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