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