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