4113   return false;
4114 }
4115 
4116 // int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4117 typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4118 static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4119 
4120 void init_pset_getloadavg_ptr(void) {
4121   pset_getloadavg_ptr =
4122     (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
4123   if (pset_getloadavg_ptr == NULL) {
4124     log_warning(os)("pset_getloadavg function not found");
4125   }
4126 }
4127 
4128 int os::Solaris::_dev_zero_fd = -1;
4129 
4130 // this is called _before_ the global arguments have been parsed
4131 void os::init(void) {
4132   _initial_pid = getpid();
4133 
4134   max_hrtime = first_hrtime = gethrtime();
4135 
4136   init_random(1234567);
4137 
4138   page_size = sysconf(_SC_PAGESIZE);
4139   if (page_size == -1) {
4140     fatal("os_solaris.cpp: os::init: sysconf failed (%s)", os::strerror(errno));
4141   }
4142   init_page_sizes((size_t) page_size);
4143 
4144   Solaris::initialize_system_info();
4145 
4146   int fd = ::open("/dev/zero", O_RDWR);
4147   if (fd < 0) {
4148     fatal("os::init: cannot open /dev/zero (%s)", os::strerror(errno));
4149   } else {
4150     Solaris::set_dev_zero_fd(fd);
4151 
4152     // Close on exec, child won't inherit.
4153     fcntl(fd, F_SETFD, FD_CLOEXEC);
4154   }
4155 
4156   clock_tics_per_sec = CLK_TCK;
4157 
4158   // check if dladdr1() exists; dladdr1 can provide more information than
4159   // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
4160   // and is available on linker patches for 5.7 and 5.8.
4161   // libdl.so must have been loaded, this call is just an entry lookup
4162   void * hdl = dlopen("libdl.so", RTLD_NOW);
4163   if (hdl) {
4164     dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
4165   }
4166 
4167   // main_thread points to the thread that created/loaded the JVM.
4168   main_thread = thr_self();
4169 
4170   // dynamic lookup of functions that may not be available in our lowest
4171   // supported Solaris release
4172   void * handle = dlopen("libc.so.1", RTLD_LAZY);
4173   if (handle != NULL) {
4174     Solaris::_pthread_setname_np =  // from 11.3
4175         (Solaris::pthread_setname_np_func_t)dlsym(handle, "pthread_setname_np");
4176   }
4177 }
4178 
4179 // To install functions for atexit system call
4180 extern "C" {
4181   static void perfMemory_exit_helper() {
4182     perfMemory_exit();
4183   }
4184 }
4185 
4186 // this is called _after_ the global arguments have been parsed
4187 jint os::init_2(void) {
4188   // try to enable extended file IO ASAP, see 6431278
4189   os::Solaris::try_enable_extended_io();
4190 
4191   // Check and sets minimum stack sizes against command line options
4192   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4193     return JNI_ERR;
4194   }
4195 
4196   Solaris::libthread_init();
4197 
4198   if (UseNUMA) {
4199     if (!Solaris::liblgrp_init()) {
4200       UseNUMA = false;
4201     } else {
4202       size_t lgrp_limit = os::numa_get_groups_num();
4203       int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit, mtInternal);
4204       size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
4205       FREE_C_HEAP_ARRAY(int, lgrp_ids);
4206       if (lgrp_num < 2) {
4207         // There's only one locality group, disable NUMA.
4208         UseNUMA = false;
4209       }
4210     }
4211     if (!UseNUMA && ForceNUMA) {
4212       UseNUMA = true;
4213     }
4214   }
4215 
4216   Solaris::signal_sets_init();
4217   Solaris::init_signal_mem();
4218   Solaris::install_signal_handlers();
4219   // Initialize data for jdk.internal.misc.Signal
4220   if (!ReduceSignalUsage) {
4221     jdk_misc_signal_init();
4222   }
4223 
4224   // initialize synchronization primitives to use either thread or
4225   // lwp synchronization (controlled by UseLWPSynchronization)
4226   Solaris::synchronization_init();
4227 
4228   if (MaxFDLimit) {
4229     // set the number of file descriptors to max. print out error
4230     // if getrlimit/setrlimit fails but continue regardless.
4231     struct rlimit nbr_files;
4232     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4233     if (status != 0) {
4234       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4235     } else {
4236       nbr_files.rlim_cur = nbr_files.rlim_max;
4237       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4238       if (status != 0) {
4239         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4240       }
4241     }
4242   }
4243 
4244   // Calculate theoretical max. size of Threads to guard gainst
4245   // artifical out-of-memory situations, where all available address-
4246   // space has been reserved by thread stacks. Default stack size is 1Mb.
4247   size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
4248     JavaThread::stack_size_at_create() : (1*K*K);
4249   assert(pre_thread_stack_size != 0, "Must have a stack");
4250   // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
4251   // we should start doing Virtual Memory banging. Currently when the threads will
4252   // have used all but 200Mb of space.
4253   size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
4254   Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
4255 
4256   // at-exit methods are called in the reverse order of their registration.
4257   // In Solaris 7 and earlier, atexit functions are called on return from
4258   // main or as a result of a call to exit(3C). There can be only 32 of
4259   // these functions registered and atexit() does not set errno. In Solaris
4260   // 8 and later, there is no limit to the number of functions registered
4261   // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
4262   // functions are called upon dlclose(3DL) in addition to return from main
4263   // and exit(3C).
4264 
4265   if (PerfAllowAtExitRegistration) {
4266     // only register atexit functions if PerfAllowAtExitRegistration is set.
4267     // atexit functions can be delayed until process exit time, which
4268     // can be problematic for embedded VM situations. Embedded VMs should
4269     // call DestroyJavaVM() to assure that VM resources are released.
4270 
4271     // note: perfMemory_exit_helper atexit function may be removed in
4272     // the future if the appropriate cleanup code can be added to the
4273     // VM_Exit VMOperation's doit method.
4274     if (atexit(perfMemory_exit_helper) != 0) {
4275       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4276     }
4277   }
4278 
4279   // Init pset_loadavg function pointer
4280   init_pset_getloadavg_ptr();
4281 
4282   return JNI_OK;
4283 }
4284 
4285 // Mark the polling page as unreadable
4286 void os::make_polling_page_unreadable(void) {
4287   if (mprotect((char *)_polling_page, page_size, PROT_NONE) != 0) {
4288     fatal("Could not disable polling page");
4289   }
4290 }
4291 
4292 // Mark the polling page as readable
4293 void os::make_polling_page_readable(void) {
4294   if (mprotect((char *)_polling_page, page_size, PROT_READ) != 0) {
4295     fatal("Could not enable polling page");
4296   }
4297 }
4298 
4299 // Is a (classpath) directory empty?
4300 bool os::dir_is_empty(const char* path) {
4301   DIR *dir = NULL;
4302   struct dirent *ptr;
4303 
4304   dir = opendir(path);
4305   if (dir == NULL) return true;
4306 
4307   // Scan the directory
4308   bool result = true;
4309   while (result && (ptr = readdir(dir)) != NULL) {
4310     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4311       result = false;
4312     }
4313   }
4314   closedir(dir);
4315   return result;
4316 }
4317 
4318 // This code originates from JDK's sysOpen and open64_w
4319 // from src/solaris/hpi/src/system_md.c
4320 
4321 int os::open(const char *path, int oflag, int mode) {
4322   if (strlen(path) > MAX_PATH - 1) {
4323     errno = ENAMETOOLONG;
4324     return -1;
4325   }
4326   int fd;
4327 
4328   fd = ::open64(path, oflag, mode);
4329   if (fd == -1) return -1;
4330 
4331   // If the open succeeded, the file might still be a directory
4332   {
4333     struct stat64 buf64;
4334     int ret = ::fstat64(fd, &buf64);
4335     int st_mode = buf64.st_mode;
4336 
4337     if (ret != -1) {
4338       if ((st_mode & S_IFMT) == S_IFDIR) {
4339         errno = EISDIR;
4340         ::close(fd);
4341         return -1;
4342       }
4343     } else {
4344       ::close(fd);
4345       return -1;
4346     }
4347   }
4348 
4349   // 32-bit Solaris systems suffer from:
4350   //
4351   // - an historical default soft limit of 256 per-process file
4352   //   descriptors that is too low for many Java programs.
4353   //
4354   // - a design flaw where file descriptors created using stdio
4355   //   fopen must be less than 256, _even_ when the first limit above
4356   //   has been raised.  This can cause calls to fopen (but not calls to
4357   //   open, for example) to fail mysteriously, perhaps in 3rd party
4358   //   native code (although the JDK itself uses fopen).  One can hardly
4359   //   criticize them for using this most standard of all functions.
4360   //
4361   // We attempt to make everything work anyways by:
4362   //
4363   // - raising the soft limit on per-process file descriptors beyond
4364   //   256
4365   //
4366   // - As of Solaris 10u4, we can request that Solaris raise the 256
4367   //   stdio fopen limit by calling function enable_extended_FILE_stdio.
4368   //   This is done in init_2 and recorded in enabled_extended_FILE_stdio
4369   //
4370   // - If we are stuck on an old (pre 10u4) Solaris system, we can
4371   //   workaround the bug by remapping non-stdio file descriptors below
4372   //   256 to ones beyond 256, which is done below.
4373   //
4374   // See:
4375   // 1085341: 32-bit stdio routines should support file descriptors >255
4376   // 6533291: Work around 32-bit Solaris stdio limit of 256 open files
4377   // 6431278: Netbeans crash on 32 bit Solaris: need to call
4378   //          enable_extended_FILE_stdio() in VM initialisation
4379   // Giri Mandalika's blog
4380   // http://technopark02.blogspot.com/2005_05_01_archive.html
4381   //
4382 #ifndef  _LP64
4383   if ((!enabled_extended_FILE_stdio) && fd < 256) {
4384     int newfd = ::fcntl(fd, F_DUPFD, 256);
4385     if (newfd != -1) {
4386       ::close(fd);
4387       fd = newfd;
4388     }
4389   }
4390 #endif // 32-bit Solaris
4391 
4392   // All file descriptors that are opened in the JVM and not
4393   // specifically destined for a subprocess should have the
4394   // close-on-exec flag set.  If we don't set it, then careless 3rd
4395   // party native code might fork and exec without closing all
4396   // appropriate file descriptors (e.g. as we do in closeDescriptors in
4397   // UNIXProcess.c), and this in turn might:
4398   //
4399   // - cause end-of-file to fail to be detected on some file
4400   //   descriptors, resulting in mysterious hangs, or
4401   //
4402   // - might cause an fopen in the subprocess to fail on a system
4403   //   suffering from bug 1085341.
4404   //
4405   // (Yes, the default setting of the close-on-exec flag is a Unix
4406   // design flaw)
4407   //
4408   // See:
4409   // 1085341: 32-bit stdio routines should support file descriptors >255
4410   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4411   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4412   //
4413 #ifdef FD_CLOEXEC
4414   {
4415     int flags = ::fcntl(fd, F_GETFD);
4416     if (flags != -1) {
4417       ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4418     }
4419   }
4420 #endif
4421 
4422   return fd;
4423 }
4424 
4425 // create binary file, rewriting existing file if required
4426 int os::create_binary_file(const char* path, bool rewrite_existing) {
4427   int oflags = O_WRONLY | O_CREAT;
4428   if (!rewrite_existing) {
4429     oflags |= O_EXCL;
4430   }
4431   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4432 }
4433 
4434 // return current position of file pointer
4435 jlong os::current_file_offset(int fd) {
4436   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4437 }
4438 
4439 // move file pointer to the specified offset
4440 jlong os::seek_to_file_offset(int fd, jlong offset) {
4441   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4442 }
4443 
4444 jlong os::lseek(int fd, jlong offset, int whence) {
4445   return (jlong) ::lseek64(fd, offset, whence);
4446 }
4447 
4448 int os::ftruncate(int fd, jlong length) {
4449   return ::ftruncate64(fd, length);
4450 }
4451 
4452 int os::fsync(int fd)  {
4453   RESTARTABLE_RETURN_INT(::fsync(fd));
4454 }
4455 
4456 int os::available(int fd, jlong *bytes) {
4457   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
4458          "Assumed _thread_in_native");
4459   jlong cur, end;
4460   int mode;
4461   struct stat64 buf64;
4462 
4463   if (::fstat64(fd, &buf64) >= 0) {
4464     mode = buf64.st_mode;
4465     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4466       int n,ioctl_return;
4467 
4468       RESTARTABLE(::ioctl(fd, FIONREAD, &n), ioctl_return);
4469       if (ioctl_return>= 0) {
4470         *bytes = n;
4471         return 1;
4472       }
4473     }
4474   }
4475   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4476     return 0;
4477   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4478     return 0;
4479   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4480     return 0;
4481   }
4482   *bytes = end - cur;
4483   return 1;
4484 }
4485 
4486 // Map a block of memory.
4487 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4488                         char *addr, size_t bytes, bool read_only,
4489                         bool allow_exec) {
4490   int prot;
4491   int flags;
4492 
4493   if (read_only) {
4494     prot = PROT_READ;
4495     flags = MAP_SHARED;
4496   } else {
4497     prot = PROT_READ | PROT_WRITE;
4498     flags = MAP_PRIVATE;
4499   }
4500 
4501   if (allow_exec) {
4502     prot |= PROT_EXEC;
4503   }
4504 
4505   if (addr != NULL) {
4506     flags |= MAP_FIXED;
4507   }
4508 
4509   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4510                                      fd, file_offset);
4511   if (mapped_address == MAP_FAILED) {
4512     return NULL;
4513   }
4514   return mapped_address;
4515 }
4516 
4517 
4518 // Remap a block of memory.
4519 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4520                           char *addr, size_t bytes, bool read_only,
4521                           bool allow_exec) {
4522   // same as map_memory() on this OS
4523   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4524                         allow_exec);
4525 }
4526 
4527 
4528 // Unmap a block of memory.
4529 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4530   return munmap(addr, bytes) == 0;
4531 }
4532 
4533 void os::pause() {
4534   char filename[MAX_PATH];
4535   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4536     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4537   } else {
4538     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4539   }
4540 
4541   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4542   if (fd != -1) {
4543     struct stat buf;
4544     ::close(fd);
4545     while (::stat(filename, &buf) == 0) {
4546       (void)::poll(NULL, 0, 100);
4547     }
4548   } else {
4549     jio_fprintf(stderr,
4550                 "Could not open pause file '%s', continuing immediately.\n", filename);
4551   }
4552 }
4553 
4554 #ifndef PRODUCT
4555 #ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4556 // Turn this on if you need to trace synch operations.
4557 // Set RECORD_SYNCH_LIMIT to a large-enough value,
4558 // and call record_synch_enable and record_synch_disable
4559 // around the computation of interest.
4560 
4561 void record_synch(char* name, bool returning);  // defined below
4562 
4563 class RecordSynch {
4564   char* _name;
4565  public:
4566   RecordSynch(char* name) :_name(name) { record_synch(_name, false); }
4567   ~RecordSynch()                       { record_synch(_name, true); }
4568 };
4569 
4570 #define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
4571 extern "C" ret name params {                                    \
4572   typedef ret name##_t params;                                  \
4573   static name##_t* implem = NULL;                               \
4574   static int callcount = 0;                                     \
4575   if (implem == NULL) {                                         \
4576     implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
4577     if (implem == NULL)  fatal(dlerror());                      \
4578   }                                                             \
4579   ++callcount;                                                  \
4580   RecordSynch _rs(#name);                                       \
4581   inner;                                                        \
4582   return implem args;                                           \
4583 }
4584 // in dbx, examine callcounts this way:
4585 // for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
4586 
4587 #define CHECK_POINTER_OK(p) \
4588   (!Universe::is_fully_initialized() || !Universe::is_reserved_heap((oop)(p)))
4589 #define CHECK_MU \
4590   if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
4591 #define CHECK_CV \
4592   if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
4593 #define CHECK_P(p) \
4594   if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
4595 
4596 #define CHECK_MUTEX(mutex_op) \
4597   CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
4598 
4599 CHECK_MUTEX(   mutex_lock)
4600 CHECK_MUTEX(  _mutex_lock)
4601 CHECK_MUTEX( mutex_unlock)
4602 CHECK_MUTEX(_mutex_unlock)
4603 CHECK_MUTEX( mutex_trylock)
4604 CHECK_MUTEX(_mutex_trylock)
4605 
4606 #define CHECK_COND(cond_op) \
4607   CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU; CHECK_CV);
4608 
4609 CHECK_COND( cond_wait);
4610 CHECK_COND(_cond_wait);
4611 CHECK_COND(_cond_wait_cancel);
4612 
4613 #define CHECK_COND2(cond_op) \
4614   CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU; CHECK_CV);
4615 
4616 CHECK_COND2( cond_timedwait);
4617 CHECK_COND2(_cond_timedwait);
4618 CHECK_COND2(_cond_timedwait_cancel);
4619 
4620 // do the _lwp_* versions too
4621 #define mutex_t lwp_mutex_t
4622 #define cond_t  lwp_cond_t
4623 CHECK_MUTEX(  _lwp_mutex_lock)
4624 CHECK_MUTEX(  _lwp_mutex_unlock)
4625 CHECK_MUTEX(  _lwp_mutex_trylock)
4626 CHECK_MUTEX( __lwp_mutex_lock)
4627 CHECK_MUTEX( __lwp_mutex_unlock)
4628 CHECK_MUTEX( __lwp_mutex_trylock)
4629 CHECK_MUTEX(___lwp_mutex_lock)
4630 CHECK_MUTEX(___lwp_mutex_unlock)
4631 
4632 CHECK_COND(  _lwp_cond_wait);
4633 CHECK_COND( __lwp_cond_wait);
4634 CHECK_COND(___lwp_cond_wait);
4635 
4636 CHECK_COND2(  _lwp_cond_timedwait);
4637 CHECK_COND2( __lwp_cond_timedwait);
4638 #undef mutex_t
4639 #undef cond_t
4640 
4641 CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4642 CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4643 CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4644 CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4645 CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4646 CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4647 CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4648 CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4649 
4650 
4651 // recording machinery:
4652 
4653 enum { RECORD_SYNCH_LIMIT = 200 };
4654 char* record_synch_name[RECORD_SYNCH_LIMIT];
4655 void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
4656 bool record_synch_returning[RECORD_SYNCH_LIMIT];
4657 thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
4658 int record_synch_count = 0;
4659 bool record_synch_enabled = false;
4660 
4661 // in dbx, examine recorded data this way:
4662 // for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
4663 
4664 void record_synch(char* name, bool returning) {
4665   if (record_synch_enabled) {
4666     if (record_synch_count < RECORD_SYNCH_LIMIT) {
4667       record_synch_name[record_synch_count] = name;
4668       record_synch_returning[record_synch_count] = returning;
4669       record_synch_thread[record_synch_count] = thr_self();
4670       record_synch_arg0ptr[record_synch_count] = &name;
4671       record_synch_count++;
4672     }
4673     // put more checking code here:
4674     // ...
4675   }
4676 }
4677 
4678 void record_synch_enable() {
4679   // start collecting trace data, if not already doing so
4680   if (!record_synch_enabled)  record_synch_count = 0;
4681   record_synch_enabled = true;
4682 }
4683 
4684 void record_synch_disable() {
4685   // stop collecting trace data
4686   record_synch_enabled = false;
4687 }
4688 
4689 #endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4690 #endif // PRODUCT
4691 
4692 const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
4693 const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
4694                                (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
4695 
4696 
4697 // JVMTI & JVM monitoring and management support
4698 // The thread_cpu_time() and current_thread_cpu_time() are only
4699 // supported if is_thread_cpu_time_supported() returns true.
4700 // They are not supported on Solaris T1.
4701 
4702 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4703 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4704 // of a thread.
4705 //
4706 // current_thread_cpu_time() and thread_cpu_time(Thread *)
4707 // returns the fast estimate available on the platform.
4708 
4709 // hrtime_t gethrvtime() return value includes
4710 // user time but does not include system time
4711 jlong os::current_thread_cpu_time() {
4712   return (jlong) gethrvtime();
4713 }
4714 
4715 jlong os::thread_cpu_time(Thread *thread) {
4716   // return user level CPU time only to be consistent with
4717   // what current_thread_cpu_time returns.
4718   // thread_cpu_time_info() must be changed if this changes
4719   return os::thread_cpu_time(thread, false /* user time only */);
4720 }
4721 
4722 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4723   if (user_sys_cpu_time) {
4724     return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
4725   } else {
4726     return os::current_thread_cpu_time();
4727   }
4728 }
4729 
4730 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4731   char proc_name[64];
4732   int count;
4733   prusage_t prusage;
4734   jlong lwp_time;
4735   int fd;
4736 
4737   sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
4738           getpid(),
4739           thread->osthread()->lwp_id());
4740   fd = ::open(proc_name, O_RDONLY);
4741   if (fd == -1) return -1;
4742 
4743   do {
4744     count = ::pread(fd,
4745                     (void *)&prusage.pr_utime,
4746                     thr_time_size,
4747                     thr_time_off);
4748   } while (count < 0 && errno == EINTR);
4749   ::close(fd);
4750   if (count < 0) return -1;
4751 
4752   if (user_sys_cpu_time) {
4753     // user + system CPU time
4754     lwp_time = (((jlong)prusage.pr_stime.tv_sec +
4755                  (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
4756                  (jlong)prusage.pr_stime.tv_nsec +
4757                  (jlong)prusage.pr_utime.tv_nsec;
4758   } else {
4759     // user level CPU time only
4760     lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
4761                 (jlong)prusage.pr_utime.tv_nsec;
4762   }
4763 
4764   return (lwp_time);
4765 }
4766 
4767 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4768   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
4769   info_ptr->may_skip_backward = false;    // elapsed time not wall time
4770   info_ptr->may_skip_forward = false;     // elapsed time not wall time
4771   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
4772 }
4773 
4774 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4775   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
4776   info_ptr->may_skip_backward = false;    // elapsed time not wall time
4777   info_ptr->may_skip_forward = false;     // elapsed time not wall time
4778   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
4779 }
4780 
4781 bool os::is_thread_cpu_time_supported() {
4782   return true;
4783 }
4784 
4785 // System loadavg support.  Returns -1 if load average cannot be obtained.
4786 // Return the load average for our processor set if the primitive exists
4787 // (Solaris 9 and later).  Otherwise just return system wide loadavg.
4788 int os::loadavg(double loadavg[], int nelem) {
4789   if (pset_getloadavg_ptr != NULL) {
4790     return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
4791   } else {
4792     return ::getloadavg(loadavg, nelem);
4793   }
4794 }
4795 
4796 //---------------------------------------------------------------------------------
4797 
4798 bool os::find(address addr, outputStream* st) {
4799   Dl_info dlinfo;
4800   memset(&dlinfo, 0, sizeof(dlinfo));
4801   if (dladdr(addr, &dlinfo) != 0) {
4802     st->print(PTR_FORMAT ": ", addr);
4803     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4804       st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
4805     } else if (dlinfo.dli_fbase != NULL) {
4806       st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
4807     } else {
4808       st->print("<absolute address>");
4809     }
4810     if (dlinfo.dli_fname != NULL) {
4811       st->print(" in %s", dlinfo.dli_fname);
4812     }
4813     if (dlinfo.dli_fbase != NULL) {
4814       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4815     }
4816     st->cr();
4817 
4818     if (Verbose) {
4819       // decode some bytes around the PC
4820       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4821       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4822       address       lowest = (address) dlinfo.dli_sname;
4823       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4824       if (begin < lowest)  begin = lowest;
4825       Dl_info dlinfo2;
4826       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4827           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4828         end = (address) dlinfo2.dli_saddr;
4829       }
4830       Disassembler::decode(begin, end, st);
4831     }
4832     return true;
4833   }
4834   return false;
4835 }
4836 
4837 // Following function has been added to support HotSparc's libjvm.so running
4838 // under Solaris production JDK 1.2.2 / 1.3.0.  These came from
4839 // src/solaris/hpi/native_threads in the EVM codebase.
4840 //
4841 // NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
4842 // libraries and should thus be removed. We will leave it behind for a while
4843 // until we no longer want to able to run on top of 1.3.0 Solaris production
4844 // JDK. See 4341971.
4845 
4846 #define STACK_SLACK 0x800
4847 
4848 extern "C" {
4849   intptr_t sysThreadAvailableStackWithSlack() {
4850     stack_t st;
4851     intptr_t retval, stack_top;
4852     retval = thr_stksegment(&st);
4853     assert(retval == 0, "incorrect return value from thr_stksegment");
4854     assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
4855     assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
4856     stack_top=(intptr_t)st.ss_sp-st.ss_size;
4857     return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
4858   }
4859 }
4860 
4861 // ObjectMonitor park-unpark infrastructure ...
4862 //
4863 // We implement Solaris and Linux PlatformEvents with the
4864 // obvious condvar-mutex-flag triple.
4865 // Another alternative that works quite well is pipes:
4866 // Each PlatformEvent consists of a pipe-pair.
4867 // The thread associated with the PlatformEvent
4868 // calls park(), which reads from the input end of the pipe.
4869 // Unpark() writes into the other end of the pipe.
4870 // The write-side of the pipe must be set NDELAY.
4871 // Unfortunately pipes consume a large # of handles.
4872 // Native solaris lwp_park() and lwp_unpark() work nicely, too.
4873 // Using pipes for the 1st few threads might be workable, however.
4874 //
4875 // park() is permitted to return spuriously.
4876 // Callers of park() should wrap the call to park() in
4877 // an appropriate loop.  A litmus test for the correct
4878 // usage of park is the following: if park() were modified
4879 // to immediately return 0 your code should still work,
4880 // albeit degenerating to a spin loop.
4881 //
4882 // In a sense, park()-unpark() just provides more polite spinning
4883 // and polling with the key difference over naive spinning being
4884 // that a parked thread needs to be explicitly unparked() in order
4885 // to wake up and to poll the underlying condition.
4886 //
4887 // Assumption:
4888 //    Only one parker can exist on an event, which is why we allocate
4889 //    them per-thread. Multiple unparkers can coexist.
4890 //
4891 // _Event transitions in park()
4892 //   -1 => -1 : illegal
4893 //    1 =>  0 : pass - return immediately
4894 //    0 => -1 : block; then set _Event to 0 before returning
4895 //
4896 // _Event transitions in unpark()
4897 //    0 => 1 : just return
4898 //    1 => 1 : just return
4899 //   -1 => either 0 or 1; must signal target thread
4900 //         That is, we can safely transition _Event from -1 to either
4901 //         0 or 1.
4902 //
4903 // _Event serves as a restricted-range semaphore.
4904 //   -1 : thread is blocked, i.e. there is a waiter
4905 //    0 : neutral: thread is running or ready,
4906 //        could have been signaled after a wait started
4907 //    1 : signaled - thread is running or ready
4908 //
4909 // Another possible encoding of _Event would be with
4910 // explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
4911 //
4912 // TODO-FIXME: add DTRACE probes for:
4913 // 1.   Tx parks
4914 // 2.   Ty unparks Tx
4915 // 3.   Tx resumes from park
4916 
4917 
4918 // value determined through experimentation
4919 #define ROUNDINGFIX 11
4920 
4921 // utility to compute the abstime argument to timedwait.
4922 // TODO-FIXME: switch from compute_abstime() to unpackTime().
4923 
4924 static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
4925   // millis is the relative timeout time
4926   // abstime will be the absolute timeout time
4927   if (millis < 0)  millis = 0;
4928   struct timeval now;
4929   int status = gettimeofday(&now, NULL);
4930   assert(status == 0, "gettimeofday");
4931   jlong seconds = millis / 1000;
4932   jlong max_wait_period;
4933 
4934   if (UseLWPSynchronization) {
4935     // forward port of fix for 4275818 (not sleeping long enough)
4936     // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
4937     // _lwp_cond_timedwait() used a round_down algorithm rather
4938     // than a round_up. For millis less than our roundfactor
4939     // it rounded down to 0 which doesn't meet the spec.
4940     // For millis > roundfactor we may return a bit sooner, but
4941     // since we can not accurately identify the patch level and
4942     // this has already been fixed in Solaris 9 and 8 we will
4943     // leave it alone rather than always rounding down.
4944 
4945     if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
4946     // It appears that when we go directly through Solaris _lwp_cond_timedwait()
4947     // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
4948     max_wait_period = 21000000;
4949   } else {
4950     max_wait_period = 50000000;
4951   }
4952   millis %= 1000;
4953   if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
4954     seconds = max_wait_period;
4955   }
4956   abstime->tv_sec = now.tv_sec  + seconds;
4957   long       usec = now.tv_usec + millis * 1000;
4958   if (usec >= 1000000) {
4959     abstime->tv_sec += 1;
4960     usec -= 1000000;
4961   }
4962   abstime->tv_nsec = usec * 1000;
4963   return abstime;
4964 }
4965 
4966 void os::PlatformEvent::park() {           // AKA: down()
4967   // Transitions for _Event:
4968   //   -1 => -1 : illegal
4969   //    1 =>  0 : pass - return immediately
4970   //    0 => -1 : block; then set _Event to 0 before returning
4971 
4972   // Invariant: Only the thread associated with the Event/PlatformEvent
4973   // may call park().
4974   assert(_nParked == 0, "invariant");
4975 
4976   int v;
4977   for (;;) {
4978     v = _Event;
4979     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
4980   }
4981   guarantee(v >= 0, "invariant");
4982   if (v == 0) {
4983     // Do this the hard way by blocking ...
4984     // See http://monaco.sfbay/detail.jsf?cr=5094058.
4985     int status = os::Solaris::mutex_lock(_mutex);
4986     assert_status(status == 0, status, "mutex_lock");
4987     guarantee(_nParked == 0, "invariant");
4988     ++_nParked;
4989     while (_Event < 0) {
4990       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4991       // Treat this the same as if the wait was interrupted
4992       // With usr/lib/lwp going to kernel, always handle ETIME
4993       status = os::Solaris::cond_wait(_cond, _mutex);
4994       if (status == ETIME) status = EINTR;
4995       assert_status(status == 0 || status == EINTR, status, "cond_wait");
4996     }
4997     --_nParked;
4998     _Event = 0;
4999     status = os::Solaris::mutex_unlock(_mutex);
5000     assert_status(status == 0, status, "mutex_unlock");
5001     // Paranoia to ensure our locked and lock-free paths interact
5002     // correctly with each other.
5003     OrderAccess::fence();
5004   }
5005 }
5006 
5007 int os::PlatformEvent::park(jlong millis) {
5008   // Transitions for _Event:
5009   //   -1 => -1 : illegal
5010   //    1 =>  0 : pass - return immediately
5011   //    0 => -1 : block; then set _Event to 0 before returning
5012 
5013   guarantee(_nParked == 0, "invariant");
5014   int v;
5015   for (;;) {
5016     v = _Event;
5017     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5018   }
5019   guarantee(v >= 0, "invariant");
5020   if (v != 0) return OS_OK;
5021 
5022   int ret = OS_TIMEOUT;
5023   timestruc_t abst;
5024   compute_abstime(&abst, millis);
5025 
5026   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5027   int status = os::Solaris::mutex_lock(_mutex);
5028   assert_status(status == 0, status, "mutex_lock");
5029   guarantee(_nParked == 0, "invariant");
5030   ++_nParked;
5031   while (_Event < 0) {
5032     int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
5033     assert_status(status == 0 || status == EINTR ||
5034                   status == ETIME || status == ETIMEDOUT,
5035                   status, "cond_timedwait");
5036     if (!FilterSpuriousWakeups) break;                // previous semantics
5037     if (status == ETIME || status == ETIMEDOUT) break;
5038     // We consume and ignore EINTR and spurious wakeups.
5039   }
5040   --_nParked;
5041   if (_Event >= 0) ret = OS_OK;
5042   _Event = 0;
5043   status = os::Solaris::mutex_unlock(_mutex);
5044   assert_status(status == 0, status, "mutex_unlock");
5045   // Paranoia to ensure our locked and lock-free paths interact
5046   // correctly with each other.
5047   OrderAccess::fence();
5048   return ret;
5049 }
5050 
5051 void os::PlatformEvent::unpark() {
5052   // Transitions for _Event:
5053   //    0 => 1 : just return
5054   //    1 => 1 : just return
5055   //   -1 => either 0 or 1; must signal target thread
5056   //         That is, we can safely transition _Event from -1 to either
5057   //         0 or 1.
5058   // See also: "Semaphores in Plan 9" by Mullender & Cox
5059   //
5060   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5061   // that it will take two back-to-back park() calls for the owning
5062   // thread to block. This has the benefit of forcing a spurious return
5063   // from the first park() call after an unpark() call which will help
5064   // shake out uses of park() and unpark() without condition variables.
5065 
5066   if (Atomic::xchg(1, &_Event) >= 0) return;
5067 
5068   // If the thread associated with the event was parked, wake it.
5069   // Wait for the thread assoc with the PlatformEvent to vacate.
5070   int status = os::Solaris::mutex_lock(_mutex);
5071   assert_status(status == 0, status, "mutex_lock");
5072   int AnyWaiters = _nParked;
5073   status = os::Solaris::mutex_unlock(_mutex);
5074   assert_status(status == 0, status, "mutex_unlock");
5075   guarantee(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5076   if (AnyWaiters != 0) {
5077     // Note that we signal() *after* dropping the lock for "immortal" Events.
5078     // This is safe and avoids a common class of  futile wakeups.  In rare
5079     // circumstances this can cause a thread to return prematurely from
5080     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5081     // will simply re-test the condition and re-park itself.
5082     // This provides particular benefit if the underlying platform does not
5083     // provide wait morphing.
5084     status = os::Solaris::cond_signal(_cond);
5085     assert_status(status == 0, status, "cond_signal");
5086   }
5087 }
5088 
5089 // JSR166
5090 // -------------------------------------------------------
5091 
5092 // The solaris and linux implementations of park/unpark are fairly
5093 // conservative for now, but can be improved. They currently use a
5094 // mutex/condvar pair, plus _counter.
5095 // Park decrements _counter if > 0, else does a condvar wait.  Unpark
5096 // sets count to 1 and signals condvar.  Only one thread ever waits
5097 // on the condvar. Contention seen when trying to park implies that someone
5098 // is unparking you, so don't wait. And spurious returns are fine, so there
5099 // is no need to track notifications.
5100 
5101 #define MAX_SECS 100000000
5102 
5103 // This code is common to linux and solaris and will be moved to a
5104 // common place in dolphin.
5105 //
5106 // The passed in time value is either a relative time in nanoseconds
5107 // or an absolute time in milliseconds. Either way it has to be unpacked
5108 // into suitable seconds and nanoseconds components and stored in the
5109 // given timespec structure.
5110 // Given time is a 64-bit value and the time_t used in the timespec is only
5111 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5112 // overflow if times way in the future are given. Further on Solaris versions
5113 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5114 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5115 // As it will be 28 years before "now + 100000000" will overflow we can
5116 // ignore overflow and just impose a hard-limit on seconds using the value
5117 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5118 // years from "now".
5119 //
5120 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5121   assert(time > 0, "convertTime");
5122 
5123   struct timeval now;
5124   int status = gettimeofday(&now, NULL);
5125   assert(status == 0, "gettimeofday");
5126 
5127   time_t max_secs = now.tv_sec + MAX_SECS;
5128 
5129   if (isAbsolute) {
5130     jlong secs = time / 1000;
5131     if (secs > max_secs) {
5132       absTime->tv_sec = max_secs;
5133     } else {
5134       absTime->tv_sec = secs;
5135     }
5136     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5137   } else {
5138     jlong secs = time / NANOSECS_PER_SEC;
5139     if (secs >= MAX_SECS) {
5140       absTime->tv_sec = max_secs;
5141       absTime->tv_nsec = 0;
5142     } else {
5143       absTime->tv_sec = now.tv_sec + secs;
5144       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5145       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5146         absTime->tv_nsec -= NANOSECS_PER_SEC;
5147         ++absTime->tv_sec; // note: this must be <= max_secs
5148       }
5149     }
5150   }
5151   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5152   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5153   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5154   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5155 }
5156 
5157 void Parker::park(bool isAbsolute, jlong time) {
5158   // Ideally we'd do something useful while spinning, such
5159   // as calling unpackTime().
5160 
5161   // Optional fast-path check:
5162   // Return immediately if a permit is available.
5163   // We depend on Atomic::xchg() having full barrier semantics
5164   // since we are doing a lock-free update to _counter.
5165   if (Atomic::xchg(0, &_counter) > 0) return;
5166 
5167   // Optional fast-exit: Check interrupt before trying to wait
5168   Thread* thread = Thread::current();
5169   assert(thread->is_Java_thread(), "Must be JavaThread");
5170   JavaThread *jt = (JavaThread *)thread;
5171   if (Thread::is_interrupted(thread, false)) {
5172     return;
5173   }
5174 
5175   // First, demultiplex/decode time arguments
5176   timespec absTime;
5177   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5178     return;
5179   }
5180   if (time > 0) {
5181     // Warning: this code might be exposed to the old Solaris time
5182     // round-down bugs.  Grep "roundingFix" for details.
5183     unpackTime(&absTime, isAbsolute, time);
5184   }
5185 
5186   // Enter safepoint region
5187   // Beware of deadlocks such as 6317397.
5188   // The per-thread Parker:: _mutex is a classic leaf-lock.
5189   // In particular a thread must never block on the Threads_lock while
5190   // holding the Parker:: mutex.  If safepoints are pending both the
5191   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5192   ThreadBlockInVM tbivm(jt);
5193 
5194   // Don't wait if cannot get lock since interference arises from
5195   // unblocking.  Also. check interrupt before trying wait
5196   if (Thread::is_interrupted(thread, false) ||
5197       os::Solaris::mutex_trylock(_mutex) != 0) {
5198     return;
5199   }
5200 
5201   int status;
5202 
5203   if (_counter > 0)  { // no wait needed
5204     _counter = 0;
5205     status = os::Solaris::mutex_unlock(_mutex);
5206     assert(status == 0, "invariant");
5207     // Paranoia to ensure our locked and lock-free paths interact
5208     // correctly with each other and Java-level accesses.
5209     OrderAccess::fence();
5210     return;
5211   }
5212 
5213   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5214   jt->set_suspend_equivalent();
5215   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5216 
5217   // Do this the hard way by blocking ...
5218   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5219   if (time == 0) {
5220     status = os::Solaris::cond_wait(_cond, _mutex);
5221   } else {
5222     status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
5223   }
5224   // Note that an untimed cond_wait() can sometimes return ETIME on older
5225   // versions of the Solaris.
5226   assert_status(status == 0 || status == EINTR ||
5227                 status == ETIME || status == ETIMEDOUT,
5228                 status, "cond_timedwait");
5229 
5230   _counter = 0;
5231   status = os::Solaris::mutex_unlock(_mutex);
5232   assert_status(status == 0, status, "mutex_unlock");
5233   // Paranoia to ensure our locked and lock-free paths interact
5234   // correctly with each other and Java-level accesses.
5235   OrderAccess::fence();
5236 
5237   // If externally suspended while waiting, re-suspend
5238   if (jt->handle_special_suspend_equivalent_condition()) {
5239     jt->java_suspend_self();
5240   }
5241 }
5242 
5243 void Parker::unpark() {
5244   int status = os::Solaris::mutex_lock(_mutex);
5245   assert(status == 0, "invariant");
5246   const int s = _counter;
5247   _counter = 1;
5248   status = os::Solaris::mutex_unlock(_mutex);
5249   assert(status == 0, "invariant");
5250 
5251   if (s < 1) {
5252     status = os::Solaris::cond_signal(_cond);
5253     assert(status == 0, "invariant");
5254   }
5255 }
5256 
5257 extern char** environ;
5258 
5259 // Run the specified command in a separate process. Return its exit value,
5260 // or -1 on failure (e.g. can't fork a new process).
5261 // Unlike system(), this function can be called from signal handler. It
5262 // doesn't block SIGINT et al.
5263 int os::fork_and_exec(char* cmd) {
5264   char * argv[4];
5265   argv[0] = (char *)"sh";
5266   argv[1] = (char *)"-c";
5267   argv[2] = cmd;
5268   argv[3] = NULL;
5269 
5270   // fork is async-safe, fork1 is not so can't use in signal handler
5271   pid_t pid;
5272   Thread* t = Thread::current_or_null_safe();
5273   if (t != NULL && t->is_inside_signal_handler()) {
5274     pid = fork();
5275   } else {
5276     pid = fork1();
5277   }
5278 
5279   if (pid < 0) {
5280     // fork failed
5281     warning("fork failed: %s", os::strerror(errno));
5282     return -1;
5283 
5284   } else if (pid == 0) {
5285     // child process
5286 
5287     // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
5288     execve("/usr/bin/sh", argv, environ);
5289 
5290     // execve failed
5291     _exit(-1);
5292 
5293   } else  {
5294     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5295     // care about the actual exit code, for now.
5296 
5297     int status;
5298 
5299     // Wait for the child process to exit.  This returns immediately if
5300     // the child has already exited. */
5301     while (waitpid(pid, &status, 0) < 0) {
5302       switch (errno) {
5303       case ECHILD: return 0;
5304       case EINTR: break;
5305       default: return -1;
5306       }
5307     }
5308 
5309     if (WIFEXITED(status)) {
5310       // The child exited normally; get its exit code.
5311       return WEXITSTATUS(status);
5312     } else if (WIFSIGNALED(status)) {
5313       // The child exited because of a signal
5314       // The best value to return is 0x80 + signal number,
5315       // because that is what all Unix shells do, and because
5316       // it allows callers to distinguish between process exit and
5317       // process death by signal.
5318       return 0x80 + WTERMSIG(status);
5319     } else {
5320       // Unknown exit code; pass it through
5321       return status;
5322     }
5323   }
5324 }
5325 
5326 size_t os::write(int fd, const void *buf, unsigned int nBytes) {
5327   size_t res;
5328   RESTARTABLE((size_t) ::write(fd, buf, (size_t) nBytes), res);
5329   return res;
5330 }
5331 
5332 int os::close(int fd) {
5333   return ::close(fd);
5334 }
5335 
5336 int os::socket_close(int fd) {
5337   return ::close(fd);
5338 }
5339 
5340 int os::recv(int fd, char* buf, size_t nBytes, uint flags) {
5341   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5342          "Assumed _thread_in_native");
5343   RESTARTABLE_RETURN_INT((int)::recv(fd, buf, nBytes, flags));
5344 }
5345 
5346 int os::send(int fd, char* buf, size_t nBytes, uint flags) {
5347   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5348          "Assumed _thread_in_native");
5349   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5350 }
5351 
5352 int os::raw_send(int fd, char* buf, size_t nBytes, uint flags) {
5353   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5354 }
5355 
5356 // As both poll and select can be interrupted by signals, we have to be
5357 // prepared to restart the system call after updating the timeout, unless
5358 // a poll() is done with timeout == -1, in which case we repeat with this
5359 // "wait forever" value.
5360 
5361 int os::connect(int fd, struct sockaddr *him, socklen_t len) {
5362   int _result;
5363   _result = ::connect(fd, him, len);
5364 
5365   // On Solaris, when a connect() call is interrupted, the connection
5366   // can be established asynchronously (see 6343810). Subsequent calls
5367   // to connect() must check the errno value which has the semantic
5368   // described below (copied from the connect() man page). Handling
5369   // of asynchronously established connections is required for both
5370   // blocking and non-blocking sockets.
5371   //     EINTR            The  connection  attempt  was   interrupted
5372   //                      before  any data arrived by the delivery of
5373   //                      a signal. The connection, however, will  be
5374   //                      established asynchronously.
5375   //
5376   //     EINPROGRESS      The socket is non-blocking, and the connec-
5377   //                      tion  cannot  be completed immediately.
5378   //
5379   //     EALREADY         The socket is non-blocking,  and a previous
5380   //                      connection  attempt  has  not yet been com-
5381   //                      pleted.
5382   //
5383   //     EISCONN          The socket is already connected.
5384   if (_result == OS_ERR && errno == EINTR) {
5385     // restarting a connect() changes its errno semantics
5386     RESTARTABLE(::connect(fd, him, len), _result);
5387     // undo these changes
5388     if (_result == OS_ERR) {
5389       if (errno == EALREADY) {
5390         errno = EINPROGRESS; // fall through
5391       } else if (errno == EISCONN) {
5392         errno = 0;
5393         return OS_OK;
5394       }
5395     }
5396   }
5397   return _result;
5398 }
5399 
5400 // Get the default path to the core file
5401 // Returns the length of the string
5402 int os::get_core_path(char* buffer, size_t bufferSize) {
5403   const char* p = get_current_directory(buffer, bufferSize);
5404 
5405   if (p == NULL) {
5406     assert(p != NULL, "failed to get current directory");
5407     return 0;
5408   }
5409 
5410   jio_snprintf(buffer, bufferSize, "%s/core or core.%d",
5411                                               p, current_process_id());
5412 
5413   return strlen(buffer);
5414 }
5415 
5416 #ifndef PRODUCT
5417 void TestReserveMemorySpecial_test() {
5418   // No tests available for this platform
5419 }
5420 #endif
5421 
5422 bool os::start_debugging(char *buf, int buflen) {
5423   int len = (int)strlen(buf);
5424   char *p = &buf[len];
5425 
5426   jio_snprintf(p, buflen-len,
5427                "\n\n"
5428                "Do you want to debug the problem?\n\n"
5429                "To debug, run 'dbx - %d'; then switch to thread " INTX_FORMAT "\n"
5430                "Enter 'yes' to launch dbx automatically (PATH must include dbx)\n"
5431                "Otherwise, press RETURN to abort...",
5432                os::current_process_id(), os::current_thread_id());
5433 
5434   bool yes = os::message_box("Unexpected Error", buf);
5435 
5436   if (yes) {
5437     // yes, user asked VM to launch debugger
5438     jio_snprintf(buf, sizeof(buf), "dbx - %d", os::current_process_id());
5439 
5440     os::fork_and_exec(buf);
5441     yes = false;
5442   }
5443   return yes;
5444 }