1 /*
   2  * Copyright (c) 1997, 2017, 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 "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_solaris.h"
  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.h"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.hpp"
  49 #include "runtime/java.hpp"
  50 #include "runtime/javaCalls.hpp"
  51 #include "runtime/mutexLocker.hpp"
  52 #include "runtime/objectMonitor.hpp"
  53 #include "runtime/orderAccess.inline.hpp"
  54 #include "runtime/osThread.hpp"
  55 #include "runtime/perfMemory.hpp"
  56 #include "runtime/sharedRuntime.hpp"
  57 #include "runtime/statSampler.hpp"
  58 #include "runtime/stubRoutines.hpp"
  59 #include "runtime/thread.inline.hpp"
  60 #include "runtime/threadCritical.hpp"
  61 #include "runtime/timer.hpp"
  62 #include "runtime/vm_version.hpp"
  63 #include "semaphore_posix.hpp"
  64 #include "services/attachListener.hpp"
  65 #include "services/memTracker.hpp"
  66 #include "services/runtimeService.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 address os::current_stack_base() {
 204   int r = thr_main();
 205   guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
 206   bool is_primordial_thread = r;
 207 
 208   // Workaround 4352906, avoid calls to thr_stksegment by
 209   // thr_main after the first one (it looks like we trash
 210   // some data, causing the value for ss_sp to be incorrect).
 211   if (!is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
 212     stack_t st = get_stack_info();
 213     if (is_primordial_thread) {
 214       // cache initial value of stack base
 215       os::Solaris::_main_stack_base = (address)st.ss_sp;
 216     }
 217     return (address)st.ss_sp;
 218   } else {
 219     guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
 220     return os::Solaris::_main_stack_base;
 221   }
 222 }
 223 
 224 size_t os::current_stack_size() {
 225   size_t size;
 226 
 227   int r = thr_main();
 228   guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
 229   if (!r) {
 230     size = get_stack_info().ss_size;
 231   } else {
 232     struct rlimit limits;
 233     getrlimit(RLIMIT_STACK, &limits);
 234     size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
 235   }
 236   // base may not be page aligned
 237   address base = current_stack_base();
 238   address bottom = align_up(base - size, os::vm_page_size());;
 239   return (size_t)(base - bottom);
 240 }
 241 
 242 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
 243   return localtime_r(clock, res);
 244 }
 245 
 246 void os::Solaris::try_enable_extended_io() {
 247   typedef int (*enable_extended_FILE_stdio_t)(int, int);
 248 
 249   if (!UseExtendedFileIO) {
 250     return;
 251   }
 252 
 253   enable_extended_FILE_stdio_t enabler =
 254     (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
 255                                          "enable_extended_FILE_stdio");
 256   if (enabler) {
 257     enabler(-1, -1);
 258   }
 259 }
 260 
 261 static int _processors_online = 0;
 262 
 263 jint os::Solaris::_os_thread_limit = 0;
 264 volatile jint os::Solaris::_os_thread_count = 0;
 265 
 266 julong os::available_memory() {
 267   return Solaris::available_memory();
 268 }
 269 
 270 julong os::Solaris::available_memory() {
 271   return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
 272 }
 273 
 274 julong os::Solaris::_physical_memory = 0;
 275 
 276 julong os::physical_memory() {
 277   return Solaris::physical_memory();
 278 }
 279 
 280 static hrtime_t first_hrtime = 0;
 281 static const hrtime_t hrtime_hz = 1000*1000*1000;
 282 static volatile hrtime_t max_hrtime = 0;
 283 
 284 
 285 void os::Solaris::initialize_system_info() {
 286   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 287   _processors_online = sysconf(_SC_NPROCESSORS_ONLN);
 288   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) *
 289                                      (julong)sysconf(_SC_PAGESIZE);
 290 }
 291 
 292 int os::active_processor_count() {
 293   int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 294   pid_t pid = getpid();
 295   psetid_t pset = PS_NONE;
 296   // Are we running in a processor set or is there any processor set around?
 297   if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
 298     uint_t pset_cpus;
 299     // Query the number of cpus available to us.
 300     if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
 301       assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
 302       _processors_online = pset_cpus;
 303       return pset_cpus;
 304     }
 305   }
 306   // Otherwise return number of online cpus
 307   return online_cpus;
 308 }
 309 
 310 static bool find_processors_in_pset(psetid_t        pset,
 311                                     processorid_t** id_array,
 312                                     uint_t*         id_length) {
 313   bool result = false;
 314   // Find the number of processors in the processor set.
 315   if (pset_info(pset, NULL, id_length, NULL) == 0) {
 316     // Make up an array to hold their ids.
 317     *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
 318     // Fill in the array with their processor ids.
 319     if (pset_info(pset, NULL, id_length, *id_array) == 0) {
 320       result = true;
 321     }
 322   }
 323   return result;
 324 }
 325 
 326 // Callers of find_processors_online() must tolerate imprecise results --
 327 // the system configuration can change asynchronously because of DR
 328 // or explicit psradm operations.
 329 //
 330 // We also need to take care that the loop (below) terminates as the
 331 // number of processors online can change between the _SC_NPROCESSORS_ONLN
 332 // request and the loop that builds the list of processor ids.   Unfortunately
 333 // there's no reliable way to determine the maximum valid processor id,
 334 // so we use a manifest constant, MAX_PROCESSOR_ID, instead.  See p_online
 335 // man pages, which claim the processor id set is "sparse, but
 336 // not too sparse".  MAX_PROCESSOR_ID is used to ensure that we eventually
 337 // exit the loop.
 338 //
 339 // In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
 340 // not available on S8.0.
 341 
 342 static bool find_processors_online(processorid_t** id_array,
 343                                    uint*           id_length) {
 344   const processorid_t MAX_PROCESSOR_ID = 100000;
 345   // Find the number of processors online.
 346   *id_length = sysconf(_SC_NPROCESSORS_ONLN);
 347   // Make up an array to hold their ids.
 348   *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
 349   // Processors need not be numbered consecutively.
 350   long found = 0;
 351   processorid_t next = 0;
 352   while (found < *id_length && next < MAX_PROCESSOR_ID) {
 353     processor_info_t info;
 354     if (processor_info(next, &info) == 0) {
 355       // NB, PI_NOINTR processors are effectively online ...
 356       if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
 357         (*id_array)[found] = next;
 358         found += 1;
 359       }
 360     }
 361     next += 1;
 362   }
 363   if (found < *id_length) {
 364     // The loop above didn't identify the expected number of processors.
 365     // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
 366     // and re-running the loop, above, but there's no guarantee of progress
 367     // if the system configuration is in flux.  Instead, we just return what
 368     // we've got.  Note that in the worst case find_processors_online() could
 369     // return an empty set.  (As a fall-back in the case of the empty set we
 370     // could just return the ID of the current processor).
 371     *id_length = found;
 372   }
 373 
 374   return true;
 375 }
 376 
 377 static bool assign_distribution(processorid_t* id_array,
 378                                 uint           id_length,
 379                                 uint*          distribution,
 380                                 uint           distribution_length) {
 381   // We assume we can assign processorid_t's to uint's.
 382   assert(sizeof(processorid_t) == sizeof(uint),
 383          "can't convert processorid_t to uint");
 384   // Quick check to see if we won't succeed.
 385   if (id_length < distribution_length) {
 386     return false;
 387   }
 388   // Assign processor ids to the distribution.
 389   // Try to shuffle processors to distribute work across boards,
 390   // assuming 4 processors per board.
 391   const uint processors_per_board = ProcessDistributionStride;
 392   // Find the maximum processor id.
 393   processorid_t max_id = 0;
 394   for (uint m = 0; m < id_length; m += 1) {
 395     max_id = MAX2(max_id, id_array[m]);
 396   }
 397   // The next id, to limit loops.
 398   const processorid_t limit_id = max_id + 1;
 399   // Make up markers for available processors.
 400   bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id, mtInternal);
 401   for (uint c = 0; c < limit_id; c += 1) {
 402     available_id[c] = false;
 403   }
 404   for (uint a = 0; a < id_length; a += 1) {
 405     available_id[id_array[a]] = true;
 406   }
 407   // Step by "boards", then by "slot", copying to "assigned".
 408   // NEEDS_CLEANUP: The assignment of processors should be stateful,
 409   //                remembering which processors have been assigned by
 410   //                previous calls, etc., so as to distribute several
 411   //                independent calls of this method.  What we'd like is
 412   //                It would be nice to have an API that let us ask
 413   //                how many processes are bound to a processor,
 414   //                but we don't have that, either.
 415   //                In the short term, "board" is static so that
 416   //                subsequent distributions don't all start at board 0.
 417   static uint board = 0;
 418   uint assigned = 0;
 419   // Until we've found enough processors ....
 420   while (assigned < distribution_length) {
 421     // ... find the next available processor in the board.
 422     for (uint slot = 0; slot < processors_per_board; slot += 1) {
 423       uint try_id = board * processors_per_board + slot;
 424       if ((try_id < limit_id) && (available_id[try_id] == true)) {
 425         distribution[assigned] = try_id;
 426         available_id[try_id] = false;
 427         assigned += 1;
 428         break;
 429       }
 430     }
 431     board += 1;
 432     if (board * processors_per_board + 0 >= limit_id) {
 433       board = 0;
 434     }
 435   }
 436   if (available_id != NULL) {
 437     FREE_C_HEAP_ARRAY(bool, available_id);
 438   }
 439   return true;
 440 }
 441 
 442 void os::set_native_thread_name(const char *name) {
 443   if (Solaris::_pthread_setname_np != NULL) {
 444     // Only the first 31 bytes of 'name' are processed by pthread_setname_np
 445     // but we explicitly copy into a size-limited buffer to avoid any
 446     // possible overflow.
 447     char buf[32];
 448     snprintf(buf, sizeof(buf), "%s", name);
 449     buf[sizeof(buf) - 1] = '\0';
 450     Solaris::_pthread_setname_np(pthread_self(), buf);
 451   }
 452 }
 453 
 454 bool os::distribute_processes(uint length, uint* distribution) {
 455   bool result = false;
 456   // Find the processor id's of all the available CPUs.
 457   processorid_t* id_array  = NULL;
 458   uint           id_length = 0;
 459   // There are some races between querying information and using it,
 460   // since processor sets can change dynamically.
 461   psetid_t pset = PS_NONE;
 462   // Are we running in a processor set?
 463   if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
 464     result = find_processors_in_pset(pset, &id_array, &id_length);
 465   } else {
 466     result = find_processors_online(&id_array, &id_length);
 467   }
 468   if (result == true) {
 469     if (id_length >= length) {
 470       result = assign_distribution(id_array, id_length, distribution, length);
 471     } else {
 472       result = false;
 473     }
 474   }
 475   if (id_array != NULL) {
 476     FREE_C_HEAP_ARRAY(processorid_t, id_array);
 477   }
 478   return result;
 479 }
 480 
 481 bool os::bind_to_processor(uint processor_id) {
 482   // We assume that a processorid_t can be stored in a uint.
 483   assert(sizeof(uint) == sizeof(processorid_t),
 484          "can't convert uint to processorid_t");
 485   int bind_result =
 486     processor_bind(P_LWPID,                       // bind LWP.
 487                    P_MYID,                        // bind current LWP.
 488                    (processorid_t) processor_id,  // id.
 489                    NULL);                         // don't return old binding.
 490   return (bind_result == 0);
 491 }
 492 
 493 // Return true if user is running as root.
 494 
 495 bool os::have_special_privileges() {
 496   static bool init = false;
 497   static bool privileges = false;
 498   if (!init) {
 499     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 500     init = true;
 501   }
 502   return privileges;
 503 }
 504 
 505 
 506 void os::init_system_properties_values() {
 507   // The next steps are taken in the product version:
 508   //
 509   // Obtain the JAVA_HOME value from the location of libjvm.so.
 510   // This library should be located at:
 511   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 512   //
 513   // If "/jre/lib/" appears at the right place in the path, then we
 514   // assume libjvm.so is installed in a JDK and we use this path.
 515   //
 516   // Otherwise exit with message: "Could not create the Java virtual machine."
 517   //
 518   // The following extra steps are taken in the debugging version:
 519   //
 520   // If "/jre/lib/" does NOT appear at the right place in the path
 521   // instead of exit check for $JAVA_HOME environment variable.
 522   //
 523   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 524   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 525   // it looks like libjvm.so is installed there
 526   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 527   //
 528   // Otherwise exit.
 529   //
 530   // Important note: if the location of libjvm.so changes this
 531   // code needs to be changed accordingly.
 532 
 533 // Base path of extensions installed on the system.
 534 #define SYS_EXT_DIR     "/usr/jdk/packages"
 535 #define EXTENSIONS_DIR  "/lib/ext"
 536 
 537   // Buffer that fits several sprintfs.
 538   // Note that the space for the colon and the trailing null are provided
 539   // by the nulls included by the sizeof operator.
 540   const size_t bufsize =
 541     MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
 542          sizeof(SYS_EXT_DIR) + sizeof("/lib/"), // invariant ld_library_path
 543          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 544   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 545 
 546   // sysclasspath, java_home, dll_dir
 547   {
 548     char *pslash;
 549     os::jvm_path(buf, bufsize);
 550 
 551     // Found the full path to libjvm.so.
 552     // Now cut the path to <java_home>/jre if we can.
 553     *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
 554     pslash = strrchr(buf, '/');
 555     if (pslash != NULL) {
 556       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 557     }
 558     Arguments::set_dll_dir(buf);
 559 
 560     if (pslash != NULL) {
 561       pslash = strrchr(buf, '/');
 562       if (pslash != NULL) {
 563         *pslash = '\0';        // Get rid of /lib.
 564       }
 565     }
 566     Arguments::set_java_home(buf);
 567     set_boot_path('/', ':');
 568   }
 569 
 570   // Where to look for native libraries.
 571   {
 572     // Use dlinfo() to determine the correct java.library.path.
 573     //
 574     // If we're launched by the Java launcher, and the user
 575     // does not set java.library.path explicitly on the commandline,
 576     // the Java launcher sets LD_LIBRARY_PATH for us and unsets
 577     // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64.  In this case
 578     // dlinfo returns LD_LIBRARY_PATH + crle settings (including
 579     // /usr/lib), which is exactly what we want.
 580     //
 581     // If the user does set java.library.path, it completely
 582     // overwrites this setting, and always has.
 583     //
 584     // If we're not launched by the Java launcher, we may
 585     // get here with any/all of the LD_LIBRARY_PATH[_32|64]
 586     // settings.  Again, dlinfo does exactly what we want.
 587 
 588     Dl_serinfo     info_sz, *info = &info_sz;
 589     Dl_serpath     *path;
 590     char           *library_path;
 591     char           *common_path = buf;
 592 
 593     // Determine search path count and required buffer size.
 594     if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
 595       FREE_C_HEAP_ARRAY(char, buf);
 596       vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
 597     }
 598 
 599     // Allocate new buffer and initialize.
 600     info = (Dl_serinfo*)NEW_C_HEAP_ARRAY(char, info_sz.dls_size, mtInternal);
 601     info->dls_size = info_sz.dls_size;
 602     info->dls_cnt = info_sz.dls_cnt;
 603 
 604     // Obtain search path information.
 605     if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
 606       FREE_C_HEAP_ARRAY(char, buf);
 607       FREE_C_HEAP_ARRAY(char, info);
 608       vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
 609     }
 610 
 611     path = &info->dls_serpath[0];
 612 
 613     // Note: Due to a legacy implementation, most of the library path
 614     // is set in the launcher. This was to accomodate linking restrictions
 615     // on legacy Solaris implementations (which are no longer supported).
 616     // Eventually, all the library path setting will be done here.
 617     //
 618     // However, to prevent the proliferation of improperly built native
 619     // libraries, the new path component /usr/jdk/packages is added here.
 620 
 621     // Construct the invariant part of ld_library_path.
 622     sprintf(common_path, SYS_EXT_DIR "/lib");
 623 
 624     // Struct size is more than sufficient for the path components obtained
 625     // through the dlinfo() call, so only add additional space for the path
 626     // components explicitly added here.
 627     size_t library_path_size = info->dls_size + strlen(common_path);
 628     library_path = (char *)NEW_C_HEAP_ARRAY(char, library_path_size, mtInternal);
 629     library_path[0] = '\0';
 630 
 631     // Construct the desired Java library path from the linker's library
 632     // search path.
 633     //
 634     // For compatibility, it is optimal that we insert the additional path
 635     // components specific to the Java VM after those components specified
 636     // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
 637     // infrastructure.
 638     if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it.
 639       strcpy(library_path, common_path);
 640     } else {
 641       int inserted = 0;
 642       int i;
 643       for (i = 0; i < info->dls_cnt; i++, path++) {
 644         uint_t flags = path->dls_flags & LA_SER_MASK;
 645         if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
 646           strcat(library_path, common_path);
 647           strcat(library_path, os::path_separator());
 648           inserted = 1;
 649         }
 650         strcat(library_path, path->dls_name);
 651         strcat(library_path, os::path_separator());
 652       }
 653       // Eliminate trailing path separator.
 654       library_path[strlen(library_path)-1] = '\0';
 655     }
 656 
 657     // happens before argument parsing - can't use a trace flag
 658     // tty->print_raw("init_system_properties_values: native lib path: ");
 659     // tty->print_raw_cr(library_path);
 660 
 661     // Callee copies into its own buffer.
 662     Arguments::set_library_path(library_path);
 663 
 664     FREE_C_HEAP_ARRAY(char, library_path);
 665     FREE_C_HEAP_ARRAY(char, info);
 666   }
 667 
 668   // Extensions directories.
 669   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 670   Arguments::set_ext_dirs(buf);
 671 
 672   FREE_C_HEAP_ARRAY(char, buf);
 673 
 674 #undef SYS_EXT_DIR
 675 #undef EXTENSIONS_DIR
 676 }
 677 
 678 void os::breakpoint() {
 679   BREAKPOINT;
 680 }
 681 
 682 bool os::obsolete_option(const JavaVMOption *option) {
 683   if (!strncmp(option->optionString, "-Xt", 3)) {
 684     return true;
 685   } else if (!strncmp(option->optionString, "-Xtm", 4)) {
 686     return true;
 687   } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
 688     return true;
 689   } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
 690     return true;
 691   }
 692   return false;
 693 }
 694 
 695 bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
 696   address  stackStart  = (address)thread->stack_base();
 697   address  stackEnd    = (address)(stackStart - (address)thread->stack_size());
 698   if (sp < stackStart && sp >= stackEnd) return true;
 699   return false;
 700 }
 701 
 702 extern "C" void breakpoint() {
 703   // use debugger to set breakpoint here
 704 }
 705 
 706 static thread_t main_thread;
 707 
 708 // Thread start routine for all newly created threads
 709 extern "C" void* thread_native_entry(void* thread_addr) {
 710   // Try to randomize the cache line index of hot stack frames.
 711   // This helps when threads of the same stack traces evict each other's
 712   // cache lines. The threads can be either from the same JVM instance, or
 713   // from different JVM instances. The benefit is especially true for
 714   // processors with hyperthreading technology.
 715   static int counter = 0;
 716   int pid = os::current_process_id();
 717   alloca(((pid ^ counter++) & 7) * 128);
 718 
 719   int prio;
 720   Thread* thread = (Thread*)thread_addr;
 721 
 722   thread->initialize_thread_current();
 723 
 724   OSThread* osthr = thread->osthread();
 725 
 726   osthr->set_lwp_id(_lwp_self());  // Store lwp in case we are bound
 727   thread->_schedctl = (void *) schedctl_init();
 728 
 729   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ").",
 730     os::current_thread_id());
 731 
 732   if (UseNUMA) {
 733     int lgrp_id = os::numa_get_group_id();
 734     if (lgrp_id != -1) {
 735       thread->set_lgrp_id(lgrp_id);
 736     }
 737   }
 738 
 739   // Our priority was set when we were created, and stored in the
 740   // osthread, but couldn't be passed through to our LWP until now.
 741   // So read back the priority and set it again.
 742 
 743   if (osthr->thread_id() != -1) {
 744     if (UseThreadPriorities) {
 745       int prio = osthr->native_priority();
 746       if (ThreadPriorityVerbose) {
 747         tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is "
 748                       INTPTR_FORMAT ", setting priority: %d\n",
 749                       osthr->thread_id(), osthr->lwp_id(), prio);
 750       }
 751       os::set_native_priority(thread, prio);
 752     }
 753   } else if (ThreadPriorityVerbose) {
 754     warning("Can't set priority in _start routine, thread id hasn't been set\n");
 755   }
 756 
 757   assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
 758 
 759   // initialize signal mask for this thread
 760   os::Solaris::hotspot_sigmask(thread);
 761 
 762   thread->run();
 763 
 764   // One less thread is executing
 765   // When the VMThread gets here, the main thread may have already exited
 766   // which frees the CodeHeap containing the Atomic::dec code
 767   if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
 768     Atomic::dec(&os::Solaris::_os_thread_count);
 769   }
 770 
 771   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ").", os::current_thread_id());
 772 
 773   // If a thread has not deleted itself ("delete this") as part of its
 774   // termination sequence, we have to ensure thread-local-storage is
 775   // cleared before we actually terminate. No threads should ever be
 776   // deleted asynchronously with respect to their termination.
 777   if (Thread::current_or_null_safe() != NULL) {
 778     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 779     thread->clear_thread_current();
 780   }
 781 
 782   if (UseDetachedThreads) {
 783     thr_exit(NULL);
 784     ShouldNotReachHere();
 785   }
 786   return NULL;
 787 }
 788 
 789 static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
 790   // Allocate the OSThread object
 791   OSThread* osthread = new OSThread(NULL, NULL);
 792   if (osthread == NULL) return NULL;
 793 
 794   // Store info on the Solaris thread into the OSThread
 795   osthread->set_thread_id(thread_id);
 796   osthread->set_lwp_id(_lwp_self());
 797   thread->_schedctl = (void *) schedctl_init();
 798 
 799   if (UseNUMA) {
 800     int lgrp_id = os::numa_get_group_id();
 801     if (lgrp_id != -1) {
 802       thread->set_lgrp_id(lgrp_id);
 803     }
 804   }
 805 
 806   if (ThreadPriorityVerbose) {
 807     tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
 808                   osthread->thread_id(), osthread->lwp_id());
 809   }
 810 
 811   // Initial thread state is INITIALIZED, not SUSPENDED
 812   osthread->set_state(INITIALIZED);
 813 
 814   return osthread;
 815 }
 816 
 817 void os::Solaris::hotspot_sigmask(Thread* thread) {
 818   //Save caller's signal mask
 819   sigset_t sigmask;
 820   pthread_sigmask(SIG_SETMASK, NULL, &sigmask);
 821   OSThread *osthread = thread->osthread();
 822   osthread->set_caller_sigmask(sigmask);
 823 
 824   pthread_sigmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
 825   if (!ReduceSignalUsage) {
 826     if (thread->is_VM_thread()) {
 827       // Only the VM thread handles BREAK_SIGNAL ...
 828       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 829     } else {
 830       // ... all other threads block BREAK_SIGNAL
 831       assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
 832       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 833     }
 834   }
 835 }
 836 
 837 bool os::create_attached_thread(JavaThread* thread) {
 838 #ifdef ASSERT
 839   thread->verify_not_published();
 840 #endif
 841   OSThread* osthread = create_os_thread(thread, thr_self());
 842   if (osthread == NULL) {
 843     return false;
 844   }
 845 
 846   // Initial thread state is RUNNABLE
 847   osthread->set_state(RUNNABLE);
 848   thread->set_osthread(osthread);
 849 
 850   // initialize signal mask for this thread
 851   // and save the caller's signal mask
 852   os::Solaris::hotspot_sigmask(thread);
 853 
 854   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ").",
 855     os::current_thread_id());
 856 
 857   return true;
 858 }
 859 
 860 bool os::create_main_thread(JavaThread* thread) {
 861 #ifdef ASSERT
 862   thread->verify_not_published();
 863 #endif
 864   if (_starting_thread == NULL) {
 865     _starting_thread = create_os_thread(thread, main_thread);
 866     if (_starting_thread == NULL) {
 867       return false;
 868     }
 869   }
 870 
 871   // The primodial thread is runnable from the start
 872   _starting_thread->set_state(RUNNABLE);
 873 
 874   thread->set_osthread(_starting_thread);
 875 
 876   // initialize signal mask for this thread
 877   // and save the caller's signal mask
 878   os::Solaris::hotspot_sigmask(thread);
 879 
 880   return true;
 881 }
 882 
 883 // Helper function to trace thread attributes, similar to os::Posix::describe_pthread_attr()
 884 static char* describe_thr_create_attributes(char* buf, size_t buflen,
 885                                             size_t stacksize, long flags) {
 886   stringStream ss(buf, buflen);
 887   ss.print("stacksize: " SIZE_FORMAT "k, ", stacksize / 1024);
 888   ss.print("flags: ");
 889   #define PRINT_FLAG(f) if (flags & f) ss.print( #f " ");
 890   #define ALL(X) \
 891     X(THR_SUSPENDED) \
 892     X(THR_DETACHED) \
 893     X(THR_BOUND) \
 894     X(THR_NEW_LWP) \
 895     X(THR_DAEMON)
 896   ALL(PRINT_FLAG)
 897   #undef ALL
 898   #undef PRINT_FLAG
 899   return buf;
 900 }
 901 
 902 // return default stack size for thr_type
 903 size_t os::Posix::default_stack_size(os::ThreadType thr_type) {
 904   // default stack size when not specified by caller is 1M (2M for LP64)
 905   size_t s = (BytesPerWord >> 2) * K * K;
 906   return s;
 907 }
 908 
 909 bool os::create_thread(Thread* thread, ThreadType thr_type,
 910                        size_t req_stack_size) {
 911   // Allocate the OSThread object
 912   OSThread* osthread = new OSThread(NULL, NULL);
 913   if (osthread == NULL) {
 914     return false;
 915   }
 916 
 917   if (ThreadPriorityVerbose) {
 918     char *thrtyp;
 919     switch (thr_type) {
 920     case vm_thread:
 921       thrtyp = (char *)"vm";
 922       break;
 923     case cgc_thread:
 924       thrtyp = (char *)"cgc";
 925       break;
 926     case pgc_thread:
 927       thrtyp = (char *)"pgc";
 928       break;
 929     case java_thread:
 930       thrtyp = (char *)"java";
 931       break;
 932     case compiler_thread:
 933       thrtyp = (char *)"compiler";
 934       break;
 935     case watcher_thread:
 936       thrtyp = (char *)"watcher";
 937       break;
 938     default:
 939       thrtyp = (char *)"unknown";
 940       break;
 941     }
 942     tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
 943   }
 944 
 945   // calculate stack size if it's not specified by caller
 946   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 947 
 948   // Initial state is ALLOCATED but not INITIALIZED
 949   osthread->set_state(ALLOCATED);
 950 
 951   if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
 952     // We got lots of threads. Check if we still have some address space left.
 953     // Need to be at least 5Mb of unreserved address space. We do check by
 954     // trying to reserve some.
 955     const size_t VirtualMemoryBangSize = 20*K*K;
 956     char* mem = os::reserve_memory(VirtualMemoryBangSize);
 957     if (mem == NULL) {
 958       delete osthread;
 959       return false;
 960     } else {
 961       // Release the memory again
 962       os::release_memory(mem, VirtualMemoryBangSize);
 963     }
 964   }
 965 
 966   // Setup osthread because the child thread may need it.
 967   thread->set_osthread(osthread);
 968 
 969   // Create the Solaris thread
 970   thread_t tid = 0;
 971   long     flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED;
 972   int      status;
 973 
 974   // Mark that we don't have an lwp or thread id yet.
 975   // In case we attempt to set the priority before the thread starts.
 976   osthread->set_lwp_id(-1);
 977   osthread->set_thread_id(-1);
 978 
 979   status = thr_create(NULL, stack_size, thread_native_entry, thread, flags, &tid);
 980 
 981   char buf[64];
 982   if (status == 0) {
 983     log_info(os, thread)("Thread started (tid: " UINTX_FORMAT ", attributes: %s). ",
 984       (uintx) tid, describe_thr_create_attributes(buf, sizeof(buf), stack_size, flags));
 985   } else {
 986     log_warning(os, thread)("Failed to start thread - thr_create failed (%s) for attributes: %s.",
 987       os::errno_name(status), describe_thr_create_attributes(buf, sizeof(buf), stack_size, flags));
 988   }
 989 
 990   if (status != 0) {
 991     thread->set_osthread(NULL);
 992     // Need to clean up stuff we've allocated so far
 993     delete osthread;
 994     return false;
 995   }
 996 
 997   Atomic::inc(&os::Solaris::_os_thread_count);
 998 
 999   // Store info on the Solaris thread into the OSThread
1000   osthread->set_thread_id(tid);
1001 
1002   // Remember that we created this thread so we can set priority on it
1003   osthread->set_vm_created();
1004 
1005   // Most thread types will set an explicit priority before starting the thread,
1006   // but for those that don't we need a valid value to read back in thread_native_entry.
1007   osthread->set_native_priority(NormPriority);
1008 
1009   // Initial thread state is INITIALIZED, not SUSPENDED
1010   osthread->set_state(INITIALIZED);
1011 
1012   // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
1013   return true;
1014 }
1015 
1016 debug_only(static bool signal_sets_initialized = false);
1017 static sigset_t unblocked_sigs, vm_sigs;
1018 
1019 bool os::Solaris::is_sig_ignored(int sig) {
1020   struct sigaction oact;
1021   sigaction(sig, (struct sigaction*)NULL, &oact);
1022   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
1023                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
1024   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
1025     return true;
1026   } else {
1027     return false;
1028   }
1029 }
1030 
1031 void os::Solaris::signal_sets_init() {
1032   // Should also have an assertion stating we are still single-threaded.
1033   assert(!signal_sets_initialized, "Already initialized");
1034   // Fill in signals that are necessarily unblocked for all threads in
1035   // the VM. Currently, we unblock the following signals:
1036   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1037   //                         by -Xrs (=ReduceSignalUsage));
1038   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1039   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1040   // the dispositions or masks wrt these signals.
1041   // Programs embedding the VM that want to use the above signals for their
1042   // own purposes must, at this time, use the "-Xrs" option to prevent
1043   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1044   // (See bug 4345157, and other related bugs).
1045   // In reality, though, unblocking these signals is really a nop, since
1046   // these signals are not blocked by default.
1047   sigemptyset(&unblocked_sigs);
1048   sigaddset(&unblocked_sigs, SIGILL);
1049   sigaddset(&unblocked_sigs, SIGSEGV);
1050   sigaddset(&unblocked_sigs, SIGBUS);
1051   sigaddset(&unblocked_sigs, SIGFPE);
1052   sigaddset(&unblocked_sigs, ASYNC_SIGNAL);
1053 
1054   if (!ReduceSignalUsage) {
1055     if (!os::Solaris::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1056       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1057     }
1058     if (!os::Solaris::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1059       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1060     }
1061     if (!os::Solaris::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1062       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1063     }
1064   }
1065   // Fill in signals that are blocked by all but the VM thread.
1066   sigemptyset(&vm_sigs);
1067   if (!ReduceSignalUsage) {
1068     sigaddset(&vm_sigs, BREAK_SIGNAL);
1069   }
1070   debug_only(signal_sets_initialized = true);
1071 
1072   // For diagnostics only used in run_periodic_checks
1073   sigemptyset(&check_signal_done);
1074 }
1075 
1076 // These are signals that are unblocked while a thread is running Java.
1077 // (For some reason, they get blocked by default.)
1078 sigset_t* os::Solaris::unblocked_signals() {
1079   assert(signal_sets_initialized, "Not initialized");
1080   return &unblocked_sigs;
1081 }
1082 
1083 // These are the signals that are blocked while a (non-VM) thread is
1084 // running Java. Only the VM thread handles these signals.
1085 sigset_t* os::Solaris::vm_signals() {
1086   assert(signal_sets_initialized, "Not initialized");
1087   return &vm_sigs;
1088 }
1089 
1090 void _handle_uncaught_cxx_exception() {
1091   VMError::report_and_die("An uncaught C++ exception");
1092 }
1093 
1094 
1095 // First crack at OS-specific initialization, from inside the new thread.
1096 void os::initialize_thread(Thread* thr) {
1097   int r = thr_main();
1098   guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
1099   if (r) {
1100     JavaThread* jt = (JavaThread *)thr;
1101     assert(jt != NULL, "Sanity check");
1102     size_t stack_size;
1103     address base = jt->stack_base();
1104     if (Arguments::created_by_java_launcher()) {
1105       // Use 2MB to allow for Solaris 7 64 bit mode.
1106       stack_size = JavaThread::stack_size_at_create() == 0
1107         ? 2048*K : JavaThread::stack_size_at_create();
1108 
1109       // There are rare cases when we may have already used more than
1110       // the basic stack size allotment before this method is invoked.
1111       // Attempt to allow for a normally sized java_stack.
1112       size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1113       stack_size += ReservedSpace::page_align_down(current_stack_offset);
1114     } else {
1115       // 6269555: If we were not created by a Java launcher, i.e. if we are
1116       // running embedded in a native application, treat the primordial thread
1117       // as much like a native attached thread as possible.  This means using
1118       // the current stack size from thr_stksegment(), unless it is too large
1119       // to reliably setup guard pages.  A reasonable max size is 8MB.
1120       size_t current_size = current_stack_size();
1121       // This should never happen, but just in case....
1122       if (current_size == 0) current_size = 2 * K * K;
1123       stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1124     }
1125     address bottom = align_up(base - stack_size, os::vm_page_size());;
1126     stack_size = (size_t)(base - bottom);
1127 
1128     assert(stack_size > 0, "Stack size calculation problem");
1129 
1130     if (stack_size > jt->stack_size()) {
1131 #ifndef PRODUCT
1132       struct rlimit limits;
1133       getrlimit(RLIMIT_STACK, &limits);
1134       size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1135       assert(size >= jt->stack_size(), "Stack size problem in main thread");
1136 #endif
1137       tty->print_cr("Stack size of %d Kb exceeds current limit of %d Kb.\n"
1138                     "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1139                     "See limit(1) to increase the stack size limit.",
1140                     stack_size / K, jt->stack_size() / K);
1141       vm_exit(1);
1142     }
1143     assert(jt->stack_size() >= stack_size,
1144            "Attempt to map more stack than was allocated");
1145     jt->set_stack_size(stack_size);
1146   }
1147 
1148   // With the T2 libthread (T1 is no longer supported) threads are always bound
1149   // and we use stackbanging in all cases.
1150 
1151   os::Solaris::init_thread_fpu_state();
1152   std::set_terminate(_handle_uncaught_cxx_exception);
1153 }
1154 
1155 
1156 
1157 // Free Solaris resources related to the OSThread
1158 void os::free_thread(OSThread* osthread) {
1159   assert(osthread != NULL, "os::free_thread but osthread not set");
1160 
1161   // We are told to free resources of the argument thread,
1162   // but we can only really operate on the current thread.
1163   assert(Thread::current()->osthread() == osthread,
1164          "os::free_thread but not current thread");
1165 
1166   // Restore caller's signal mask
1167   sigset_t sigmask = osthread->caller_sigmask();
1168   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1169 
1170   delete osthread;
1171 }
1172 
1173 void os::pd_start_thread(Thread* thread) {
1174   int status = thr_continue(thread->osthread()->thread_id());
1175   assert_status(status == 0, status, "thr_continue failed");
1176 }
1177 
1178 
1179 intx os::current_thread_id() {
1180   return (intx)thr_self();
1181 }
1182 
1183 static pid_t _initial_pid = 0;
1184 
1185 int os::current_process_id() {
1186   return (int)(_initial_pid ? _initial_pid : getpid());
1187 }
1188 
1189 // gethrtime() should be monotonic according to the documentation,
1190 // but some virtualized platforms are known to break this guarantee.
1191 // getTimeNanos() must be guaranteed not to move backwards, so we
1192 // are forced to add a check here.
1193 inline hrtime_t getTimeNanos() {
1194   const hrtime_t now = gethrtime();
1195   const hrtime_t prev = max_hrtime;
1196   if (now <= prev) {
1197     return prev;   // same or retrograde time;
1198   }
1199   const hrtime_t obsv = Atomic::cmpxchg(now, (volatile jlong*)&max_hrtime, prev);
1200   assert(obsv >= prev, "invariant");   // Monotonicity
1201   // If the CAS succeeded then we're done and return "now".
1202   // If the CAS failed and the observed value "obsv" is >= now then
1203   // we should return "obsv".  If the CAS failed and now > obsv > prv then
1204   // some other thread raced this thread and installed a new value, in which case
1205   // we could either (a) retry the entire operation, (b) retry trying to install now
1206   // or (c) just return obsv.  We use (c).   No loop is required although in some cases
1207   // we might discard a higher "now" value in deference to a slightly lower but freshly
1208   // installed obsv value.   That's entirely benign -- it admits no new orderings compared
1209   // to (a) or (b) -- and greatly reduces coherence traffic.
1210   // We might also condition (c) on the magnitude of the delta between obsv and now.
1211   // Avoiding excessive CAS operations to hot RW locations is critical.
1212   // See https://blogs.oracle.com/dave/entry/cas_and_cache_trivia_invalidate
1213   return (prev == obsv) ? now : obsv;
1214 }
1215 
1216 // Time since start-up in seconds to a fine granularity.
1217 // Used by VMSelfDestructTimer and the MemProfiler.
1218 double os::elapsedTime() {
1219   return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1220 }
1221 
1222 jlong os::elapsed_counter() {
1223   return (jlong)(getTimeNanos() - first_hrtime);
1224 }
1225 
1226 jlong os::elapsed_frequency() {
1227   return hrtime_hz;
1228 }
1229 
1230 // Return the real, user, and system times in seconds from an
1231 // arbitrary fixed point in the past.
1232 bool os::getTimesSecs(double* process_real_time,
1233                       double* process_user_time,
1234                       double* process_system_time) {
1235   struct tms ticks;
1236   clock_t real_ticks = times(&ticks);
1237 
1238   if (real_ticks == (clock_t) (-1)) {
1239     return false;
1240   } else {
1241     double ticks_per_second = (double) clock_tics_per_sec;
1242     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1243     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1244     // For consistency return the real time from getTimeNanos()
1245     // converted to seconds.
1246     *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1247 
1248     return true;
1249   }
1250 }
1251 
1252 bool os::supports_vtime() { return true; }
1253 bool os::enable_vtime() { return false; }
1254 bool os::vtime_enabled() { return false; }
1255 
1256 double os::elapsedVTime() {
1257   return (double)gethrvtime() / (double)hrtime_hz;
1258 }
1259 
1260 // Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1261 jlong os::javaTimeMillis() {
1262   timeval t;
1263   if (gettimeofday(&t, NULL) == -1) {
1264     fatal("os::javaTimeMillis: gettimeofday (%s)", os::strerror(errno));
1265   }
1266   return jlong(t.tv_sec) * 1000  +  jlong(t.tv_usec) / 1000;
1267 }
1268 
1269 // Must return seconds+nanos since Jan 1 1970. This must use the same
1270 // time source as javaTimeMillis and can't use get_nsec_fromepoch as
1271 // we need better than 1ms accuracy
1272 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1273   timeval t;
1274   if (gettimeofday(&t, NULL) == -1) {
1275     fatal("os::javaTimeSystemUTC: gettimeofday (%s)", os::strerror(errno));
1276   }
1277   seconds = jlong(t.tv_sec);
1278   nanos = jlong(t.tv_usec) * 1000;
1279 }
1280 
1281 
1282 jlong os::javaTimeNanos() {
1283   return (jlong)getTimeNanos();
1284 }
1285 
1286 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1287   info_ptr->max_value = ALL_64_BITS;      // gethrtime() uses all 64 bits
1288   info_ptr->may_skip_backward = false;    // not subject to resetting or drifting
1289   info_ptr->may_skip_forward = false;     // not subject to resetting or drifting
1290   info_ptr->kind = JVMTI_TIMER_ELAPSED;   // elapsed not CPU time
1291 }
1292 
1293 char * os::local_time_string(char *buf, size_t buflen) {
1294   struct tm t;
1295   time_t long_time;
1296   time(&long_time);
1297   localtime_r(&long_time, &t);
1298   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1299                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1300                t.tm_hour, t.tm_min, t.tm_sec);
1301   return buf;
1302 }
1303 
1304 // Note: os::shutdown() might be called very early during initialization, or
1305 // called from signal handler. Before adding something to os::shutdown(), make
1306 // sure it is async-safe and can handle partially initialized VM.
1307 void os::shutdown() {
1308 
1309   // allow PerfMemory to attempt cleanup of any persistent resources
1310   perfMemory_exit();
1311 
1312   // needs to remove object in file system
1313   AttachListener::abort();
1314 
1315   // flush buffered output, finish log files
1316   ostream_abort();
1317 
1318   // Check for abort hook
1319   abort_hook_t abort_hook = Arguments::abort_hook();
1320   if (abort_hook != NULL) {
1321     abort_hook();
1322   }
1323 }
1324 
1325 // Note: os::abort() might be called very early during initialization, or
1326 // called from signal handler. Before adding something to os::abort(), make
1327 // sure it is async-safe and can handle partially initialized VM.
1328 void os::abort(bool dump_core, void* siginfo, const void* context) {
1329   os::shutdown();
1330   if (dump_core) {
1331 #ifndef PRODUCT
1332     fdStream out(defaultStream::output_fd());
1333     out.print_raw("Current thread is ");
1334     char buf[16];
1335     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1336     out.print_raw_cr(buf);
1337     out.print_raw_cr("Dumping core ...");
1338 #endif
1339     ::abort(); // dump core (for debugging)
1340   }
1341 
1342   ::exit(1);
1343 }
1344 
1345 // Die immediately, no exit hook, no abort hook, no cleanup.
1346 void os::die() {
1347   ::abort(); // dump core (for debugging)
1348 }
1349 
1350 // DLL functions
1351 
1352 const char* os::dll_file_extension() { return ".so"; }
1353 
1354 // This must be hard coded because it's the system's temporary
1355 // directory not the java application's temp directory, ala java.io.tmpdir.
1356 const char* os::get_temp_directory() { return "/tmp"; }
1357 
1358 static bool file_exists(const char* filename) {
1359   struct stat statbuf;
1360   if (filename == NULL || strlen(filename) == 0) {
1361     return false;
1362   }
1363   return os::stat(filename, &statbuf) == 0;
1364 }
1365 
1366 bool os::dll_build_name(char* buffer, size_t buflen,
1367                         const char* pname, const char* fname) {
1368   bool retval = false;
1369   const size_t pnamelen = pname ? strlen(pname) : 0;
1370 
1371   // Return error on buffer overflow.
1372   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1373     return retval;
1374   }
1375 
1376   if (pnamelen == 0) {
1377     snprintf(buffer, buflen, "lib%s.so", fname);
1378     retval = true;
1379   } else if (strchr(pname, *os::path_separator()) != NULL) {
1380     int n;
1381     char** pelements = split_path(pname, &n);
1382     if (pelements == NULL) {
1383       return false;
1384     }
1385     for (int i = 0; i < n; i++) {
1386       // really shouldn't be NULL but what the heck, check can't hurt
1387       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1388         continue; // skip the empty path values
1389       }
1390       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1391       if (file_exists(buffer)) {
1392         retval = true;
1393         break;
1394       }
1395     }
1396     // release the storage
1397     for (int i = 0; i < n; i++) {
1398       if (pelements[i] != NULL) {
1399         FREE_C_HEAP_ARRAY(char, pelements[i]);
1400       }
1401     }
1402     if (pelements != NULL) {
1403       FREE_C_HEAP_ARRAY(char*, pelements);
1404     }
1405   } else {
1406     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1407     retval = true;
1408   }
1409   return retval;
1410 }
1411 
1412 // check if addr is inside libjvm.so
1413 bool os::address_is_in_vm(address addr) {
1414   static address libjvm_base_addr;
1415   Dl_info dlinfo;
1416 
1417   if (libjvm_base_addr == NULL) {
1418     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1419       libjvm_base_addr = (address)dlinfo.dli_fbase;
1420     }
1421     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1422   }
1423 
1424   if (dladdr((void *)addr, &dlinfo) != 0) {
1425     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1426   }
1427 
1428   return false;
1429 }
1430 
1431 typedef int (*dladdr1_func_type)(void *, Dl_info *, void **, int);
1432 static dladdr1_func_type dladdr1_func = NULL;
1433 
1434 bool os::dll_address_to_function_name(address addr, char *buf,
1435                                       int buflen, int * offset,
1436                                       bool demangle) {
1437   // buf is not optional, but offset is optional
1438   assert(buf != NULL, "sanity check");
1439 
1440   Dl_info dlinfo;
1441 
1442   // dladdr1_func was initialized in os::init()
1443   if (dladdr1_func != NULL) {
1444     // yes, we have dladdr1
1445 
1446     // Support for dladdr1 is checked at runtime; it may be
1447     // available even if the vm is built on a machine that does
1448     // not have dladdr1 support.  Make sure there is a value for
1449     // RTLD_DL_SYMENT.
1450 #ifndef RTLD_DL_SYMENT
1451   #define RTLD_DL_SYMENT 1
1452 #endif
1453 #ifdef _LP64
1454     Elf64_Sym * info;
1455 #else
1456     Elf32_Sym * info;
1457 #endif
1458     if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1459                      RTLD_DL_SYMENT) != 0) {
1460       // see if we have a matching symbol that covers our address
1461       if (dlinfo.dli_saddr != NULL &&
1462           (char *)dlinfo.dli_saddr + info->st_size > (char *)addr) {
1463         if (dlinfo.dli_sname != NULL) {
1464           if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1465             jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1466           }
1467           if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1468           return true;
1469         }
1470       }
1471       // no matching symbol so try for just file info
1472       if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1473         if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1474                             buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1475           return true;
1476         }
1477       }
1478     }
1479     buf[0] = '\0';
1480     if (offset != NULL) *offset  = -1;
1481     return false;
1482   }
1483 
1484   // no, only dladdr is available
1485   if (dladdr((void *)addr, &dlinfo) != 0) {
1486     // see if we have a matching symbol
1487     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1488       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1489         jio_snprintf(buf, buflen, dlinfo.dli_sname);
1490       }
1491       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1492       return true;
1493     }
1494     // no matching symbol so try for just file info
1495     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1496       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1497                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1498         return true;
1499       }
1500     }
1501   }
1502   buf[0] = '\0';
1503   if (offset != NULL) *offset  = -1;
1504   return false;
1505 }
1506 
1507 bool os::dll_address_to_library_name(address addr, char* buf,
1508                                      int buflen, int* offset) {
1509   // buf is not optional, but offset is optional
1510   assert(buf != NULL, "sanity check");
1511 
1512   Dl_info dlinfo;
1513 
1514   if (dladdr((void*)addr, &dlinfo) != 0) {
1515     if (dlinfo.dli_fname != NULL) {
1516       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1517     }
1518     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1519       *offset = addr - (address)dlinfo.dli_fbase;
1520     }
1521     return true;
1522   }
1523 
1524   buf[0] = '\0';
1525   if (offset) *offset = -1;
1526   return false;
1527 }
1528 
1529 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1530   Dl_info dli;
1531   // Sanity check?
1532   if (dladdr(CAST_FROM_FN_PTR(void *, os::get_loaded_modules_info), &dli) == 0 ||
1533       dli.dli_fname == NULL) {
1534     return 1;
1535   }
1536 
1537   void * handle = dlopen(dli.dli_fname, RTLD_LAZY);
1538   if (handle == NULL) {
1539     return 1;
1540   }
1541 
1542   Link_map *map;
1543   dlinfo(handle, RTLD_DI_LINKMAP, &map);
1544   if (map == NULL) {
1545     dlclose(handle);
1546     return 1;
1547   }
1548 
1549   while (map->l_prev != NULL) {
1550     map = map->l_prev;
1551   }
1552 
1553   while (map != NULL) {
1554     // Iterate through all map entries and call callback with fields of interest
1555     if(callback(map->l_name, (address)map->l_addr, (address)0, param)) {
1556       dlclose(handle);
1557       return 1;
1558     }
1559     map = map->l_next;
1560   }
1561 
1562   dlclose(handle);
1563   return 0;
1564 }
1565 
1566 int _print_dll_info_cb(const char * name, address base_address, address top_address, void * param) {
1567   outputStream * out = (outputStream *) param;
1568   out->print_cr(PTR_FORMAT " \t%s", base_address, name);
1569   return 0;
1570 }
1571 
1572 void os::print_dll_info(outputStream * st) {
1573   st->print_cr("Dynamic libraries:"); st->flush();
1574   if (get_loaded_modules_info(_print_dll_info_cb, (void *)st)) {
1575     st->print_cr("Error: Cannot print dynamic libraries.");
1576   }
1577 }
1578 
1579 // Loads .dll/.so and
1580 // in case of error it checks if .dll/.so was built for the
1581 // same architecture as Hotspot is running on
1582 
1583 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1584   void * result= ::dlopen(filename, RTLD_LAZY);
1585   if (result != NULL) {
1586     // Successful loading
1587     return result;
1588   }
1589 
1590   Elf32_Ehdr elf_head;
1591 
1592   // Read system error message into ebuf
1593   // It may or may not be overwritten below
1594   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1595   ebuf[ebuflen-1]='\0';
1596   int diag_msg_max_length=ebuflen-strlen(ebuf);
1597   char* diag_msg_buf=ebuf+strlen(ebuf);
1598 
1599   if (diag_msg_max_length==0) {
1600     // No more space in ebuf for additional diagnostics message
1601     return NULL;
1602   }
1603 
1604 
1605   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1606 
1607   if (file_descriptor < 0) {
1608     // Can't open library, report dlerror() message
1609     return NULL;
1610   }
1611 
1612   bool failed_to_read_elf_head=
1613     (sizeof(elf_head)!=
1614      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1615 
1616   ::close(file_descriptor);
1617   if (failed_to_read_elf_head) {
1618     // file i/o error - report dlerror() msg
1619     return NULL;
1620   }
1621 
1622   typedef struct {
1623     Elf32_Half  code;         // Actual value as defined in elf.h
1624     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1625     char        elf_class;    // 32 or 64 bit
1626     char        endianess;    // MSB or LSB
1627     char*       name;         // String representation
1628   } arch_t;
1629 
1630   static const arch_t arch_array[]={
1631     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1632     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1633     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1634     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1635     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1636     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1637     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1638     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1639     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1640     {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM 32"}
1641   };
1642 
1643 #if  (defined IA32)
1644   static  Elf32_Half running_arch_code=EM_386;
1645 #elif   (defined AMD64)
1646   static  Elf32_Half running_arch_code=EM_X86_64;
1647 #elif  (defined IA64)
1648   static  Elf32_Half running_arch_code=EM_IA_64;
1649 #elif  (defined __sparc) && (defined _LP64)
1650   static  Elf32_Half running_arch_code=EM_SPARCV9;
1651 #elif  (defined __sparc) && (!defined _LP64)
1652   static  Elf32_Half running_arch_code=EM_SPARC;
1653 #elif  (defined __powerpc64__)
1654   static  Elf32_Half running_arch_code=EM_PPC64;
1655 #elif  (defined __powerpc__)
1656   static  Elf32_Half running_arch_code=EM_PPC;
1657 #elif (defined ARM)
1658   static  Elf32_Half running_arch_code=EM_ARM;
1659 #else
1660   #error Method os::dll_load requires that one of following is defined:\
1661        IA32, AMD64, IA64, __sparc, __powerpc__, ARM, ARM
1662 #endif
1663 
1664   // Identify compatability class for VM's architecture and library's architecture
1665   // Obtain string descriptions for architectures
1666 
1667   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1668   int running_arch_index=-1;
1669 
1670   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1671     if (running_arch_code == arch_array[i].code) {
1672       running_arch_index    = i;
1673     }
1674     if (lib_arch.code == arch_array[i].code) {
1675       lib_arch.compat_class = arch_array[i].compat_class;
1676       lib_arch.name         = arch_array[i].name;
1677     }
1678   }
1679 
1680   assert(running_arch_index != -1,
1681          "Didn't find running architecture code (running_arch_code) in arch_array");
1682   if (running_arch_index == -1) {
1683     // Even though running architecture detection failed
1684     // we may still continue with reporting dlerror() message
1685     return NULL;
1686   }
1687 
1688   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1689     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1690     return NULL;
1691   }
1692 
1693   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1694     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1695     return NULL;
1696   }
1697 
1698   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1699     if (lib_arch.name!=NULL) {
1700       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1701                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1702                  lib_arch.name, arch_array[running_arch_index].name);
1703     } else {
1704       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1705                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1706                  lib_arch.code,
1707                  arch_array[running_arch_index].name);
1708     }
1709   }
1710 
1711   return NULL;
1712 }
1713 
1714 void* os::dll_lookup(void* handle, const char* name) {
1715   return dlsym(handle, name);
1716 }
1717 
1718 void* os::get_default_process_handle() {
1719   return (void*)::dlopen(NULL, RTLD_LAZY);
1720 }
1721 
1722 int os::stat(const char *path, struct stat *sbuf) {
1723   char pathbuf[MAX_PATH];
1724   if (strlen(path) > MAX_PATH - 1) {
1725     errno = ENAMETOOLONG;
1726     return -1;
1727   }
1728   os::native_path(strcpy(pathbuf, path));
1729   return ::stat(pathbuf, sbuf);
1730 }
1731 
1732 static inline time_t get_mtime(const char* filename) {
1733   struct stat st;
1734   int ret = os::stat(filename, &st);
1735   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
1736   return st.st_mtime;
1737 }
1738 
1739 int os::compare_file_modified_times(const char* file1, const char* file2) {
1740   time_t t1 = get_mtime(file1);
1741   time_t t2 = get_mtime(file2);
1742   return t1 - t2;
1743 }
1744 
1745 static bool _print_ascii_file(const char* filename, outputStream* st) {
1746   int fd = ::open(filename, O_RDONLY);
1747   if (fd == -1) {
1748     return false;
1749   }
1750 
1751   char buf[32];
1752   int bytes;
1753   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1754     st->print_raw(buf, bytes);
1755   }
1756 
1757   ::close(fd);
1758 
1759   return true;
1760 }
1761 
1762 void os::print_os_info_brief(outputStream* st) {
1763   os::Solaris::print_distro_info(st);
1764 
1765   os::Posix::print_uname_info(st);
1766 
1767   os::Solaris::print_libversion_info(st);
1768 }
1769 
1770 void os::print_os_info(outputStream* st) {
1771   st->print("OS:");
1772 
1773   os::Solaris::print_distro_info(st);
1774 
1775   os::Posix::print_uname_info(st);
1776 
1777   os::Solaris::print_libversion_info(st);
1778 
1779   os::Posix::print_rlimit_info(st);
1780 
1781   os::Posix::print_load_average(st);
1782 }
1783 
1784 void os::Solaris::print_distro_info(outputStream* st) {
1785   if (!_print_ascii_file("/etc/release", st)) {
1786     st->print("Solaris");
1787   }
1788   st->cr();
1789 }
1790 
1791 void os::get_summary_os_info(char* buf, size_t buflen) {
1792   strncpy(buf, "Solaris", buflen);  // default to plain solaris
1793   FILE* fp = fopen("/etc/release", "r");
1794   if (fp != NULL) {
1795     char tmp[256];
1796     // Only get the first line and chop out everything but the os name.
1797     if (fgets(tmp, sizeof(tmp), fp)) {
1798       char* ptr = tmp;
1799       // skip past whitespace characters
1800       while (*ptr != '\0' && (*ptr == ' ' || *ptr == '\t' || *ptr == '\n')) ptr++;
1801       if (*ptr != '\0') {
1802         char* nl = strchr(ptr, '\n');
1803         if (nl != NULL) *nl = '\0';
1804         strncpy(buf, ptr, buflen);
1805       }
1806     }
1807     fclose(fp);
1808   }
1809 }
1810 
1811 void os::Solaris::print_libversion_info(outputStream* st) {
1812   st->print("  (T2 libthread)");
1813   st->cr();
1814 }
1815 
1816 static bool check_addr0(outputStream* st) {
1817   jboolean status = false;
1818   const int read_chunk = 200;
1819   int ret = 0;
1820   int nmap = 0;
1821   int fd = ::open("/proc/self/map",O_RDONLY);
1822   if (fd >= 0) {
1823     prmap_t *p = NULL;
1824     char *mbuff = (char *) calloc(read_chunk, sizeof(prmap_t));
1825     if (NULL == mbuff) {
1826       ::close(fd);
1827       return status;
1828     }
1829     while ((ret = ::read(fd, mbuff, read_chunk*sizeof(prmap_t))) > 0) {
1830       //check if read() has not read partial data
1831       if( 0 != ret % sizeof(prmap_t)){
1832         break;
1833       }
1834       nmap = ret / sizeof(prmap_t);
1835       p = (prmap_t *)mbuff;
1836       for(int i = 0; i < nmap; i++){
1837         if (p->pr_vaddr == 0x0) {
1838           st->print("Warning: Address: " PTR_FORMAT ", Size: " SIZE_FORMAT "K, ",p->pr_vaddr, p->pr_size/1024);
1839           st->print("Mapped file: %s, ", p->pr_mapname[0] == '\0' ? "None" : p->pr_mapname);
1840           st->print("Access: ");
1841           st->print("%s",(p->pr_mflags & MA_READ)  ? "r" : "-");
1842           st->print("%s",(p->pr_mflags & MA_WRITE) ? "w" : "-");
1843           st->print("%s",(p->pr_mflags & MA_EXEC)  ? "x" : "-");
1844           st->cr();
1845           status = true;
1846         }
1847         p++;
1848       }
1849     }
1850     free(mbuff);
1851     ::close(fd);
1852   }
1853   return status;
1854 }
1855 
1856 void os::get_summary_cpu_info(char* buf, size_t buflen) {
1857   // Get MHz with system call. We don't seem to already have this.
1858   processor_info_t stats;
1859   processorid_t id = getcpuid();
1860   int clock = 0;
1861   if (processor_info(id, &stats) != -1) {
1862     clock = stats.pi_clock;  // pi_processor_type isn't more informative than below
1863   }
1864 #ifdef AMD64
1865   snprintf(buf, buflen, "x86 64 bit %d MHz", clock);
1866 #else
1867   // must be sparc
1868   snprintf(buf, buflen, "Sparcv9 64 bit %d MHz", clock);
1869 #endif
1870 }
1871 
1872 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
1873   // Nothing to do for now.
1874 }
1875 
1876 void os::print_memory_info(outputStream* st) {
1877   st->print("Memory:");
1878   st->print(" %dk page", os::vm_page_size()>>10);
1879   st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
1880   st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
1881   st->cr();
1882   (void) check_addr0(st);
1883 }
1884 
1885 // Moved from whole group, because we need them here for diagnostic
1886 // prints.
1887 static int Maxsignum = 0;
1888 static int *ourSigFlags = NULL;
1889 
1890 int os::Solaris::get_our_sigflags(int sig) {
1891   assert(ourSigFlags!=NULL, "signal data structure not initialized");
1892   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
1893   return ourSigFlags[sig];
1894 }
1895 
1896 void os::Solaris::set_our_sigflags(int sig, int flags) {
1897   assert(ourSigFlags!=NULL, "signal data structure not initialized");
1898   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
1899   ourSigFlags[sig] = flags;
1900 }
1901 
1902 
1903 static const char* get_signal_handler_name(address handler,
1904                                            char* buf, int buflen) {
1905   int offset;
1906   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
1907   if (found) {
1908     // skip directory names
1909     const char *p1, *p2;
1910     p1 = buf;
1911     size_t len = strlen(os::file_separator());
1912     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
1913     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
1914   } else {
1915     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
1916   }
1917   return buf;
1918 }
1919 
1920 static void print_signal_handler(outputStream* st, int sig,
1921                                  char* buf, size_t buflen) {
1922   struct sigaction sa;
1923 
1924   sigaction(sig, NULL, &sa);
1925 
1926   st->print("%s: ", os::exception_name(sig, buf, buflen));
1927 
1928   address handler = (sa.sa_flags & SA_SIGINFO)
1929                   ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
1930                   : CAST_FROM_FN_PTR(address, sa.sa_handler);
1931 
1932   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
1933     st->print("SIG_DFL");
1934   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
1935     st->print("SIG_IGN");
1936   } else {
1937     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
1938   }
1939 
1940   st->print(", sa_mask[0]=");
1941   os::Posix::print_signal_set_short(st, &sa.sa_mask);
1942 
1943   address rh = VMError::get_resetted_sighandler(sig);
1944   // May be, handler was resetted by VMError?
1945   if (rh != NULL) {
1946     handler = rh;
1947     sa.sa_flags = VMError::get_resetted_sigflags(sig);
1948   }
1949 
1950   st->print(", sa_flags=");
1951   os::Posix::print_sa_flags(st, sa.sa_flags);
1952 
1953   // Check: is it our handler?
1954   if (handler == CAST_FROM_FN_PTR(address, signalHandler)) {
1955     // It is our signal handler
1956     // check for flags
1957     if (sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
1958       st->print(
1959                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
1960                 os::Solaris::get_our_sigflags(sig));
1961     }
1962   }
1963   st->cr();
1964 }
1965 
1966 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
1967   st->print_cr("Signal Handlers:");
1968   print_signal_handler(st, SIGSEGV, buf, buflen);
1969   print_signal_handler(st, SIGBUS , buf, buflen);
1970   print_signal_handler(st, SIGFPE , buf, buflen);
1971   print_signal_handler(st, SIGPIPE, buf, buflen);
1972   print_signal_handler(st, SIGXFSZ, buf, buflen);
1973   print_signal_handler(st, SIGILL , buf, buflen);
1974   print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
1975   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
1976   print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
1977   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
1978   print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
1979 }
1980 
1981 static char saved_jvm_path[MAXPATHLEN] = { 0 };
1982 
1983 // Find the full path to the current module, libjvm.so
1984 void os::jvm_path(char *buf, jint buflen) {
1985   // Error checking.
1986   if (buflen < MAXPATHLEN) {
1987     assert(false, "must use a large-enough buffer");
1988     buf[0] = '\0';
1989     return;
1990   }
1991   // Lazy resolve the path to current module.
1992   if (saved_jvm_path[0] != 0) {
1993     strcpy(buf, saved_jvm_path);
1994     return;
1995   }
1996 
1997   Dl_info dlinfo;
1998   int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
1999   assert(ret != 0, "cannot locate libjvm");
2000   if (ret != 0 && dlinfo.dli_fname != NULL) {
2001     if (os::Posix::realpath((char *)dlinfo.dli_fname, buf, buflen) == NULL) {
2002       return;
2003     }
2004   } else {
2005     buf[0] = '\0';
2006     return;
2007   }
2008 
2009   if (Arguments::sun_java_launcher_is_altjvm()) {
2010     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2011     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2012     // If "/jre/lib/" appears at the right place in the string, then
2013     // assume we are installed in a JDK and we're done.  Otherwise, check
2014     // for a JAVA_HOME environment variable and fix up the path so it
2015     // looks like libjvm.so is installed there (append a fake suffix
2016     // hotspot/libjvm.so).
2017     const char *p = buf + strlen(buf) - 1;
2018     for (int count = 0; p > buf && count < 5; ++count) {
2019       for (--p; p > buf && *p != '/'; --p)
2020         /* empty */ ;
2021     }
2022 
2023     if (strncmp(p, "/jre/lib/", 9) != 0) {
2024       // Look for JAVA_HOME in the environment.
2025       char* java_home_var = ::getenv("JAVA_HOME");
2026       if (java_home_var != NULL && java_home_var[0] != 0) {
2027         char* jrelib_p;
2028         int   len;
2029 
2030         // Check the current module name "libjvm.so".
2031         p = strrchr(buf, '/');
2032         assert(strstr(p, "/libjvm") == p, "invalid library name");
2033 
2034         if (os::Posix::realpath(java_home_var, buf, buflen) == NULL) {
2035           return;
2036         }
2037         // determine if this is a legacy image or modules image
2038         // modules image doesn't have "jre" subdirectory
2039         len = strlen(buf);
2040         assert(len < buflen, "Ran out of buffer space");
2041         jrelib_p = buf + len;
2042         snprintf(jrelib_p, buflen-len, "/jre/lib");
2043         if (0 != access(buf, F_OK)) {
2044           snprintf(jrelib_p, buflen-len, "/lib");
2045         }
2046 
2047         if (0 == access(buf, F_OK)) {
2048           // Use current module name "libjvm.so"
2049           len = strlen(buf);
2050           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2051         } else {
2052           // Go back to path of .so
2053           if (os::Posix::realpath((char *)dlinfo.dli_fname, buf, buflen) == NULL) {
2054             return;
2055           }
2056         }
2057       }
2058     }
2059   }
2060 
2061   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2062   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2063 }
2064 
2065 
2066 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2067   // no prefix required, not even "_"
2068 }
2069 
2070 
2071 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2072   // no suffix required
2073 }
2074 
2075 // This method is a copy of JDK's sysGetLastErrorString
2076 // from src/solaris/hpi/src/system_md.c
2077 
2078 size_t os::lasterror(char *buf, size_t len) {
2079   if (errno == 0)  return 0;
2080 
2081   const char *s = os::strerror(errno);
2082   size_t n = ::strlen(s);
2083   if (n >= len) {
2084     n = len - 1;
2085   }
2086   ::strncpy(buf, s, n);
2087   buf[n] = '\0';
2088   return n;
2089 }
2090 
2091 
2092 // sun.misc.Signal
2093 
2094 extern "C" {
2095   static void UserHandler(int sig, void *siginfo, void *context) {
2096     // Ctrl-C is pressed during error reporting, likely because the error
2097     // handler fails to abort. Let VM die immediately.
2098     if (sig == SIGINT && is_error_reported()) {
2099       os::die();
2100     }
2101 
2102     os::signal_notify(sig);
2103     // We do not need to reinstate the signal handler each time...
2104   }
2105 }
2106 
2107 void* os::user_handler() {
2108   return CAST_FROM_FN_PTR(void*, UserHandler);
2109 }
2110 
2111 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2112   struct timespec ts;
2113   unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2114 
2115   return ts;
2116 }
2117 
2118 extern "C" {
2119   typedef void (*sa_handler_t)(int);
2120   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2121 }
2122 
2123 void* os::signal(int signal_number, void* handler) {
2124   struct sigaction sigAct, oldSigAct;
2125   sigfillset(&(sigAct.sa_mask));
2126   sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2127   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2128 
2129   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2130     // -1 means registration failed
2131     return (void *)-1;
2132   }
2133 
2134   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2135 }
2136 
2137 void os::signal_raise(int signal_number) {
2138   raise(signal_number);
2139 }
2140 
2141 // The following code is moved from os.cpp for making this
2142 // code platform specific, which it is by its very nature.
2143 
2144 // a counter for each possible signal value
2145 static int Sigexit = 0;
2146 static jint *pending_signals = NULL;
2147 static int *preinstalled_sigs = NULL;
2148 static struct sigaction *chainedsigactions = NULL;
2149 static sema_t sig_sem;
2150 typedef int (*version_getting_t)();
2151 version_getting_t os::Solaris::get_libjsig_version = NULL;
2152 
2153 int os::sigexitnum_pd() {
2154   assert(Sigexit > 0, "signal memory not yet initialized");
2155   return Sigexit;
2156 }
2157 
2158 void os::Solaris::init_signal_mem() {
2159   // Initialize signal structures
2160   Maxsignum = SIGRTMAX;
2161   Sigexit = Maxsignum+1;
2162   assert(Maxsignum >0, "Unable to obtain max signal number");
2163 
2164   // pending_signals has one int per signal
2165   // The additional signal is for SIGEXIT - exit signal to signal_thread
2166   pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1), mtInternal);
2167   memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2168 
2169   if (UseSignalChaining) {
2170     chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2171                                                    * (Maxsignum + 1), mtInternal);
2172     memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2173     preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2174     memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2175   }
2176   ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2177   memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2178 }
2179 
2180 void os::signal_init_pd() {
2181   int ret;
2182 
2183   ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2184   assert(ret == 0, "sema_init() failed");
2185 }
2186 
2187 void os::signal_notify(int signal_number) {
2188   int ret;
2189 
2190   Atomic::inc(&pending_signals[signal_number]);
2191   ret = ::sema_post(&sig_sem);
2192   assert(ret == 0, "sema_post() failed");
2193 }
2194 
2195 static int check_pending_signals(bool wait_for_signal) {
2196   int ret;
2197   while (true) {
2198     for (int i = 0; i < Sigexit + 1; i++) {
2199       jint n = pending_signals[i];
2200       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2201         return i;
2202       }
2203     }
2204     if (!wait_for_signal) {
2205       return -1;
2206     }
2207     JavaThread *thread = JavaThread::current();
2208     ThreadBlockInVM tbivm(thread);
2209 
2210     bool threadIsSuspended;
2211     do {
2212       thread->set_suspend_equivalent();
2213       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2214       while ((ret = ::sema_wait(&sig_sem)) == EINTR)
2215         ;
2216       assert(ret == 0, "sema_wait() failed");
2217 
2218       // were we externally suspended while we were waiting?
2219       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2220       if (threadIsSuspended) {
2221         // The semaphore has been incremented, but while we were waiting
2222         // another thread suspended us. We don't want to continue running
2223         // while suspended because that would surprise the thread that
2224         // suspended us.
2225         ret = ::sema_post(&sig_sem);
2226         assert(ret == 0, "sema_post() failed");
2227 
2228         thread->java_suspend_self();
2229       }
2230     } while (threadIsSuspended);
2231   }
2232 }
2233 
2234 int os::signal_lookup() {
2235   return check_pending_signals(false);
2236 }
2237 
2238 int os::signal_wait() {
2239   return check_pending_signals(true);
2240 }
2241 
2242 ////////////////////////////////////////////////////////////////////////////////
2243 // Virtual Memory
2244 
2245 static int page_size = -1;
2246 
2247 // The mmap MAP_ALIGN flag is supported on Solaris 9 and later.  init_2() will
2248 // clear this var if support is not available.
2249 static bool has_map_align = true;
2250 
2251 int os::vm_page_size() {
2252   assert(page_size != -1, "must call os::init");
2253   return page_size;
2254 }
2255 
2256 // Solaris allocates memory by pages.
2257 int os::vm_allocation_granularity() {
2258   assert(page_size != -1, "must call os::init");
2259   return page_size;
2260 }
2261 
2262 static bool recoverable_mmap_error(int err) {
2263   // See if the error is one we can let the caller handle. This
2264   // list of errno values comes from the Solaris mmap(2) man page.
2265   switch (err) {
2266   case EBADF:
2267   case EINVAL:
2268   case ENOTSUP:
2269     // let the caller deal with these errors
2270     return true;
2271 
2272   default:
2273     // Any remaining errors on this OS can cause our reserved mapping
2274     // to be lost. That can cause confusion where different data
2275     // structures think they have the same memory mapped. The worst
2276     // scenario is if both the VM and a library think they have the
2277     // same memory mapped.
2278     return false;
2279   }
2280 }
2281 
2282 static void warn_fail_commit_memory(char* addr, size_t bytes, bool exec,
2283                                     int err) {
2284   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2285           ", %d) failed; error='%s' (errno=%d)", addr, bytes, exec,
2286           os::strerror(err), err);
2287 }
2288 
2289 static void warn_fail_commit_memory(char* addr, size_t bytes,
2290                                     size_t alignment_hint, bool exec,
2291                                     int err) {
2292   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2293           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, bytes,
2294           alignment_hint, exec, os::strerror(err), err);
2295 }
2296 
2297 int os::Solaris::commit_memory_impl(char* addr, size_t bytes, bool exec) {
2298   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2299   size_t size = bytes;
2300   char *res = Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2301   if (res != NULL) {
2302     if (UseNUMAInterleaving) {
2303       numa_make_global(addr, bytes);
2304     }
2305     return 0;
2306   }
2307 
2308   int err = errno;  // save errno from mmap() call in mmap_chunk()
2309 
2310   if (!recoverable_mmap_error(err)) {
2311     warn_fail_commit_memory(addr, bytes, exec, err);
2312     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "committing reserved memory.");
2313   }
2314 
2315   return err;
2316 }
2317 
2318 bool os::pd_commit_memory(char* addr, size_t bytes, bool exec) {
2319   return Solaris::commit_memory_impl(addr, bytes, exec) == 0;
2320 }
2321 
2322 void os::pd_commit_memory_or_exit(char* addr, size_t bytes, bool exec,
2323                                   const char* mesg) {
2324   assert(mesg != NULL, "mesg must be specified");
2325   int err = os::Solaris::commit_memory_impl(addr, bytes, exec);
2326   if (err != 0) {
2327     // the caller wants all commit errors to exit with the specified mesg:
2328     warn_fail_commit_memory(addr, bytes, exec, err);
2329     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2330   }
2331 }
2332 
2333 size_t os::Solaris::page_size_for_alignment(size_t alignment) {
2334   assert(is_aligned(alignment, (size_t) vm_page_size()),
2335          SIZE_FORMAT " is not aligned to " SIZE_FORMAT,
2336          alignment, (size_t) vm_page_size());
2337 
2338   for (int i = 0; _page_sizes[i] != 0; i++) {
2339     if (is_aligned(alignment, _page_sizes[i])) {
2340       return _page_sizes[i];
2341     }
2342   }
2343 
2344   return (size_t) vm_page_size();
2345 }
2346 
2347 int os::Solaris::commit_memory_impl(char* addr, size_t bytes,
2348                                     size_t alignment_hint, bool exec) {
2349   int err = Solaris::commit_memory_impl(addr, bytes, exec);
2350   if (err == 0 && UseLargePages && alignment_hint > 0) {
2351     assert(is_aligned(bytes, alignment_hint),
2352            SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, alignment_hint);
2353 
2354     // The syscall memcntl requires an exact page size (see man memcntl for details).
2355     size_t page_size = page_size_for_alignment(alignment_hint);
2356     if (page_size > (size_t) vm_page_size()) {
2357       (void)Solaris::setup_large_pages(addr, bytes, page_size);
2358     }
2359   }
2360   return err;
2361 }
2362 
2363 bool os::pd_commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2364                           bool exec) {
2365   return Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec) == 0;
2366 }
2367 
2368 void os::pd_commit_memory_or_exit(char* addr, size_t bytes,
2369                                   size_t alignment_hint, bool exec,
2370                                   const char* mesg) {
2371   assert(mesg != NULL, "mesg must be specified");
2372   int err = os::Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec);
2373   if (err != 0) {
2374     // the caller wants all commit errors to exit with the specified mesg:
2375     warn_fail_commit_memory(addr, bytes, alignment_hint, exec, err);
2376     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2377   }
2378 }
2379 
2380 // Uncommit the pages in a specified region.
2381 void os::pd_free_memory(char* addr, size_t bytes, size_t alignment_hint) {
2382   if (madvise(addr, bytes, MADV_FREE) < 0) {
2383     debug_only(warning("MADV_FREE failed."));
2384     return;
2385   }
2386 }
2387 
2388 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2389   return os::commit_memory(addr, size, !ExecMem);
2390 }
2391 
2392 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2393   return os::uncommit_memory(addr, size);
2394 }
2395 
2396 // Change the page size in a given range.
2397 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2398   assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2399   assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2400   if (UseLargePages) {
2401     size_t page_size = Solaris::page_size_for_alignment(alignment_hint);
2402     if (page_size > (size_t) vm_page_size()) {
2403       Solaris::setup_large_pages(addr, bytes, page_size);
2404     }
2405   }
2406 }
2407 
2408 // Tell the OS to make the range local to the first-touching LWP
2409 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2410   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2411   if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2412     debug_only(warning("MADV_ACCESS_LWP failed."));
2413   }
2414 }
2415 
2416 // Tell the OS that this range would be accessed from different LWPs.
2417 void os::numa_make_global(char *addr, size_t bytes) {
2418   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2419   if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2420     debug_only(warning("MADV_ACCESS_MANY failed."));
2421   }
2422 }
2423 
2424 // Get the number of the locality groups.
2425 size_t os::numa_get_groups_num() {
2426   size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2427   return n != -1 ? n : 1;
2428 }
2429 
2430 // Get a list of leaf locality groups. A leaf lgroup is group that
2431 // doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2432 // board. An LWP is assigned to one of these groups upon creation.
2433 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2434   if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2435     ids[0] = 0;
2436     return 1;
2437   }
2438   int result_size = 0, top = 1, bottom = 0, cur = 0;
2439   for (int k = 0; k < size; k++) {
2440     int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2441                                    (Solaris::lgrp_id_t*)&ids[top], size - top);
2442     if (r == -1) {
2443       ids[0] = 0;
2444       return 1;
2445     }
2446     if (!r) {
2447       // That's a leaf node.
2448       assert(bottom <= cur, "Sanity check");
2449       // Check if the node has memory
2450       if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2451                                   NULL, 0, LGRP_RSRC_MEM) > 0) {
2452         ids[bottom++] = ids[cur];
2453       }
2454     }
2455     top += r;
2456     cur++;
2457   }
2458   if (bottom == 0) {
2459     // Handle a situation, when the OS reports no memory available.
2460     // Assume UMA architecture.
2461     ids[0] = 0;
2462     return 1;
2463   }
2464   return bottom;
2465 }
2466 
2467 // Detect the topology change. Typically happens during CPU plugging-unplugging.
2468 bool os::numa_topology_changed() {
2469   int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2470   if (is_stale != -1 && is_stale) {
2471     Solaris::lgrp_fini(Solaris::lgrp_cookie());
2472     Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2473     assert(c != 0, "Failure to initialize LGRP API");
2474     Solaris::set_lgrp_cookie(c);
2475     return true;
2476   }
2477   return false;
2478 }
2479 
2480 // Get the group id of the current LWP.
2481 int os::numa_get_group_id() {
2482   int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2483   if (lgrp_id == -1) {
2484     return 0;
2485   }
2486   const int size = os::numa_get_groups_num();
2487   int *ids = (int*)alloca(size * sizeof(int));
2488 
2489   // Get the ids of all lgroups with memory; r is the count.
2490   int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2491                                   (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2492   if (r <= 0) {
2493     return 0;
2494   }
2495   return ids[os::random() % r];
2496 }
2497 
2498 // Request information about the page.
2499 bool os::get_page_info(char *start, page_info* info) {
2500   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2501   uint64_t addr = (uintptr_t)start;
2502   uint64_t outdata[2];
2503   uint_t validity = 0;
2504 
2505   if (meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2506     return false;
2507   }
2508 
2509   info->size = 0;
2510   info->lgrp_id = -1;
2511 
2512   if ((validity & 1) != 0) {
2513     if ((validity & 2) != 0) {
2514       info->lgrp_id = outdata[0];
2515     }
2516     if ((validity & 4) != 0) {
2517       info->size = outdata[1];
2518     }
2519     return true;
2520   }
2521   return false;
2522 }
2523 
2524 // Scan the pages from start to end until a page different than
2525 // the one described in the info parameter is encountered.
2526 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2527                      page_info* page_found) {
2528   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2529   const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2530   uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT + 1];
2531   uint_t validity[MAX_MEMINFO_CNT];
2532 
2533   size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2534   uint64_t p = (uint64_t)start;
2535   while (p < (uint64_t)end) {
2536     addrs[0] = p;
2537     size_t addrs_count = 1;
2538     while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] + page_size < (uint64_t)end) {
2539       addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2540       addrs_count++;
2541     }
2542 
2543     if (meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2544       return NULL;
2545     }
2546 
2547     size_t i = 0;
2548     for (; i < addrs_count; i++) {
2549       if ((validity[i] & 1) != 0) {
2550         if ((validity[i] & 4) != 0) {
2551           if (outdata[types * i + 1] != page_expected->size) {
2552             break;
2553           }
2554         } else if (page_expected->size != 0) {
2555           break;
2556         }
2557 
2558         if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2559           if (outdata[types * i] != page_expected->lgrp_id) {
2560             break;
2561           }
2562         }
2563       } else {
2564         return NULL;
2565       }
2566     }
2567 
2568     if (i < addrs_count) {
2569       if ((validity[i] & 2) != 0) {
2570         page_found->lgrp_id = outdata[types * i];
2571       } else {
2572         page_found->lgrp_id = -1;
2573       }
2574       if ((validity[i] & 4) != 0) {
2575         page_found->size = outdata[types * i + 1];
2576       } else {
2577         page_found->size = 0;
2578       }
2579       return (char*)addrs[i];
2580     }
2581 
2582     p = addrs[addrs_count - 1] + page_size;
2583   }
2584   return end;
2585 }
2586 
2587 bool os::pd_uncommit_memory(char* addr, size_t bytes) {
2588   size_t size = bytes;
2589   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2590   // uncommitted page. Otherwise, the read/write might succeed if we
2591   // have enough swap space to back the physical page.
2592   return
2593     NULL != Solaris::mmap_chunk(addr, size,
2594                                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
2595                                 PROT_NONE);
2596 }
2597 
2598 char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
2599   char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
2600 
2601   if (b == MAP_FAILED) {
2602     return NULL;
2603   }
2604   return b;
2605 }
2606 
2607 char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes,
2608                              size_t alignment_hint, bool fixed) {
2609   char* addr = requested_addr;
2610   int flags = MAP_PRIVATE | MAP_NORESERVE;
2611 
2612   assert(!(fixed && (alignment_hint > 0)),
2613          "alignment hint meaningless with fixed mmap");
2614 
2615   if (fixed) {
2616     flags |= MAP_FIXED;
2617   } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
2618     flags |= MAP_ALIGN;
2619     addr = (char*) alignment_hint;
2620   }
2621 
2622   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2623   // uncommitted page. Otherwise, the read/write might succeed if we
2624   // have enough swap space to back the physical page.
2625   return mmap_chunk(addr, bytes, flags, PROT_NONE);
2626 }
2627 
2628 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2629                             size_t alignment_hint) {
2630   char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint,
2631                                   (requested_addr != NULL));
2632 
2633   guarantee(requested_addr == NULL || requested_addr == addr,
2634             "OS failed to return requested mmap address.");
2635   return addr;
2636 }
2637 
2638 // Reserve memory at an arbitrary address, only if that area is
2639 // available (and not reserved for something else).
2640 
2641 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2642   const int max_tries = 10;
2643   char* base[max_tries];
2644   size_t size[max_tries];
2645 
2646   // Solaris adds a gap between mmap'ed regions.  The size of the gap
2647   // is dependent on the requested size and the MMU.  Our initial gap
2648   // value here is just a guess and will be corrected later.
2649   bool had_top_overlap = false;
2650   bool have_adjusted_gap = false;
2651   size_t gap = 0x400000;
2652 
2653   // Assert only that the size is a multiple of the page size, since
2654   // that's all that mmap requires, and since that's all we really know
2655   // about at this low abstraction level.  If we need higher alignment,
2656   // we can either pass an alignment to this method or verify alignment
2657   // in one of the methods further up the call chain.  See bug 5044738.
2658   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2659 
2660   // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
2661   // Give it a try, if the kernel honors the hint we can return immediately.
2662   char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
2663 
2664   volatile int err = errno;
2665   if (addr == requested_addr) {
2666     return addr;
2667   } else if (addr != NULL) {
2668     pd_unmap_memory(addr, bytes);
2669   }
2670 
2671   if (log_is_enabled(Warning, os)) {
2672     char buf[256];
2673     buf[0] = '\0';
2674     if (addr == NULL) {
2675       jio_snprintf(buf, sizeof(buf), ": %s", os::strerror(err));
2676     }
2677     log_info(os)("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
2678             PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
2679             "%s", bytes, requested_addr, addr, buf);
2680   }
2681 
2682   // Address hint method didn't work.  Fall back to the old method.
2683   // In theory, once SNV becomes our oldest supported platform, this
2684   // code will no longer be needed.
2685   //
2686   // Repeatedly allocate blocks until the block is allocated at the
2687   // right spot. Give up after max_tries.
2688   int i;
2689   for (i = 0; i < max_tries; ++i) {
2690     base[i] = reserve_memory(bytes);
2691 
2692     if (base[i] != NULL) {
2693       // Is this the block we wanted?
2694       if (base[i] == requested_addr) {
2695         size[i] = bytes;
2696         break;
2697       }
2698 
2699       // check that the gap value is right
2700       if (had_top_overlap && !have_adjusted_gap) {
2701         size_t actual_gap = base[i-1] - base[i] - bytes;
2702         if (gap != actual_gap) {
2703           // adjust the gap value and retry the last 2 allocations
2704           assert(i > 0, "gap adjustment code problem");
2705           have_adjusted_gap = true;  // adjust the gap only once, just in case
2706           gap = actual_gap;
2707           log_info(os)("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
2708           unmap_memory(base[i], bytes);
2709           unmap_memory(base[i-1], size[i-1]);
2710           i-=2;
2711           continue;
2712         }
2713       }
2714 
2715       // Does this overlap the block we wanted? Give back the overlapped
2716       // parts and try again.
2717       //
2718       // There is still a bug in this code: if top_overlap == bytes,
2719       // the overlap is offset from requested region by the value of gap.
2720       // In this case giving back the overlapped part will not work,
2721       // because we'll give back the entire block at base[i] and
2722       // therefore the subsequent allocation will not generate a new gap.
2723       // This could be fixed with a new algorithm that used larger
2724       // or variable size chunks to find the requested region -
2725       // but such a change would introduce additional complications.
2726       // It's rare enough that the planets align for this bug,
2727       // so we'll just wait for a fix for 6204603/5003415 which
2728       // will provide a mmap flag to allow us to avoid this business.
2729 
2730       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2731       if (top_overlap >= 0 && top_overlap < bytes) {
2732         had_top_overlap = true;
2733         unmap_memory(base[i], top_overlap);
2734         base[i] += top_overlap;
2735         size[i] = bytes - top_overlap;
2736       } else {
2737         size_t bottom_overlap = base[i] + bytes - requested_addr;
2738         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2739           if (bottom_overlap == 0) {
2740             log_info(os)("attempt_reserve_memory_at: possible alignment bug");
2741           }
2742           unmap_memory(requested_addr, bottom_overlap);
2743           size[i] = bytes - bottom_overlap;
2744         } else {
2745           size[i] = bytes;
2746         }
2747       }
2748     }
2749   }
2750 
2751   // Give back the unused reserved pieces.
2752 
2753   for (int j = 0; j < i; ++j) {
2754     if (base[j] != NULL) {
2755       unmap_memory(base[j], size[j]);
2756     }
2757   }
2758 
2759   return (i < max_tries) ? requested_addr : NULL;
2760 }
2761 
2762 bool os::pd_release_memory(char* addr, size_t bytes) {
2763   size_t size = bytes;
2764   return munmap(addr, size) == 0;
2765 }
2766 
2767 static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
2768   assert(addr == (char*)align_down((uintptr_t)addr, os::vm_page_size()),
2769          "addr must be page aligned");
2770   int retVal = mprotect(addr, bytes, prot);
2771   return retVal == 0;
2772 }
2773 
2774 // Protect memory (Used to pass readonly pages through
2775 // JNI GetArray<type>Elements with empty arrays.)
2776 // Also, used for serialization page and for compressed oops null pointer
2777 // checking.
2778 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2779                         bool is_committed) {
2780   unsigned int p = 0;
2781   switch (prot) {
2782   case MEM_PROT_NONE: p = PROT_NONE; break;
2783   case MEM_PROT_READ: p = PROT_READ; break;
2784   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2785   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2786   default:
2787     ShouldNotReachHere();
2788   }
2789   // is_committed is unused.
2790   return solaris_mprotect(addr, bytes, p);
2791 }
2792 
2793 // guard_memory and unguard_memory only happens within stack guard pages.
2794 // Since ISM pertains only to the heap, guard and unguard memory should not
2795 /// happen with an ISM region.
2796 bool os::guard_memory(char* addr, size_t bytes) {
2797   return solaris_mprotect(addr, bytes, PROT_NONE);
2798 }
2799 
2800 bool os::unguard_memory(char* addr, size_t bytes) {
2801   return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
2802 }
2803 
2804 // Large page support
2805 static size_t _large_page_size = 0;
2806 
2807 // Insertion sort for small arrays (descending order).
2808 static void insertion_sort_descending(size_t* array, int len) {
2809   for (int i = 0; i < len; i++) {
2810     size_t val = array[i];
2811     for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
2812       size_t tmp = array[key];
2813       array[key] = array[key - 1];
2814       array[key - 1] = tmp;
2815     }
2816   }
2817 }
2818 
2819 bool os::Solaris::mpss_sanity_check(bool warn, size_t* page_size) {
2820   const unsigned int usable_count = VM_Version::page_size_count();
2821   if (usable_count == 1) {
2822     return false;
2823   }
2824 
2825   // Find the right getpagesizes interface.  When solaris 11 is the minimum
2826   // build platform, getpagesizes() (without the '2') can be called directly.
2827   typedef int (*gps_t)(size_t[], int);
2828   gps_t gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes2"));
2829   if (gps_func == NULL) {
2830     gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes"));
2831     if (gps_func == NULL) {
2832       if (warn) {
2833         warning("MPSS is not supported by the operating system.");
2834       }
2835       return false;
2836     }
2837   }
2838 
2839   // Fill the array of page sizes.
2840   int n = (*gps_func)(_page_sizes, page_sizes_max);
2841   assert(n > 0, "Solaris bug?");
2842 
2843   if (n == page_sizes_max) {
2844     // Add a sentinel value (necessary only if the array was completely filled
2845     // since it is static (zeroed at initialization)).
2846     _page_sizes[--n] = 0;
2847     DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
2848   }
2849   assert(_page_sizes[n] == 0, "missing sentinel");
2850   trace_page_sizes("available page sizes", _page_sizes, n);
2851 
2852   if (n == 1) return false;     // Only one page size available.
2853 
2854   // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
2855   // select up to usable_count elements.  First sort the array, find the first
2856   // acceptable value, then copy the usable sizes to the top of the array and
2857   // trim the rest.  Make sure to include the default page size :-).
2858   //
2859   // A better policy could get rid of the 4M limit by taking the sizes of the
2860   // important VM memory regions (java heap and possibly the code cache) into
2861   // account.
2862   insertion_sort_descending(_page_sizes, n);
2863   const size_t size_limit =
2864     FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
2865   int beg;
2866   for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */;
2867   const int end = MIN2((int)usable_count, n) - 1;
2868   for (int cur = 0; cur < end; ++cur, ++beg) {
2869     _page_sizes[cur] = _page_sizes[beg];
2870   }
2871   _page_sizes[end] = vm_page_size();
2872   _page_sizes[end + 1] = 0;
2873 
2874   if (_page_sizes[end] > _page_sizes[end - 1]) {
2875     // Default page size is not the smallest; sort again.
2876     insertion_sort_descending(_page_sizes, end + 1);
2877   }
2878   *page_size = _page_sizes[0];
2879 
2880   trace_page_sizes("usable page sizes", _page_sizes, end + 1);
2881   return true;
2882 }
2883 
2884 void os::large_page_init() {
2885   if (UseLargePages) {
2886     // print a warning if any large page related flag is specified on command line
2887     bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
2888                            !FLAG_IS_DEFAULT(LargePageSizeInBytes);
2889 
2890     UseLargePages = Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
2891   }
2892 }
2893 
2894 bool os::Solaris::is_valid_page_size(size_t bytes) {
2895   for (int i = 0; _page_sizes[i] != 0; i++) {
2896     if (_page_sizes[i] == bytes) {
2897       return true;
2898     }
2899   }
2900   return false;
2901 }
2902 
2903 bool os::Solaris::setup_large_pages(caddr_t start, size_t bytes, size_t align) {
2904   assert(is_valid_page_size(align), SIZE_FORMAT " is not a valid page size", align);
2905   assert(is_aligned((void*) start, align),
2906          PTR_FORMAT " is not aligned to " SIZE_FORMAT, p2i((void*) start), align);
2907   assert(is_aligned(bytes, align),
2908          SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, align);
2909 
2910   // Signal to OS that we want large pages for addresses
2911   // from addr, addr + bytes
2912   struct memcntl_mha mpss_struct;
2913   mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
2914   mpss_struct.mha_pagesize = align;
2915   mpss_struct.mha_flags = 0;
2916   // Upon successful completion, memcntl() returns 0
2917   if (memcntl(start, bytes, MC_HAT_ADVISE, (caddr_t) &mpss_struct, 0, 0)) {
2918     debug_only(warning("Attempt to use MPSS failed."));
2919     return false;
2920   }
2921   return true;
2922 }
2923 
2924 char* os::reserve_memory_special(size_t size, size_t alignment, char* addr, bool exec) {
2925   fatal("os::reserve_memory_special should not be called on Solaris.");
2926   return NULL;
2927 }
2928 
2929 bool os::release_memory_special(char* base, size_t bytes) {
2930   fatal("os::release_memory_special should not be called on Solaris.");
2931   return false;
2932 }
2933 
2934 size_t os::large_page_size() {
2935   return _large_page_size;
2936 }
2937 
2938 // MPSS allows application to commit large page memory on demand; with ISM
2939 // the entire memory region must be allocated as shared memory.
2940 bool os::can_commit_large_page_memory() {
2941   return true;
2942 }
2943 
2944 bool os::can_execute_large_page_memory() {
2945   return true;
2946 }
2947 
2948 // Read calls from inside the vm need to perform state transitions
2949 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2950   size_t res;
2951   JavaThread* thread = (JavaThread*)Thread::current();
2952   assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
2953   ThreadBlockInVM tbiv(thread);
2954   RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
2955   return res;
2956 }
2957 
2958 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
2959   size_t res;
2960   JavaThread* thread = (JavaThread*)Thread::current();
2961   assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
2962   ThreadBlockInVM tbiv(thread);
2963   RESTARTABLE(::pread(fd, buf, (size_t) nBytes, offset), res);
2964   return res;
2965 }
2966 
2967 size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
2968   size_t res;
2969   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
2970          "Assumed _thread_in_native");
2971   RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
2972   return res;
2973 }
2974 
2975 void os::naked_short_sleep(jlong ms) {
2976   assert(ms < 1000, "Un-interruptable sleep, short time use only");
2977 
2978   // usleep is deprecated and removed from POSIX, in favour of nanosleep, but
2979   // Solaris requires -lrt for this.
2980   usleep((ms * 1000));
2981 
2982   return;
2983 }
2984 
2985 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2986 void os::infinite_sleep() {
2987   while (true) {    // sleep forever ...
2988     ::sleep(100);   // ... 100 seconds at a time
2989   }
2990 }
2991 
2992 // Used to convert frequent JVM_Yield() to nops
2993 bool os::dont_yield() {
2994   if (DontYieldALot) {
2995     static hrtime_t last_time = 0;
2996     hrtime_t diff = getTimeNanos() - last_time;
2997 
2998     if (diff < DontYieldALotInterval * 1000000) {
2999       return true;
3000     }
3001 
3002     last_time += diff;
3003 
3004     return false;
3005   } else {
3006     return false;
3007   }
3008 }
3009 
3010 // Note that yield semantics are defined by the scheduling class to which
3011 // the thread currently belongs.  Typically, yield will _not yield to
3012 // other equal or higher priority threads that reside on the dispatch queues
3013 // of other CPUs.
3014 
3015 void os::naked_yield() {
3016   thr_yield();
3017 }
3018 
3019 // Interface for setting lwp priorities.  We are using T2 libthread,
3020 // which forces the use of bound threads, so all of our threads will
3021 // be assigned to real lwp's.  Using the thr_setprio function is
3022 // meaningless in this mode so we must adjust the real lwp's priority.
3023 // The routines below implement the getting and setting of lwp priorities.
3024 //
3025 // Note: There are three priority scales used on Solaris.  Java priotities
3026 //       which range from 1 to 10, libthread "thr_setprio" scale which range
3027 //       from 0 to 127, and the current scheduling class of the process we
3028 //       are running in.  This is typically from -60 to +60.
3029 //       The setting of the lwp priorities in done after a call to thr_setprio
3030 //       so Java priorities are mapped to libthread priorities and we map from
3031 //       the latter to lwp priorities.  We don't keep priorities stored in
3032 //       Java priorities since some of our worker threads want to set priorities
3033 //       higher than all Java threads.
3034 //
3035 // For related information:
3036 // (1)  man -s 2 priocntl
3037 // (2)  man -s 4 priocntl
3038 // (3)  man dispadmin
3039 // =    librt.so
3040 // =    libthread/common/rtsched.c - thrp_setlwpprio().
3041 // =    ps -cL <pid> ... to validate priority.
3042 // =    sched_get_priority_min and _max
3043 //              pthread_create
3044 //              sched_setparam
3045 //              pthread_setschedparam
3046 //
3047 // Assumptions:
3048 // +    We assume that all threads in the process belong to the same
3049 //              scheduling class.   IE. an homogenous process.
3050 // +    Must be root or in IA group to change change "interactive" attribute.
3051 //              Priocntl() will fail silently.  The only indication of failure is when
3052 //              we read-back the value and notice that it hasn't changed.
3053 // +    Interactive threads enter the runq at the head, non-interactive at the tail.
3054 // +    For RT, change timeslice as well.  Invariant:
3055 //              constant "priority integral"
3056 //              Konst == TimeSlice * (60-Priority)
3057 //              Given a priority, compute appropriate timeslice.
3058 // +    Higher numerical values have higher priority.
3059 
3060 // sched class attributes
3061 typedef struct {
3062   int   schedPolicy;              // classID
3063   int   maxPrio;
3064   int   minPrio;
3065 } SchedInfo;
3066 
3067 
3068 static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
3069 
3070 #ifdef ASSERT
3071 static int  ReadBackValidate = 1;
3072 #endif
3073 static int  myClass     = 0;
3074 static int  myMin       = 0;
3075 static int  myMax       = 0;
3076 static int  myCur       = 0;
3077 static bool priocntl_enable = false;
3078 
3079 static const int criticalPrio = FXCriticalPriority;
3080 static int java_MaxPriority_to_os_priority = 0; // Saved mapping
3081 
3082 
3083 // lwp_priocntl_init
3084 //
3085 // Try to determine the priority scale for our process.
3086 //
3087 // Return errno or 0 if OK.
3088 //
3089 static int lwp_priocntl_init() {
3090   int rslt;
3091   pcinfo_t ClassInfo;
3092   pcparms_t ParmInfo;
3093   int i;
3094 
3095   if (!UseThreadPriorities) return 0;
3096 
3097   // If ThreadPriorityPolicy is 1, switch tables
3098   if (ThreadPriorityPolicy == 1) {
3099     for (i = 0; i < CriticalPriority+1; i++)
3100       os::java_to_os_priority[i] = prio_policy1[i];
3101   }
3102   if (UseCriticalJavaThreadPriority) {
3103     // MaxPriority always maps to the FX scheduling class and criticalPrio.
3104     // See set_native_priority() and set_lwp_class_and_priority().
3105     // Save original MaxPriority mapping in case attempt to
3106     // use critical priority fails.
3107     java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
3108     // Set negative to distinguish from other priorities
3109     os::java_to_os_priority[MaxPriority] = -criticalPrio;
3110   }
3111 
3112   // Get IDs for a set of well-known scheduling classes.
3113   // TODO-FIXME: GETCLINFO returns the current # of classes in the
3114   // the system.  We should have a loop that iterates over the
3115   // classID values, which are known to be "small" integers.
3116 
3117   strcpy(ClassInfo.pc_clname, "TS");
3118   ClassInfo.pc_cid = -1;
3119   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3120   if (rslt < 0) return errno;
3121   assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3122   tsLimits.schedPolicy = ClassInfo.pc_cid;
3123   tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3124   tsLimits.minPrio = -tsLimits.maxPrio;
3125 
3126   strcpy(ClassInfo.pc_clname, "IA");
3127   ClassInfo.pc_cid = -1;
3128   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3129   if (rslt < 0) return errno;
3130   assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3131   iaLimits.schedPolicy = ClassInfo.pc_cid;
3132   iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3133   iaLimits.minPrio = -iaLimits.maxPrio;
3134 
3135   strcpy(ClassInfo.pc_clname, "RT");
3136   ClassInfo.pc_cid = -1;
3137   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3138   if (rslt < 0) return errno;
3139   assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3140   rtLimits.schedPolicy = ClassInfo.pc_cid;
3141   rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3142   rtLimits.minPrio = 0;
3143 
3144   strcpy(ClassInfo.pc_clname, "FX");
3145   ClassInfo.pc_cid = -1;
3146   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3147   if (rslt < 0) return errno;
3148   assert(ClassInfo.pc_cid != -1, "cid for FX class is -1");
3149   fxLimits.schedPolicy = ClassInfo.pc_cid;
3150   fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
3151   fxLimits.minPrio = 0;
3152 
3153   // Query our "current" scheduling class.
3154   // This will normally be IA, TS or, rarely, FX or RT.
3155   memset(&ParmInfo, 0, sizeof(ParmInfo));
3156   ParmInfo.pc_cid = PC_CLNULL;
3157   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3158   if (rslt < 0) return errno;
3159   myClass = ParmInfo.pc_cid;
3160 
3161   // We now know our scheduling classId, get specific information
3162   // about the class.
3163   ClassInfo.pc_cid = myClass;
3164   ClassInfo.pc_clname[0] = 0;
3165   rslt = priocntl((idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
3166   if (rslt < 0) return errno;
3167 
3168   if (ThreadPriorityVerbose) {
3169     tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3170   }
3171 
3172   memset(&ParmInfo, 0, sizeof(pcparms_t));
3173   ParmInfo.pc_cid = PC_CLNULL;
3174   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3175   if (rslt < 0) return errno;
3176 
3177   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3178     myMin = rtLimits.minPrio;
3179     myMax = rtLimits.maxPrio;
3180   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3181     iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3182     myMin = iaLimits.minPrio;
3183     myMax = iaLimits.maxPrio;
3184     myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3185   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3186     tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3187     myMin = tsLimits.minPrio;
3188     myMax = tsLimits.maxPrio;
3189     myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3190   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3191     fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
3192     myMin = fxLimits.minPrio;
3193     myMax = fxLimits.maxPrio;
3194     myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
3195   } else {
3196     // No clue - punt
3197     if (ThreadPriorityVerbose) {
3198       tty->print_cr("Unknown scheduling class: %s ... \n",
3199                     ClassInfo.pc_clname);
3200     }
3201     return EINVAL;      // no clue, punt
3202   }
3203 
3204   if (ThreadPriorityVerbose) {
3205     tty->print_cr("Thread priority Range: [%d..%d]\n", myMin, myMax);
3206   }
3207 
3208   priocntl_enable = true;  // Enable changing priorities
3209   return 0;
3210 }
3211 
3212 #define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3213 #define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3214 #define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3215 #define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
3216 
3217 
3218 // scale_to_lwp_priority
3219 //
3220 // Convert from the libthread "thr_setprio" scale to our current
3221 // lwp scheduling class scale.
3222 //
3223 static int scale_to_lwp_priority(int rMin, int rMax, int x) {
3224   int v;
3225 
3226   if (x == 127) return rMax;            // avoid round-down
3227   v = (((x*(rMax-rMin)))/128)+rMin;
3228   return v;
3229 }
3230 
3231 
3232 // set_lwp_class_and_priority
3233 int set_lwp_class_and_priority(int ThreadID, int lwpid,
3234                                int newPrio, int new_class, bool scale) {
3235   int rslt;
3236   int Actual, Expected, prv;
3237   pcparms_t ParmInfo;                   // for GET-SET
3238 #ifdef ASSERT
3239   pcparms_t ReadBack;                   // for readback
3240 #endif
3241 
3242   // Set priority via PC_GETPARMS, update, PC_SETPARMS
3243   // Query current values.
3244   // TODO: accelerate this by eliminating the PC_GETPARMS call.
3245   // Cache "pcparms_t" in global ParmCache.
3246   // TODO: elide set-to-same-value
3247 
3248   // If something went wrong on init, don't change priorities.
3249   if (!priocntl_enable) {
3250     if (ThreadPriorityVerbose) {
3251       tty->print_cr("Trying to set priority but init failed, ignoring");
3252     }
3253     return EINVAL;
3254   }
3255 
3256   // If lwp hasn't started yet, just return
3257   // the _start routine will call us again.
3258   if (lwpid <= 0) {
3259     if (ThreadPriorityVerbose) {
3260       tty->print_cr("deferring the set_lwp_class_and_priority of thread "
3261                     INTPTR_FORMAT " to %d, lwpid not set",
3262                     ThreadID, newPrio);
3263     }
3264     return 0;
3265   }
3266 
3267   if (ThreadPriorityVerbose) {
3268     tty->print_cr ("set_lwp_class_and_priority("
3269                    INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3270                    ThreadID, lwpid, newPrio);
3271   }
3272 
3273   memset(&ParmInfo, 0, sizeof(pcparms_t));
3274   ParmInfo.pc_cid = PC_CLNULL;
3275   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3276   if (rslt < 0) return errno;
3277 
3278   int cur_class = ParmInfo.pc_cid;
3279   ParmInfo.pc_cid = (id_t)new_class;
3280 
3281   if (new_class == rtLimits.schedPolicy) {
3282     rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3283     rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
3284                                                        rtLimits.maxPrio, newPrio)
3285                                : newPrio;
3286     rtInfo->rt_tqsecs  = RT_NOCHANGE;
3287     rtInfo->rt_tqnsecs = RT_NOCHANGE;
3288     if (ThreadPriorityVerbose) {
3289       tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3290     }
3291   } else if (new_class == iaLimits.schedPolicy) {
3292     iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3293     int maxClamped     = MIN2(iaLimits.maxPrio,
3294                               cur_class == new_class
3295                               ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
3296     iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
3297                                                        maxClamped, newPrio)
3298                                : newPrio;
3299     iaInfo->ia_uprilim = cur_class == new_class
3300                            ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
3301     iaInfo->ia_mode    = IA_NOCHANGE;
3302     if (ThreadPriorityVerbose) {
3303       tty->print_cr("IA: [%d...%d] %d->%d\n",
3304                     iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3305     }
3306   } else if (new_class == tsLimits.schedPolicy) {
3307     tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3308     int maxClamped     = MIN2(tsLimits.maxPrio,
3309                               cur_class == new_class
3310                               ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
3311     tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
3312                                                        maxClamped, newPrio)
3313                                : newPrio;
3314     tsInfo->ts_uprilim = cur_class == new_class
3315                            ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
3316     if (ThreadPriorityVerbose) {
3317       tty->print_cr("TS: [%d...%d] %d->%d\n",
3318                     tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3319     }
3320   } else if (new_class == fxLimits.schedPolicy) {
3321     fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
3322     int maxClamped     = MIN2(fxLimits.maxPrio,
3323                               cur_class == new_class
3324                               ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
3325     fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
3326                                                        maxClamped, newPrio)
3327                                : newPrio;
3328     fxInfo->fx_uprilim = cur_class == new_class
3329                            ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
3330     fxInfo->fx_tqsecs  = FX_NOCHANGE;
3331     fxInfo->fx_tqnsecs = FX_NOCHANGE;
3332     if (ThreadPriorityVerbose) {
3333       tty->print_cr("FX: [%d...%d] %d->%d\n",
3334                     fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
3335     }
3336   } else {
3337     if (ThreadPriorityVerbose) {
3338       tty->print_cr("Unknown new scheduling class %d\n", new_class);
3339     }
3340     return EINVAL;    // no clue, punt
3341   }
3342 
3343   rslt = priocntl(P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3344   if (ThreadPriorityVerbose && rslt) {
3345     tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3346   }
3347   if (rslt < 0) return errno;
3348 
3349 #ifdef ASSERT
3350   // Sanity check: read back what we just attempted to set.
3351   // In theory it could have changed in the interim ...
3352   //
3353   // The priocntl system call is tricky.
3354   // Sometimes it'll validate the priority value argument and
3355   // return EINVAL if unhappy.  At other times it fails silently.
3356   // Readbacks are prudent.
3357 
3358   if (!ReadBackValidate) return 0;
3359 
3360   memset(&ReadBack, 0, sizeof(pcparms_t));
3361   ReadBack.pc_cid = PC_CLNULL;
3362   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3363   assert(rslt >= 0, "priocntl failed");
3364   Actual = Expected = 0xBAD;
3365   assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3366   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3367     Actual   = RTPRI(ReadBack)->rt_pri;
3368     Expected = RTPRI(ParmInfo)->rt_pri;
3369   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3370     Actual   = IAPRI(ReadBack)->ia_upri;
3371     Expected = IAPRI(ParmInfo)->ia_upri;
3372   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3373     Actual   = TSPRI(ReadBack)->ts_upri;
3374     Expected = TSPRI(ParmInfo)->ts_upri;
3375   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3376     Actual   = FXPRI(ReadBack)->fx_upri;
3377     Expected = FXPRI(ParmInfo)->fx_upri;
3378   } else {
3379     if (ThreadPriorityVerbose) {
3380       tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
3381                     ParmInfo.pc_cid);
3382     }
3383   }
3384 
3385   if (Actual != Expected) {
3386     if (ThreadPriorityVerbose) {
3387       tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
3388                      lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
3389     }
3390   }
3391 #endif
3392 
3393   return 0;
3394 }
3395 
3396 // Solaris only gives access to 128 real priorities at a time,
3397 // so we expand Java's ten to fill this range.  This would be better
3398 // if we dynamically adjusted relative priorities.
3399 //
3400 // The ThreadPriorityPolicy option allows us to select 2 different
3401 // priority scales.
3402 //
3403 // ThreadPriorityPolicy=0
3404 // Since the Solaris' default priority is MaximumPriority, we do not
3405 // set a priority lower than Max unless a priority lower than
3406 // NormPriority is requested.
3407 //
3408 // ThreadPriorityPolicy=1
3409 // This mode causes the priority table to get filled with
3410 // linear values.  NormPriority get's mapped to 50% of the
3411 // Maximum priority an so on.  This will cause VM threads
3412 // to get unfair treatment against other Solaris processes
3413 // which do not explicitly alter their thread priorities.
3414 
3415 int os::java_to_os_priority[CriticalPriority + 1] = {
3416   -99999,         // 0 Entry should never be used
3417 
3418   0,              // 1 MinPriority
3419   32,             // 2
3420   64,             // 3
3421 
3422   96,             // 4
3423   127,            // 5 NormPriority
3424   127,            // 6
3425 
3426   127,            // 7
3427   127,            // 8
3428   127,            // 9 NearMaxPriority
3429 
3430   127,            // 10 MaxPriority
3431 
3432   -criticalPrio   // 11 CriticalPriority
3433 };
3434 
3435 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3436   OSThread* osthread = thread->osthread();
3437 
3438   // Save requested priority in case the thread hasn't been started
3439   osthread->set_native_priority(newpri);
3440 
3441   // Check for critical priority request
3442   bool fxcritical = false;
3443   if (newpri == -criticalPrio) {
3444     fxcritical = true;
3445     newpri = criticalPrio;
3446   }
3447 
3448   assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
3449   if (!UseThreadPriorities) return OS_OK;
3450 
3451   int status = 0;
3452 
3453   if (!fxcritical) {
3454     // Use thr_setprio only if we have a priority that thr_setprio understands
3455     status = thr_setprio(thread->osthread()->thread_id(), newpri);
3456   }
3457 
3458   int lwp_status =
3459           set_lwp_class_and_priority(osthread->thread_id(),
3460                                      osthread->lwp_id(),
3461                                      newpri,
3462                                      fxcritical ? fxLimits.schedPolicy : myClass,
3463                                      !fxcritical);
3464   if (lwp_status != 0 && fxcritical) {
3465     // Try again, this time without changing the scheduling class
3466     newpri = java_MaxPriority_to_os_priority;
3467     lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
3468                                             osthread->lwp_id(),
3469                                             newpri, myClass, false);
3470   }
3471   status |= lwp_status;
3472   return (status == 0) ? OS_OK : OS_ERR;
3473 }
3474 
3475 
3476 OSReturn os::get_native_priority(const Thread* const thread,
3477                                  int *priority_ptr) {
3478   int p;
3479   if (!UseThreadPriorities) {
3480     *priority_ptr = NormalPriority;
3481     return OS_OK;
3482   }
3483   int status = thr_getprio(thread->osthread()->thread_id(), &p);
3484   if (status != 0) {
3485     return OS_ERR;
3486   }
3487   *priority_ptr = p;
3488   return OS_OK;
3489 }
3490 
3491 
3492 // Hint to the underlying OS that a task switch would not be good.
3493 // Void return because it's a hint and can fail.
3494 void os::hint_no_preempt() {
3495   schedctl_start(schedctl_init());
3496 }
3497 
3498 static void resume_clear_context(OSThread *osthread) {
3499   osthread->set_ucontext(NULL);
3500 }
3501 
3502 static void suspend_save_context(OSThread *osthread, ucontext_t* context) {
3503   osthread->set_ucontext(context);
3504 }
3505 
3506 static PosixSemaphore sr_semaphore;
3507 
3508 void os::Solaris::SR_handler(Thread* thread, ucontext_t* uc) {
3509   // Save and restore errno to avoid confusing native code with EINTR
3510   // after sigsuspend.
3511   int old_errno = errno;
3512 
3513   OSThread* osthread = thread->osthread();
3514   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3515 
3516   os::SuspendResume::State current = osthread->sr.state();
3517   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3518     suspend_save_context(osthread, uc);
3519 
3520     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3521     os::SuspendResume::State state = osthread->sr.suspended();
3522     if (state == os::SuspendResume::SR_SUSPENDED) {
3523       sigset_t suspend_set;  // signals for sigsuspend()
3524 
3525       // get current set of blocked signals and unblock resume signal
3526       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3527       sigdelset(&suspend_set, ASYNC_SIGNAL);
3528 
3529       sr_semaphore.signal();
3530       // wait here until we are resumed
3531       while (1) {
3532         sigsuspend(&suspend_set);
3533 
3534         os::SuspendResume::State result = osthread->sr.running();
3535         if (result == os::SuspendResume::SR_RUNNING) {
3536           sr_semaphore.signal();
3537           break;
3538         }
3539       }
3540 
3541     } else if (state == os::SuspendResume::SR_RUNNING) {
3542       // request was cancelled, continue
3543     } else {
3544       ShouldNotReachHere();
3545     }
3546 
3547     resume_clear_context(osthread);
3548   } else if (current == os::SuspendResume::SR_RUNNING) {
3549     // request was cancelled, continue
3550   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3551     // ignore
3552   } else {
3553     // ignore
3554   }
3555 
3556   errno = old_errno;
3557 }
3558 
3559 void os::print_statistics() {
3560 }
3561 
3562 bool os::message_box(const char* title, const char* message) {
3563   int i;
3564   fdStream err(defaultStream::error_fd());
3565   for (i = 0; i < 78; i++) err.print_raw("=");
3566   err.cr();
3567   err.print_raw_cr(title);
3568   for (i = 0; i < 78; i++) err.print_raw("-");
3569   err.cr();
3570   err.print_raw_cr(message);
3571   for (i = 0; i < 78; i++) err.print_raw("=");
3572   err.cr();
3573 
3574   char buf[16];
3575   // Prevent process from exiting upon "read error" without consuming all CPU
3576   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3577 
3578   return buf[0] == 'y' || buf[0] == 'Y';
3579 }
3580 
3581 static int sr_notify(OSThread* osthread) {
3582   int status = thr_kill(osthread->thread_id(), ASYNC_SIGNAL);
3583   assert_status(status == 0, status, "thr_kill");
3584   return status;
3585 }
3586 
3587 // "Randomly" selected value for how long we want to spin
3588 // before bailing out on suspending a thread, also how often
3589 // we send a signal to a thread we want to resume
3590 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3591 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3592 
3593 static bool do_suspend(OSThread* osthread) {
3594   assert(osthread->sr.is_running(), "thread should be running");
3595   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3596 
3597   // mark as suspended and send signal
3598   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3599     // failed to switch, state wasn't running?
3600     ShouldNotReachHere();
3601     return false;
3602   }
3603 
3604   if (sr_notify(osthread) != 0) {
3605     ShouldNotReachHere();
3606   }
3607 
3608   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3609   while (true) {
3610     if (sr_semaphore.timedwait(0, 2000 * NANOSECS_PER_MILLISEC)) {
3611       break;
3612     } else {
3613       // timeout
3614       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3615       if (cancelled == os::SuspendResume::SR_RUNNING) {
3616         return false;
3617       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3618         // make sure that we consume the signal on the semaphore as well
3619         sr_semaphore.wait();
3620         break;
3621       } else {
3622         ShouldNotReachHere();
3623         return false;
3624       }
3625     }
3626   }
3627 
3628   guarantee(osthread->sr.is_suspended(), "Must be suspended");
3629   return true;
3630 }
3631 
3632 static void do_resume(OSThread* osthread) {
3633   assert(osthread->sr.is_suspended(), "thread should be suspended");
3634   assert(!sr_semaphore.trywait(), "invalid semaphore state");
3635 
3636   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3637     // failed to switch to WAKEUP_REQUEST
3638     ShouldNotReachHere();
3639     return;
3640   }
3641 
3642   while (true) {
3643     if (sr_notify(osthread) == 0) {
3644       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3645         if (osthread->sr.is_running()) {
3646           return;
3647         }
3648       }
3649     } else {
3650       ShouldNotReachHere();
3651     }
3652   }
3653 
3654   guarantee(osthread->sr.is_running(), "Must be running!");
3655 }
3656 
3657 void os::SuspendedThreadTask::internal_do_task() {
3658   if (do_suspend(_thread->osthread())) {
3659     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
3660     do_task(context);
3661     do_resume(_thread->osthread());
3662   }
3663 }
3664 
3665 class PcFetcher : public os::SuspendedThreadTask {
3666  public:
3667   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
3668   ExtendedPC result();
3669  protected:
3670   void do_task(const os::SuspendedThreadTaskContext& context);
3671  private:
3672   ExtendedPC _epc;
3673 };
3674 
3675 ExtendedPC PcFetcher::result() {
3676   guarantee(is_done(), "task is not done yet.");
3677   return _epc;
3678 }
3679 
3680 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
3681   Thread* thread = context.thread();
3682   OSThread* osthread = thread->osthread();
3683   if (osthread->ucontext() != NULL) {
3684     _epc = os::Solaris::ucontext_get_pc((const ucontext_t *) context.ucontext());
3685   } else {
3686     // NULL context is unexpected, double-check this is the VMThread
3687     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3688   }
3689 }
3690 
3691 // A lightweight implementation that does not suspend the target thread and
3692 // thus returns only a hint. Used for profiling only!
3693 ExtendedPC os::get_thread_pc(Thread* thread) {
3694   // Make sure that it is called by the watcher and the Threads lock is owned.
3695   assert(Thread::current()->is_Watcher_thread(), "Must be watcher and own Threads_lock");
3696   // For now, is only used to profile the VM Thread
3697   assert(thread->is_VM_thread(), "Can only be called for VMThread");
3698   PcFetcher fetcher(thread);
3699   fetcher.run();
3700   return fetcher.result();
3701 }
3702 
3703 
3704 // This does not do anything on Solaris. This is basically a hook for being
3705 // able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
3706 void os::os_exception_wrapper(java_call_t f, JavaValue* value,
3707                               const methodHandle& method, JavaCallArguments* args,
3708                               Thread* thread) {
3709   f(value, method, args, thread);
3710 }
3711 
3712 // This routine may be used by user applications as a "hook" to catch signals.
3713 // The user-defined signal handler must pass unrecognized signals to this
3714 // routine, and if it returns true (non-zero), then the signal handler must
3715 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3716 // routine will never retun false (zero), but instead will execute a VM panic
3717 // routine kill the process.
3718 //
3719 // If this routine returns false, it is OK to call it again.  This allows
3720 // the user-defined signal handler to perform checks either before or after
3721 // the VM performs its own checks.  Naturally, the user code would be making
3722 // a serious error if it tried to handle an exception (such as a null check
3723 // or breakpoint) that the VM was generating for its own correct operation.
3724 //
3725 // This routine may recognize any of the following kinds of signals:
3726 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
3727 // ASYNC_SIGNAL.
3728 // It should be consulted by handlers for any of those signals.
3729 //
3730 // The caller of this routine must pass in the three arguments supplied
3731 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3732 // field of the structure passed to sigaction().  This routine assumes that
3733 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3734 //
3735 // Note that the VM will print warnings if it detects conflicting signal
3736 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3737 //
3738 extern "C" JNIEXPORT int JVM_handle_solaris_signal(int signo,
3739                                                    siginfo_t* siginfo,
3740                                                    void* ucontext,
3741                                                    int abort_if_unrecognized);
3742 
3743 
3744 void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
3745   int orig_errno = errno;  // Preserve errno value over signal handler.
3746   JVM_handle_solaris_signal(sig, info, ucVoid, true);
3747   errno = orig_errno;
3748 }
3749 
3750 // This boolean allows users to forward their own non-matching signals
3751 // to JVM_handle_solaris_signal, harmlessly.
3752 bool os::Solaris::signal_handlers_are_installed = false;
3753 
3754 // For signal-chaining
3755 bool os::Solaris::libjsig_is_loaded = false;
3756 typedef struct sigaction *(*get_signal_t)(int);
3757 get_signal_t os::Solaris::get_signal_action = NULL;
3758 
3759 struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
3760   struct sigaction *actp = NULL;
3761 
3762   if ((libjsig_is_loaded)  && (sig <= Maxsignum)) {
3763     // Retrieve the old signal handler from libjsig
3764     actp = (*get_signal_action)(sig);
3765   }
3766   if (actp == NULL) {
3767     // Retrieve the preinstalled signal handler from jvm
3768     actp = get_preinstalled_handler(sig);
3769   }
3770 
3771   return actp;
3772 }
3773 
3774 static bool call_chained_handler(struct sigaction *actp, int sig,
3775                                  siginfo_t *siginfo, void *context) {
3776   // Call the old signal handler
3777   if (actp->sa_handler == SIG_DFL) {
3778     // It's more reasonable to let jvm treat it as an unexpected exception
3779     // instead of taking the default action.
3780     return false;
3781   } else if (actp->sa_handler != SIG_IGN) {
3782     if ((actp->sa_flags & SA_NODEFER) == 0) {
3783       // automaticlly block the signal
3784       sigaddset(&(actp->sa_mask), sig);
3785     }
3786 
3787     sa_handler_t hand;
3788     sa_sigaction_t sa;
3789     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3790     // retrieve the chained handler
3791     if (siginfo_flag_set) {
3792       sa = actp->sa_sigaction;
3793     } else {
3794       hand = actp->sa_handler;
3795     }
3796 
3797     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3798       actp->sa_handler = SIG_DFL;
3799     }
3800 
3801     // try to honor the signal mask
3802     sigset_t oset;
3803     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3804 
3805     // call into the chained handler
3806     if (siginfo_flag_set) {
3807       (*sa)(sig, siginfo, context);
3808     } else {
3809       (*hand)(sig);
3810     }
3811 
3812     // restore the signal mask
3813     pthread_sigmask(SIG_SETMASK, &oset, 0);
3814   }
3815   // Tell jvm's signal handler the signal is taken care of.
3816   return true;
3817 }
3818 
3819 bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3820   bool chained = false;
3821   // signal-chaining
3822   if (UseSignalChaining) {
3823     struct sigaction *actp = get_chained_signal_action(sig);
3824     if (actp != NULL) {
3825       chained = call_chained_handler(actp, sig, siginfo, context);
3826     }
3827   }
3828   return chained;
3829 }
3830 
3831 struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
3832   assert((chainedsigactions != (struct sigaction *)NULL) &&
3833          (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3834   if (preinstalled_sigs[sig] != 0) {
3835     return &chainedsigactions[sig];
3836   }
3837   return NULL;
3838 }
3839 
3840 void os::Solaris::save_preinstalled_handler(int sig,
3841                                             struct sigaction& oldAct) {
3842   assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
3843   assert((chainedsigactions != (struct sigaction *)NULL) &&
3844          (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3845   chainedsigactions[sig] = oldAct;
3846   preinstalled_sigs[sig] = 1;
3847 }
3848 
3849 void os::Solaris::set_signal_handler(int sig, bool set_installed,
3850                                      bool oktochain) {
3851   // Check for overwrite.
3852   struct sigaction oldAct;
3853   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3854   void* oldhand =
3855       oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3856                           : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3857   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3858       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3859       oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
3860     if (AllowUserSignalHandlers || !set_installed) {
3861       // Do not overwrite; user takes responsibility to forward to us.
3862       return;
3863     } else if (UseSignalChaining) {
3864       if (oktochain) {
3865         // save the old handler in jvm
3866         save_preinstalled_handler(sig, oldAct);
3867       } else {
3868         vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal.");
3869       }
3870       // libjsig also interposes the sigaction() call below and saves the
3871       // old sigaction on it own.
3872     } else {
3873       fatal("Encountered unexpected pre-existing sigaction handler "
3874             "%#lx for signal %d.", (long)oldhand, sig);
3875     }
3876   }
3877 
3878   struct sigaction sigAct;
3879   sigfillset(&(sigAct.sa_mask));
3880   sigAct.sa_handler = SIG_DFL;
3881 
3882   sigAct.sa_sigaction = signalHandler;
3883   // Handle SIGSEGV on alternate signal stack if
3884   // not using stack banging
3885   if (!UseStackBanging && sig == SIGSEGV) {
3886     sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
3887   } else {
3888     sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
3889   }
3890   os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
3891 
3892   sigaction(sig, &sigAct, &oldAct);
3893 
3894   void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3895                                        : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3896   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3897 }
3898 
3899 
3900 #define DO_SIGNAL_CHECK(sig)                      \
3901   do {                                            \
3902     if (!sigismember(&check_signal_done, sig)) {  \
3903       os::Solaris::check_signal_handler(sig);     \
3904     }                                             \
3905   } while (0)
3906 
3907 // This method is a periodic task to check for misbehaving JNI applications
3908 // under CheckJNI, we can add any periodic checks here
3909 
3910 void os::run_periodic_checks() {
3911   // A big source of grief is hijacking virt. addr 0x0 on Solaris,
3912   // thereby preventing a NULL checks.
3913   if (!check_addr0_done) check_addr0_done = check_addr0(tty);
3914 
3915   if (check_signals == false) return;
3916 
3917   // SEGV and BUS if overridden could potentially prevent
3918   // generation of hs*.log in the event of a crash, debugging
3919   // such a case can be very challenging, so we absolutely
3920   // check for the following for a good measure:
3921   DO_SIGNAL_CHECK(SIGSEGV);
3922   DO_SIGNAL_CHECK(SIGILL);
3923   DO_SIGNAL_CHECK(SIGFPE);
3924   DO_SIGNAL_CHECK(SIGBUS);
3925   DO_SIGNAL_CHECK(SIGPIPE);
3926   DO_SIGNAL_CHECK(SIGXFSZ);
3927   DO_SIGNAL_CHECK(ASYNC_SIGNAL);
3928 
3929   // ReduceSignalUsage allows the user to override these handlers
3930   // see comments at the very top and jvm_solaris.h
3931   if (!ReduceSignalUsage) {
3932     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3933     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3934     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3935     DO_SIGNAL_CHECK(BREAK_SIGNAL);
3936   }
3937 }
3938 
3939 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3940 
3941 static os_sigaction_t os_sigaction = NULL;
3942 
3943 void os::Solaris::check_signal_handler(int sig) {
3944   char buf[O_BUFLEN];
3945   address jvmHandler = NULL;
3946 
3947   struct sigaction act;
3948   if (os_sigaction == NULL) {
3949     // only trust the default sigaction, in case it has been interposed
3950     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3951     if (os_sigaction == NULL) return;
3952   }
3953 
3954   os_sigaction(sig, (struct sigaction*)NULL, &act);
3955 
3956   address thisHandler = (act.sa_flags & SA_SIGINFO)
3957     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3958     : CAST_FROM_FN_PTR(address, act.sa_handler);
3959 
3960 
3961   switch (sig) {
3962   case SIGSEGV:
3963   case SIGBUS:
3964   case SIGFPE:
3965   case SIGPIPE:
3966   case SIGXFSZ:
3967   case SIGILL:
3968   case ASYNC_SIGNAL:
3969     jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
3970     break;
3971 
3972   case SHUTDOWN1_SIGNAL:
3973   case SHUTDOWN2_SIGNAL:
3974   case SHUTDOWN3_SIGNAL:
3975   case BREAK_SIGNAL:
3976     jvmHandler = (address)user_handler();
3977     break;
3978 
3979   default:
3980       return;
3981   }
3982 
3983   if (thisHandler != jvmHandler) {
3984     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3985     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3986     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3987     // No need to check this sig any longer
3988     sigaddset(&check_signal_done, sig);
3989     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
3990     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
3991       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
3992                     exception_name(sig, buf, O_BUFLEN));
3993     }
3994   } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
3995     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3996     tty->print("expected:");
3997     os::Posix::print_sa_flags(tty, os::Solaris::get_our_sigflags(sig));
3998     tty->cr();
3999     tty->print("  found:");
4000     os::Posix::print_sa_flags(tty, act.sa_flags);
4001     tty->cr();
4002     // No need to check this sig any longer
4003     sigaddset(&check_signal_done, sig);
4004   }
4005 
4006   // Print all the signal handler state
4007   if (sigismember(&check_signal_done, sig)) {
4008     print_signal_handlers(tty, buf, O_BUFLEN);
4009   }
4010 
4011 }
4012 
4013 void os::Solaris::install_signal_handlers() {
4014   signal_handlers_are_installed = true;
4015 
4016   // signal-chaining
4017   typedef void (*signal_setting_t)();
4018   signal_setting_t begin_signal_setting = NULL;
4019   signal_setting_t end_signal_setting = NULL;
4020   begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4021                                         dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4022   if (begin_signal_setting != NULL) {
4023     end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4024                                         dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4025     get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4026                                        dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4027     get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
4028                                          dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
4029     libjsig_is_loaded = true;
4030     if (os::Solaris::get_libjsig_version != NULL) {
4031       int libjsigversion =  (*os::Solaris::get_libjsig_version)();
4032       assert(libjsigversion == JSIG_VERSION_1_4_1, "libjsig version mismatch");
4033     }
4034     assert(UseSignalChaining, "should enable signal-chaining");
4035   }
4036   if (libjsig_is_loaded) {
4037     // Tell libjsig jvm is setting signal handlers
4038     (*begin_signal_setting)();
4039   }
4040 
4041   set_signal_handler(SIGSEGV, true, true);
4042   set_signal_handler(SIGPIPE, true, true);
4043   set_signal_handler(SIGXFSZ, true, true);
4044   set_signal_handler(SIGBUS, true, true);
4045   set_signal_handler(SIGILL, true, true);
4046   set_signal_handler(SIGFPE, true, true);
4047   set_signal_handler(ASYNC_SIGNAL, true, true);
4048 
4049   if (libjsig_is_loaded) {
4050     // Tell libjsig jvm finishes setting signal handlers
4051     (*end_signal_setting)();
4052   }
4053 
4054   // We don't activate signal checker if libjsig is in place, we trust ourselves
4055   // and if UserSignalHandler is installed all bets are off.
4056   // Log that signal checking is off only if -verbose:jni is specified.
4057   if (CheckJNICalls) {
4058     if (libjsig_is_loaded) {
4059       if (PrintJNIResolving) {
4060         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4061       }
4062       check_signals = false;
4063     }
4064     if (AllowUserSignalHandlers) {
4065       if (PrintJNIResolving) {
4066         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4067       }
4068       check_signals = false;
4069     }
4070   }
4071 }
4072 
4073 
4074 void report_error(const char* file_name, int line_no, const char* title,
4075                   const char* format, ...);
4076 
4077 // (Static) wrappers for the liblgrp API
4078 os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4079 os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4080 os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4081 os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4082 os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4083 os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4084 os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4085 os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4086 os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4087 
4088 static address resolve_symbol_lazy(const char* name) {
4089   address addr = (address) dlsym(RTLD_DEFAULT, name);
4090   if (addr == NULL) {
4091     // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4092     addr = (address) dlsym(RTLD_NEXT, name);
4093   }
4094   return addr;
4095 }
4096 
4097 static address resolve_symbol(const char* name) {
4098   address addr = resolve_symbol_lazy(name);
4099   if (addr == NULL) {
4100     fatal(dlerror());
4101   }
4102   return addr;
4103 }
4104 
4105 void os::Solaris::libthread_init() {
4106   address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4107 
4108   lwp_priocntl_init();
4109 
4110   // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4111   if (func == NULL) {
4112     func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4113     // Guarantee that this VM is running on an new enough OS (5.6 or
4114     // later) that it will have a new enough libthread.so.
4115     guarantee(func != NULL, "libthread.so is too old.");
4116   }
4117 
4118   int size;
4119   void (*handler_info_func)(address *, int *);
4120   handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4121   handler_info_func(&handler_start, &size);
4122   handler_end = handler_start + size;
4123 }
4124 
4125 
4126 int_fnP_mutex_tP os::Solaris::_mutex_lock;
4127 int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4128 int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4129 int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4130 int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4131 int os::Solaris::_mutex_scope = USYNC_THREAD;
4132 
4133 int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4134 int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4135 int_fnP_cond_tP os::Solaris::_cond_signal;
4136 int_fnP_cond_tP os::Solaris::_cond_broadcast;
4137 int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4138 int_fnP_cond_tP os::Solaris::_cond_destroy;
4139 int os::Solaris::_cond_scope = USYNC_THREAD;
4140 
4141 void os::Solaris::synchronization_init() {
4142   if (UseLWPSynchronization) {
4143     os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4144     os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4145     os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4146     os::Solaris::set_mutex_init(lwp_mutex_init);
4147     os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4148     os::Solaris::set_mutex_scope(USYNC_THREAD);
4149 
4150     os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4151     os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4152     os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4153     os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4154     os::Solaris::set_cond_init(lwp_cond_init);
4155     os::Solaris::set_cond_destroy(lwp_cond_destroy);
4156     os::Solaris::set_cond_scope(USYNC_THREAD);
4157   } else {
4158     os::Solaris::set_mutex_scope(USYNC_THREAD);
4159     os::Solaris::set_cond_scope(USYNC_THREAD);
4160 
4161     if (UsePthreads) {
4162       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4163       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4164       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4165       os::Solaris::set_mutex_init(pthread_mutex_default_init);
4166       os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4167 
4168       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4169       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4170       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4171       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4172       os::Solaris::set_cond_init(pthread_cond_default_init);
4173       os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4174     } else {
4175       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4176       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4177       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4178       os::Solaris::set_mutex_init(::mutex_init);
4179       os::Solaris::set_mutex_destroy(::mutex_destroy);
4180 
4181       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4182       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4183       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4184       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4185       os::Solaris::set_cond_init(::cond_init);
4186       os::Solaris::set_cond_destroy(::cond_destroy);
4187     }
4188   }
4189 }
4190 
4191 bool os::Solaris::liblgrp_init() {
4192   void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4193   if (handle != NULL) {
4194     os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4195     os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4196     os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4197     os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4198     os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4199     os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4200     os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4201     os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4202                                                       dlsym(handle, "lgrp_cookie_stale")));
4203 
4204     lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4205     set_lgrp_cookie(c);
4206     return true;
4207   }
4208   return false;
4209 }
4210 
4211 // int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4212 typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4213 static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4214 
4215 void init_pset_getloadavg_ptr(void) {
4216   pset_getloadavg_ptr =
4217     (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
4218   if (pset_getloadavg_ptr == NULL) {
4219     log_warning(os)("pset_getloadavg function not found");
4220   }
4221 }
4222 
4223 int os::Solaris::_dev_zero_fd = -1;
4224 
4225 // this is called _before_ the global arguments have been parsed
4226 void os::init(void) {
4227   _initial_pid = getpid();
4228 
4229   max_hrtime = first_hrtime = gethrtime();
4230 
4231   init_random(1234567);
4232 
4233   page_size = sysconf(_SC_PAGESIZE);
4234   if (page_size == -1) {
4235     fatal("os_solaris.cpp: os::init: sysconf failed (%s)", os::strerror(errno));
4236   }
4237   init_page_sizes((size_t) page_size);
4238 
4239   Solaris::initialize_system_info();
4240 
4241   int fd = ::open("/dev/zero", O_RDWR);
4242   if (fd < 0) {
4243     fatal("os::init: cannot open /dev/zero (%s)", os::strerror(errno));
4244   } else {
4245     Solaris::set_dev_zero_fd(fd);
4246 
4247     // Close on exec, child won't inherit.
4248     fcntl(fd, F_SETFD, FD_CLOEXEC);
4249   }
4250 
4251   clock_tics_per_sec = CLK_TCK;
4252 
4253   // check if dladdr1() exists; dladdr1 can provide more information than
4254   // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
4255   // and is available on linker patches for 5.7 and 5.8.
4256   // libdl.so must have been loaded, this call is just an entry lookup
4257   void * hdl = dlopen("libdl.so", RTLD_NOW);
4258   if (hdl) {
4259     dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
4260   }
4261 
4262   // (Solaris only) this switches to calls that actually do locking.
4263   ThreadCritical::initialize();
4264 
4265   main_thread = thr_self();
4266 
4267   // dynamic lookup of functions that may not be available in our lowest
4268   // supported Solaris release
4269   void * handle = dlopen("libc.so.1", RTLD_LAZY);
4270   if (handle != NULL) {
4271     Solaris::_pthread_setname_np =  // from 11.3
4272         (Solaris::pthread_setname_np_func_t)dlsym(handle, "pthread_setname_np");
4273   }
4274 }
4275 
4276 // To install functions for atexit system call
4277 extern "C" {
4278   static void perfMemory_exit_helper() {
4279     perfMemory_exit();
4280   }
4281 }
4282 
4283 // this is called _after_ the global arguments have been parsed
4284 jint os::init_2(void) {
4285   // try to enable extended file IO ASAP, see 6431278
4286   os::Solaris::try_enable_extended_io();
4287 
4288   // Allocate a single page and mark it as readable for safepoint polling.  Also
4289   // use this first mmap call to check support for MAP_ALIGN.
4290   address polling_page = (address)Solaris::mmap_chunk((char*)page_size,
4291                                                       page_size,
4292                                                       MAP_PRIVATE | MAP_ALIGN,
4293                                                       PROT_READ);
4294   if (polling_page == NULL) {
4295     has_map_align = false;
4296     polling_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE,
4297                                                 PROT_READ);
4298   }
4299 
4300   os::set_polling_page(polling_page);
4301   log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4302 
4303   if (!UseMembar) {
4304     address mem_serialize_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE, PROT_READ | PROT_WRITE);
4305     guarantee(mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4306     os::set_memory_serialize_page(mem_serialize_page);
4307     log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4308   }
4309 
4310   // Check and sets minimum stack sizes against command line options
4311   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4312     return JNI_ERR;
4313   }
4314 
4315   Solaris::libthread_init();
4316 
4317   if (UseNUMA) {
4318     if (!Solaris::liblgrp_init()) {
4319       UseNUMA = false;
4320     } else {
4321       size_t lgrp_limit = os::numa_get_groups_num();
4322       int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit, mtInternal);
4323       size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
4324       FREE_C_HEAP_ARRAY(int, lgrp_ids);
4325       if (lgrp_num < 2) {
4326         // There's only one locality group, disable NUMA.
4327         UseNUMA = false;
4328       }
4329     }
4330     if (!UseNUMA && ForceNUMA) {
4331       UseNUMA = true;
4332     }
4333   }
4334 
4335   Solaris::signal_sets_init();
4336   Solaris::init_signal_mem();
4337   Solaris::install_signal_handlers();
4338 
4339   // initialize synchronization primitives to use either thread or
4340   // lwp synchronization (controlled by UseLWPSynchronization)
4341   Solaris::synchronization_init();
4342 
4343   if (MaxFDLimit) {
4344     // set the number of file descriptors to max. print out error
4345     // if getrlimit/setrlimit fails but continue regardless.
4346     struct rlimit nbr_files;
4347     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4348     if (status != 0) {
4349       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4350     } else {
4351       nbr_files.rlim_cur = nbr_files.rlim_max;
4352       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4353       if (status != 0) {
4354         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4355       }
4356     }
4357   }
4358 
4359   // Calculate theoretical max. size of Threads to guard gainst
4360   // artifical out-of-memory situations, where all available address-
4361   // space has been reserved by thread stacks. Default stack size is 1Mb.
4362   size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
4363     JavaThread::stack_size_at_create() : (1*K*K);
4364   assert(pre_thread_stack_size != 0, "Must have a stack");
4365   // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
4366   // we should start doing Virtual Memory banging. Currently when the threads will
4367   // have used all but 200Mb of space.
4368   size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
4369   Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
4370 
4371   // at-exit methods are called in the reverse order of their registration.
4372   // In Solaris 7 and earlier, atexit functions are called on return from
4373   // main or as a result of a call to exit(3C). There can be only 32 of
4374   // these functions registered and atexit() does not set errno. In Solaris
4375   // 8 and later, there is no limit to the number of functions registered
4376   // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
4377   // functions are called upon dlclose(3DL) in addition to return from main
4378   // and exit(3C).
4379 
4380   if (PerfAllowAtExitRegistration) {
4381     // only register atexit functions if PerfAllowAtExitRegistration is set.
4382     // atexit functions can be delayed until process exit time, which
4383     // can be problematic for embedded VM situations. Embedded VMs should
4384     // call DestroyJavaVM() to assure that VM resources are released.
4385 
4386     // note: perfMemory_exit_helper atexit function may be removed in
4387     // the future if the appropriate cleanup code can be added to the
4388     // VM_Exit VMOperation's doit method.
4389     if (atexit(perfMemory_exit_helper) != 0) {
4390       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4391     }
4392   }
4393 
4394   // Init pset_loadavg function pointer
4395   init_pset_getloadavg_ptr();
4396 
4397   return JNI_OK;
4398 }
4399 
4400 // Mark the polling page as unreadable
4401 void os::make_polling_page_unreadable(void) {
4402   if (mprotect((char *)_polling_page, page_size, PROT_NONE) != 0) {
4403     fatal("Could not disable polling page");
4404   }
4405 }
4406 
4407 // Mark the polling page as readable
4408 void os::make_polling_page_readable(void) {
4409   if (mprotect((char *)_polling_page, page_size, PROT_READ) != 0) {
4410     fatal("Could not enable polling page");
4411   }
4412 }
4413 
4414 // Is a (classpath) directory empty?
4415 bool os::dir_is_empty(const char* path) {
4416   DIR *dir = NULL;
4417   struct dirent *ptr;
4418 
4419   dir = opendir(path);
4420   if (dir == NULL) return true;
4421 
4422   // Scan the directory
4423   bool result = true;
4424   char buf[sizeof(struct dirent) + MAX_PATH];
4425   struct dirent *dbuf = (struct dirent *) buf;
4426   while (result && (ptr = readdir(dir, dbuf)) != NULL) {
4427     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4428       result = false;
4429     }
4430   }
4431   closedir(dir);
4432   return result;
4433 }
4434 
4435 // This code originates from JDK's sysOpen and open64_w
4436 // from src/solaris/hpi/src/system_md.c
4437 
4438 int os::open(const char *path, int oflag, int mode) {
4439   if (strlen(path) > MAX_PATH - 1) {
4440     errno = ENAMETOOLONG;
4441     return -1;
4442   }
4443   int fd;
4444 
4445   fd = ::open64(path, oflag, mode);
4446   if (fd == -1) return -1;
4447 
4448   // If the open succeeded, the file might still be a directory
4449   {
4450     struct stat64 buf64;
4451     int ret = ::fstat64(fd, &buf64);
4452     int st_mode = buf64.st_mode;
4453 
4454     if (ret != -1) {
4455       if ((st_mode & S_IFMT) == S_IFDIR) {
4456         errno = EISDIR;
4457         ::close(fd);
4458         return -1;
4459       }
4460     } else {
4461       ::close(fd);
4462       return -1;
4463     }
4464   }
4465 
4466   // 32-bit Solaris systems suffer from:
4467   //
4468   // - an historical default soft limit of 256 per-process file
4469   //   descriptors that is too low for many Java programs.
4470   //
4471   // - a design flaw where file descriptors created using stdio
4472   //   fopen must be less than 256, _even_ when the first limit above
4473   //   has been raised.  This can cause calls to fopen (but not calls to
4474   //   open, for example) to fail mysteriously, perhaps in 3rd party
4475   //   native code (although the JDK itself uses fopen).  One can hardly
4476   //   criticize them for using this most standard of all functions.
4477   //
4478   // We attempt to make everything work anyways by:
4479   //
4480   // - raising the soft limit on per-process file descriptors beyond
4481   //   256
4482   //
4483   // - As of Solaris 10u4, we can request that Solaris raise the 256
4484   //   stdio fopen limit by calling function enable_extended_FILE_stdio.
4485   //   This is done in init_2 and recorded in enabled_extended_FILE_stdio
4486   //
4487   // - If we are stuck on an old (pre 10u4) Solaris system, we can
4488   //   workaround the bug by remapping non-stdio file descriptors below
4489   //   256 to ones beyond 256, which is done below.
4490   //
4491   // See:
4492   // 1085341: 32-bit stdio routines should support file descriptors >255
4493   // 6533291: Work around 32-bit Solaris stdio limit of 256 open files
4494   // 6431278: Netbeans crash on 32 bit Solaris: need to call
4495   //          enable_extended_FILE_stdio() in VM initialisation
4496   // Giri Mandalika's blog
4497   // http://technopark02.blogspot.com/2005_05_01_archive.html
4498   //
4499 #ifndef  _LP64
4500   if ((!enabled_extended_FILE_stdio) && fd < 256) {
4501     int newfd = ::fcntl(fd, F_DUPFD, 256);
4502     if (newfd != -1) {
4503       ::close(fd);
4504       fd = newfd;
4505     }
4506   }
4507 #endif // 32-bit Solaris
4508 
4509   // All file descriptors that are opened in the JVM and not
4510   // specifically destined for a subprocess should have the
4511   // close-on-exec flag set.  If we don't set it, then careless 3rd
4512   // party native code might fork and exec without closing all
4513   // appropriate file descriptors (e.g. as we do in closeDescriptors in
4514   // UNIXProcess.c), and this in turn might:
4515   //
4516   // - cause end-of-file to fail to be detected on some file
4517   //   descriptors, resulting in mysterious hangs, or
4518   //
4519   // - might cause an fopen in the subprocess to fail on a system
4520   //   suffering from bug 1085341.
4521   //
4522   // (Yes, the default setting of the close-on-exec flag is a Unix
4523   // design flaw)
4524   //
4525   // See:
4526   // 1085341: 32-bit stdio routines should support file descriptors >255
4527   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4528   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4529   //
4530 #ifdef FD_CLOEXEC
4531   {
4532     int flags = ::fcntl(fd, F_GETFD);
4533     if (flags != -1) {
4534       ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4535     }
4536   }
4537 #endif
4538 
4539   return fd;
4540 }
4541 
4542 // create binary file, rewriting existing file if required
4543 int os::create_binary_file(const char* path, bool rewrite_existing) {
4544   int oflags = O_WRONLY | O_CREAT;
4545   if (!rewrite_existing) {
4546     oflags |= O_EXCL;
4547   }
4548   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4549 }
4550 
4551 // return current position of file pointer
4552 jlong os::current_file_offset(int fd) {
4553   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4554 }
4555 
4556 // move file pointer to the specified offset
4557 jlong os::seek_to_file_offset(int fd, jlong offset) {
4558   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4559 }
4560 
4561 jlong os::lseek(int fd, jlong offset, int whence) {
4562   return (jlong) ::lseek64(fd, offset, whence);
4563 }
4564 
4565 char * os::native_path(char *path) {
4566   return path;
4567 }
4568 
4569 int os::ftruncate(int fd, jlong length) {
4570   return ::ftruncate64(fd, length);
4571 }
4572 
4573 int os::fsync(int fd)  {
4574   RESTARTABLE_RETURN_INT(::fsync(fd));
4575 }
4576 
4577 int os::available(int fd, jlong *bytes) {
4578   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
4579          "Assumed _thread_in_native");
4580   jlong cur, end;
4581   int mode;
4582   struct stat64 buf64;
4583 
4584   if (::fstat64(fd, &buf64) >= 0) {
4585     mode = buf64.st_mode;
4586     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4587       int n,ioctl_return;
4588 
4589       RESTARTABLE(::ioctl(fd, FIONREAD, &n), ioctl_return);
4590       if (ioctl_return>= 0) {
4591         *bytes = n;
4592         return 1;
4593       }
4594     }
4595   }
4596   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4597     return 0;
4598   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4599     return 0;
4600   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4601     return 0;
4602   }
4603   *bytes = end - cur;
4604   return 1;
4605 }
4606 
4607 // Map a block of memory.
4608 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4609                         char *addr, size_t bytes, bool read_only,
4610                         bool allow_exec) {
4611   int prot;
4612   int flags;
4613 
4614   if (read_only) {
4615     prot = PROT_READ;
4616     flags = MAP_SHARED;
4617   } else {
4618     prot = PROT_READ | PROT_WRITE;
4619     flags = MAP_PRIVATE;
4620   }
4621 
4622   if (allow_exec) {
4623     prot |= PROT_EXEC;
4624   }
4625 
4626   if (addr != NULL) {
4627     flags |= MAP_FIXED;
4628   }
4629 
4630   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4631                                      fd, file_offset);
4632   if (mapped_address == MAP_FAILED) {
4633     return NULL;
4634   }
4635   return mapped_address;
4636 }
4637 
4638 
4639 // Remap a block of memory.
4640 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4641                           char *addr, size_t bytes, bool read_only,
4642                           bool allow_exec) {
4643   // same as map_memory() on this OS
4644   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4645                         allow_exec);
4646 }
4647 
4648 
4649 // Unmap a block of memory.
4650 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4651   return munmap(addr, bytes) == 0;
4652 }
4653 
4654 void os::pause() {
4655   char filename[MAX_PATH];
4656   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4657     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4658   } else {
4659     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4660   }
4661 
4662   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4663   if (fd != -1) {
4664     struct stat buf;
4665     ::close(fd);
4666     while (::stat(filename, &buf) == 0) {
4667       (void)::poll(NULL, 0, 100);
4668     }
4669   } else {
4670     jio_fprintf(stderr,
4671                 "Could not open pause file '%s', continuing immediately.\n", filename);
4672   }
4673 }
4674 
4675 #ifndef PRODUCT
4676 #ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4677 // Turn this on if you need to trace synch operations.
4678 // Set RECORD_SYNCH_LIMIT to a large-enough value,
4679 // and call record_synch_enable and record_synch_disable
4680 // around the computation of interest.
4681 
4682 void record_synch(char* name, bool returning);  // defined below
4683 
4684 class RecordSynch {
4685   char* _name;
4686  public:
4687   RecordSynch(char* name) :_name(name) { record_synch(_name, false); }
4688   ~RecordSynch()                       { record_synch(_name, true); }
4689 };
4690 
4691 #define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
4692 extern "C" ret name params {                                    \
4693   typedef ret name##_t params;                                  \
4694   static name##_t* implem = NULL;                               \
4695   static int callcount = 0;                                     \
4696   if (implem == NULL) {                                         \
4697     implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
4698     if (implem == NULL)  fatal(dlerror());                      \
4699   }                                                             \
4700   ++callcount;                                                  \
4701   RecordSynch _rs(#name);                                       \
4702   inner;                                                        \
4703   return implem args;                                           \
4704 }
4705 // in dbx, examine callcounts this way:
4706 // for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
4707 
4708 #define CHECK_POINTER_OK(p) \
4709   (!Universe::is_fully_initialized() || !Universe::is_reserved_heap((oop)(p)))
4710 #define CHECK_MU \
4711   if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
4712 #define CHECK_CV \
4713   if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
4714 #define CHECK_P(p) \
4715   if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
4716 
4717 #define CHECK_MUTEX(mutex_op) \
4718   CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
4719 
4720 CHECK_MUTEX(   mutex_lock)
4721 CHECK_MUTEX(  _mutex_lock)
4722 CHECK_MUTEX( mutex_unlock)
4723 CHECK_MUTEX(_mutex_unlock)
4724 CHECK_MUTEX( mutex_trylock)
4725 CHECK_MUTEX(_mutex_trylock)
4726 
4727 #define CHECK_COND(cond_op) \
4728   CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU; CHECK_CV);
4729 
4730 CHECK_COND( cond_wait);
4731 CHECK_COND(_cond_wait);
4732 CHECK_COND(_cond_wait_cancel);
4733 
4734 #define CHECK_COND2(cond_op) \
4735   CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU; CHECK_CV);
4736 
4737 CHECK_COND2( cond_timedwait);
4738 CHECK_COND2(_cond_timedwait);
4739 CHECK_COND2(_cond_timedwait_cancel);
4740 
4741 // do the _lwp_* versions too
4742 #define mutex_t lwp_mutex_t
4743 #define cond_t  lwp_cond_t
4744 CHECK_MUTEX(  _lwp_mutex_lock)
4745 CHECK_MUTEX(  _lwp_mutex_unlock)
4746 CHECK_MUTEX(  _lwp_mutex_trylock)
4747 CHECK_MUTEX( __lwp_mutex_lock)
4748 CHECK_MUTEX( __lwp_mutex_unlock)
4749 CHECK_MUTEX( __lwp_mutex_trylock)
4750 CHECK_MUTEX(___lwp_mutex_lock)
4751 CHECK_MUTEX(___lwp_mutex_unlock)
4752 
4753 CHECK_COND(  _lwp_cond_wait);
4754 CHECK_COND( __lwp_cond_wait);
4755 CHECK_COND(___lwp_cond_wait);
4756 
4757 CHECK_COND2(  _lwp_cond_timedwait);
4758 CHECK_COND2( __lwp_cond_timedwait);
4759 #undef mutex_t
4760 #undef cond_t
4761 
4762 CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4763 CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4764 CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4765 CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4766 CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4767 CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4768 CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4769 CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4770 
4771 
4772 // recording machinery:
4773 
4774 enum { RECORD_SYNCH_LIMIT = 200 };
4775 char* record_synch_name[RECORD_SYNCH_LIMIT];
4776 void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
4777 bool record_synch_returning[RECORD_SYNCH_LIMIT];
4778 thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
4779 int record_synch_count = 0;
4780 bool record_synch_enabled = false;
4781 
4782 // in dbx, examine recorded data this way:
4783 // for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
4784 
4785 void record_synch(char* name, bool returning) {
4786   if (record_synch_enabled) {
4787     if (record_synch_count < RECORD_SYNCH_LIMIT) {
4788       record_synch_name[record_synch_count] = name;
4789       record_synch_returning[record_synch_count] = returning;
4790       record_synch_thread[record_synch_count] = thr_self();
4791       record_synch_arg0ptr[record_synch_count] = &name;
4792       record_synch_count++;
4793     }
4794     // put more checking code here:
4795     // ...
4796   }
4797 }
4798 
4799 void record_synch_enable() {
4800   // start collecting trace data, if not already doing so
4801   if (!record_synch_enabled)  record_synch_count = 0;
4802   record_synch_enabled = true;
4803 }
4804 
4805 void record_synch_disable() {
4806   // stop collecting trace data
4807   record_synch_enabled = false;
4808 }
4809 
4810 #endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4811 #endif // PRODUCT
4812 
4813 const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
4814 const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
4815                                (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
4816 
4817 
4818 // JVMTI & JVM monitoring and management support
4819 // The thread_cpu_time() and current_thread_cpu_time() are only
4820 // supported if is_thread_cpu_time_supported() returns true.
4821 // They are not supported on Solaris T1.
4822 
4823 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4824 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4825 // of a thread.
4826 //
4827 // current_thread_cpu_time() and thread_cpu_time(Thread *)
4828 // returns the fast estimate available on the platform.
4829 
4830 // hrtime_t gethrvtime() return value includes
4831 // user time but does not include system time
4832 jlong os::current_thread_cpu_time() {
4833   return (jlong) gethrvtime();
4834 }
4835 
4836 jlong os::thread_cpu_time(Thread *thread) {
4837   // return user level CPU time only to be consistent with
4838   // what current_thread_cpu_time returns.
4839   // thread_cpu_time_info() must be changed if this changes
4840   return os::thread_cpu_time(thread, false /* user time only */);
4841 }
4842 
4843 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4844   if (user_sys_cpu_time) {
4845     return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
4846   } else {
4847     return os::current_thread_cpu_time();
4848   }
4849 }
4850 
4851 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4852   char proc_name[64];
4853   int count;
4854   prusage_t prusage;
4855   jlong lwp_time;
4856   int fd;
4857 
4858   sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
4859           getpid(),
4860           thread->osthread()->lwp_id());
4861   fd = ::open(proc_name, O_RDONLY);
4862   if (fd == -1) return -1;
4863 
4864   do {
4865     count = ::pread(fd,
4866                     (void *)&prusage.pr_utime,
4867                     thr_time_size,
4868                     thr_time_off);
4869   } while (count < 0 && errno == EINTR);
4870   ::close(fd);
4871   if (count < 0) return -1;
4872 
4873   if (user_sys_cpu_time) {
4874     // user + system CPU time
4875     lwp_time = (((jlong)prusage.pr_stime.tv_sec +
4876                  (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
4877                  (jlong)prusage.pr_stime.tv_nsec +
4878                  (jlong)prusage.pr_utime.tv_nsec;
4879   } else {
4880     // user level CPU time only
4881     lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
4882                 (jlong)prusage.pr_utime.tv_nsec;
4883   }
4884 
4885   return (lwp_time);
4886 }
4887 
4888 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4889   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
4890   info_ptr->may_skip_backward = false;    // elapsed time not wall time
4891   info_ptr->may_skip_forward = false;     // elapsed time not wall time
4892   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
4893 }
4894 
4895 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4896   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
4897   info_ptr->may_skip_backward = false;    // elapsed time not wall time
4898   info_ptr->may_skip_forward = false;     // elapsed time not wall time
4899   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
4900 }
4901 
4902 bool os::is_thread_cpu_time_supported() {
4903   return true;
4904 }
4905 
4906 // System loadavg support.  Returns -1 if load average cannot be obtained.
4907 // Return the load average for our processor set if the primitive exists
4908 // (Solaris 9 and later).  Otherwise just return system wide loadavg.
4909 int os::loadavg(double loadavg[], int nelem) {
4910   if (pset_getloadavg_ptr != NULL) {
4911     return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
4912   } else {
4913     return ::getloadavg(loadavg, nelem);
4914   }
4915 }
4916 
4917 //---------------------------------------------------------------------------------
4918 
4919 bool os::find(address addr, outputStream* st) {
4920   Dl_info dlinfo;
4921   memset(&dlinfo, 0, sizeof(dlinfo));
4922   if (dladdr(addr, &dlinfo) != 0) {
4923     st->print(PTR_FORMAT ": ", addr);
4924     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4925       st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
4926     } else if (dlinfo.dli_fbase != NULL) {
4927       st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
4928     } else {
4929       st->print("<absolute address>");
4930     }
4931     if (dlinfo.dli_fname != NULL) {
4932       st->print(" in %s", dlinfo.dli_fname);
4933     }
4934     if (dlinfo.dli_fbase != NULL) {
4935       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4936     }
4937     st->cr();
4938 
4939     if (Verbose) {
4940       // decode some bytes around the PC
4941       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4942       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4943       address       lowest = (address) dlinfo.dli_sname;
4944       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4945       if (begin < lowest)  begin = lowest;
4946       Dl_info dlinfo2;
4947       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4948           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4949         end = (address) dlinfo2.dli_saddr;
4950       }
4951       Disassembler::decode(begin, end, st);
4952     }
4953     return true;
4954   }
4955   return false;
4956 }
4957 
4958 // Following function has been added to support HotSparc's libjvm.so running
4959 // under Solaris production JDK 1.2.2 / 1.3.0.  These came from
4960 // src/solaris/hpi/native_threads in the EVM codebase.
4961 //
4962 // NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
4963 // libraries and should thus be removed. We will leave it behind for a while
4964 // until we no longer want to able to run on top of 1.3.0 Solaris production
4965 // JDK. See 4341971.
4966 
4967 #define STACK_SLACK 0x800
4968 
4969 extern "C" {
4970   intptr_t sysThreadAvailableStackWithSlack() {
4971     stack_t st;
4972     intptr_t retval, stack_top;
4973     retval = thr_stksegment(&st);
4974     assert(retval == 0, "incorrect return value from thr_stksegment");
4975     assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
4976     assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
4977     stack_top=(intptr_t)st.ss_sp-st.ss_size;
4978     return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
4979   }
4980 }
4981 
4982 // ObjectMonitor park-unpark infrastructure ...
4983 //
4984 // We implement Solaris and Linux PlatformEvents with the
4985 // obvious condvar-mutex-flag triple.
4986 // Another alternative that works quite well is pipes:
4987 // Each PlatformEvent consists of a pipe-pair.
4988 // The thread associated with the PlatformEvent
4989 // calls park(), which reads from the input end of the pipe.
4990 // Unpark() writes into the other end of the pipe.
4991 // The write-side of the pipe must be set NDELAY.
4992 // Unfortunately pipes consume a large # of handles.
4993 // Native solaris lwp_park() and lwp_unpark() work nicely, too.
4994 // Using pipes for the 1st few threads might be workable, however.
4995 //
4996 // park() is permitted to return spuriously.
4997 // Callers of park() should wrap the call to park() in
4998 // an appropriate loop.  A litmus test for the correct
4999 // usage of park is the following: if park() were modified
5000 // to immediately return 0 your code should still work,
5001 // albeit degenerating to a spin loop.
5002 //
5003 // In a sense, park()-unpark() just provides more polite spinning
5004 // and polling with the key difference over naive spinning being
5005 // that a parked thread needs to be explicitly unparked() in order
5006 // to wake up and to poll the underlying condition.
5007 //
5008 // Assumption:
5009 //    Only one parker can exist on an event, which is why we allocate
5010 //    them per-thread. Multiple unparkers can coexist.
5011 //
5012 // _Event transitions in park()
5013 //   -1 => -1 : illegal
5014 //    1 =>  0 : pass - return immediately
5015 //    0 => -1 : block; then set _Event to 0 before returning
5016 //
5017 // _Event transitions in unpark()
5018 //    0 => 1 : just return
5019 //    1 => 1 : just return
5020 //   -1 => either 0 or 1; must signal target thread
5021 //         That is, we can safely transition _Event from -1 to either
5022 //         0 or 1.
5023 //
5024 // _Event serves as a restricted-range semaphore.
5025 //   -1 : thread is blocked, i.e. there is a waiter
5026 //    0 : neutral: thread is running or ready,
5027 //        could have been signaled after a wait started
5028 //    1 : signaled - thread is running or ready
5029 //
5030 // Another possible encoding of _Event would be with
5031 // explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
5032 //
5033 // TODO-FIXME: add DTRACE probes for:
5034 // 1.   Tx parks
5035 // 2.   Ty unparks Tx
5036 // 3.   Tx resumes from park
5037 
5038 
5039 // value determined through experimentation
5040 #define ROUNDINGFIX 11
5041 
5042 // utility to compute the abstime argument to timedwait.
5043 // TODO-FIXME: switch from compute_abstime() to unpackTime().
5044 
5045 static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
5046   // millis is the relative timeout time
5047   // abstime will be the absolute timeout time
5048   if (millis < 0)  millis = 0;
5049   struct timeval now;
5050   int status = gettimeofday(&now, NULL);
5051   assert(status == 0, "gettimeofday");
5052   jlong seconds = millis / 1000;
5053   jlong max_wait_period;
5054 
5055   if (UseLWPSynchronization) {
5056     // forward port of fix for 4275818 (not sleeping long enough)
5057     // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
5058     // _lwp_cond_timedwait() used a round_down algorithm rather
5059     // than a round_up. For millis less than our roundfactor
5060     // it rounded down to 0 which doesn't meet the spec.
5061     // For millis > roundfactor we may return a bit sooner, but
5062     // since we can not accurately identify the patch level and
5063     // this has already been fixed in Solaris 9 and 8 we will
5064     // leave it alone rather than always rounding down.
5065 
5066     if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
5067     // It appears that when we go directly through Solaris _lwp_cond_timedwait()
5068     // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
5069     max_wait_period = 21000000;
5070   } else {
5071     max_wait_period = 50000000;
5072   }
5073   millis %= 1000;
5074   if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
5075     seconds = max_wait_period;
5076   }
5077   abstime->tv_sec = now.tv_sec  + seconds;
5078   long       usec = now.tv_usec + millis * 1000;
5079   if (usec >= 1000000) {
5080     abstime->tv_sec += 1;
5081     usec -= 1000000;
5082   }
5083   abstime->tv_nsec = usec * 1000;
5084   return abstime;
5085 }
5086 
5087 void os::PlatformEvent::park() {           // AKA: down()
5088   // Transitions for _Event:
5089   //   -1 => -1 : illegal
5090   //    1 =>  0 : pass - return immediately
5091   //    0 => -1 : block; then set _Event to 0 before returning
5092 
5093   // Invariant: Only the thread associated with the Event/PlatformEvent
5094   // may call park().
5095   assert(_nParked == 0, "invariant");
5096 
5097   int v;
5098   for (;;) {
5099     v = _Event;
5100     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5101   }
5102   guarantee(v >= 0, "invariant");
5103   if (v == 0) {
5104     // Do this the hard way by blocking ...
5105     // See http://monaco.sfbay/detail.jsf?cr=5094058.
5106     int status = os::Solaris::mutex_lock(_mutex);
5107     assert_status(status == 0, status, "mutex_lock");
5108     guarantee(_nParked == 0, "invariant");
5109     ++_nParked;
5110     while (_Event < 0) {
5111       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5112       // Treat this the same as if the wait was interrupted
5113       // With usr/lib/lwp going to kernel, always handle ETIME
5114       status = os::Solaris::cond_wait(_cond, _mutex);
5115       if (status == ETIME) status = EINTR;
5116       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5117     }
5118     --_nParked;
5119     _Event = 0;
5120     status = os::Solaris::mutex_unlock(_mutex);
5121     assert_status(status == 0, status, "mutex_unlock");
5122     // Paranoia to ensure our locked and lock-free paths interact
5123     // correctly with each other.
5124     OrderAccess::fence();
5125   }
5126 }
5127 
5128 int os::PlatformEvent::park(jlong millis) {
5129   // Transitions for _Event:
5130   //   -1 => -1 : illegal
5131   //    1 =>  0 : pass - return immediately
5132   //    0 => -1 : block; then set _Event to 0 before returning
5133 
5134   guarantee(_nParked == 0, "invariant");
5135   int v;
5136   for (;;) {
5137     v = _Event;
5138     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5139   }
5140   guarantee(v >= 0, "invariant");
5141   if (v != 0) return OS_OK;
5142 
5143   int ret = OS_TIMEOUT;
5144   timestruc_t abst;
5145   compute_abstime(&abst, millis);
5146 
5147   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5148   int status = os::Solaris::mutex_lock(_mutex);
5149   assert_status(status == 0, status, "mutex_lock");
5150   guarantee(_nParked == 0, "invariant");
5151   ++_nParked;
5152   while (_Event < 0) {
5153     int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
5154     assert_status(status == 0 || status == EINTR ||
5155                   status == ETIME || status == ETIMEDOUT,
5156                   status, "cond_timedwait");
5157     if (!FilterSpuriousWakeups) break;                // previous semantics
5158     if (status == ETIME || status == ETIMEDOUT) break;
5159     // We consume and ignore EINTR and spurious wakeups.
5160   }
5161   --_nParked;
5162   if (_Event >= 0) ret = OS_OK;
5163   _Event = 0;
5164   status = os::Solaris::mutex_unlock(_mutex);
5165   assert_status(status == 0, status, "mutex_unlock");
5166   // Paranoia to ensure our locked and lock-free paths interact
5167   // correctly with each other.
5168   OrderAccess::fence();
5169   return ret;
5170 }
5171 
5172 void os::PlatformEvent::unpark() {
5173   // Transitions for _Event:
5174   //    0 => 1 : just return
5175   //    1 => 1 : just return
5176   //   -1 => either 0 or 1; must signal target thread
5177   //         That is, we can safely transition _Event from -1 to either
5178   //         0 or 1.
5179   // See also: "Semaphores in Plan 9" by Mullender & Cox
5180   //
5181   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5182   // that it will take two back-to-back park() calls for the owning
5183   // thread to block. This has the benefit of forcing a spurious return
5184   // from the first park() call after an unpark() call which will help
5185   // shake out uses of park() and unpark() without condition variables.
5186 
5187   if (Atomic::xchg(1, &_Event) >= 0) return;
5188 
5189   // If the thread associated with the event was parked, wake it.
5190   // Wait for the thread assoc with the PlatformEvent to vacate.
5191   int status = os::Solaris::mutex_lock(_mutex);
5192   assert_status(status == 0, status, "mutex_lock");
5193   int AnyWaiters = _nParked;
5194   status = os::Solaris::mutex_unlock(_mutex);
5195   assert_status(status == 0, status, "mutex_unlock");
5196   guarantee(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5197   if (AnyWaiters != 0) {
5198     // Note that we signal() *after* dropping the lock for "immortal" Events.
5199     // This is safe and avoids a common class of  futile wakeups.  In rare
5200     // circumstances this can cause a thread to return prematurely from
5201     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5202     // will simply re-test the condition and re-park itself.
5203     // This provides particular benefit if the underlying platform does not
5204     // provide wait morphing.
5205     status = os::Solaris::cond_signal(_cond);
5206     assert_status(status == 0, status, "cond_signal");
5207   }
5208 }
5209 
5210 // JSR166
5211 // -------------------------------------------------------
5212 
5213 // The solaris and linux implementations of park/unpark are fairly
5214 // conservative for now, but can be improved. They currently use a
5215 // mutex/condvar pair, plus _counter.
5216 // Park decrements _counter if > 0, else does a condvar wait.  Unpark
5217 // sets count to 1 and signals condvar.  Only one thread ever waits
5218 // on the condvar. Contention seen when trying to park implies that someone
5219 // is unparking you, so don't wait. And spurious returns are fine, so there
5220 // is no need to track notifications.
5221 
5222 #define MAX_SECS 100000000
5223 
5224 // This code is common to linux and solaris and will be moved to a
5225 // common place in dolphin.
5226 //
5227 // The passed in time value is either a relative time in nanoseconds
5228 // or an absolute time in milliseconds. Either way it has to be unpacked
5229 // into suitable seconds and nanoseconds components and stored in the
5230 // given timespec structure.
5231 // Given time is a 64-bit value and the time_t used in the timespec is only
5232 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5233 // overflow if times way in the future are given. Further on Solaris versions
5234 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5235 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5236 // As it will be 28 years before "now + 100000000" will overflow we can
5237 // ignore overflow and just impose a hard-limit on seconds using the value
5238 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5239 // years from "now".
5240 //
5241 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5242   assert(time > 0, "convertTime");
5243 
5244   struct timeval now;
5245   int status = gettimeofday(&now, NULL);
5246   assert(status == 0, "gettimeofday");
5247 
5248   time_t max_secs = now.tv_sec + MAX_SECS;
5249 
5250   if (isAbsolute) {
5251     jlong secs = time / 1000;
5252     if (secs > max_secs) {
5253       absTime->tv_sec = max_secs;
5254     } else {
5255       absTime->tv_sec = secs;
5256     }
5257     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5258   } else {
5259     jlong secs = time / NANOSECS_PER_SEC;
5260     if (secs >= MAX_SECS) {
5261       absTime->tv_sec = max_secs;
5262       absTime->tv_nsec = 0;
5263     } else {
5264       absTime->tv_sec = now.tv_sec + secs;
5265       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5266       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5267         absTime->tv_nsec -= NANOSECS_PER_SEC;
5268         ++absTime->tv_sec; // note: this must be <= max_secs
5269       }
5270     }
5271   }
5272   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5273   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5274   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5275   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5276 }
5277 
5278 void Parker::park(bool isAbsolute, jlong time) {
5279   // Ideally we'd do something useful while spinning, such
5280   // as calling unpackTime().
5281 
5282   // Optional fast-path check:
5283   // Return immediately if a permit is available.
5284   // We depend on Atomic::xchg() having full barrier semantics
5285   // since we are doing a lock-free update to _counter.
5286   if (Atomic::xchg(0, &_counter) > 0) return;
5287 
5288   // Optional fast-exit: Check interrupt before trying to wait
5289   Thread* thread = Thread::current();
5290   assert(thread->is_Java_thread(), "Must be JavaThread");
5291   JavaThread *jt = (JavaThread *)thread;
5292   if (Thread::is_interrupted(thread, false)) {
5293     return;
5294   }
5295 
5296   // First, demultiplex/decode time arguments
5297   timespec absTime;
5298   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5299     return;
5300   }
5301   if (time > 0) {
5302     // Warning: this code might be exposed to the old Solaris time
5303     // round-down bugs.  Grep "roundingFix" for details.
5304     unpackTime(&absTime, isAbsolute, time);
5305   }
5306 
5307   // Enter safepoint region
5308   // Beware of deadlocks such as 6317397.
5309   // The per-thread Parker:: _mutex is a classic leaf-lock.
5310   // In particular a thread must never block on the Threads_lock while
5311   // holding the Parker:: mutex.  If safepoints are pending both the
5312   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5313   ThreadBlockInVM tbivm(jt);
5314 
5315   // Don't wait if cannot get lock since interference arises from
5316   // unblocking.  Also. check interrupt before trying wait
5317   if (Thread::is_interrupted(thread, false) ||
5318       os::Solaris::mutex_trylock(_mutex) != 0) {
5319     return;
5320   }
5321 
5322   int status;
5323 
5324   if (_counter > 0)  { // no wait needed
5325     _counter = 0;
5326     status = os::Solaris::mutex_unlock(_mutex);
5327     assert(status == 0, "invariant");
5328     // Paranoia to ensure our locked and lock-free paths interact
5329     // correctly with each other and Java-level accesses.
5330     OrderAccess::fence();
5331     return;
5332   }
5333 
5334   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5335   jt->set_suspend_equivalent();
5336   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5337 
5338   // Do this the hard way by blocking ...
5339   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5340   if (time == 0) {
5341     status = os::Solaris::cond_wait(_cond, _mutex);
5342   } else {
5343     status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
5344   }
5345   // Note that an untimed cond_wait() can sometimes return ETIME on older
5346   // versions of the Solaris.
5347   assert_status(status == 0 || status == EINTR ||
5348                 status == ETIME || status == ETIMEDOUT,
5349                 status, "cond_timedwait");
5350 
5351   _counter = 0;
5352   status = os::Solaris::mutex_unlock(_mutex);
5353   assert_status(status == 0, status, "mutex_unlock");
5354   // Paranoia to ensure our locked and lock-free paths interact
5355   // correctly with each other and Java-level accesses.
5356   OrderAccess::fence();
5357 
5358   // If externally suspended while waiting, re-suspend
5359   if (jt->handle_special_suspend_equivalent_condition()) {
5360     jt->java_suspend_self();
5361   }
5362 }
5363 
5364 void Parker::unpark() {
5365   int status = os::Solaris::mutex_lock(_mutex);
5366   assert(status == 0, "invariant");
5367   const int s = _counter;
5368   _counter = 1;
5369   status = os::Solaris::mutex_unlock(_mutex);
5370   assert(status == 0, "invariant");
5371 
5372   if (s < 1) {
5373     status = os::Solaris::cond_signal(_cond);
5374     assert(status == 0, "invariant");
5375   }
5376 }
5377 
5378 extern char** environ;
5379 
5380 // Run the specified command in a separate process. Return its exit value,
5381 // or -1 on failure (e.g. can't fork a new process).
5382 // Unlike system(), this function can be called from signal handler. It
5383 // doesn't block SIGINT et al.
5384 int os::fork_and_exec(char* cmd) {
5385   char * argv[4];
5386   argv[0] = (char *)"sh";
5387   argv[1] = (char *)"-c";
5388   argv[2] = cmd;
5389   argv[3] = NULL;
5390 
5391   // fork is async-safe, fork1 is not so can't use in signal handler
5392   pid_t pid;
5393   Thread* t = Thread::current_or_null_safe();
5394   if (t != NULL && t->is_inside_signal_handler()) {
5395     pid = fork();
5396   } else {
5397     pid = fork1();
5398   }
5399 
5400   if (pid < 0) {
5401     // fork failed
5402     warning("fork failed: %s", os::strerror(errno));
5403     return -1;
5404 
5405   } else if (pid == 0) {
5406     // child process
5407 
5408     // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
5409     execve("/usr/bin/sh", argv, environ);
5410 
5411     // execve failed
5412     _exit(-1);
5413 
5414   } else  {
5415     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5416     // care about the actual exit code, for now.
5417 
5418     int status;
5419 
5420     // Wait for the child process to exit.  This returns immediately if
5421     // the child has already exited. */
5422     while (waitpid(pid, &status, 0) < 0) {
5423       switch (errno) {
5424       case ECHILD: return 0;
5425       case EINTR: break;
5426       default: return -1;
5427       }
5428     }
5429 
5430     if (WIFEXITED(status)) {
5431       // The child exited normally; get its exit code.
5432       return WEXITSTATUS(status);
5433     } else if (WIFSIGNALED(status)) {
5434       // The child exited because of a signal
5435       // The best value to return is 0x80 + signal number,
5436       // because that is what all Unix shells do, and because
5437       // it allows callers to distinguish between process exit and
5438       // process death by signal.
5439       return 0x80 + WTERMSIG(status);
5440     } else {
5441       // Unknown exit code; pass it through
5442       return status;
5443     }
5444   }
5445 }
5446 
5447 // is_headless_jre()
5448 //
5449 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5450 // in order to report if we are running in a headless jre
5451 //
5452 // Since JDK8 xawt/libmawt.so was moved into the same directory
5453 // as libawt.so, and renamed libawt_xawt.so
5454 //
5455 bool os::is_headless_jre() {
5456   struct stat statbuf;
5457   char buf[MAXPATHLEN];
5458   char libmawtpath[MAXPATHLEN];
5459   const char *xawtstr  = "/xawt/libmawt.so";
5460   const char *new_xawtstr = "/libawt_xawt.so";
5461   char *p;
5462 
5463   // Get path to libjvm.so
5464   os::jvm_path(buf, sizeof(buf));
5465 
5466   // Get rid of libjvm.so
5467   p = strrchr(buf, '/');
5468   if (p == NULL) {
5469     return false;
5470   } else {
5471     *p = '\0';
5472   }
5473 
5474   // Get rid of client or server
5475   p = strrchr(buf, '/');
5476   if (p == NULL) {
5477     return false;
5478   } else {
5479     *p = '\0';
5480   }
5481 
5482   // check xawt/libmawt.so
5483   strcpy(libmawtpath, buf);
5484   strcat(libmawtpath, xawtstr);
5485   if (::stat(libmawtpath, &statbuf) == 0) return false;
5486 
5487   // check libawt_xawt.so
5488   strcpy(libmawtpath, buf);
5489   strcat(libmawtpath, new_xawtstr);
5490   if (::stat(libmawtpath, &statbuf) == 0) return false;
5491 
5492   return true;
5493 }
5494 
5495 size_t os::write(int fd, const void *buf, unsigned int nBytes) {
5496   size_t res;
5497   RESTARTABLE((size_t) ::write(fd, buf, (size_t) nBytes), res);
5498   return res;
5499 }
5500 
5501 int os::close(int fd) {
5502   return ::close(fd);
5503 }
5504 
5505 int os::socket_close(int fd) {
5506   return ::close(fd);
5507 }
5508 
5509 int os::recv(int fd, char* buf, size_t nBytes, uint flags) {
5510   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5511          "Assumed _thread_in_native");
5512   RESTARTABLE_RETURN_INT((int)::recv(fd, buf, nBytes, flags));
5513 }
5514 
5515 int os::send(int fd, char* buf, size_t nBytes, uint flags) {
5516   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5517          "Assumed _thread_in_native");
5518   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5519 }
5520 
5521 int os::raw_send(int fd, char* buf, size_t nBytes, uint flags) {
5522   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5523 }
5524 
5525 // As both poll and select can be interrupted by signals, we have to be
5526 // prepared to restart the system call after updating the timeout, unless
5527 // a poll() is done with timeout == -1, in which case we repeat with this
5528 // "wait forever" value.
5529 
5530 int os::connect(int fd, struct sockaddr *him, socklen_t len) {
5531   int _result;
5532   _result = ::connect(fd, him, len);
5533 
5534   // On Solaris, when a connect() call is interrupted, the connection
5535   // can be established asynchronously (see 6343810). Subsequent calls
5536   // to connect() must check the errno value which has the semantic
5537   // described below (copied from the connect() man page). Handling
5538   // of asynchronously established connections is required for both
5539   // blocking and non-blocking sockets.
5540   //     EINTR            The  connection  attempt  was   interrupted
5541   //                      before  any data arrived by the delivery of
5542   //                      a signal. The connection, however, will  be
5543   //                      established asynchronously.
5544   //
5545   //     EINPROGRESS      The socket is non-blocking, and the connec-
5546   //                      tion  cannot  be completed immediately.
5547   //
5548   //     EALREADY         The socket is non-blocking,  and a previous
5549   //                      connection  attempt  has  not yet been com-
5550   //                      pleted.
5551   //
5552   //     EISCONN          The socket is already connected.
5553   if (_result == OS_ERR && errno == EINTR) {
5554     // restarting a connect() changes its errno semantics
5555     RESTARTABLE(::connect(fd, him, len), _result);
5556     // undo these changes
5557     if (_result == OS_ERR) {
5558       if (errno == EALREADY) {
5559         errno = EINPROGRESS; // fall through
5560       } else if (errno == EISCONN) {
5561         errno = 0;
5562         return OS_OK;
5563       }
5564     }
5565   }
5566   return _result;
5567 }
5568 
5569 // Get the default path to the core file
5570 // Returns the length of the string
5571 int os::get_core_path(char* buffer, size_t bufferSize) {
5572   const char* p = get_current_directory(buffer, bufferSize);
5573 
5574   if (p == NULL) {
5575     assert(p != NULL, "failed to get current directory");
5576     return 0;
5577   }
5578 
5579   jio_snprintf(buffer, bufferSize, "%s/core or core.%d",
5580                                               p, current_process_id());
5581 
5582   return strlen(buffer);
5583 }
5584 
5585 #ifndef PRODUCT
5586 void TestReserveMemorySpecial_test() {
5587   // No tests available for this platform
5588 }
5589 #endif
5590 
5591 bool os::start_debugging(char *buf, int buflen) {
5592   int len = (int)strlen(buf);
5593   char *p = &buf[len];
5594 
5595   jio_snprintf(p, buflen-len,
5596                "\n\n"
5597                "Do you want to debug the problem?\n\n"
5598                "To debug, run 'dbx - %d'; then switch to thread " INTX_FORMAT "\n"
5599                "Enter 'yes' to launch dbx automatically (PATH must include dbx)\n"
5600                "Otherwise, press RETURN to abort...",
5601                os::current_process_id(), os::current_thread_id());
5602 
5603   bool yes = os::message_box("Unexpected Error", buf);
5604 
5605   if (yes) {
5606     // yes, user asked VM to launch debugger
5607     jio_snprintf(buf, sizeof(buf), "dbx - %d", os::current_process_id());
5608 
5609     os::fork_and_exec(buf);
5610     yes = false;
5611   }
5612   return yes;
5613 }