1 /*
   2  * Copyright (c) 1999, 2014, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright 2012, 2014 SAP AG. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 // According to the AIX OS doc #pragma alloca must be used
  27 // with C++ compiler before referencing the function alloca()
  28 #pragma alloca
  29 
  30 // no precompiled headers
  31 #include "classfile/classLoader.hpp"
  32 #include "classfile/systemDictionary.hpp"
  33 #include "classfile/vmSymbols.hpp"
  34 #include "code/icBuffer.hpp"
  35 #include "code/vtableStubs.hpp"
  36 #include "compiler/compileBroker.hpp"
  37 #include "interpreter/interpreter.hpp"
  38 #include "jvm_aix.h"
  39 #include "libperfstat_aix.hpp"
  40 #include "loadlib_aix.hpp"
  41 #include "memory/allocation.inline.hpp"
  42 #include "memory/filemap.hpp"
  43 #include "mutex_aix.inline.hpp"
  44 #include "oops/oop.inline.hpp"
  45 #include "os_share_aix.hpp"
  46 #include "porting_aix.hpp"
  47 #include "prims/jniFastGetField.hpp"
  48 #include "prims/jvm.h"
  49 #include "prims/jvm_misc.hpp"
  50 #include "runtime/arguments.hpp"
  51 #include "runtime/extendedPC.hpp"
  52 #include "runtime/globals.hpp"
  53 #include "runtime/interfaceSupport.hpp"
  54 #include "runtime/java.hpp"
  55 #include "runtime/javaCalls.hpp"
  56 #include "runtime/mutexLocker.hpp"
  57 #include "runtime/objectMonitor.hpp"
  58 #include "runtime/orderAccess.inline.hpp"
  59 #include "runtime/osThread.hpp"
  60 #include "runtime/perfMemory.hpp"
  61 #include "runtime/sharedRuntime.hpp"
  62 #include "runtime/statSampler.hpp"
  63 #include "runtime/stubRoutines.hpp"
  64 #include "runtime/thread.inline.hpp"
  65 #include "runtime/threadCritical.hpp"
  66 #include "runtime/timer.hpp"
  67 #include "services/attachListener.hpp"
  68 #include "services/runtimeService.hpp"
  69 #include "utilities/decoder.hpp"
  70 #include "utilities/defaultStream.hpp"
  71 #include "utilities/events.hpp"
  72 #include "utilities/growableArray.hpp"
  73 #include "utilities/vmError.hpp"
  74 
  75 // put OS-includes here (sorted alphabetically)
  76 #include <errno.h>
  77 #include <fcntl.h>
  78 #include <inttypes.h>
  79 #include <poll.h>
  80 #include <procinfo.h>
  81 #include <pthread.h>
  82 #include <pwd.h>
  83 #include <semaphore.h>
  84 #include <signal.h>
  85 #include <stdint.h>
  86 #include <stdio.h>
  87 #include <string.h>
  88 #include <unistd.h>
  89 #include <sys/ioctl.h>
  90 #include <sys/ipc.h>
  91 #include <sys/mman.h>
  92 #include <sys/resource.h>
  93 #include <sys/select.h>
  94 #include <sys/shm.h>
  95 #include <sys/socket.h>
  96 #include <sys/stat.h>
  97 #include <sys/sysinfo.h>
  98 #include <sys/systemcfg.h>
  99 #include <sys/time.h>
 100 #include <sys/times.h>
 101 #include <sys/types.h>
 102 #include <sys/utsname.h>
 103 #include <sys/vminfo.h>
 104 #include <sys/wait.h>
 105 
 106 // Add missing declarations (should be in procinfo.h but isn't until AIX 6.1).
 107 #if !defined(_AIXVERSION_610)
 108 extern "C" {
 109   int getthrds64(pid_t ProcessIdentifier,
 110                  struct thrdentry64* ThreadBuffer,
 111                  int ThreadSize,
 112                  tid64_t* IndexPointer,
 113                  int Count);
 114 }
 115 #endif
 116 
 117 // Excerpts from systemcfg.h definitions newer than AIX 5.3
 118 #ifndef PV_7
 119 # define PV_7 0x200000          // Power PC 7
 120 # define PV_7_Compat 0x208000   // Power PC 7
 121 #endif
 122 
 123 #define MAX_PATH (2 * K)
 124 
 125 // for timer info max values which include all bits
 126 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 127 // for multipage initialization error analysis (in 'g_multipage_error')
 128 #define ERROR_MP_OS_TOO_OLD                          100
 129 #define ERROR_MP_EXTSHM_ACTIVE                       101
 130 #define ERROR_MP_VMGETINFO_FAILED                    102
 131 #define ERROR_MP_VMGETINFO_CLAIMS_NO_SUPPORT_FOR_64K 103
 132 
 133 // the semantics in this file are thus that codeptr_t is a *real code ptr*
 134 // This means that any function taking codeptr_t as arguments will assume
 135 // a real codeptr and won't handle function descriptors (eg getFuncName),
 136 // whereas functions taking address as args will deal with function
 137 // descriptors (eg os::dll_address_to_library_name)
 138 typedef unsigned int* codeptr_t;
 139 
 140 // typedefs for stackslots, stack pointers, pointers to op codes
 141 typedef unsigned long stackslot_t;
 142 typedef stackslot_t* stackptr_t;
 143 
 144 // query dimensions of the stack of the calling thread
 145 static void query_stack_dimensions(address* p_stack_base, size_t* p_stack_size);
 146 
 147 // function to check a given stack pointer against given stack limits
 148 inline bool is_valid_stackpointer(stackptr_t sp, stackptr_t stack_base, size_t stack_size) {
 149   if (((uintptr_t)sp) & 0x7) {
 150     return false;
 151   }
 152   if (sp > stack_base) {
 153     return false;
 154   }
 155   if (sp < (stackptr_t) ((address)stack_base - stack_size)) {
 156     return false;
 157   }
 158   return true;
 159 }
 160 
 161 // returns true if function is a valid codepointer
 162 inline bool is_valid_codepointer(codeptr_t p) {
 163   if (!p) {
 164     return false;
 165   }
 166   if (((uintptr_t)p) & 0x3) {
 167     return false;
 168   }
 169   if (LoadedLibraries::find_for_text_address((address)p) == NULL) {
 170     return false;
 171   }
 172   return true;
 173 }
 174 
 175 // macro to check a given stack pointer against given stack limits and to die if test fails
 176 #define CHECK_STACK_PTR(sp, stack_base, stack_size) { \
 177     guarantee(is_valid_stackpointer((stackptr_t)(sp), (stackptr_t)(stack_base), stack_size), "Stack Pointer Invalid"); \
 178 }
 179 
 180 // macro to check the current stack pointer against given stacklimits
 181 #define CHECK_CURRENT_STACK_PTR(stack_base, stack_size) { \
 182   address sp; \
 183   sp = os::current_stack_pointer(); \
 184   CHECK_STACK_PTR(sp, stack_base, stack_size); \
 185 }
 186 
 187 ////////////////////////////////////////////////////////////////////////////////
 188 // global variables (for a description see os_aix.hpp)
 189 
 190 julong    os::Aix::_physical_memory = 0;
 191 pthread_t os::Aix::_main_thread = ((pthread_t)0);
 192 int       os::Aix::_page_size = -1;
 193 int       os::Aix::_on_pase = -1;
 194 int       os::Aix::_os_version = -1;
 195 int       os::Aix::_stack_page_size = -1;
 196 size_t    os::Aix::_shm_default_page_size = -1;
 197 int       os::Aix::_can_use_64K_pages = -1;
 198 int       os::Aix::_can_use_16M_pages = -1;
 199 int       os::Aix::_xpg_sus_mode = -1;
 200 int       os::Aix::_extshm = -1;
 201 int       os::Aix::_logical_cpus = -1;
 202 
 203 ////////////////////////////////////////////////////////////////////////////////
 204 // local variables
 205 
 206 static int      g_multipage_error  = -1;   // error analysis for multipage initialization
 207 static jlong    initial_time_count = 0;
 208 static int      clock_tics_per_sec = 100;
 209 static sigset_t check_signal_done;         // For diagnostics to print a message once (see run_periodic_checks)
 210 static bool     check_signals      = true;
 211 static pid_t    _initial_pid       = 0;
 212 static int      SR_signum          = SIGUSR2; // Signal used to suspend/resume a thread (must be > SIGSEGV, see 4355769)
 213 static sigset_t SR_sigset;
 214 static pthread_mutex_t dl_mutex;           // Used to protect dlsym() calls */
 215 
 216 julong os::available_memory() {
 217   return Aix::available_memory();
 218 }
 219 
 220 julong os::Aix::available_memory() {
 221   os::Aix::meminfo_t mi;
 222   if (os::Aix::get_meminfo(&mi)) {
 223     return mi.real_free;
 224   } else {
 225     return 0xFFFFFFFFFFFFFFFFLL;
 226   }
 227 }
 228 
 229 julong os::physical_memory() {
 230   return Aix::physical_memory();
 231 }
 232 
 233 ////////////////////////////////////////////////////////////////////////////////
 234 // environment support
 235 
 236 bool os::getenv(const char* name, char* buf, int len) {
 237   const char* val = ::getenv(name);
 238   if (val != NULL && strlen(val) < (size_t)len) {
 239     strcpy(buf, val);
 240     return true;
 241   }
 242   if (len > 0) buf[0] = 0;  // return a null string
 243   return false;
 244 }
 245 
 246 
 247 // Return true if user is running as root.
 248 
 249 bool os::have_special_privileges() {
 250   static bool init = false;
 251   static bool privileges = false;
 252   if (!init) {
 253     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 254     init = true;
 255   }
 256   return privileges;
 257 }
 258 
 259 // Helper function, emulates disclaim64 using multiple 32bit disclaims
 260 // because we cannot use disclaim64() on AS/400 and old AIX releases.
 261 static bool my_disclaim64(char* addr, size_t size) {
 262 
 263   if (size == 0) {
 264     return true;
 265   }
 266 
 267   // Maximum size 32bit disclaim() accepts. (Theoretically 4GB, but I just do not trust that.)
 268   const unsigned int maxDisclaimSize = 0x80000000;
 269 
 270   const unsigned int numFullDisclaimsNeeded = (size / maxDisclaimSize);
 271   const unsigned int lastDisclaimSize = (size % maxDisclaimSize);
 272 
 273   char* p = addr;
 274 
 275   for (int i = 0; i < numFullDisclaimsNeeded; i ++) {
 276     if (::disclaim(p, maxDisclaimSize, DISCLAIM_ZEROMEM) != 0) {
 277       //if (Verbose)
 278       fprintf(stderr, "Cannot disclaim %p - %p (errno %d)\n", p, p + maxDisclaimSize, errno);
 279       return false;
 280     }
 281     p += maxDisclaimSize;
 282   }
 283 
 284   if (lastDisclaimSize > 0) {
 285     if (::disclaim(p, lastDisclaimSize, DISCLAIM_ZEROMEM) != 0) {
 286       //if (Verbose)
 287         fprintf(stderr, "Cannot disclaim %p - %p (errno %d)\n", p, p + lastDisclaimSize, errno);
 288       return false;
 289     }
 290   }
 291 
 292   return true;
 293 }
 294 
 295 // Cpu architecture string
 296 #if defined(PPC32)
 297 static char cpu_arch[] = "ppc";
 298 #elif defined(PPC64)
 299 static char cpu_arch[] = "ppc64";
 300 #else
 301 #error Add appropriate cpu_arch setting
 302 #endif
 303 
 304 
 305 // Given an address, returns the size of the page backing that address.
 306 size_t os::Aix::query_pagesize(void* addr) {
 307 
 308   vm_page_info pi;
 309   pi.addr = (uint64_t)addr;
 310   if (::vmgetinfo(&pi, VM_PAGE_INFO, sizeof(pi)) == 0) {
 311     return pi.pagesize;
 312   } else {
 313     fprintf(stderr, "vmgetinfo failed to retrieve page size for address %p (errno %d).\n", addr, errno);
 314     assert(false, "vmgetinfo failed to retrieve page size");
 315     return SIZE_4K;
 316   }
 317 
 318 }
 319 
 320 // Returns the kernel thread id of the currently running thread.
 321 pid_t os::Aix::gettid() {
 322   return (pid_t) thread_self();
 323 }
 324 
 325 void os::Aix::initialize_system_info() {
 326 
 327   // get the number of online(logical) cpus instead of configured
 328   os::_processor_count = sysconf(_SC_NPROCESSORS_ONLN);
 329   assert(_processor_count > 0, "_processor_count must be > 0");
 330 
 331   // retrieve total physical storage
 332   os::Aix::meminfo_t mi;
 333   if (!os::Aix::get_meminfo(&mi)) {
 334     fprintf(stderr, "os::Aix::get_meminfo failed.\n"); fflush(stderr);
 335     assert(false, "os::Aix::get_meminfo failed.");
 336   }
 337   _physical_memory = (julong) mi.real_total;
 338 }
 339 
 340 // Helper function for tracing page sizes.
 341 static const char* describe_pagesize(size_t pagesize) {
 342   switch (pagesize) {
 343     case SIZE_4K : return "4K";
 344     case SIZE_64K: return "64K";
 345     case SIZE_16M: return "16M";
 346     case SIZE_16G: return "16G";
 347     default:
 348       assert(false, "surprise");
 349       return "??";
 350   }
 351 }
 352 
 353 // Retrieve information about multipage size support. Will initialize
 354 // Aix::_page_size, Aix::_stack_page_size, Aix::_can_use_64K_pages,
 355 // Aix::_can_use_16M_pages.
 356 // Must be called before calling os::large_page_init().
 357 void os::Aix::query_multipage_support() {
 358 
 359   guarantee(_page_size == -1 &&
 360             _stack_page_size == -1 &&
 361             _can_use_64K_pages == -1 &&
 362             _can_use_16M_pages == -1 &&
 363             g_multipage_error == -1,
 364             "do not call twice");
 365 
 366   _page_size = ::sysconf(_SC_PAGESIZE);
 367 
 368   // This really would surprise me.
 369   assert(_page_size == SIZE_4K, "surprise!");
 370 
 371 
 372   // Query default data page size (default page size for C-Heap, pthread stacks and .bss).
 373   // Default data page size is influenced either by linker options (-bdatapsize)
 374   // or by environment variable LDR_CNTRL (suboption DATAPSIZE). If none is given,
 375   // default should be 4K.
 376   size_t data_page_size = SIZE_4K;
 377   {
 378     void* p = ::malloc(SIZE_16M);
 379     guarantee(p != NULL, "malloc failed");
 380     data_page_size = os::Aix::query_pagesize(p);
 381     ::free(p);
 382   }
 383 
 384   // query default shm page size (LDR_CNTRL SHMPSIZE)
 385   {
 386     const int shmid = ::shmget(IPC_PRIVATE, 1, IPC_CREAT | S_IRUSR | S_IWUSR);
 387     guarantee(shmid != -1, "shmget failed");
 388     void* p = ::shmat(shmid, NULL, 0);
 389     ::shmctl(shmid, IPC_RMID, NULL);
 390     guarantee(p != (void*) -1, "shmat failed");
 391     _shm_default_page_size = os::Aix::query_pagesize(p);
 392     ::shmdt(p);
 393   }
 394 
 395   // before querying the stack page size, make sure we are not running as primordial
 396   // thread (because primordial thread's stack may have different page size than
 397   // pthread thread stacks). Running a VM on the primordial thread won't work for a
 398   // number of reasons so we may just as well guarantee it here
 399   guarantee(!os::Aix::is_primordial_thread(), "Must not be called for primordial thread");
 400 
 401   // query stack page size
 402   {
 403     int dummy = 0;
 404     _stack_page_size = os::Aix::query_pagesize(&dummy);
 405     // everything else would surprise me and should be looked into
 406     guarantee(_stack_page_size == SIZE_4K || _stack_page_size == SIZE_64K, "Wrong page size");
 407     // also, just for completeness: pthread stacks are allocated from C heap, so
 408     // stack page size should be the same as data page size
 409     guarantee(_stack_page_size == data_page_size, "stack page size should be the same as data page size");
 410   }
 411 
 412   // EXTSHM is bad: among other things, it prevents setting pagesize dynamically
 413   // for system V shm.
 414   if (Aix::extshm()) {
 415     if (Verbose) {
 416       fprintf(stderr, "EXTSHM is active - will disable large page support.\n"
 417                       "Please make sure EXTSHM is OFF for large page support.\n");
 418     }
 419     g_multipage_error = ERROR_MP_EXTSHM_ACTIVE;
 420     _can_use_64K_pages = _can_use_16M_pages = 0;
 421     goto query_multipage_support_end;
 422   }
 423 
 424   // now check which page sizes the OS claims it supports, and of those, which actually can be used.
 425   {
 426     const int MAX_PAGE_SIZES = 4;
 427     psize_t sizes[MAX_PAGE_SIZES];
 428     const int num_psizes = ::vmgetinfo(sizes, VMINFO_GETPSIZES, MAX_PAGE_SIZES);
 429     if (num_psizes == -1) {
 430       if (Verbose) {
 431         fprintf(stderr, "vmgetinfo(VMINFO_GETPSIZES) failed (errno: %d)\n", errno);
 432         fprintf(stderr, "disabling multipage support.\n");
 433       }
 434       g_multipage_error = ERROR_MP_VMGETINFO_FAILED;
 435       _can_use_64K_pages = _can_use_16M_pages = 0;
 436       goto query_multipage_support_end;
 437     }
 438     guarantee(num_psizes > 0, "vmgetinfo(.., VMINFO_GETPSIZES, ...) failed.");
 439     assert(num_psizes <= MAX_PAGE_SIZES, "Surprise! more than 4 page sizes?");
 440     if (Verbose) {
 441       fprintf(stderr, "vmgetinfo(.., VMINFO_GETPSIZES, ...) returns %d supported page sizes: ", num_psizes);
 442       for (int i = 0; i < num_psizes; i ++) {
 443         fprintf(stderr, " %s ", describe_pagesize(sizes[i]));
 444       }
 445       fprintf(stderr, " .\n");
 446     }
 447 
 448     // Can we use 64K, 16M pages?
 449     _can_use_64K_pages = 0;
 450     _can_use_16M_pages = 0;
 451     for (int i = 0; i < num_psizes; i ++) {
 452       if (sizes[i] == SIZE_64K) {
 453         _can_use_64K_pages = 1;
 454       } else if (sizes[i] == SIZE_16M) {
 455         _can_use_16M_pages = 1;
 456       }
 457     }
 458 
 459     if (!_can_use_64K_pages) {
 460       g_multipage_error = ERROR_MP_VMGETINFO_CLAIMS_NO_SUPPORT_FOR_64K;
 461     }
 462 
 463     // Double-check for 16M pages: Even if AIX claims to be able to use 16M pages,
 464     // there must be an actual 16M page pool, and we must run with enough rights.
 465     if (_can_use_16M_pages) {
 466       const int shmid = ::shmget(IPC_PRIVATE, SIZE_16M, IPC_CREAT | S_IRUSR | S_IWUSR);
 467       guarantee(shmid != -1, "shmget failed");
 468       struct shmid_ds shm_buf = { 0 };
 469       shm_buf.shm_pagesize = SIZE_16M;
 470       const bool can_set_pagesize = ::shmctl(shmid, SHM_PAGESIZE, &shm_buf) == 0 ? true : false;
 471       const int en = errno;
 472       ::shmctl(shmid, IPC_RMID, NULL);
 473       if (!can_set_pagesize) {
 474         if (Verbose) {
 475           fprintf(stderr, "Failed to allocate even one misely 16M page. shmctl failed with %d (%s).\n"
 476                           "Will deactivate 16M support.\n", en, strerror(en));
 477         }
 478         _can_use_16M_pages = 0;
 479       }
 480     }
 481 
 482   } // end: check which pages can be used for shared memory
 483 
 484 query_multipage_support_end:
 485 
 486   guarantee(_page_size != -1 &&
 487             _stack_page_size != -1 &&
 488             _can_use_64K_pages != -1 &&
 489             _can_use_16M_pages != -1, "Page sizes not properly initialized");
 490 
 491   if (_can_use_64K_pages) {
 492     g_multipage_error = 0;
 493   }
 494 
 495   if (Verbose) {
 496     fprintf(stderr, "Data page size (C-Heap, bss, etc): %s\n", describe_pagesize(data_page_size));
 497     fprintf(stderr, "Thread stack page size (pthread): %s\n", describe_pagesize(_stack_page_size));
 498     fprintf(stderr, "Default shared memory page size: %s\n", describe_pagesize(_shm_default_page_size));
 499     fprintf(stderr, "Can use 64K pages dynamically with shared meory: %s\n", (_can_use_64K_pages ? "yes" :"no"));
 500     fprintf(stderr, "Can use 16M pages dynamically with shared memory: %s\n", (_can_use_16M_pages ? "yes" :"no"));
 501     fprintf(stderr, "Multipage error details: %d\n", g_multipage_error);
 502   }
 503 
 504 } // end os::Aix::query_multipage_support()
 505 
 506 // The code for this method was initially derived from the version in os_linux.cpp.
 507 void os::init_system_properties_values() {
 508 
 509 #define DEFAULT_LIBPATH "/usr/lib:/lib"
 510 #define EXTENSIONS_DIR  "/lib/ext"
 511 #define ENDORSED_DIR    "/lib/endorsed"
 512 
 513   // Buffer that fits several sprintfs.
 514   // Note that the space for the trailing null is provided
 515   // by the nulls included by the sizeof operator.
 516   const size_t bufsize =
 517     MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
 518          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR), // extensions dir
 519          (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
 520   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 521 
 522   // sysclasspath, java_home, dll_dir
 523   {
 524     char *pslash;
 525     os::jvm_path(buf, bufsize);
 526 
 527     // Found the full path to libjvm.so.
 528     // Now cut the path to <java_home>/jre if we can.
 529     *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
 530     pslash = strrchr(buf, '/');
 531     if (pslash != NULL) {
 532       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 533     }
 534     Arguments::set_dll_dir(buf);
 535 
 536     if (pslash != NULL) {
 537       pslash = strrchr(buf, '/');
 538       if (pslash != NULL) {
 539         *pslash = '\0';          // Get rid of /<arch>.
 540         pslash = strrchr(buf, '/');
 541         if (pslash != NULL) {
 542           *pslash = '\0';        // Get rid of /lib.
 543         }
 544       }
 545     }
 546     Arguments::set_java_home(buf);
 547     set_boot_path('/', ':');
 548   }
 549 
 550   // Where to look for native libraries.
 551 
 552   // On Aix we get the user setting of LIBPATH.
 553   // Eventually, all the library path setting will be done here.
 554   // Get the user setting of LIBPATH.
 555   const char *v = ::getenv("LIBPATH");
 556   const char *v_colon = ":";
 557   if (v == NULL) { v = ""; v_colon = ""; }
 558 
 559   // Concatenate user and invariant part of ld_library_path.
 560   // That's +1 for the colon and +1 for the trailing '\0'.
 561   char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, strlen(v) + 1 + sizeof(DEFAULT_LIBPATH) + 1, mtInternal);
 562   sprintf(ld_library_path, "%s%s" DEFAULT_LIBPATH, v, v_colon);
 563   Arguments::set_library_path(ld_library_path);
 564   FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
 565 
 566   // Extensions directories.
 567   sprintf(buf, "%s" EXTENSIONS_DIR, Arguments::get_java_home());
 568   Arguments::set_ext_dirs(buf);
 569 
 570   // Endorsed standards default directory.
 571   sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 572   Arguments::set_endorsed_dirs(buf);
 573 
 574   FREE_C_HEAP_ARRAY(char, buf, mtInternal);
 575 
 576 #undef DEFAULT_LIBPATH
 577 #undef EXTENSIONS_DIR
 578 #undef ENDORSED_DIR
 579 }
 580 
 581 ////////////////////////////////////////////////////////////////////////////////
 582 // breakpoint support
 583 
 584 void os::breakpoint() {
 585   BREAKPOINT;
 586 }
 587 
 588 extern "C" void breakpoint() {
 589   // use debugger to set breakpoint here
 590 }
 591 
 592 ////////////////////////////////////////////////////////////////////////////////
 593 // signal support
 594 
 595 debug_only(static bool signal_sets_initialized = false);
 596 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 597 
 598 bool os::Aix::is_sig_ignored(int sig) {
 599   struct sigaction oact;
 600   sigaction(sig, (struct sigaction*)NULL, &oact);
 601   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
 602     : CAST_FROM_FN_PTR(void*, oact.sa_handler);
 603   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 604     return true;
 605   else
 606     return false;
 607 }
 608 
 609 void os::Aix::signal_sets_init() {
 610   // Should also have an assertion stating we are still single-threaded.
 611   assert(!signal_sets_initialized, "Already initialized");
 612   // Fill in signals that are necessarily unblocked for all threads in
 613   // the VM. Currently, we unblock the following signals:
 614   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 615   //                         by -Xrs (=ReduceSignalUsage));
 616   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 617   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 618   // the dispositions or masks wrt these signals.
 619   // Programs embedding the VM that want to use the above signals for their
 620   // own purposes must, at this time, use the "-Xrs" option to prevent
 621   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 622   // (See bug 4345157, and other related bugs).
 623   // In reality, though, unblocking these signals is really a nop, since
 624   // these signals are not blocked by default.
 625   sigemptyset(&unblocked_sigs);
 626   sigemptyset(&allowdebug_blocked_sigs);
 627   sigaddset(&unblocked_sigs, SIGILL);
 628   sigaddset(&unblocked_sigs, SIGSEGV);
 629   sigaddset(&unblocked_sigs, SIGBUS);
 630   sigaddset(&unblocked_sigs, SIGFPE);
 631   sigaddset(&unblocked_sigs, SIGTRAP);
 632   sigaddset(&unblocked_sigs, SIGDANGER);
 633   sigaddset(&unblocked_sigs, SR_signum);
 634 
 635   if (!ReduceSignalUsage) {
 636    if (!os::Aix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 637      sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 638      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 639    }
 640    if (!os::Aix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 641      sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 642      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 643    }
 644    if (!os::Aix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 645      sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 646      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 647    }
 648   }
 649   // Fill in signals that are blocked by all but the VM thread.
 650   sigemptyset(&vm_sigs);
 651   if (!ReduceSignalUsage)
 652     sigaddset(&vm_sigs, BREAK_SIGNAL);
 653   debug_only(signal_sets_initialized = true);
 654 }
 655 
 656 // These are signals that are unblocked while a thread is running Java.
 657 // (For some reason, they get blocked by default.)
 658 sigset_t* os::Aix::unblocked_signals() {
 659   assert(signal_sets_initialized, "Not initialized");
 660   return &unblocked_sigs;
 661 }
 662 
 663 // These are the signals that are blocked while a (non-VM) thread is
 664 // running Java. Only the VM thread handles these signals.
 665 sigset_t* os::Aix::vm_signals() {
 666   assert(signal_sets_initialized, "Not initialized");
 667   return &vm_sigs;
 668 }
 669 
 670 // These are signals that are blocked during cond_wait to allow debugger in
 671 sigset_t* os::Aix::allowdebug_blocked_signals() {
 672   assert(signal_sets_initialized, "Not initialized");
 673   return &allowdebug_blocked_sigs;
 674 }
 675 
 676 void os::Aix::hotspot_sigmask(Thread* thread) {
 677 
 678   //Save caller's signal mask before setting VM signal mask
 679   sigset_t caller_sigmask;
 680   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 681 
 682   OSThread* osthread = thread->osthread();
 683   osthread->set_caller_sigmask(caller_sigmask);
 684 
 685   pthread_sigmask(SIG_UNBLOCK, os::Aix::unblocked_signals(), NULL);
 686 
 687   if (!ReduceSignalUsage) {
 688     if (thread->is_VM_thread()) {
 689       // Only the VM thread handles BREAK_SIGNAL ...
 690       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 691     } else {
 692       // ... all other threads block BREAK_SIGNAL
 693       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 694     }
 695   }
 696 }
 697 
 698 // retrieve memory information.
 699 // Returns false if something went wrong;
 700 // content of pmi undefined in this case.
 701 bool os::Aix::get_meminfo(meminfo_t* pmi) {
 702 
 703   assert(pmi, "get_meminfo: invalid parameter");
 704 
 705   memset(pmi, 0, sizeof(meminfo_t));
 706 
 707   if (os::Aix::on_pase()) {
 708 
 709     Unimplemented();
 710     return false;
 711 
 712   } else {
 713 
 714     // On AIX, I use the (dynamically loaded) perfstat library to retrieve memory statistics
 715     // See:
 716     // http://publib.boulder.ibm.com/infocenter/systems/index.jsp
 717     //        ?topic=/com.ibm.aix.basetechref/doc/basetrf1/perfstat_memtot.htm
 718     // http://publib.boulder.ibm.com/infocenter/systems/index.jsp
 719     //        ?topic=/com.ibm.aix.files/doc/aixfiles/libperfstat.h.htm
 720 
 721     perfstat_memory_total_t psmt;
 722     memset (&psmt, '\0', sizeof(psmt));
 723     const int rc = libperfstat::perfstat_memory_total(NULL, &psmt, sizeof(psmt), 1);
 724     if (rc == -1) {
 725       fprintf(stderr, "perfstat_memory_total() failed (errno=%d)\n", errno);
 726       assert(0, "perfstat_memory_total() failed");
 727       return false;
 728     }
 729 
 730     assert(rc == 1, "perfstat_memory_total() - weird return code");
 731 
 732     // excerpt from
 733     // http://publib.boulder.ibm.com/infocenter/systems/index.jsp
 734     //        ?topic=/com.ibm.aix.files/doc/aixfiles/libperfstat.h.htm
 735     // The fields of perfstat_memory_total_t:
 736     // u_longlong_t virt_total         Total virtual memory (in 4 KB pages).
 737     // u_longlong_t real_total         Total real memory (in 4 KB pages).
 738     // u_longlong_t real_free          Free real memory (in 4 KB pages).
 739     // u_longlong_t pgsp_total         Total paging space (in 4 KB pages).
 740     // u_longlong_t pgsp_free          Free paging space (in 4 KB pages).
 741 
 742     pmi->virt_total = psmt.virt_total * 4096;
 743     pmi->real_total = psmt.real_total * 4096;
 744     pmi->real_free = psmt.real_free * 4096;
 745     pmi->pgsp_total = psmt.pgsp_total * 4096;
 746     pmi->pgsp_free = psmt.pgsp_free * 4096;
 747 
 748     return true;
 749 
 750   }
 751 } // end os::Aix::get_meminfo
 752 
 753 // Retrieve global cpu information.
 754 // Returns false if something went wrong;
 755 // the content of pci is undefined in this case.
 756 bool os::Aix::get_cpuinfo(cpuinfo_t* pci) {
 757   assert(pci, "get_cpuinfo: invalid parameter");
 758   memset(pci, 0, sizeof(cpuinfo_t));
 759 
 760   perfstat_cpu_total_t psct;
 761   memset (&psct, '\0', sizeof(psct));
 762 
 763   if (-1 == libperfstat::perfstat_cpu_total(NULL, &psct, sizeof(perfstat_cpu_total_t), 1)) {
 764     fprintf(stderr, "perfstat_cpu_total() failed (errno=%d)\n", errno);
 765     assert(0, "perfstat_cpu_total() failed");
 766     return false;
 767   }
 768 
 769   // global cpu information
 770   strcpy (pci->description, psct.description);
 771   pci->processorHZ = psct.processorHZ;
 772   pci->ncpus = psct.ncpus;
 773   os::Aix::_logical_cpus = psct.ncpus;
 774   for (int i = 0; i < 3; i++) {
 775     pci->loadavg[i] = (double) psct.loadavg[i] / (1 << SBITS);
 776   }
 777 
 778   // get the processor version from _system_configuration
 779   switch (_system_configuration.version) {
 780   case PV_7:
 781     strcpy(pci->version, "Power PC 7");
 782     break;
 783   case PV_6_1:
 784     strcpy(pci->version, "Power PC 6 DD1.x");
 785     break;
 786   case PV_6:
 787     strcpy(pci->version, "Power PC 6");
 788     break;
 789   case PV_5:
 790     strcpy(pci->version, "Power PC 5");
 791     break;
 792   case PV_5_2:
 793     strcpy(pci->version, "Power PC 5_2");
 794     break;
 795   case PV_5_3:
 796     strcpy(pci->version, "Power PC 5_3");
 797     break;
 798   case PV_5_Compat:
 799     strcpy(pci->version, "PV_5_Compat");
 800     break;
 801   case PV_6_Compat:
 802     strcpy(pci->version, "PV_6_Compat");
 803     break;
 804   case PV_7_Compat:
 805     strcpy(pci->version, "PV_7_Compat");
 806     break;
 807   default:
 808     strcpy(pci->version, "unknown");
 809   }
 810 
 811   return true;
 812 
 813 } //end os::Aix::get_cpuinfo
 814 
 815 //////////////////////////////////////////////////////////////////////////////
 816 // detecting pthread library
 817 
 818 void os::Aix::libpthread_init() {
 819   return;
 820 }
 821 
 822 //////////////////////////////////////////////////////////////////////////////
 823 // create new thread
 824 
 825 // Thread start routine for all newly created threads
 826 static void *java_start(Thread *thread) {
 827 
 828   // find out my own stack dimensions
 829   {
 830     // actually, this should do exactly the same as thread->record_stack_base_and_size...
 831     address base = 0;
 832     size_t size = 0;
 833     query_stack_dimensions(&base, &size);
 834     thread->set_stack_base(base);
 835     thread->set_stack_size(size);
 836   }
 837 
 838   // Do some sanity checks.
 839   CHECK_CURRENT_STACK_PTR(thread->stack_base(), thread->stack_size());
 840 
 841   // Try to randomize the cache line index of hot stack frames.
 842   // This helps when threads of the same stack traces evict each other's
 843   // cache lines. The threads can be either from the same JVM instance, or
 844   // from different JVM instances. The benefit is especially true for
 845   // processors with hyperthreading technology.
 846 
 847   static int counter = 0;
 848   int pid = os::current_process_id();
 849   alloca(((pid ^ counter++) & 7) * 128);
 850 
 851   ThreadLocalStorage::set_thread(thread);
 852 
 853   OSThread* osthread = thread->osthread();
 854 
 855   // thread_id is kernel thread id (similar to Solaris LWP id)
 856   osthread->set_thread_id(os::Aix::gettid());
 857 
 858   // initialize signal mask for this thread
 859   os::Aix::hotspot_sigmask(thread);
 860 
 861   // initialize floating point control register
 862   os::Aix::init_thread_fpu_state();
 863 
 864   assert(osthread->get_state() == RUNNABLE, "invalid os thread state");
 865 
 866   // call one more level start routine
 867   thread->run();
 868 
 869   return 0;
 870 }
 871 
 872 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 873 
 874   // We want the whole function to be synchronized.
 875   ThreadCritical cs;
 876 
 877   assert(thread->osthread() == NULL, "caller responsible");
 878 
 879   // Allocate the OSThread object
 880   OSThread* osthread = new OSThread(NULL, NULL);
 881   if (osthread == NULL) {
 882     return false;
 883   }
 884 
 885   // set the correct thread state
 886   osthread->set_thread_type(thr_type);
 887 
 888   // Initial state is ALLOCATED but not INITIALIZED
 889   osthread->set_state(ALLOCATED);
 890 
 891   thread->set_osthread(osthread);
 892 
 893   // init thread attributes
 894   pthread_attr_t attr;
 895   pthread_attr_init(&attr);
 896   guarantee(pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED) == 0, "???");
 897 
 898   // Make sure we run in 1:1 kernel-user-thread mode.
 899   if (os::Aix::on_aix()) {
 900     guarantee(pthread_attr_setscope(&attr, PTHREAD_SCOPE_SYSTEM) == 0, "???");
 901     guarantee(pthread_attr_setinheritsched(&attr, PTHREAD_EXPLICIT_SCHED) == 0, "???");
 902   } // end: aix
 903 
 904   // Start in suspended state, and in os::thread_start, wake the thread up.
 905   guarantee(pthread_attr_setsuspendstate_np(&attr, PTHREAD_CREATE_SUSPENDED_NP) == 0, "???");
 906 
 907   // calculate stack size if it's not specified by caller
 908   if (os::Aix::supports_variable_stack_size()) {
 909     if (stack_size == 0) {
 910       stack_size = os::Aix::default_stack_size(thr_type);
 911 
 912       switch (thr_type) {
 913       case os::java_thread:
 914         // Java threads use ThreadStackSize whose default value can be changed with the flag -Xss.
 915         assert(JavaThread::stack_size_at_create() > 0, "this should be set");
 916         stack_size = JavaThread::stack_size_at_create();
 917         break;
 918       case os::compiler_thread:
 919         if (CompilerThreadStackSize > 0) {
 920           stack_size = (size_t)(CompilerThreadStackSize * K);
 921           break;
 922         } // else fall through:
 923           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 924       case os::vm_thread:
 925       case os::pgc_thread:
 926       case os::cgc_thread:
 927       case os::watcher_thread:
 928         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 929         break;
 930       }
 931     }
 932 
 933     stack_size = MAX2(stack_size, os::Aix::min_stack_allowed);
 934     pthread_attr_setstacksize(&attr, stack_size);
 935   } //else let thread_create() pick the default value (96 K on AIX)
 936 
 937   pthread_t tid;
 938   int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 939 
 940   pthread_attr_destroy(&attr);
 941 
 942   if (ret != 0) {
 943     if (PrintMiscellaneous && (Verbose || WizardMode)) {
 944       perror("pthread_create()");
 945     }
 946     // Need to clean up stuff we've allocated so far
 947     thread->set_osthread(NULL);
 948     delete osthread;
 949     return false;
 950   }
 951 
 952   // Store pthread info into the OSThread
 953   osthread->set_pthread_id(tid);
 954 
 955   return true;
 956 }
 957 
 958 /////////////////////////////////////////////////////////////////////////////
 959 // attach existing thread
 960 
 961 // bootstrap the main thread
 962 bool os::create_main_thread(JavaThread* thread) {
 963   assert(os::Aix::_main_thread == pthread_self(), "should be called inside main thread");
 964   return create_attached_thread(thread);
 965 }
 966 
 967 bool os::create_attached_thread(JavaThread* thread) {
 968 #ifdef ASSERT
 969     thread->verify_not_published();
 970 #endif
 971 
 972   // Allocate the OSThread object
 973   OSThread* osthread = new OSThread(NULL, NULL);
 974 
 975   if (osthread == NULL) {
 976     return false;
 977   }
 978 
 979   // Store pthread info into the OSThread
 980   osthread->set_thread_id(os::Aix::gettid());
 981   osthread->set_pthread_id(::pthread_self());
 982 
 983   // initialize floating point control register
 984   os::Aix::init_thread_fpu_state();
 985 
 986   // some sanity checks
 987   CHECK_CURRENT_STACK_PTR(thread->stack_base(), thread->stack_size());
 988 
 989   // Initial thread state is RUNNABLE
 990   osthread->set_state(RUNNABLE);
 991 
 992   thread->set_osthread(osthread);
 993 
 994   if (UseNUMA) {
 995     int lgrp_id = os::numa_get_group_id();
 996     if (lgrp_id != -1) {
 997       thread->set_lgrp_id(lgrp_id);
 998     }
 999   }
1000 
1001   // initialize signal mask for this thread
1002   // and save the caller's signal mask
1003   os::Aix::hotspot_sigmask(thread);
1004 
1005   return true;
1006 }
1007 
1008 void os::pd_start_thread(Thread* thread) {
1009   int status = pthread_continue_np(thread->osthread()->pthread_id());
1010   assert(status == 0, "thr_continue failed");
1011 }
1012 
1013 // Free OS resources related to the OSThread
1014 void os::free_thread(OSThread* osthread) {
1015   assert(osthread != NULL, "osthread not set");
1016 
1017   if (Thread::current()->osthread() == osthread) {
1018     // Restore caller's signal mask
1019     sigset_t sigmask = osthread->caller_sigmask();
1020     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1021    }
1022 
1023   delete osthread;
1024 }
1025 
1026 //////////////////////////////////////////////////////////////////////////////
1027 // thread local storage
1028 
1029 int os::allocate_thread_local_storage() {
1030   pthread_key_t key;
1031   int rslt = pthread_key_create(&key, NULL);
1032   assert(rslt == 0, "cannot allocate thread local storage");
1033   return (int)key;
1034 }
1035 
1036 // Note: This is currently not used by VM, as we don't destroy TLS key
1037 // on VM exit.
1038 void os::free_thread_local_storage(int index) {
1039   int rslt = pthread_key_delete((pthread_key_t)index);
1040   assert(rslt == 0, "invalid index");
1041 }
1042 
1043 void os::thread_local_storage_at_put(int index, void* value) {
1044   int rslt = pthread_setspecific((pthread_key_t)index, value);
1045   assert(rslt == 0, "pthread_setspecific failed");
1046 }
1047 
1048 extern "C" Thread* get_thread() {
1049   return ThreadLocalStorage::thread();
1050 }
1051 
1052 ////////////////////////////////////////////////////////////////////////////////
1053 // time support
1054 
1055 // Time since start-up in seconds to a fine granularity.
1056 // Used by VMSelfDestructTimer and the MemProfiler.
1057 double os::elapsedTime() {
1058   return (double)(os::elapsed_counter()) * 0.000001;
1059 }
1060 
1061 jlong os::elapsed_counter() {
1062   timeval time;
1063   int status = gettimeofday(&time, NULL);
1064   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1065 }
1066 
1067 jlong os::elapsed_frequency() {
1068   return (1000 * 1000);
1069 }
1070 
1071 // For now, we say that linux does not support vtime. I have no idea
1072 // whether it can actually be made to (DLD, 9/13/05).
1073 
1074 bool os::supports_vtime() { return false; }
1075 bool os::enable_vtime()   { return false; }
1076 bool os::vtime_enabled()  { return false; }
1077 double os::elapsedVTime() {
1078   // better than nothing, but not much
1079   return elapsedTime();
1080 }
1081 
1082 jlong os::javaTimeMillis() {
1083   timeval time;
1084   int status = gettimeofday(&time, NULL);
1085   assert(status != -1, "aix error at gettimeofday()");
1086   return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1087 }
1088 
1089 // We need to manually declare mread_real_time,
1090 // because IBM didn't provide a prototype in time.h.
1091 // (they probably only ever tested in C, not C++)
1092 extern "C"
1093 int mread_real_time(timebasestruct_t *t, size_t size_of_timebasestruct_t);
1094 
1095 jlong os::javaTimeNanos() {
1096   if (os::Aix::on_pase()) {
1097     Unimplemented();
1098     return 0;
1099   }
1100   else {
1101     // On AIX use the precision of processors real time clock
1102     // or time base registers.
1103     timebasestruct_t time;
1104     int rc;
1105 
1106     // If the CPU has a time register, it will be used and
1107     // we have to convert to real time first. After convertion we have following data:
1108     // time.tb_high [seconds since 00:00:00 UTC on 1.1.1970]
1109     // time.tb_low  [nanoseconds after the last full second above]
1110     // We better use mread_real_time here instead of read_real_time
1111     // to ensure that we will get a monotonic increasing time.
1112     if (mread_real_time(&time, TIMEBASE_SZ) != RTC_POWER) {
1113       rc = time_base_to_time(&time, TIMEBASE_SZ);
1114       assert(rc != -1, "aix error at time_base_to_time()");
1115     }
1116     return jlong(time.tb_high) * (1000 * 1000 * 1000) + jlong(time.tb_low);
1117   }
1118 }
1119 
1120 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1121   info_ptr->max_value = ALL_64_BITS;
1122   // mread_real_time() is monotonic (see 'os::javaTimeNanos()')
1123   info_ptr->may_skip_backward = false;
1124   info_ptr->may_skip_forward = false;
1125   info_ptr->kind = JVMTI_TIMER_ELAPSED;    // elapsed not CPU time
1126 }
1127 
1128 // Return the real, user, and system times in seconds from an
1129 // arbitrary fixed point in the past.
1130 bool os::getTimesSecs(double* process_real_time,
1131                       double* process_user_time,
1132                       double* process_system_time) {
1133   struct tms ticks;
1134   clock_t real_ticks = times(&ticks);
1135 
1136   if (real_ticks == (clock_t) (-1)) {
1137     return false;
1138   } else {
1139     double ticks_per_second = (double) clock_tics_per_sec;
1140     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1141     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1142     *process_real_time = ((double) real_ticks) / ticks_per_second;
1143 
1144     return true;
1145   }
1146 }
1147 
1148 
1149 char * os::local_time_string(char *buf, size_t buflen) {
1150   struct tm t;
1151   time_t long_time;
1152   time(&long_time);
1153   localtime_r(&long_time, &t);
1154   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1155                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1156                t.tm_hour, t.tm_min, t.tm_sec);
1157   return buf;
1158 }
1159 
1160 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1161   return localtime_r(clock, res);
1162 }
1163 
1164 ////////////////////////////////////////////////////////////////////////////////
1165 // runtime exit support
1166 
1167 // Note: os::shutdown() might be called very early during initialization, or
1168 // called from signal handler. Before adding something to os::shutdown(), make
1169 // sure it is async-safe and can handle partially initialized VM.
1170 void os::shutdown() {
1171 
1172   // allow PerfMemory to attempt cleanup of any persistent resources
1173   perfMemory_exit();
1174 
1175   // needs to remove object in file system
1176   AttachListener::abort();
1177 
1178   // flush buffered output, finish log files
1179   ostream_abort();
1180 
1181   // Check for abort hook
1182   abort_hook_t abort_hook = Arguments::abort_hook();
1183   if (abort_hook != NULL) {
1184     abort_hook();
1185   }
1186 
1187 }
1188 
1189 // Note: os::abort() might be called very early during initialization, or
1190 // called from signal handler. Before adding something to os::abort(), make
1191 // sure it is async-safe and can handle partially initialized VM.
1192 void os::abort(bool dump_core) {
1193   os::shutdown();
1194   if (dump_core) {
1195 #ifndef PRODUCT
1196     fdStream out(defaultStream::output_fd());
1197     out.print_raw("Current thread is ");
1198     char buf[16];
1199     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1200     out.print_raw_cr(buf);
1201     out.print_raw_cr("Dumping core ...");
1202 #endif
1203     ::abort(); // dump core
1204   }
1205 
1206   ::exit(1);
1207 }
1208 
1209 // Die immediately, no exit hook, no abort hook, no cleanup.
1210 void os::die() {
1211   ::abort();
1212 }
1213 
1214 // Unused on Aix for now.
1215 void os::set_error_file(const char *logfile) {}
1216 
1217 
1218 // This method is a copy of JDK's sysGetLastErrorString
1219 // from src/solaris/hpi/src/system_md.c
1220 
1221 size_t os::lasterror(char *buf, size_t len) {
1222 
1223   if (errno == 0)  return 0;
1224 
1225   const char *s = ::strerror(errno);
1226   size_t n = ::strlen(s);
1227   if (n >= len) {
1228     n = len - 1;
1229   }
1230   ::strncpy(buf, s, n);
1231   buf[n] = '\0';
1232   return n;
1233 }
1234 
1235 intx os::current_thread_id() { return (intx)pthread_self(); }
1236 int os::current_process_id() {
1237 
1238   // This implementation returns a unique pid, the pid of the
1239   // launcher thread that starts the vm 'process'.
1240 
1241   // Under POSIX, getpid() returns the same pid as the
1242   // launcher thread rather than a unique pid per thread.
1243   // Use gettid() if you want the old pre NPTL behaviour.
1244 
1245   // if you are looking for the result of a call to getpid() that
1246   // returns a unique pid for the calling thread, then look at the
1247   // OSThread::thread_id() method in osThread_linux.hpp file
1248 
1249   return (int)(_initial_pid ? _initial_pid : getpid());
1250 }
1251 
1252 // DLL functions
1253 
1254 const char* os::dll_file_extension() { return ".so"; }
1255 
1256 // This must be hard coded because it's the system's temporary
1257 // directory not the java application's temp directory, ala java.io.tmpdir.
1258 const char* os::get_temp_directory() { return "/tmp"; }
1259 
1260 static bool file_exists(const char* filename) {
1261   struct stat statbuf;
1262   if (filename == NULL || strlen(filename) == 0) {
1263     return false;
1264   }
1265   return os::stat(filename, &statbuf) == 0;
1266 }
1267 
1268 bool os::dll_build_name(char* buffer, size_t buflen,
1269                         const char* pname, const char* fname) {
1270   bool retval = false;
1271   // Copied from libhpi
1272   const size_t pnamelen = pname ? strlen(pname) : 0;
1273 
1274   // Return error on buffer overflow.
1275   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1276     *buffer = '\0';
1277     return retval;
1278   }
1279 
1280   if (pnamelen == 0) {
1281     snprintf(buffer, buflen, "lib%s.so", fname);
1282     retval = true;
1283   } else if (strchr(pname, *os::path_separator()) != NULL) {
1284     int n;
1285     char** pelements = split_path(pname, &n);
1286     for (int i = 0; i < n; i++) {
1287       // Really shouldn't be NULL, but check can't hurt
1288       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1289         continue; // skip the empty path values
1290       }
1291       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1292       if (file_exists(buffer)) {
1293         retval = true;
1294         break;
1295       }
1296     }
1297     // release the storage
1298     for (int i = 0; i < n; i++) {
1299       if (pelements[i] != NULL) {
1300         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1301       }
1302     }
1303     if (pelements != NULL) {
1304       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1305     }
1306   } else {
1307     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1308     retval = true;
1309   }
1310   return retval;
1311 }
1312 
1313 // Check if addr is inside libjvm.so.
1314 bool os::address_is_in_vm(address addr) {
1315 
1316   // Input could be a real pc or a function pointer literal. The latter
1317   // would be a function descriptor residing in the data segment of a module.
1318 
1319   const LoadedLibraryModule* lib = LoadedLibraries::find_for_text_address(addr);
1320   if (lib) {
1321     if (strcmp(lib->get_shortname(), "libjvm.so") == 0) {
1322       return true;
1323     } else {
1324       return false;
1325     }
1326   } else {
1327     lib = LoadedLibraries::find_for_data_address(addr);
1328     if (lib) {
1329       if (strcmp(lib->get_shortname(), "libjvm.so") == 0) {
1330         return true;
1331       } else {
1332         return false;
1333       }
1334     } else {
1335       return false;
1336     }
1337   }
1338 }
1339 
1340 // Resolve an AIX function descriptor literal to a code pointer.
1341 // If the input is a valid code pointer to a text segment of a loaded module,
1342 //   it is returned unchanged.
1343 // If the input is a valid AIX function descriptor, it is resolved to the
1344 //   code entry point.
1345 // If the input is neither a valid function descriptor nor a valid code pointer,
1346 //   NULL is returned.
1347 static address resolve_function_descriptor_to_code_pointer(address p) {
1348 
1349   const LoadedLibraryModule* lib = LoadedLibraries::find_for_text_address(p);
1350   if (lib) {
1351     // its a real code pointer
1352     return p;
1353   } else {
1354     lib = LoadedLibraries::find_for_data_address(p);
1355     if (lib) {
1356       // pointer to data segment, potential function descriptor
1357       address code_entry = (address)(((FunctionDescriptor*)p)->entry());
1358       if (LoadedLibraries::find_for_text_address(code_entry)) {
1359         // Its a function descriptor
1360         return code_entry;
1361       }
1362     }
1363   }
1364   return NULL;
1365 }
1366 
1367 bool os::dll_address_to_function_name(address addr, char *buf,
1368                                       int buflen, int *offset) {
1369   if (offset) {
1370     *offset = -1;
1371   }
1372   if (buf) {
1373     buf[0] = '\0';
1374   }
1375 
1376   // Resolve function ptr literals first.
1377   addr = resolve_function_descriptor_to_code_pointer(addr);
1378   if (!addr) {
1379     return false;
1380   }
1381 
1382   // Go through Decoder::decode to call getFuncName which reads the name from the traceback table.
1383   return Decoder::decode(addr, buf, buflen, offset);
1384 }
1385 
1386 static int getModuleName(codeptr_t pc,                    // [in] program counter
1387                          char* p_name, size_t namelen,    // [out] optional: function name
1388                          char* p_errmsg, size_t errmsglen // [out] optional: user provided buffer for error messages
1389                          ) {
1390 
1391   // initialize output parameters
1392   if (p_name && namelen > 0) {
1393     *p_name = '\0';
1394   }
1395   if (p_errmsg && errmsglen > 0) {
1396     *p_errmsg = '\0';
1397   }
1398 
1399   const LoadedLibraryModule* const lib = LoadedLibraries::find_for_text_address((address)pc);
1400   if (lib) {
1401     if (p_name && namelen > 0) {
1402       sprintf(p_name, "%.*s", namelen, lib->get_shortname());
1403     }
1404     return 0;
1405   }
1406 
1407   if (Verbose) {
1408     fprintf(stderr, "pc outside any module");
1409   }
1410 
1411   return -1;
1412 
1413 }
1414 
1415 bool os::dll_address_to_library_name(address addr, char* buf,
1416                                      int buflen, int* offset) {
1417   if (offset) {
1418     *offset = -1;
1419   }
1420   if (buf) {
1421       buf[0] = '\0';
1422   }
1423 
1424   // Resolve function ptr literals first.
1425   addr = resolve_function_descriptor_to_code_pointer(addr);
1426   if (!addr) {
1427     return false;
1428   }
1429 
1430   if (::getModuleName((codeptr_t) addr, buf, buflen, 0, 0) == 0) {
1431     return true;
1432   }
1433   return false;
1434 }
1435 
1436 // Loads .dll/.so and in case of error it checks if .dll/.so was built
1437 // for the same architecture as Hotspot is running on
1438 void *os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1439 
1440   if (ebuf && ebuflen > 0) {
1441     ebuf[0] = '\0';
1442     ebuf[ebuflen - 1] = '\0';
1443   }
1444 
1445   if (!filename || strlen(filename) == 0) {
1446     ::strncpy(ebuf, "dll_load: empty filename specified", ebuflen - 1);
1447     return NULL;
1448   }
1449 
1450   // RTLD_LAZY is currently not implemented. The dl is loaded immediately with all its dependants.
1451   void * result= ::dlopen(filename, RTLD_LAZY);
1452   if (result != NULL) {
1453     // Reload dll cache. Don't do this in signal handling.
1454     LoadedLibraries::reload();
1455     return result;
1456   } else {
1457     // error analysis when dlopen fails
1458     const char* const error_report = ::dlerror();
1459     if (error_report && ebuf && ebuflen > 0) {
1460       snprintf(ebuf, ebuflen - 1, "%s, LIBPATH=%s, LD_LIBRARY_PATH=%s : %s",
1461                filename, ::getenv("LIBPATH"), ::getenv("LD_LIBRARY_PATH"), error_report);
1462     }
1463   }
1464   return NULL;
1465 }
1466 
1467 // Glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1468 // chances are you might want to run the generated bits against glibc-2.0
1469 // libdl.so, so always use locking for any version of glibc.
1470 void* os::dll_lookup(void* handle, const char* name) {
1471   pthread_mutex_lock(&dl_mutex);
1472   void* res = dlsym(handle, name);
1473   pthread_mutex_unlock(&dl_mutex);
1474   return res;
1475 }
1476 
1477 void* os::get_default_process_handle() {
1478   return (void*)::dlopen(NULL, RTLD_LAZY);
1479 }
1480 
1481 void os::print_dll_info(outputStream *st) {
1482   st->print_cr("Dynamic libraries:");
1483   LoadedLibraries::print(st);
1484 }
1485 
1486 void os::print_os_info(outputStream* st) {
1487   st->print("OS:");
1488 
1489   st->print("uname:");
1490   struct utsname name;
1491   uname(&name);
1492   st->print(name.sysname); st->print(" ");
1493   st->print(name.nodename); st->print(" ");
1494   st->print(name.release); st->print(" ");
1495   st->print(name.version); st->print(" ");
1496   st->print(name.machine);
1497   st->cr();
1498 
1499   // rlimit
1500   st->print("rlimit:");
1501   struct rlimit rlim;
1502 
1503   st->print(" STACK ");
1504   getrlimit(RLIMIT_STACK, &rlim);
1505   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1506   else st->print("%uk", rlim.rlim_cur >> 10);
1507 
1508   st->print(", CORE ");
1509   getrlimit(RLIMIT_CORE, &rlim);
1510   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1511   else st->print("%uk", rlim.rlim_cur >> 10);
1512 
1513   st->print(", NPROC ");
1514   st->print("%d", sysconf(_SC_CHILD_MAX));
1515 
1516   st->print(", NOFILE ");
1517   getrlimit(RLIMIT_NOFILE, &rlim);
1518   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1519   else st->print("%d", rlim.rlim_cur);
1520 
1521   st->print(", AS ");
1522   getrlimit(RLIMIT_AS, &rlim);
1523   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1524   else st->print("%uk", rlim.rlim_cur >> 10);
1525 
1526   // Print limits on DATA, because it limits the C-heap.
1527   st->print(", DATA ");
1528   getrlimit(RLIMIT_DATA, &rlim);
1529   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1530   else st->print("%uk", rlim.rlim_cur >> 10);
1531   st->cr();
1532 
1533   // load average
1534   st->print("load average:");
1535   double loadavg[3] = {-1.L, -1.L, -1.L};
1536   os::loadavg(loadavg, 3);
1537   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
1538   st->cr();
1539 }
1540 
1541 void os::print_memory_info(outputStream* st) {
1542 
1543   st->print_cr("Memory:");
1544 
1545   st->print_cr("  default page size: %s", describe_pagesize(os::vm_page_size()));
1546   st->print_cr("  default stack page size: %s", describe_pagesize(os::vm_page_size()));
1547   st->print_cr("  default shm page size: %s", describe_pagesize(os::Aix::shm_default_page_size()));
1548   st->print_cr("  can use 64K pages dynamically: %s", (os::Aix::can_use_64K_pages() ? "yes" :"no"));
1549   st->print_cr("  can use 16M pages dynamically: %s", (os::Aix::can_use_16M_pages() ? "yes" :"no"));
1550   if (g_multipage_error != 0) {
1551     st->print_cr("  multipage error: %d", g_multipage_error);
1552   }
1553 
1554   // print out LDR_CNTRL because it affects the default page sizes
1555   const char* const ldr_cntrl = ::getenv("LDR_CNTRL");
1556   st->print_cr("  LDR_CNTRL=%s.", ldr_cntrl ? ldr_cntrl : "<unset>");
1557 
1558   const char* const extshm = ::getenv("EXTSHM");
1559   st->print_cr("  EXTSHM=%s.", extshm ? extshm : "<unset>");
1560 
1561   // Call os::Aix::get_meminfo() to retrieve memory statistics.
1562   os::Aix::meminfo_t mi;
1563   if (os::Aix::get_meminfo(&mi)) {
1564     char buffer[256];
1565     if (os::Aix::on_aix()) {
1566       jio_snprintf(buffer, sizeof(buffer),
1567                    "  physical total : %llu\n"
1568                    "  physical free  : %llu\n"
1569                    "  swap total     : %llu\n"
1570                    "  swap free      : %llu\n",
1571                    mi.real_total,
1572                    mi.real_free,
1573                    mi.pgsp_total,
1574                    mi.pgsp_free);
1575     } else {
1576       Unimplemented();
1577     }
1578     st->print_raw(buffer);
1579   } else {
1580     st->print_cr("  (no more information available)");
1581   }
1582 }
1583 
1584 void os::pd_print_cpu_info(outputStream* st) {
1585   // cpu
1586   st->print("CPU:");
1587   st->print("total %d", os::processor_count());
1588   // It's not safe to query number of active processors after crash
1589   // st->print("(active %d)", os::active_processor_count());
1590   st->print(" %s", VM_Version::cpu_features());
1591   st->cr();
1592 }
1593 
1594 void os::print_siginfo(outputStream* st, void* siginfo) {
1595   // Use common posix version.
1596   os::Posix::print_siginfo_brief(st, (const siginfo_t*) siginfo);
1597   st->cr();
1598 }
1599 
1600 
1601 static void print_signal_handler(outputStream* st, int sig,
1602                                  char* buf, size_t buflen);
1603 
1604 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
1605   st->print_cr("Signal Handlers:");
1606   print_signal_handler(st, SIGSEGV, buf, buflen);
1607   print_signal_handler(st, SIGBUS , buf, buflen);
1608   print_signal_handler(st, SIGFPE , buf, buflen);
1609   print_signal_handler(st, SIGPIPE, buf, buflen);
1610   print_signal_handler(st, SIGXFSZ, buf, buflen);
1611   print_signal_handler(st, SIGILL , buf, buflen);
1612   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
1613   print_signal_handler(st, SR_signum, buf, buflen);
1614   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
1615   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
1616   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
1617   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
1618   print_signal_handler(st, SIGTRAP, buf, buflen);
1619   print_signal_handler(st, SIGDANGER, buf, buflen);
1620 }
1621 
1622 static char saved_jvm_path[MAXPATHLEN] = {0};
1623 
1624 // Find the full path to the current module, libjvm.so or libjvm_g.so
1625 void os::jvm_path(char *buf, jint buflen) {
1626   // Error checking.
1627   if (buflen < MAXPATHLEN) {
1628     assert(false, "must use a large-enough buffer");
1629     buf[0] = '\0';
1630     return;
1631   }
1632   // Lazy resolve the path to current module.
1633   if (saved_jvm_path[0] != 0) {
1634     strcpy(buf, saved_jvm_path);
1635     return;
1636   }
1637 
1638   Dl_info dlinfo;
1639   int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
1640   assert(ret != 0, "cannot locate libjvm");
1641   char* rp = realpath((char *)dlinfo.dli_fname, buf);
1642   assert(rp != NULL, "error in realpath(): maybe the 'path' argument is too long?");
1643 
1644   strcpy(saved_jvm_path, buf);
1645 }
1646 
1647 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
1648   // no prefix required, not even "_"
1649 }
1650 
1651 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
1652   // no suffix required
1653 }
1654 
1655 ////////////////////////////////////////////////////////////////////////////////
1656 // sun.misc.Signal support
1657 
1658 static volatile jint sigint_count = 0;
1659 
1660 static void
1661 UserHandler(int sig, void *siginfo, void *context) {
1662   // 4511530 - sem_post is serialized and handled by the manager thread. When
1663   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
1664   // don't want to flood the manager thread with sem_post requests.
1665   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
1666     return;
1667 
1668   // Ctrl-C is pressed during error reporting, likely because the error
1669   // handler fails to abort. Let VM die immediately.
1670   if (sig == SIGINT && is_error_reported()) {
1671     os::die();
1672   }
1673 
1674   os::signal_notify(sig);
1675 }
1676 
1677 void* os::user_handler() {
1678   return CAST_FROM_FN_PTR(void*, UserHandler);
1679 }
1680 
1681 extern "C" {
1682   typedef void (*sa_handler_t)(int);
1683   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
1684 }
1685 
1686 void* os::signal(int signal_number, void* handler) {
1687   struct sigaction sigAct, oldSigAct;
1688 
1689   sigfillset(&(sigAct.sa_mask));
1690 
1691   // Do not block out synchronous signals in the signal handler.
1692   // Blocking synchronous signals only makes sense if you can really
1693   // be sure that those signals won't happen during signal handling,
1694   // when the blocking applies.  Normal signal handlers are lean and
1695   // do not cause signals. But our signal handlers tend to be "risky"
1696   // - secondary SIGSEGV, SIGILL, SIGBUS' may and do happen.
1697   // On AIX, PASE there was a case where a SIGSEGV happened, followed
1698   // by a SIGILL, which was blocked due to the signal mask. The process
1699   // just hung forever. Better to crash from a secondary signal than to hang.
1700   sigdelset(&(sigAct.sa_mask), SIGSEGV);
1701   sigdelset(&(sigAct.sa_mask), SIGBUS);
1702   sigdelset(&(sigAct.sa_mask), SIGILL);
1703   sigdelset(&(sigAct.sa_mask), SIGFPE);
1704   sigdelset(&(sigAct.sa_mask), SIGTRAP);
1705 
1706   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
1707 
1708   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
1709 
1710   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
1711     // -1 means registration failed
1712     return (void *)-1;
1713   }
1714 
1715   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
1716 }
1717 
1718 void os::signal_raise(int signal_number) {
1719   ::raise(signal_number);
1720 }
1721 
1722 //
1723 // The following code is moved from os.cpp for making this
1724 // code platform specific, which it is by its very nature.
1725 //
1726 
1727 // Will be modified when max signal is changed to be dynamic
1728 int os::sigexitnum_pd() {
1729   return NSIG;
1730 }
1731 
1732 // a counter for each possible signal value
1733 static volatile jint pending_signals[NSIG+1] = { 0 };
1734 
1735 // Linux(POSIX) specific hand shaking semaphore.
1736 static sem_t sig_sem;
1737 
1738 void os::signal_init_pd() {
1739   // Initialize signal structures
1740   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
1741 
1742   // Initialize signal semaphore
1743   int rc = ::sem_init(&sig_sem, 0, 0);
1744   guarantee(rc != -1, "sem_init failed");
1745 }
1746 
1747 void os::signal_notify(int sig) {
1748   Atomic::inc(&pending_signals[sig]);
1749   ::sem_post(&sig_sem);
1750 }
1751 
1752 static int check_pending_signals(bool wait) {
1753   Atomic::store(0, &sigint_count);
1754   for (;;) {
1755     for (int i = 0; i < NSIG + 1; i++) {
1756       jint n = pending_signals[i];
1757       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
1758         return i;
1759       }
1760     }
1761     if (!wait) {
1762       return -1;
1763     }
1764     JavaThread *thread = JavaThread::current();
1765     ThreadBlockInVM tbivm(thread);
1766 
1767     bool threadIsSuspended;
1768     do {
1769       thread->set_suspend_equivalent();
1770       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
1771 
1772       ::sem_wait(&sig_sem);
1773 
1774       // were we externally suspended while we were waiting?
1775       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
1776       if (threadIsSuspended) {
1777         //
1778         // The semaphore has been incremented, but while we were waiting
1779         // another thread suspended us. We don't want to continue running
1780         // while suspended because that would surprise the thread that
1781         // suspended us.
1782         //
1783         ::sem_post(&sig_sem);
1784 
1785         thread->java_suspend_self();
1786       }
1787     } while (threadIsSuspended);
1788   }
1789 }
1790 
1791 int os::signal_lookup() {
1792   return check_pending_signals(false);
1793 }
1794 
1795 int os::signal_wait() {
1796   return check_pending_signals(true);
1797 }
1798 
1799 ////////////////////////////////////////////////////////////////////////////////
1800 // Virtual Memory
1801 
1802 // AddrRange describes an immutable address range
1803 //
1804 // This is a helper class for the 'shared memory bookkeeping' below.
1805 class AddrRange {
1806   friend class ShmBkBlock;
1807 
1808   char* _start;
1809   size_t _size;
1810 
1811 public:
1812 
1813   AddrRange(char* start, size_t size)
1814     : _start(start), _size(size)
1815   {}
1816 
1817   AddrRange(const AddrRange& r)
1818     : _start(r.start()), _size(r.size())
1819   {}
1820 
1821   char* start() const { return _start; }
1822   size_t size() const { return _size; }
1823   char* end() const { return _start + _size; }
1824   bool is_empty() const { return _size == 0 ? true : false; }
1825 
1826   static AddrRange empty_range() { return AddrRange(NULL, 0); }
1827 
1828   bool contains(const char* p) const {
1829     return start() <= p && end() > p;
1830   }
1831 
1832   bool contains(const AddrRange& range) const {
1833     return start() <= range.start() && end() >= range.end();
1834   }
1835 
1836   bool intersects(const AddrRange& range) const {
1837     return (range.start() <= start() && range.end() > start()) ||
1838            (range.start() < end() && range.end() >= end()) ||
1839            contains(range);
1840   }
1841 
1842   bool is_same_range(const AddrRange& range) const {
1843     return start() == range.start() && size() == range.size();
1844   }
1845 
1846   // return the closest inside range consisting of whole pages
1847   AddrRange find_closest_aligned_range(size_t pagesize) const {
1848     if (pagesize == 0 || is_empty()) {
1849       return empty_range();
1850     }
1851     char* const from = (char*)align_size_up((intptr_t)_start, pagesize);
1852     char* const to = (char*)align_size_down((intptr_t)end(), pagesize);
1853     if (from > to) {
1854       return empty_range();
1855     }
1856     return AddrRange(from, to - from);
1857   }
1858 };
1859 
1860 ////////////////////////////////////////////////////////////////////////////
1861 // shared memory bookkeeping
1862 //
1863 // the os::reserve_memory() API and friends hand out different kind of memory, depending
1864 // on need and circumstances. Memory may be allocated with mmap() or with shmget/shmat.
1865 //
1866 // But these memory types have to be treated differently. For example, to uncommit
1867 // mmap-based memory, msync(MS_INVALIDATE) is needed, to uncommit shmat-based memory,
1868 // disclaim64() is needed.
1869 //
1870 // Therefore we need to keep track of the allocated memory segments and their
1871 // properties.
1872 
1873 // ShmBkBlock: base class for all blocks in the shared memory bookkeeping
1874 class ShmBkBlock : public CHeapObj<mtInternal> {
1875 
1876   ShmBkBlock* _next;
1877 
1878 protected:
1879 
1880   AddrRange _range;
1881   const size_t _pagesize;
1882   const bool _pinned;
1883 
1884 public:
1885 
1886   ShmBkBlock(AddrRange range, size_t pagesize, bool pinned)
1887     : _range(range), _pagesize(pagesize), _pinned(pinned) , _next(NULL) {
1888 
1889     assert(_pagesize == SIZE_4K || _pagesize == SIZE_64K || _pagesize == SIZE_16M, "invalid page size");
1890     assert(!_range.is_empty(), "invalid range");
1891   }
1892 
1893   virtual void print(outputStream* st) const {
1894     st->print("0x%p ... 0x%p (%llu) - %d %s pages - %s",
1895               _range.start(), _range.end(), _range.size(),
1896               _range.size() / _pagesize, describe_pagesize(_pagesize),
1897               _pinned ? "pinned" : "");
1898   }
1899 
1900   enum Type { MMAP, SHMAT };
1901   virtual Type getType() = 0;
1902 
1903   char* base() const { return _range.start(); }
1904   size_t size() const { return _range.size(); }
1905 
1906   void setAddrRange(AddrRange range) {
1907     _range = range;
1908   }
1909 
1910   bool containsAddress(const char* p) const {
1911     return _range.contains(p);
1912   }
1913 
1914   bool containsRange(const char* p, size_t size) const {
1915     return _range.contains(AddrRange((char*)p, size));
1916   }
1917 
1918   bool isSameRange(const char* p, size_t size) const {
1919     return _range.is_same_range(AddrRange((char*)p, size));
1920   }
1921 
1922   virtual bool disclaim(char* p, size_t size) = 0;
1923   virtual bool release() = 0;
1924 
1925   // blocks live in a list.
1926   ShmBkBlock* next() const { return _next; }
1927   void set_next(ShmBkBlock* blk) { _next = blk; }
1928 
1929 }; // end: ShmBkBlock
1930 
1931 
1932 // ShmBkMappedBlock: describes an block allocated with mmap()
1933 class ShmBkMappedBlock : public ShmBkBlock {
1934 public:
1935 
1936   ShmBkMappedBlock(AddrRange range)
1937     : ShmBkBlock(range, SIZE_4K, false) {} // mmap: always 4K, never pinned
1938 
1939   void print(outputStream* st) const {
1940     ShmBkBlock::print(st);
1941     st->print_cr(" - mmap'ed");
1942   }
1943 
1944   Type getType() {
1945     return MMAP;
1946   }
1947 
1948   bool disclaim(char* p, size_t size) {
1949 
1950     AddrRange r(p, size);
1951 
1952     guarantee(_range.contains(r), "invalid disclaim");
1953 
1954     // only disclaim whole ranges.
1955     const AddrRange r2 = r.find_closest_aligned_range(_pagesize);
1956     if (r2.is_empty()) {
1957       return true;
1958     }
1959 
1960     const int rc = ::msync(r2.start(), r2.size(), MS_INVALIDATE);
1961 
1962     if (rc != 0) {
1963       warning("msync(0x%p, %llu, MS_INVALIDATE) failed (%d)\n", r2.start(), r2.size(), errno);
1964     }
1965 
1966     return rc == 0 ? true : false;
1967   }
1968 
1969   bool release() {
1970     // mmap'ed blocks are released using munmap
1971     if (::munmap(_range.start(), _range.size()) != 0) {
1972       warning("munmap(0x%p, %llu) failed (%d)\n", _range.start(), _range.size(), errno);
1973       return false;
1974     }
1975     return true;
1976   }
1977 }; // end: ShmBkMappedBlock
1978 
1979 // ShmBkShmatedBlock: describes an block allocated with shmget/shmat()
1980 class ShmBkShmatedBlock : public ShmBkBlock {
1981 public:
1982 
1983   ShmBkShmatedBlock(AddrRange range, size_t pagesize, bool pinned)
1984     : ShmBkBlock(range, pagesize, pinned) {}
1985 
1986   void print(outputStream* st) const {
1987     ShmBkBlock::print(st);
1988     st->print_cr(" - shmat'ed");
1989   }
1990 
1991   Type getType() {
1992     return SHMAT;
1993   }
1994 
1995   bool disclaim(char* p, size_t size) {
1996 
1997     AddrRange r(p, size);
1998 
1999     if (_pinned) {
2000       return true;
2001     }
2002 
2003     // shmat'ed blocks are disclaimed using disclaim64
2004     guarantee(_range.contains(r), "invalid disclaim");
2005 
2006     // only disclaim whole ranges.
2007     const AddrRange r2 = r.find_closest_aligned_range(_pagesize);
2008     if (r2.is_empty()) {
2009       return true;
2010     }
2011 
2012     const bool rc = my_disclaim64(r2.start(), r2.size());
2013 
2014     if (Verbose && !rc) {
2015       warning("failed to disclaim shm %p-%p\n", r2.start(), r2.end());
2016     }
2017 
2018     return rc;
2019   }
2020 
2021   bool release() {
2022     bool rc = false;
2023     if (::shmdt(_range.start()) != 0) {
2024       warning("shmdt(0x%p) failed (%d)\n", _range.start(), errno);
2025     } else {
2026       rc = true;
2027     }
2028     return rc;
2029   }
2030 
2031 }; // end: ShmBkShmatedBlock
2032 
2033 static ShmBkBlock* g_shmbk_list = NULL;
2034 static volatile jint g_shmbk_table_lock = 0;
2035 
2036 // keep some usage statistics
2037 static struct {
2038   int nodes;    // number of nodes in list
2039   size_t bytes; // reserved - not committed - bytes.
2040   int reserves; // how often reserve was called
2041   int lookups;  // how often a lookup was made
2042 } g_shmbk_stats = { 0, 0, 0, 0 };
2043 
2044 // add information about a shared memory segment to the bookkeeping
2045 static void shmbk_register(ShmBkBlock* p_block) {
2046   guarantee(p_block, "logic error");
2047   p_block->set_next(g_shmbk_list);
2048   g_shmbk_list = p_block;
2049   g_shmbk_stats.reserves ++;
2050   g_shmbk_stats.bytes += p_block->size();
2051   g_shmbk_stats.nodes ++;
2052 }
2053 
2054 // remove information about a shared memory segment by its starting address
2055 static void shmbk_unregister(ShmBkBlock* p_block) {
2056   ShmBkBlock* p = g_shmbk_list;
2057   ShmBkBlock* prev = NULL;
2058   while (p) {
2059     if (p == p_block) {
2060       if (prev) {
2061         prev->set_next(p->next());
2062       } else {
2063         g_shmbk_list = p->next();
2064       }
2065       g_shmbk_stats.nodes --;
2066       g_shmbk_stats.bytes -= p->size();
2067       return;
2068     }
2069     prev = p;
2070     p = p->next();
2071   }
2072   assert(false, "should not happen");
2073 }
2074 
2075 // given a pointer, return shared memory bookkeeping record for the segment it points into
2076 // using the returned block info must happen under lock protection
2077 static ShmBkBlock* shmbk_find_by_containing_address(const char* addr) {
2078   g_shmbk_stats.lookups ++;
2079   ShmBkBlock* p = g_shmbk_list;
2080   while (p) {
2081     if (p->containsAddress(addr)) {
2082       return p;
2083     }
2084     p = p->next();
2085   }
2086   return NULL;
2087 }
2088 
2089 // dump all information about all memory segments allocated with os::reserve_memory()
2090 void shmbk_dump_info() {
2091   tty->print_cr("-- shared mem bookkeeping (alive: %d segments, %llu bytes, "
2092     "total reserves: %d total lookups: %d)",
2093     g_shmbk_stats.nodes, g_shmbk_stats.bytes, g_shmbk_stats.reserves, g_shmbk_stats.lookups);
2094   const ShmBkBlock* p = g_shmbk_list;
2095   int i = 0;
2096   while (p) {
2097     p->print(tty);
2098     p = p->next();
2099     i ++;
2100   }
2101 }
2102 
2103 #define LOCK_SHMBK     { ThreadCritical _LOCK_SHMBK;
2104 #define UNLOCK_SHMBK   }
2105 
2106 // End: shared memory bookkeeping
2107 ////////////////////////////////////////////////////////////////////////////////////////////////////
2108 
2109 int os::vm_page_size() {
2110   // Seems redundant as all get out
2111   assert(os::Aix::page_size() != -1, "must call os::init");
2112   return os::Aix::page_size();
2113 }
2114 
2115 // Aix allocates memory by pages.
2116 int os::vm_allocation_granularity() {
2117   assert(os::Aix::page_size() != -1, "must call os::init");
2118   return os::Aix::page_size();
2119 }
2120 
2121 int os::Aix::commit_memory_impl(char* addr, size_t size, bool exec) {
2122 
2123   // Commit is a noop. There is no explicit commit
2124   // needed on AIX. Memory is committed when touched.
2125   //
2126   // Debug : check address range for validity
2127 #ifdef ASSERT
2128   LOCK_SHMBK
2129     ShmBkBlock* const block = shmbk_find_by_containing_address(addr);
2130     if (!block) {
2131       fprintf(stderr, "invalid pointer: " INTPTR_FORMAT "\n", addr);
2132       shmbk_dump_info();
2133       assert(false, "invalid pointer");
2134       return false;
2135     } else if (!block->containsRange(addr, size)) {
2136       fprintf(stderr, "invalid range: " INTPTR_FORMAT " .. " INTPTR_FORMAT "\n", addr, addr + size);
2137       shmbk_dump_info();
2138       assert(false, "invalid range");
2139       return false;
2140     }
2141   UNLOCK_SHMBK
2142 #endif // ASSERT
2143 
2144   return 0;
2145 }
2146 
2147 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2148   return os::Aix::commit_memory_impl(addr, size, exec) == 0;
2149 }
2150 
2151 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2152                                   const char* mesg) {
2153   assert(mesg != NULL, "mesg must be specified");
2154   os::Aix::commit_memory_impl(addr, size, exec);
2155 }
2156 
2157 int os::Aix::commit_memory_impl(char* addr, size_t size,
2158                                 size_t alignment_hint, bool exec) {
2159   return os::Aix::commit_memory_impl(addr, size, exec);
2160 }
2161 
2162 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2163                           bool exec) {
2164   return os::Aix::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2165 }
2166 
2167 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2168                                   size_t alignment_hint, bool exec,
2169                                   const char* mesg) {
2170   os::Aix::commit_memory_impl(addr, size, alignment_hint, exec);
2171 }
2172 
2173 bool os::pd_uncommit_memory(char* addr, size_t size) {
2174 
2175   // Delegate to ShmBkBlock class which knows how to uncommit its memory.
2176 
2177   bool rc = false;
2178   LOCK_SHMBK
2179     ShmBkBlock* const block = shmbk_find_by_containing_address(addr);
2180     if (!block) {
2181       fprintf(stderr, "invalid pointer: 0x%p.\n", addr);
2182       shmbk_dump_info();
2183       assert(false, "invalid pointer");
2184       return false;
2185     } else if (!block->containsRange(addr, size)) {
2186       fprintf(stderr, "invalid range: 0x%p .. 0x%p.\n", addr, addr + size);
2187       shmbk_dump_info();
2188       assert(false, "invalid range");
2189       return false;
2190     }
2191     rc = block->disclaim(addr, size);
2192   UNLOCK_SHMBK
2193 
2194   if (Verbose && !rc) {
2195     warning("failed to disclaim 0x%p .. 0x%p (0x%llX bytes).", addr, addr + size, size);
2196   }
2197   return rc;
2198 }
2199 
2200 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2201   return os::guard_memory(addr, size);
2202 }
2203 
2204 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2205   return os::unguard_memory(addr, size);
2206 }
2207 
2208 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2209 }
2210 
2211 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2212 }
2213 
2214 void os::numa_make_global(char *addr, size_t bytes) {
2215 }
2216 
2217 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2218 }
2219 
2220 bool os::numa_topology_changed() {
2221   return false;
2222 }
2223 
2224 size_t os::numa_get_groups_num() {
2225   return 1;
2226 }
2227 
2228 int os::numa_get_group_id() {
2229   return 0;
2230 }
2231 
2232 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2233   if (size > 0) {
2234     ids[0] = 0;
2235     return 1;
2236   }
2237   return 0;
2238 }
2239 
2240 bool os::get_page_info(char *start, page_info* info) {
2241   return false;
2242 }
2243 
2244 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2245   return end;
2246 }
2247 
2248 // Flags for reserve_shmatted_memory:
2249 #define RESSHM_WISHADDR_OR_FAIL                     1
2250 #define RESSHM_TRY_16M_PAGES                        2
2251 #define RESSHM_16M_PAGES_OR_FAIL                    4
2252 
2253 // Result of reserve_shmatted_memory:
2254 struct shmatted_memory_info_t {
2255   char* addr;
2256   size_t pagesize;
2257   bool pinned;
2258 };
2259 
2260 // Reserve a section of shmatted memory.
2261 // params:
2262 // bytes [in]: size of memory, in bytes
2263 // requested_addr [in]: wish address.
2264 //                      NULL = no wish.
2265 //                      If RESSHM_WISHADDR_OR_FAIL is set in flags and wish address cannot
2266 //                      be obtained, function will fail. Otherwise wish address is treated as hint and
2267 //                      another pointer is returned.
2268 // flags [in]:          some flags. Valid flags are:
2269 //                      RESSHM_WISHADDR_OR_FAIL - fail if wish address is given and cannot be obtained.
2270 //                      RESSHM_TRY_16M_PAGES - try to allocate from 16M page pool
2271 //                          (requires UseLargePages and Use16MPages)
2272 //                      RESSHM_16M_PAGES_OR_FAIL - if you cannot allocate from 16M page pool, fail.
2273 //                          Otherwise any other page size will do.
2274 // p_info [out] :       holds information about the created shared memory segment.
2275 static bool reserve_shmatted_memory(size_t bytes, char* requested_addr, int flags, shmatted_memory_info_t* p_info) {
2276 
2277   assert(p_info, "parameter error");
2278 
2279   // init output struct.
2280   p_info->addr = NULL;
2281 
2282   // neither should we be here for EXTSHM=ON.
2283   if (os::Aix::extshm()) {
2284     ShouldNotReachHere();
2285   }
2286 
2287   // extract flags. sanity checks.
2288   const bool wishaddr_or_fail =
2289     flags & RESSHM_WISHADDR_OR_FAIL;
2290   const bool try_16M_pages =
2291     flags & RESSHM_TRY_16M_PAGES;
2292   const bool f16M_pages_or_fail =
2293     flags & RESSHM_16M_PAGES_OR_FAIL;
2294 
2295   // first check: if a wish address is given and it is mandatory, but not aligned to segment boundary,
2296   // shmat will fail anyway, so save some cycles by failing right away
2297   if (requested_addr && ((uintptr_t)requested_addr % SIZE_256M == 0)) {
2298     if (wishaddr_or_fail) {
2299       return false;
2300     } else {
2301       requested_addr = NULL;
2302     }
2303   }
2304 
2305   char* addr = NULL;
2306 
2307   // Align size of shm up to the largest possible page size, to avoid errors later on when we try to change
2308   // pagesize dynamically.
2309   const size_t size = align_size_up(bytes, SIZE_16M);
2310 
2311   // reserve the shared segment
2312   int shmid = shmget(IPC_PRIVATE, size, IPC_CREAT | S_IRUSR | S_IWUSR);
2313   if (shmid == -1) {
2314     warning("shmget(.., %lld, ..) failed (errno: %d).", size, errno);
2315     return false;
2316   }
2317 
2318   // Important note:
2319   // It is very important that we, upon leaving this function, do not leave a shm segment alive.
2320   // We must right after attaching it remove it from the system. System V shm segments are global and
2321   // survive the process.
2322   // So, from here on: Do not assert. Do not return. Always do a "goto cleanup_shm".
2323 
2324   // try forcing the page size
2325   size_t pagesize = -1; // unknown so far
2326 
2327   if (UseLargePages) {
2328 
2329     struct shmid_ds shmbuf;
2330     memset(&shmbuf, 0, sizeof(shmbuf));
2331 
2332     // First, try to take from 16M page pool if...
2333     if (os::Aix::can_use_16M_pages()  // we can ...
2334         && Use16MPages                // we are not explicitly forbidden to do so (-XX:-Use16MPages)..
2335         && try_16M_pages) {           // caller wants us to.
2336       shmbuf.shm_pagesize = SIZE_16M;
2337       if (shmctl(shmid, SHM_PAGESIZE, &shmbuf) == 0) {
2338         pagesize = SIZE_16M;
2339       } else {
2340         warning("Failed to allocate %d 16M pages. 16M page pool might be exhausted. (shmctl failed with %d)",
2341                 size / SIZE_16M, errno);
2342         if (f16M_pages_or_fail) {
2343           goto cleanup_shm;
2344         }
2345       }
2346     }
2347 
2348     // Nothing yet? Try setting 64K pages. Note that I never saw this fail, but in theory it might,
2349     // because the 64K page pool may also be exhausted.
2350     if (pagesize == -1) {
2351       shmbuf.shm_pagesize = SIZE_64K;
2352       if (shmctl(shmid, SHM_PAGESIZE, &shmbuf) == 0) {
2353         pagesize = SIZE_64K;
2354       } else {
2355         warning("Failed to allocate %d 64K pages. (shmctl failed with %d)",
2356                 size / SIZE_64K, errno);
2357         // here I give up. leave page_size -1 - later, after attaching, we will query the
2358         // real page size of the attached memory. (in theory, it may be something different
2359         // from 4K if LDR_CNTRL SHM_PSIZE is set)
2360       }
2361     }
2362   }
2363 
2364   // sanity point
2365   assert(pagesize == -1 || pagesize == SIZE_16M || pagesize == SIZE_64K, "wrong page size");
2366 
2367   // Now attach the shared segment.
2368   addr = (char*) shmat(shmid, requested_addr, 0);
2369   if (addr == (char*)-1) {
2370     // How to handle attach failure:
2371     // If it failed for a specific wish address, tolerate this: in that case, if wish address was
2372     // mandatory, fail, if not, retry anywhere.
2373     // If it failed for any other reason, treat that as fatal error.
2374     addr = NULL;
2375     if (requested_addr) {
2376       if (wishaddr_or_fail) {
2377         goto cleanup_shm;
2378       } else {
2379         addr = (char*) shmat(shmid, NULL, 0);
2380         if (addr == (char*)-1) { // fatal
2381           addr = NULL;
2382           warning("shmat failed (errno: %d)", errno);
2383           goto cleanup_shm;
2384         }
2385       }
2386     } else { // fatal
2387       addr = NULL;
2388       warning("shmat failed (errno: %d)", errno);
2389       goto cleanup_shm;
2390     }
2391   }
2392 
2393   // sanity point
2394   assert(addr && addr != (char*) -1, "wrong address");
2395 
2396   // after successful Attach remove the segment - right away.
2397   if (::shmctl(shmid, IPC_RMID, NULL) == -1) {
2398     warning("shmctl(%u, IPC_RMID) failed (%d)\n", shmid, errno);
2399     guarantee(false, "failed to remove shared memory segment!");
2400   }
2401   shmid = -1;
2402 
2403   // query the real page size. In case setting the page size did not work (see above), the system
2404   // may have given us something other then 4K (LDR_CNTRL)
2405   {
2406     const size_t real_pagesize = os::Aix::query_pagesize(addr);
2407     if (pagesize != -1) {
2408       assert(pagesize == real_pagesize, "unexpected pagesize after shmat");
2409     } else {
2410       pagesize = real_pagesize;
2411     }
2412   }
2413 
2414   // Now register the reserved block with internal book keeping.
2415   LOCK_SHMBK
2416     const bool pinned = pagesize >= SIZE_16M ? true : false;
2417     ShmBkShmatedBlock* const p_block = new ShmBkShmatedBlock(AddrRange(addr, size), pagesize, pinned);
2418     assert(p_block, "");
2419     shmbk_register(p_block);
2420   UNLOCK_SHMBK
2421 
2422 cleanup_shm:
2423 
2424   // if we have not done so yet, remove the shared memory segment. This is very important.
2425   if (shmid != -1) {
2426     if (::shmctl(shmid, IPC_RMID, NULL) == -1) {
2427       warning("shmctl(%u, IPC_RMID) failed (%d)\n", shmid, errno);
2428       guarantee(false, "failed to remove shared memory segment!");
2429     }
2430     shmid = -1;
2431   }
2432 
2433   // trace
2434   if (Verbose && !addr) {
2435     if (requested_addr != NULL) {
2436       warning("failed to shm-allocate 0x%llX bytes at wish address 0x%p.", size, requested_addr);
2437     } else {
2438       warning("failed to shm-allocate 0x%llX bytes at any address.", size);
2439     }
2440   }
2441 
2442   // hand info to caller
2443   if (addr) {
2444     p_info->addr = addr;
2445     p_info->pagesize = pagesize;
2446     p_info->pinned = pagesize == SIZE_16M ? true : false;
2447   }
2448 
2449   // sanity test:
2450   if (requested_addr && addr && wishaddr_or_fail) {
2451     guarantee(addr == requested_addr, "shmat error");
2452   }
2453 
2454   // just one more test to really make sure we have no dangling shm segments.
2455   guarantee(shmid == -1, "dangling shm segments");
2456 
2457   return addr ? true : false;
2458 
2459 } // end: reserve_shmatted_memory
2460 
2461 // Reserve memory using mmap. Behaves the same as reserve_shmatted_memory():
2462 // will return NULL in case of an error.
2463 static char* reserve_mmaped_memory(size_t bytes, char* requested_addr) {
2464 
2465   // if a wish address is given, but not aligned to 4K page boundary, mmap will fail.
2466   if (requested_addr && ((uintptr_t)requested_addr % os::vm_page_size() != 0)) {
2467     warning("Wish address 0x%p not aligned to page boundary.", requested_addr);
2468     return NULL;
2469   }
2470 
2471   const size_t size = align_size_up(bytes, SIZE_4K);
2472 
2473   // Note: MAP_SHARED (instead of MAP_PRIVATE) needed to be able to
2474   // msync(MS_INVALIDATE) (see os::uncommit_memory)
2475   int flags = MAP_ANONYMOUS | MAP_SHARED;
2476 
2477   // MAP_FIXED is needed to enforce requested_addr - manpage is vague about what
2478   // it means if wishaddress is given but MAP_FIXED is not set.
2479   //
2480   // Note however that this changes semantics in SPEC1170 mode insofar as MAP_FIXED
2481   // clobbers the address range, which is probably not what the caller wants. That's
2482   // why I assert here (again) that the SPEC1170 compat mode is off.
2483   // If we want to be able to run under SPEC1170, we have to do some porting and
2484   // testing.
2485   if (requested_addr != NULL) {
2486     assert(!os::Aix::xpg_sus_mode(), "SPEC1170 mode not allowed.");
2487     flags |= MAP_FIXED;
2488   }
2489 
2490   char* addr = (char*)::mmap(requested_addr, size, PROT_READ|PROT_WRITE|PROT_EXEC, flags, -1, 0);
2491 
2492   if (addr == MAP_FAILED) {
2493     // attach failed: tolerate for specific wish addresses. Not being able to attach
2494     // anywhere is a fatal error.
2495     if (requested_addr == NULL) {
2496       // It's ok to fail here if the machine has not enough memory.
2497       warning("mmap(NULL, 0x%llX, ..) failed (%d)", size, errno);
2498     }
2499     addr = NULL;
2500     goto cleanup_mmap;
2501   }
2502 
2503   // If we did request a specific address and that address was not available, fail.
2504   if (addr && requested_addr) {
2505     guarantee(addr == requested_addr, "unexpected");
2506   }
2507 
2508   // register this mmap'ed segment with book keeping
2509   LOCK_SHMBK
2510     ShmBkMappedBlock* const p_block = new ShmBkMappedBlock(AddrRange(addr, size));
2511     assert(p_block, "");
2512     shmbk_register(p_block);
2513   UNLOCK_SHMBK
2514 
2515 cleanup_mmap:
2516 
2517   // trace
2518   if (Verbose) {
2519     if (addr) {
2520       fprintf(stderr, "mmap-allocated 0x%p .. 0x%p (0x%llX bytes)\n", addr, addr + bytes, bytes);
2521     }
2522     else {
2523       if (requested_addr != NULL) {
2524         warning("failed to mmap-allocate 0x%llX bytes at wish address 0x%p.", bytes, requested_addr);
2525       } else {
2526         warning("failed to mmap-allocate 0x%llX bytes at any address.", bytes);
2527       }
2528     }
2529   }
2530 
2531   return addr;
2532 
2533 } // end: reserve_mmaped_memory
2534 
2535 // Reserves and attaches a shared memory segment.
2536 // Will assert if a wish address is given and could not be obtained.
2537 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, size_t alignment_hint) {
2538   return os::attempt_reserve_memory_at(bytes, requested_addr);
2539 }
2540 
2541 bool os::pd_release_memory(char* addr, size_t size) {
2542 
2543   // delegate to ShmBkBlock class which knows how to uncommit its memory.
2544 
2545   bool rc = false;
2546   LOCK_SHMBK
2547     ShmBkBlock* const block = shmbk_find_by_containing_address(addr);
2548     if (!block) {
2549       fprintf(stderr, "invalid pointer: 0x%p.\n", addr);
2550       shmbk_dump_info();
2551       assert(false, "invalid pointer");
2552       return false;
2553     }
2554     else if (!block->isSameRange(addr, size)) {
2555       if (block->getType() == ShmBkBlock::MMAP) {
2556         // Release only the same range or a the beginning or the end of a range.
2557         if (block->base() == addr && size < block->size()) {
2558           ShmBkMappedBlock* const b = new ShmBkMappedBlock(AddrRange(block->base() + size, block->size() - size));
2559           assert(b, "");
2560           shmbk_register(b);
2561           block->setAddrRange(AddrRange(addr, size));
2562         }
2563         else if (addr > block->base() && addr + size == block->base() + block->size()) {
2564           ShmBkMappedBlock* const b = new ShmBkMappedBlock(AddrRange(block->base(), block->size() - size));
2565           assert(b, "");
2566           shmbk_register(b);
2567           block->setAddrRange(AddrRange(addr, size));
2568         }
2569         else {
2570           fprintf(stderr, "invalid mmap range: 0x%p .. 0x%p.\n", addr, addr + size);
2571           shmbk_dump_info();
2572           assert(false, "invalid mmap range");
2573           return false;
2574         }
2575       }
2576       else {
2577         // Release only the same range. No partial release allowed.
2578         // Soften the requirement a bit, because the user may think he owns a smaller size
2579         // than the block is due to alignment etc.
2580         if (block->base() != addr || block->size() < size) {
2581           fprintf(stderr, "invalid shmget range: 0x%p .. 0x%p.\n", addr, addr + size);
2582           shmbk_dump_info();
2583           assert(false, "invalid shmget range");
2584           return false;
2585         }
2586       }
2587     }
2588     rc = block->release();
2589     assert(rc, "release failed");
2590     // remove block from bookkeeping
2591     shmbk_unregister(block);
2592     delete block;
2593   UNLOCK_SHMBK
2594 
2595   if (!rc) {
2596     warning("failed to released %lu bytes at 0x%p", size, addr);
2597   }
2598 
2599   return rc;
2600 }
2601 
2602 static bool checked_mprotect(char* addr, size_t size, int prot) {
2603 
2604   // Little problem here: if SPEC1170 behaviour is off, mprotect() on AIX will
2605   // not tell me if protection failed when trying to protect an un-protectable range.
2606   //
2607   // This means if the memory was allocated using shmget/shmat, protection wont work
2608   // but mprotect will still return 0:
2609   //
2610   // See http://publib.boulder.ibm.com/infocenter/pseries/v5r3/index.jsp?topic=/com.ibm.aix.basetechref/doc/basetrf1/mprotect.htm
2611 
2612   bool rc = ::mprotect(addr, size, prot) == 0 ? true : false;
2613 
2614   if (!rc) {
2615     const char* const s_errno = strerror(errno);
2616     warning("mprotect(" PTR_FORMAT "-" PTR_FORMAT ", 0x%X) failed (%s).", addr, addr + size, prot, s_errno);
2617     return false;
2618   }
2619 
2620   // mprotect success check
2621   //
2622   // Mprotect said it changed the protection but can I believe it?
2623   //
2624   // To be sure I need to check the protection afterwards. Try to
2625   // read from protected memory and check whether that causes a segfault.
2626   //
2627   if (!os::Aix::xpg_sus_mode()) {
2628 
2629     if (StubRoutines::SafeFetch32_stub()) {
2630 
2631       const bool read_protected =
2632         (SafeFetch32((int*)addr, 0x12345678) == 0x12345678 &&
2633          SafeFetch32((int*)addr, 0x76543210) == 0x76543210) ? true : false;
2634 
2635       if (prot & PROT_READ) {
2636         rc = !read_protected;
2637       } else {
2638         rc = read_protected;
2639       }
2640     }
2641   }
2642   if (!rc) {
2643     assert(false, "mprotect failed.");
2644   }
2645   return rc;
2646 }
2647 
2648 // Set protections specified
2649 bool os::protect_memory(char* addr, size_t size, ProtType prot, bool is_committed) {
2650   unsigned int p = 0;
2651   switch (prot) {
2652   case MEM_PROT_NONE: p = PROT_NONE; break;
2653   case MEM_PROT_READ: p = PROT_READ; break;
2654   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2655   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2656   default:
2657     ShouldNotReachHere();
2658   }
2659   // is_committed is unused.
2660   return checked_mprotect(addr, size, p);
2661 }
2662 
2663 bool os::guard_memory(char* addr, size_t size) {
2664   return checked_mprotect(addr, size, PROT_NONE);
2665 }
2666 
2667 bool os::unguard_memory(char* addr, size_t size) {
2668   return checked_mprotect(addr, size, PROT_READ|PROT_WRITE|PROT_EXEC);
2669 }
2670 
2671 // Large page support
2672 
2673 static size_t _large_page_size = 0;
2674 
2675 // Enable large page support if OS allows that.
2676 void os::large_page_init() {
2677 
2678   // Note: os::Aix::query_multipage_support must run first.
2679 
2680   if (!UseLargePages) {
2681     return;
2682   }
2683 
2684   if (!Aix::can_use_64K_pages()) {
2685     assert(!Aix::can_use_16M_pages(), "64K is a precondition for 16M.");
2686     UseLargePages = false;
2687     return;
2688   }
2689 
2690   if (!Aix::can_use_16M_pages() && Use16MPages) {
2691     fprintf(stderr, "Cannot use 16M pages. Please ensure that there is a 16M page pool "
2692             " and that the VM runs with CAP_BYPASS_RAC_VMM and CAP_PROPAGATE capabilities.\n");
2693   }
2694 
2695   // Do not report 16M page alignment as part of os::_page_sizes if we are
2696   // explicitly forbidden from using 16M pages. Doing so would increase the
2697   // alignment the garbage collector calculates with, slightly increasing
2698   // heap usage. We should only pay for 16M alignment if we really want to
2699   // use 16M pages.
2700   if (Use16MPages && Aix::can_use_16M_pages()) {
2701     _large_page_size = SIZE_16M;
2702     _page_sizes[0] = SIZE_16M;
2703     _page_sizes[1] = SIZE_64K;
2704     _page_sizes[2] = SIZE_4K;
2705     _page_sizes[3] = 0;
2706   } else if (Aix::can_use_64K_pages()) {
2707     _large_page_size = SIZE_64K;
2708     _page_sizes[0] = SIZE_64K;
2709     _page_sizes[1] = SIZE_4K;
2710     _page_sizes[2] = 0;
2711   }
2712 
2713   if (Verbose) {
2714     ("Default large page size is 0x%llX.", _large_page_size);
2715   }
2716 } // end: os::large_page_init()
2717 
2718 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
2719   // "exec" is passed in but not used. Creating the shared image for
2720   // the code cache doesn't have an SHM_X executable permission to check.
2721   Unimplemented();
2722   return 0;
2723 }
2724 
2725 bool os::release_memory_special(char* base, size_t bytes) {
2726   // detaching the SHM segment will also delete it, see reserve_memory_special()
2727   Unimplemented();
2728   return false;
2729 }
2730 
2731 size_t os::large_page_size() {
2732   return _large_page_size;
2733 }
2734 
2735 bool os::can_commit_large_page_memory() {
2736   // Well, sadly we cannot commit anything at all (see comment in
2737   // os::commit_memory) but we claim to so we can make use of large pages
2738   return true;
2739 }
2740 
2741 bool os::can_execute_large_page_memory() {
2742   // We can do that
2743   return true;
2744 }
2745 
2746 // Reserve memory at an arbitrary address, only if that area is
2747 // available (and not reserved for something else).
2748 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2749 
2750   bool use_mmap = false;
2751 
2752   // mmap: smaller graining, no large page support
2753   // shm: large graining (256M), large page support, limited number of shm segments
2754   //
2755   // Prefer mmap wherever we either do not need large page support or have OS limits
2756 
2757   if (!UseLargePages || bytes < SIZE_16M) {
2758     use_mmap = true;
2759   }
2760 
2761   char* addr = NULL;
2762   if (use_mmap) {
2763     addr = reserve_mmaped_memory(bytes, requested_addr);
2764   } else {
2765     // shmat: wish address is mandatory, and do not try 16M pages here.
2766     shmatted_memory_info_t info;
2767     const int flags = RESSHM_WISHADDR_OR_FAIL;
2768     if (reserve_shmatted_memory(bytes, requested_addr, flags, &info)) {
2769       addr = info.addr;
2770     }
2771   }
2772 
2773   return addr;
2774 }
2775 
2776 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2777   return ::read(fd, buf, nBytes);
2778 }
2779 
2780 void os::naked_short_sleep(jlong ms) {
2781   struct timespec req;
2782 
2783   assert(ms < 1000, "Un-interruptable sleep, short time use only");
2784   req.tv_sec = 0;
2785   if (ms > 0) {
2786     req.tv_nsec = (ms % 1000) * 1000000;
2787   }
2788   else {
2789     req.tv_nsec = 1;
2790   }
2791 
2792   nanosleep(&req, NULL);
2793 
2794   return;
2795 }
2796 
2797 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2798 void os::infinite_sleep() {
2799   while (true) {    // sleep forever ...
2800     ::sleep(100);   // ... 100 seconds at a time
2801   }
2802 }
2803 
2804 // Used to convert frequent JVM_Yield() to nops
2805 bool os::dont_yield() {
2806   return DontYieldALot;
2807 }
2808 
2809 void os::yield() {
2810   sched_yield();
2811 }
2812 
2813 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN; }
2814 
2815 ////////////////////////////////////////////////////////////////////////////////
2816 // thread priority support
2817 
2818 // From AIX manpage to pthread_setschedparam
2819 // (see: http://publib.boulder.ibm.com/infocenter/pseries/v5r3/index.jsp?
2820 //    topic=/com.ibm.aix.basetechref/doc/basetrf1/pthread_setschedparam.htm):
2821 //
2822 // "If schedpolicy is SCHED_OTHER, then sched_priority must be in the
2823 // range from 40 to 80, where 40 is the least favored priority and 80
2824 // is the most favored."
2825 //
2826 // (Actually, I doubt this even has an impact on AIX, as we do kernel
2827 // scheduling there; however, this still leaves iSeries.)
2828 //
2829 // We use the same values for AIX and PASE.
2830 int os::java_to_os_priority[CriticalPriority + 1] = {
2831   54,             // 0 Entry should never be used
2832 
2833   55,             // 1 MinPriority
2834   55,             // 2
2835   56,             // 3
2836 
2837   56,             // 4
2838   57,             // 5 NormPriority
2839   57,             // 6
2840 
2841   58,             // 7
2842   58,             // 8
2843   59,             // 9 NearMaxPriority
2844 
2845   60,             // 10 MaxPriority
2846 
2847   60              // 11 CriticalPriority
2848 };
2849 
2850 OSReturn os::set_native_priority(Thread* thread, int newpri) {
2851   if (!UseThreadPriorities) return OS_OK;
2852   pthread_t thr = thread->osthread()->pthread_id();
2853   int policy = SCHED_OTHER;
2854   struct sched_param param;
2855   param.sched_priority = newpri;
2856   int ret = pthread_setschedparam(thr, policy, &param);
2857 
2858   if (Verbose) {
2859     if (ret == 0) {
2860       fprintf(stderr, "changed priority of thread %d to %d\n", (int)thr, newpri);
2861     } else {
2862       fprintf(stderr, "Could not changed priority for thread %d to %d (error %d, %s)\n",
2863               (int)thr, newpri, ret, strerror(ret));
2864     }
2865   }
2866   return (ret == 0) ? OS_OK : OS_ERR;
2867 }
2868 
2869 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
2870   if (!UseThreadPriorities) {
2871     *priority_ptr = java_to_os_priority[NormPriority];
2872     return OS_OK;
2873   }
2874   pthread_t thr = thread->osthread()->pthread_id();
2875   int policy = SCHED_OTHER;
2876   struct sched_param param;
2877   int ret = pthread_getschedparam(thr, &policy, &param);
2878   *priority_ptr = param.sched_priority;
2879 
2880   return (ret == 0) ? OS_OK : OS_ERR;
2881 }
2882 
2883 // Hint to the underlying OS that a task switch would not be good.
2884 // Void return because it's a hint and can fail.
2885 void os::hint_no_preempt() {}
2886 
2887 ////////////////////////////////////////////////////////////////////////////////
2888 // suspend/resume support
2889 
2890 //  the low-level signal-based suspend/resume support is a remnant from the
2891 //  old VM-suspension that used to be for java-suspension, safepoints etc,
2892 //  within hotspot. Now there is a single use-case for this:
2893 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
2894 //      that runs in the watcher thread.
2895 //  The remaining code is greatly simplified from the more general suspension
2896 //  code that used to be used.
2897 //
2898 //  The protocol is quite simple:
2899 //  - suspend:
2900 //      - sends a signal to the target thread
2901 //      - polls the suspend state of the osthread using a yield loop
2902 //      - target thread signal handler (SR_handler) sets suspend state
2903 //        and blocks in sigsuspend until continued
2904 //  - resume:
2905 //      - sets target osthread state to continue
2906 //      - sends signal to end the sigsuspend loop in the SR_handler
2907 //
2908 //  Note that the SR_lock plays no role in this suspend/resume protocol.
2909 //
2910 
2911 static void resume_clear_context(OSThread *osthread) {
2912   osthread->set_ucontext(NULL);
2913   osthread->set_siginfo(NULL);
2914 }
2915 
2916 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
2917   osthread->set_ucontext(context);
2918   osthread->set_siginfo(siginfo);
2919 }
2920 
2921 //
2922 // Handler function invoked when a thread's execution is suspended or
2923 // resumed. We have to be careful that only async-safe functions are
2924 // called here (Note: most pthread functions are not async safe and
2925 // should be avoided.)
2926 //
2927 // Note: sigwait() is a more natural fit than sigsuspend() from an
2928 // interface point of view, but sigwait() prevents the signal hander
2929 // from being run. libpthread would get very confused by not having
2930 // its signal handlers run and prevents sigwait()'s use with the
2931 // mutex granting granting signal.
2932 //
2933 // Currently only ever called on the VMThread and JavaThreads (PC sampling).
2934 //
2935 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
2936   // Save and restore errno to avoid confusing native code with EINTR
2937   // after sigsuspend.
2938   int old_errno = errno;
2939 
2940   Thread* thread = Thread::current();
2941   OSThread* osthread = thread->osthread();
2942   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
2943 
2944   os::SuspendResume::State current = osthread->sr.state();
2945   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
2946     suspend_save_context(osthread, siginfo, context);
2947 
2948     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
2949     os::SuspendResume::State state = osthread->sr.suspended();
2950     if (state == os::SuspendResume::SR_SUSPENDED) {
2951       sigset_t suspend_set;  // signals for sigsuspend()
2952 
2953       // get current set of blocked signals and unblock resume signal
2954       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
2955       sigdelset(&suspend_set, SR_signum);
2956 
2957       // wait here until we are resumed
2958       while (1) {
2959         sigsuspend(&suspend_set);
2960 
2961         os::SuspendResume::State result = osthread->sr.running();
2962         if (result == os::SuspendResume::SR_RUNNING) {
2963           break;
2964         }
2965       }
2966 
2967     } else if (state == os::SuspendResume::SR_RUNNING) {
2968       // request was cancelled, continue
2969     } else {
2970       ShouldNotReachHere();
2971     }
2972 
2973     resume_clear_context(osthread);
2974   } else if (current == os::SuspendResume::SR_RUNNING) {
2975     // request was cancelled, continue
2976   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
2977     // ignore
2978   } else {
2979     ShouldNotReachHere();
2980   }
2981 
2982   errno = old_errno;
2983 }
2984 
2985 
2986 static int SR_initialize() {
2987   struct sigaction act;
2988   char *s;
2989   // Get signal number to use for suspend/resume
2990   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
2991     int sig = ::strtol(s, 0, 10);
2992     if (sig > 0 || sig < NSIG) {
2993       SR_signum = sig;
2994     }
2995   }
2996 
2997   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
2998         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
2999 
3000   sigemptyset(&SR_sigset);
3001   sigaddset(&SR_sigset, SR_signum);
3002 
3003   // Set up signal handler for suspend/resume.
3004   act.sa_flags = SA_RESTART|SA_SIGINFO;
3005   act.sa_handler = (void (*)(int)) SR_handler;
3006 
3007   // SR_signum is blocked by default.
3008   // 4528190 - We also need to block pthread restart signal (32 on all
3009   // supported Linux platforms). Note that LinuxThreads need to block
3010   // this signal for all threads to work properly. So we don't have
3011   // to use hard-coded signal number when setting up the mask.
3012   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3013 
3014   if (sigaction(SR_signum, &act, 0) == -1) {
3015     return -1;
3016   }
3017 
3018   // Save signal flag
3019   os::Aix::set_our_sigflags(SR_signum, act.sa_flags);
3020   return 0;
3021 }
3022 
3023 static int SR_finalize() {
3024   return 0;
3025 }
3026 
3027 static int sr_notify(OSThread* osthread) {
3028   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3029   assert_status(status == 0, status, "pthread_kill");
3030   return status;
3031 }
3032 
3033 // "Randomly" selected value for how long we want to spin
3034 // before bailing out on suspending a thread, also how often
3035 // we send a signal to a thread we want to resume
3036 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3037 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3038 
3039 // returns true on success and false on error - really an error is fatal
3040 // but this seems the normal response to library errors
3041 static bool do_suspend(OSThread* osthread) {
3042   assert(osthread->sr.is_running(), "thread should be running");
3043   // mark as suspended and send signal
3044 
3045   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3046     // failed to switch, state wasn't running?
3047     ShouldNotReachHere();
3048     return false;
3049   }
3050 
3051   if (sr_notify(osthread) != 0) {
3052     // try to cancel, switch to running
3053 
3054     os::SuspendResume::State result = osthread->sr.cancel_suspend();
3055     if (result == os::SuspendResume::SR_RUNNING) {
3056       // cancelled
3057       return false;
3058     } else if (result == os::SuspendResume::SR_SUSPENDED) {
3059       // somehow managed to suspend
3060       return true;
3061     } else {
3062       ShouldNotReachHere();
3063       return false;
3064     }
3065   }
3066 
3067   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3068 
3069   for (int n = 0; !osthread->sr.is_suspended(); n++) {
3070     for (int i = 0; i < RANDOMLY_LARGE_INTEGER2 && !osthread->sr.is_suspended(); i++) {
3071       os::yield();
3072     }
3073 
3074     // timeout, try to cancel the request
3075     if (n >= RANDOMLY_LARGE_INTEGER) {
3076       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3077       if (cancelled == os::SuspendResume::SR_RUNNING) {
3078         return false;
3079       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3080         return true;
3081       } else {
3082         ShouldNotReachHere();
3083         return false;
3084       }
3085     }
3086   }
3087 
3088   guarantee(osthread->sr.is_suspended(), "Must be suspended");
3089   return true;
3090 }
3091 
3092 static void do_resume(OSThread* osthread) {
3093   //assert(osthread->sr.is_suspended(), "thread should be suspended");
3094 
3095   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3096     // failed to switch to WAKEUP_REQUEST
3097     ShouldNotReachHere();
3098     return;
3099   }
3100 
3101   while (!osthread->sr.is_running()) {
3102     if (sr_notify(osthread) == 0) {
3103       for (int n = 0; n < RANDOMLY_LARGE_INTEGER && !osthread->sr.is_running(); n++) {
3104         for (int i = 0; i < 100 && !osthread->sr.is_running(); i++) {
3105           os::yield();
3106         }
3107       }
3108     } else {
3109       ShouldNotReachHere();
3110     }
3111   }
3112 
3113   guarantee(osthread->sr.is_running(), "Must be running!");
3114 }
3115 
3116 ///////////////////////////////////////////////////////////////////////////////////
3117 // signal handling (except suspend/resume)
3118 
3119 // This routine may be used by user applications as a "hook" to catch signals.
3120 // The user-defined signal handler must pass unrecognized signals to this
3121 // routine, and if it returns true (non-zero), then the signal handler must
3122 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3123 // routine will never retun false (zero), but instead will execute a VM panic
3124 // routine kill the process.
3125 //
3126 // If this routine returns false, it is OK to call it again. This allows
3127 // the user-defined signal handler to perform checks either before or after
3128 // the VM performs its own checks. Naturally, the user code would be making
3129 // a serious error if it tried to handle an exception (such as a null check
3130 // or breakpoint) that the VM was generating for its own correct operation.
3131 //
3132 // This routine may recognize any of the following kinds of signals:
3133 //   SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3134 // It should be consulted by handlers for any of those signals.
3135 //
3136 // The caller of this routine must pass in the three arguments supplied
3137 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3138 // field of the structure passed to sigaction(). This routine assumes that
3139 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3140 //
3141 // Note that the VM will print warnings if it detects conflicting signal
3142 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3143 //
3144 extern "C" JNIEXPORT int
3145 JVM_handle_aix_signal(int signo, siginfo_t* siginfo, void* ucontext, int abort_if_unrecognized);
3146 
3147 // Set thread signal mask (for some reason on AIX sigthreadmask() seems
3148 // to be the thing to call; documentation is not terribly clear about whether
3149 // pthread_sigmask also works, and if it does, whether it does the same.
3150 bool set_thread_signal_mask(int how, const sigset_t* set, sigset_t* oset) {
3151   const int rc = ::pthread_sigmask(how, set, oset);
3152   // return value semantics differ slightly for error case:
3153   // pthread_sigmask returns error number, sigthreadmask -1 and sets global errno
3154   // (so, pthread_sigmask is more theadsafe for error handling)
3155   // But success is always 0.
3156   return rc == 0 ? true : false;
3157 }
3158 
3159 // Function to unblock all signals which are, according
3160 // to POSIX, typical program error signals. If they happen while being blocked,
3161 // they typically will bring down the process immediately.
3162 bool unblock_program_error_signals() {
3163   sigset_t set;
3164   ::sigemptyset(&set);
3165   ::sigaddset(&set, SIGILL);
3166   ::sigaddset(&set, SIGBUS);
3167   ::sigaddset(&set, SIGFPE);
3168   ::sigaddset(&set, SIGSEGV);
3169   return set_thread_signal_mask(SIG_UNBLOCK, &set, NULL);
3170 }
3171 
3172 // Renamed from 'signalHandler' to avoid collision with other shared libs.
3173 void javaSignalHandler(int sig, siginfo_t* info, void* uc) {
3174   assert(info != NULL && uc != NULL, "it must be old kernel");
3175 
3176   // Never leave program error signals blocked;
3177   // on all our platforms they would bring down the process immediately when
3178   // getting raised while being blocked.
3179   unblock_program_error_signals();
3180 
3181   JVM_handle_aix_signal(sig, info, uc, true);
3182 }
3183 
3184 
3185 // This boolean allows users to forward their own non-matching signals
3186 // to JVM_handle_aix_signal, harmlessly.
3187 bool os::Aix::signal_handlers_are_installed = false;
3188 
3189 // For signal-chaining
3190 struct sigaction os::Aix::sigact[MAXSIGNUM];
3191 unsigned int os::Aix::sigs = 0;
3192 bool os::Aix::libjsig_is_loaded = false;
3193 typedef struct sigaction *(*get_signal_t)(int);
3194 get_signal_t os::Aix::get_signal_action = NULL;
3195 
3196 struct sigaction* os::Aix::get_chained_signal_action(int sig) {
3197   struct sigaction *actp = NULL;
3198 
3199   if (libjsig_is_loaded) {
3200     // Retrieve the old signal handler from libjsig
3201     actp = (*get_signal_action)(sig);
3202   }
3203   if (actp == NULL) {
3204     // Retrieve the preinstalled signal handler from jvm
3205     actp = get_preinstalled_handler(sig);
3206   }
3207 
3208   return actp;
3209 }
3210 
3211 static bool call_chained_handler(struct sigaction *actp, int sig,
3212                                  siginfo_t *siginfo, void *context) {
3213   // Call the old signal handler
3214   if (actp->sa_handler == SIG_DFL) {
3215     // It's more reasonable to let jvm treat it as an unexpected exception
3216     // instead of taking the default action.
3217     return false;
3218   } else if (actp->sa_handler != SIG_IGN) {
3219     if ((actp->sa_flags & SA_NODEFER) == 0) {
3220       // automaticlly block the signal
3221       sigaddset(&(actp->sa_mask), sig);
3222     }
3223 
3224     sa_handler_t hand = NULL;
3225     sa_sigaction_t sa = NULL;
3226     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3227     // retrieve the chained handler
3228     if (siginfo_flag_set) {
3229       sa = actp->sa_sigaction;
3230     } else {
3231       hand = actp->sa_handler;
3232     }
3233 
3234     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3235       actp->sa_handler = SIG_DFL;
3236     }
3237 
3238     // try to honor the signal mask
3239     sigset_t oset;
3240     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3241 
3242     // call into the chained handler
3243     if (siginfo_flag_set) {
3244       (*sa)(sig, siginfo, context);
3245     } else {
3246       (*hand)(sig);
3247     }
3248 
3249     // restore the signal mask
3250     pthread_sigmask(SIG_SETMASK, &oset, 0);
3251   }
3252   // Tell jvm's signal handler the signal is taken care of.
3253   return true;
3254 }
3255 
3256 bool os::Aix::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3257   bool chained = false;
3258   // signal-chaining
3259   if (UseSignalChaining) {
3260     struct sigaction *actp = get_chained_signal_action(sig);
3261     if (actp != NULL) {
3262       chained = call_chained_handler(actp, sig, siginfo, context);
3263     }
3264   }
3265   return chained;
3266 }
3267 
3268 struct sigaction* os::Aix::get_preinstalled_handler(int sig) {
3269   if ((((unsigned int)1 << sig) & sigs) != 0) {
3270     return &sigact[sig];
3271   }
3272   return NULL;
3273 }
3274 
3275 void os::Aix::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3276   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3277   sigact[sig] = oldAct;
3278   sigs |= (unsigned int)1 << sig;
3279 }
3280 
3281 // for diagnostic
3282 int os::Aix::sigflags[MAXSIGNUM];
3283 
3284 int os::Aix::get_our_sigflags(int sig) {
3285   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3286   return sigflags[sig];
3287 }
3288 
3289 void os::Aix::set_our_sigflags(int sig, int flags) {
3290   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3291   sigflags[sig] = flags;
3292 }
3293 
3294 void os::Aix::set_signal_handler(int sig, bool set_installed) {
3295   // Check for overwrite.
3296   struct sigaction oldAct;
3297   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3298 
3299   void* oldhand = oldAct.sa_sigaction
3300     ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3301     : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3302   // Renamed 'signalHandler' to avoid collision with other shared libs.
3303   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3304       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3305       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)javaSignalHandler)) {
3306     if (AllowUserSignalHandlers || !set_installed) {
3307       // Do not overwrite; user takes responsibility to forward to us.
3308       return;
3309     } else if (UseSignalChaining) {
3310       // save the old handler in jvm
3311       save_preinstalled_handler(sig, oldAct);
3312       // libjsig also interposes the sigaction() call below and saves the
3313       // old sigaction on it own.
3314     } else {
3315       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3316                     "%#lx for signal %d.", (long)oldhand, sig));
3317     }
3318   }
3319 
3320   struct sigaction sigAct;
3321   sigfillset(&(sigAct.sa_mask));
3322   if (!set_installed) {
3323     sigAct.sa_handler = SIG_DFL;
3324     sigAct.sa_flags = SA_RESTART;
3325   } else {
3326     // Renamed 'signalHandler' to avoid collision with other shared libs.
3327     sigAct.sa_sigaction = javaSignalHandler;
3328     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3329   }
3330   // Save flags, which are set by ours
3331   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3332   sigflags[sig] = sigAct.sa_flags;
3333 
3334   int ret = sigaction(sig, &sigAct, &oldAct);
3335   assert(ret == 0, "check");
3336 
3337   void* oldhand2 = oldAct.sa_sigaction
3338                  ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3339                  : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3340   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3341 }
3342 
3343 // install signal handlers for signals that HotSpot needs to
3344 // handle in order to support Java-level exception handling.
3345 void os::Aix::install_signal_handlers() {
3346   if (!signal_handlers_are_installed) {
3347     signal_handlers_are_installed = true;
3348 
3349     // signal-chaining
3350     typedef void (*signal_setting_t)();
3351     signal_setting_t begin_signal_setting = NULL;
3352     signal_setting_t end_signal_setting = NULL;
3353     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3354                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3355     if (begin_signal_setting != NULL) {
3356       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3357                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3358       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3359                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3360       libjsig_is_loaded = true;
3361       assert(UseSignalChaining, "should enable signal-chaining");
3362     }
3363     if (libjsig_is_loaded) {
3364       // Tell libjsig jvm is setting signal handlers
3365       (*begin_signal_setting)();
3366     }
3367 
3368     set_signal_handler(SIGSEGV, true);
3369     set_signal_handler(SIGPIPE, true);
3370     set_signal_handler(SIGBUS, true);
3371     set_signal_handler(SIGILL, true);
3372     set_signal_handler(SIGFPE, true);
3373     set_signal_handler(SIGTRAP, true);
3374     set_signal_handler(SIGXFSZ, true);
3375     set_signal_handler(SIGDANGER, true);
3376 
3377     if (libjsig_is_loaded) {
3378       // Tell libjsig jvm finishes setting signal handlers
3379       (*end_signal_setting)();
3380     }
3381 
3382     // We don't activate signal checker if libjsig is in place, we trust ourselves
3383     // and if UserSignalHandler is installed all bets are off.
3384     // Log that signal checking is off only if -verbose:jni is specified.
3385     if (CheckJNICalls) {
3386       if (libjsig_is_loaded) {
3387         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3388         check_signals = false;
3389       }
3390       if (AllowUserSignalHandlers) {
3391         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3392         check_signals = false;
3393       }
3394       // need to initialize check_signal_done
3395       ::sigemptyset(&check_signal_done);
3396     }
3397   }
3398 }
3399 
3400 static const char* get_signal_handler_name(address handler,
3401                                            char* buf, int buflen) {
3402   int offset;
3403   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3404   if (found) {
3405     // skip directory names
3406     const char *p1, *p2;
3407     p1 = buf;
3408     size_t len = strlen(os::file_separator());
3409     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3410     // The way os::dll_address_to_library_name is implemented on Aix
3411     // right now, it always returns -1 for the offset which is not
3412     // terribly informative.
3413     // Will fix that. For now, omit the offset.
3414     jio_snprintf(buf, buflen, "%s", p1);
3415   } else {
3416     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3417   }
3418   return buf;
3419 }
3420 
3421 static void print_signal_handler(outputStream* st, int sig,
3422                                  char* buf, size_t buflen) {
3423   struct sigaction sa;
3424   sigaction(sig, NULL, &sa);
3425 
3426   st->print("%s: ", os::exception_name(sig, buf, buflen));
3427 
3428   address handler = (sa.sa_flags & SA_SIGINFO)
3429     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3430     : CAST_FROM_FN_PTR(address, sa.sa_handler);
3431 
3432   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3433     st->print("SIG_DFL");
3434   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3435     st->print("SIG_IGN");
3436   } else {
3437     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3438   }
3439 
3440   // Print readable mask.
3441   st->print(", sa_mask[0]=");
3442   os::Posix::print_signal_set_short(st, &sa.sa_mask);
3443 
3444   address rh = VMError::get_resetted_sighandler(sig);
3445   // May be, handler was resetted by VMError?
3446   if (rh != NULL) {
3447     handler = rh;
3448     sa.sa_flags = VMError::get_resetted_sigflags(sig);
3449   }
3450 
3451   // Print textual representation of sa_flags.
3452   st->print(", sa_flags=");
3453   os::Posix::print_sa_flags(st, sa.sa_flags);
3454 
3455   // Check: is it our handler?
3456   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)javaSignalHandler) ||
3457       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3458     // It is our signal handler.
3459     // Check for flags, reset system-used one!
3460     if ((int)sa.sa_flags != os::Aix::get_our_sigflags(sig)) {
3461       st->print(", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3462                 os::Aix::get_our_sigflags(sig));
3463     }
3464   }
3465   st->cr();
3466 }
3467 
3468 
3469 #define DO_SIGNAL_CHECK(sig) \
3470   if (!sigismember(&check_signal_done, sig)) \
3471     os::Aix::check_signal_handler(sig)
3472 
3473 // This method is a periodic task to check for misbehaving JNI applications
3474 // under CheckJNI, we can add any periodic checks here
3475 
3476 void os::run_periodic_checks() {
3477 
3478   if (check_signals == false) return;
3479 
3480   // SEGV and BUS if overridden could potentially prevent
3481   // generation of hs*.log in the event of a crash, debugging
3482   // such a case can be very challenging, so we absolutely
3483   // check the following for a good measure:
3484   DO_SIGNAL_CHECK(SIGSEGV);
3485   DO_SIGNAL_CHECK(SIGILL);
3486   DO_SIGNAL_CHECK(SIGFPE);
3487   DO_SIGNAL_CHECK(SIGBUS);
3488   DO_SIGNAL_CHECK(SIGPIPE);
3489   DO_SIGNAL_CHECK(SIGXFSZ);
3490   if (UseSIGTRAP) {
3491     DO_SIGNAL_CHECK(SIGTRAP);
3492   }
3493   DO_SIGNAL_CHECK(SIGDANGER);
3494 
3495   // ReduceSignalUsage allows the user to override these handlers
3496   // see comments at the very top and jvm_solaris.h
3497   if (!ReduceSignalUsage) {
3498     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3499     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3500     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3501     DO_SIGNAL_CHECK(BREAK_SIGNAL);
3502   }
3503 
3504   DO_SIGNAL_CHECK(SR_signum);
3505   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3506 }
3507 
3508 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3509 
3510 static os_sigaction_t os_sigaction = NULL;
3511 
3512 void os::Aix::check_signal_handler(int sig) {
3513   char buf[O_BUFLEN];
3514   address jvmHandler = NULL;
3515 
3516   struct sigaction act;
3517   if (os_sigaction == NULL) {
3518     // only trust the default sigaction, in case it has been interposed
3519     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3520     if (os_sigaction == NULL) return;
3521   }
3522 
3523   os_sigaction(sig, (struct sigaction*)NULL, &act);
3524 
3525   address thisHandler = (act.sa_flags & SA_SIGINFO)
3526     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3527     : CAST_FROM_FN_PTR(address, act.sa_handler);
3528 
3529 
3530   switch(sig) {
3531   case SIGSEGV:
3532   case SIGBUS:
3533   case SIGFPE:
3534   case SIGPIPE:
3535   case SIGILL:
3536   case SIGXFSZ:
3537     // Renamed 'signalHandler' to avoid collision with other shared libs.
3538     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)javaSignalHandler);
3539     break;
3540 
3541   case SHUTDOWN1_SIGNAL:
3542   case SHUTDOWN2_SIGNAL:
3543   case SHUTDOWN3_SIGNAL:
3544   case BREAK_SIGNAL:
3545     jvmHandler = (address)user_handler();
3546     break;
3547 
3548   case INTERRUPT_SIGNAL:
3549     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3550     break;
3551 
3552   default:
3553     if (sig == SR_signum) {
3554       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3555     } else {
3556       return;
3557     }
3558     break;
3559   }
3560 
3561   if (thisHandler != jvmHandler) {
3562     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3563     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3564     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3565     // No need to check this sig any longer
3566     sigaddset(&check_signal_done, sig);
3567     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
3568     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
3569       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
3570                     exception_name(sig, buf, O_BUFLEN));
3571     }
3572   } else if (os::Aix::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Aix::get_our_sigflags(sig)) {
3573     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3574     tty->print("expected:" PTR32_FORMAT, os::Aix::get_our_sigflags(sig));
3575     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
3576     // No need to check this sig any longer
3577     sigaddset(&check_signal_done, sig);
3578   }
3579 
3580   // Dump all the signal
3581   if (sigismember(&check_signal_done, sig)) {
3582     print_signal_handlers(tty, buf, O_BUFLEN);
3583   }
3584 }
3585 
3586 extern bool signal_name(int signo, char* buf, size_t len);
3587 
3588 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3589   if (0 < exception_code && exception_code <= SIGRTMAX) {
3590     // signal
3591     if (!signal_name(exception_code, buf, size)) {
3592       jio_snprintf(buf, size, "SIG%d", exception_code);
3593     }
3594     return buf;
3595   } else {
3596     return NULL;
3597   }
3598 }
3599 
3600 // To install functions for atexit system call
3601 extern "C" {
3602   static void perfMemory_exit_helper() {
3603     perfMemory_exit();
3604   }
3605 }
3606 
3607 // This is called _before_ the most of global arguments have been parsed.
3608 void os::init(void) {
3609   // This is basic, we want to know if that ever changes.
3610   // (shared memory boundary is supposed to be a 256M aligned)
3611   assert(SHMLBA == ((uint64_t)0x10000000ULL)/*256M*/, "unexpected");
3612 
3613   // First off, we need to know whether we run on AIX or PASE, and
3614   // the OS level we run on.
3615   os::Aix::initialize_os_info();
3616 
3617   // Scan environment (SPEC1170 behaviour, etc)
3618   os::Aix::scan_environment();
3619 
3620   // Check which pages are supported by AIX.
3621   os::Aix::query_multipage_support();
3622 
3623   // Next, we need to initialize libo4 and libperfstat libraries.
3624   if (os::Aix::on_pase()) {
3625     os::Aix::initialize_libo4();
3626   } else {
3627     os::Aix::initialize_libperfstat();
3628   }
3629 
3630   // Reset the perfstat information provided by ODM.
3631   if (os::Aix::on_aix()) {
3632     libperfstat::perfstat_reset();
3633   }
3634 
3635   // Now initialze basic system properties. Note that for some of the values we
3636   // need libperfstat etc.
3637   os::Aix::initialize_system_info();
3638 
3639   // Initialize large page support.
3640   if (UseLargePages) {
3641     os::large_page_init();
3642     if (!UseLargePages) {
3643       // initialize os::_page_sizes
3644       _page_sizes[0] = Aix::page_size();
3645       _page_sizes[1] = 0;
3646       if (Verbose) {
3647         fprintf(stderr, "Large Page initialization failed: setting UseLargePages=0.\n");
3648       }
3649     }
3650   } else {
3651     // initialize os::_page_sizes
3652     _page_sizes[0] = Aix::page_size();
3653     _page_sizes[1] = 0;
3654   }
3655 
3656   // debug trace
3657   if (Verbose) {
3658     fprintf(stderr, "os::vm_page_size 0x%llX\n", os::vm_page_size());
3659     fprintf(stderr, "os::large_page_size 0x%llX\n", os::large_page_size());
3660     fprintf(stderr, "os::_page_sizes = ( ");
3661     for (int i = 0; _page_sizes[i]; i ++) {
3662       fprintf(stderr, " %s ", describe_pagesize(_page_sizes[i]));
3663     }
3664     fprintf(stderr, ")\n");
3665   }
3666 
3667   _initial_pid = getpid();
3668 
3669   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3670 
3671   init_random(1234567);
3672 
3673   ThreadCritical::initialize();
3674 
3675   // Main_thread points to the aboriginal thread.
3676   Aix::_main_thread = pthread_self();
3677 
3678   initial_time_count = os::elapsed_counter();
3679   pthread_mutex_init(&dl_mutex, NULL);
3680 }
3681 
3682 // this is called _after_ the global arguments have been parsed
3683 jint os::init_2(void) {
3684 
3685   if (Verbose) {
3686     fprintf(stderr, "processor count: %d\n", os::_processor_count);
3687     fprintf(stderr, "physical memory: %lu\n", Aix::_physical_memory);
3688   }
3689 
3690   // initially build up the loaded dll map
3691   LoadedLibraries::reload();
3692 
3693   const int page_size = Aix::page_size();
3694   const int map_size = page_size;
3695 
3696   address map_address = (address) MAP_FAILED;
3697   const int prot  = PROT_READ;
3698   const int flags = MAP_PRIVATE|MAP_ANONYMOUS;
3699 
3700   // use optimized addresses for the polling page,
3701   // e.g. map it to a special 32-bit address.
3702   if (OptimizePollingPageLocation) {
3703     // architecture-specific list of address wishes:
3704     address address_wishes[] = {
3705       // AIX: addresses lower than 0x30000000 don't seem to work on AIX.
3706       // PPC64: all address wishes are non-negative 32 bit values where
3707       // the lower 16 bits are all zero. we can load these addresses
3708       // with a single ppc_lis instruction.
3709       (address) 0x30000000, (address) 0x31000000,
3710       (address) 0x32000000, (address) 0x33000000,
3711       (address) 0x40000000, (address) 0x41000000,
3712       (address) 0x42000000, (address) 0x43000000,
3713       (address) 0x50000000, (address) 0x51000000,
3714       (address) 0x52000000, (address) 0x53000000,
3715       (address) 0x60000000, (address) 0x61000000,
3716       (address) 0x62000000, (address) 0x63000000
3717     };
3718     int address_wishes_length = sizeof(address_wishes)/sizeof(address);
3719 
3720     // iterate over the list of address wishes:
3721     for (int i=0; i<address_wishes_length; i++) {
3722       // try to map with current address wish.
3723       // AIX: AIX needs MAP_FIXED if we provide an address and mmap will
3724       // fail if the address is already mapped.
3725       map_address = (address) ::mmap(address_wishes[i] - (ssize_t)page_size,
3726                                      map_size, prot,
3727                                      flags | MAP_FIXED,
3728                                      -1, 0);
3729       if (Verbose) {
3730         fprintf(stderr, "SafePoint Polling Page address: %p (wish) => %p\n",
3731                 address_wishes[i], map_address + (ssize_t)page_size);
3732       }
3733 
3734       if (map_address + (ssize_t)page_size == address_wishes[i]) {
3735         // map succeeded and map_address is at wished address, exit loop.
3736         break;
3737       }
3738 
3739       if (map_address != (address) MAP_FAILED) {
3740         // map succeeded, but polling_page is not at wished address, unmap and continue.
3741         ::munmap(map_address, map_size);
3742         map_address = (address) MAP_FAILED;
3743       }
3744       // map failed, continue loop.
3745     }
3746   } // end OptimizePollingPageLocation
3747 
3748   if (map_address == (address) MAP_FAILED) {
3749     map_address = (address) ::mmap(NULL, map_size, prot, flags, -1, 0);
3750   }
3751   guarantee(map_address != MAP_FAILED, "os::init_2: failed to allocate polling page");
3752   os::set_polling_page(map_address);
3753 
3754   if (!UseMembar) {
3755     address mem_serialize_page = (address) ::mmap(NULL, Aix::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3756     guarantee(mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3757     os::set_memory_serialize_page(mem_serialize_page);
3758 
3759 #ifndef PRODUCT
3760     if (Verbose && PrintMiscellaneous)
3761       tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3762 #endif
3763   }
3764 
3765   // initialize suspend/resume support - must do this before signal_sets_init()
3766   if (SR_initialize() != 0) {
3767     perror("SR_initialize failed");
3768     return JNI_ERR;
3769   }
3770 
3771   Aix::signal_sets_init();
3772   Aix::install_signal_handlers();
3773 
3774   // Check minimum allowable stack size for thread creation and to initialize
3775   // the java system classes, including StackOverflowError - depends on page
3776   // size. Add a page for compiler2 recursion in main thread.
3777   // Add in 2*BytesPerWord times page size to account for VM stack during
3778   // class initialization depending on 32 or 64 bit VM.
3779   os::Aix::min_stack_allowed = MAX2(os::Aix::min_stack_allowed,
3780             (size_t)(StackYellowPages+StackRedPages+StackShadowPages +
3781                      2*BytesPerWord COMPILER2_PRESENT(+1)) * Aix::page_size());
3782 
3783   size_t threadStackSizeInBytes = ThreadStackSize * K;
3784   if (threadStackSizeInBytes != 0 &&
3785       threadStackSizeInBytes < os::Aix::min_stack_allowed) {
3786         tty->print_cr("\nThe stack size specified is too small, "
3787                       "Specify at least %dk",
3788                       os::Aix::min_stack_allowed / K);
3789         return JNI_ERR;
3790   }
3791 
3792   // Make the stack size a multiple of the page size so that
3793   // the yellow/red zones can be guarded.
3794   // note that this can be 0, if no default stacksize was set
3795   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, vm_page_size()));
3796 
3797   Aix::libpthread_init();
3798 
3799   if (MaxFDLimit) {
3800     // set the number of file descriptors to max. print out error
3801     // if getrlimit/setrlimit fails but continue regardless.
3802     struct rlimit nbr_files;
3803     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3804     if (status != 0) {
3805       if (PrintMiscellaneous && (Verbose || WizardMode))
3806         perror("os::init_2 getrlimit failed");
3807     } else {
3808       nbr_files.rlim_cur = nbr_files.rlim_max;
3809       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3810       if (status != 0) {
3811         if (PrintMiscellaneous && (Verbose || WizardMode))
3812           perror("os::init_2 setrlimit failed");
3813       }
3814     }
3815   }
3816 
3817   if (PerfAllowAtExitRegistration) {
3818     // only register atexit functions if PerfAllowAtExitRegistration is set.
3819     // atexit functions can be delayed until process exit time, which
3820     // can be problematic for embedded VM situations. Embedded VMs should
3821     // call DestroyJavaVM() to assure that VM resources are released.
3822 
3823     // note: perfMemory_exit_helper atexit function may be removed in
3824     // the future if the appropriate cleanup code can be added to the
3825     // VM_Exit VMOperation's doit method.
3826     if (atexit(perfMemory_exit_helper) != 0) {
3827       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
3828     }
3829   }
3830 
3831   return JNI_OK;
3832 }
3833 
3834 // this is called at the end of vm_initialization
3835 void os::init_3(void) {
3836   return;
3837 }
3838 
3839 // Mark the polling page as unreadable
3840 void os::make_polling_page_unreadable(void) {
3841   if (!guard_memory((char*)_polling_page, Aix::page_size())) {
3842     fatal("Could not disable polling page");
3843   }
3844 };
3845 
3846 // Mark the polling page as readable
3847 void os::make_polling_page_readable(void) {
3848   // Changed according to os_linux.cpp.
3849   if (!checked_mprotect((char *)_polling_page, Aix::page_size(), PROT_READ)) {
3850     fatal(err_msg("Could not enable polling page at " PTR_FORMAT, _polling_page));
3851   }
3852 };
3853 
3854 int os::active_processor_count() {
3855   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
3856   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
3857   return online_cpus;
3858 }
3859 
3860 void os::set_native_thread_name(const char *name) {
3861   // Not yet implemented.
3862   return;
3863 }
3864 
3865 bool os::distribute_processes(uint length, uint* distribution) {
3866   // Not yet implemented.
3867   return false;
3868 }
3869 
3870 bool os::bind_to_processor(uint processor_id) {
3871   // Not yet implemented.
3872   return false;
3873 }
3874 
3875 void os::SuspendedThreadTask::internal_do_task() {
3876   if (do_suspend(_thread->osthread())) {
3877     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
3878     do_task(context);
3879     do_resume(_thread->osthread());
3880   }
3881 }
3882 
3883 class PcFetcher : public os::SuspendedThreadTask {
3884 public:
3885   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
3886   ExtendedPC result();
3887 protected:
3888   void do_task(const os::SuspendedThreadTaskContext& context);
3889 private:
3890   ExtendedPC _epc;
3891 };
3892 
3893 ExtendedPC PcFetcher::result() {
3894   guarantee(is_done(), "task is not done yet.");
3895   return _epc;
3896 }
3897 
3898 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
3899   Thread* thread = context.thread();
3900   OSThread* osthread = thread->osthread();
3901   if (osthread->ucontext() != NULL) {
3902     _epc = os::Aix::ucontext_get_pc((ucontext_t *) context.ucontext());
3903   } else {
3904     // NULL context is unexpected, double-check this is the VMThread.
3905     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3906   }
3907 }
3908 
3909 // Suspends the target using the signal mechanism and then grabs the PC before
3910 // resuming the target. Used by the flat-profiler only
3911 ExtendedPC os::get_thread_pc(Thread* thread) {
3912   // Make sure that it is called by the watcher for the VMThread.
3913   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
3914   assert(thread->is_VM_thread(), "Can only be called for VMThread");
3915 
3916   PcFetcher fetcher(thread);
3917   fetcher.run();
3918   return fetcher.result();
3919 }
3920 
3921 // Not neede on Aix.
3922 // int os::Aix::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) {
3923 // }
3924 
3925 ////////////////////////////////////////////////////////////////////////////////
3926 // debug support
3927 
3928 static address same_page(address x, address y) {
3929   intptr_t page_bits = -os::vm_page_size();
3930   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
3931     return x;
3932   else if (x > y)
3933     return (address)(intptr_t(y) | ~page_bits) + 1;
3934   else
3935     return (address)(intptr_t(y) & page_bits);
3936 }
3937 
3938 bool os::find(address addr, outputStream* st) {
3939 
3940   st->print(PTR_FORMAT ": ", addr);
3941 
3942   const LoadedLibraryModule* lib = LoadedLibraries::find_for_text_address(addr);
3943   if (lib) {
3944     lib->print(st);
3945     return true;
3946   } else {
3947     lib = LoadedLibraries::find_for_data_address(addr);
3948     if (lib) {
3949       lib->print(st);
3950       return true;
3951     } else {
3952       st->print_cr("(outside any module)");
3953     }
3954   }
3955 
3956   return false;
3957 }
3958 
3959 ////////////////////////////////////////////////////////////////////////////////
3960 // misc
3961 
3962 // This does not do anything on Aix. This is basically a hook for being
3963 // able to use structured exception handling (thread-local exception filters)
3964 // on, e.g., Win32.
3965 void
3966 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
3967                          JavaCallArguments* args, Thread* thread) {
3968   f(value, method, args, thread);
3969 }
3970 
3971 void os::print_statistics() {
3972 }
3973 
3974 int os::message_box(const char* title, const char* message) {
3975   int i;
3976   fdStream err(defaultStream::error_fd());
3977   for (i = 0; i < 78; i++) err.print_raw("=");
3978   err.cr();
3979   err.print_raw_cr(title);
3980   for (i = 0; i < 78; i++) err.print_raw("-");
3981   err.cr();
3982   err.print_raw_cr(message);
3983   for (i = 0; i < 78; i++) err.print_raw("=");
3984   err.cr();
3985 
3986   char buf[16];
3987   // Prevent process from exiting upon "read error" without consuming all CPU
3988   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3989 
3990   return buf[0] == 'y' || buf[0] == 'Y';
3991 }
3992 
3993 int os::stat(const char *path, struct stat *sbuf) {
3994   char pathbuf[MAX_PATH];
3995   if (strlen(path) > MAX_PATH - 1) {
3996     errno = ENAMETOOLONG;
3997     return -1;
3998   }
3999   os::native_path(strcpy(pathbuf, path));
4000   return ::stat(pathbuf, sbuf);
4001 }
4002 
4003 bool os::check_heap(bool force) {
4004   return true;
4005 }
4006 
4007 // int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4008 //   return ::vsnprintf(buf, count, format, args);
4009 // }
4010 
4011 // Is a (classpath) directory empty?
4012 bool os::dir_is_empty(const char* path) {
4013   DIR *dir = NULL;
4014   struct dirent *ptr;
4015 
4016   dir = opendir(path);
4017   if (dir == NULL) return true;
4018 
4019   /* Scan the directory */
4020   bool result = true;
4021   char buf[sizeof(struct dirent) + MAX_PATH];
4022   while (result && (ptr = ::readdir(dir)) != NULL) {
4023     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4024       result = false;
4025     }
4026   }
4027   closedir(dir);
4028   return result;
4029 }
4030 
4031 // This code originates from JDK's sysOpen and open64_w
4032 // from src/solaris/hpi/src/system_md.c
4033 
4034 #ifndef O_DELETE
4035 #define O_DELETE 0x10000
4036 #endif
4037 
4038 // Open a file. Unlink the file immediately after open returns
4039 // if the specified oflag has the O_DELETE flag set.
4040 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4041 
4042 int os::open(const char *path, int oflag, int mode) {
4043 
4044   if (strlen(path) > MAX_PATH - 1) {
4045     errno = ENAMETOOLONG;
4046     return -1;
4047   }
4048   int fd;
4049   int o_delete = (oflag & O_DELETE);
4050   oflag = oflag & ~O_DELETE;
4051 
4052   fd = ::open64(path, oflag, mode);
4053   if (fd == -1) return -1;
4054 
4055   // If the open succeeded, the file might still be a directory.
4056   {
4057     struct stat64 buf64;
4058     int ret = ::fstat64(fd, &buf64);
4059     int st_mode = buf64.st_mode;
4060 
4061     if (ret != -1) {
4062       if ((st_mode & S_IFMT) == S_IFDIR) {
4063         errno = EISDIR;
4064         ::close(fd);
4065         return -1;
4066       }
4067     } else {
4068       ::close(fd);
4069       return -1;
4070     }
4071   }
4072 
4073   // All file descriptors that are opened in the JVM and not
4074   // specifically destined for a subprocess should have the
4075   // close-on-exec flag set. If we don't set it, then careless 3rd
4076   // party native code might fork and exec without closing all
4077   // appropriate file descriptors (e.g. as we do in closeDescriptors in
4078   // UNIXProcess.c), and this in turn might:
4079   //
4080   // - cause end-of-file to fail to be detected on some file
4081   //   descriptors, resulting in mysterious hangs, or
4082   //
4083   // - might cause an fopen in the subprocess to fail on a system
4084   //   suffering from bug 1085341.
4085   //
4086   // (Yes, the default setting of the close-on-exec flag is a Unix
4087   // design flaw.)
4088   //
4089   // See:
4090   // 1085341: 32-bit stdio routines should support file descriptors >255
4091   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4092   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4093 #ifdef FD_CLOEXEC
4094   {
4095     int flags = ::fcntl(fd, F_GETFD);
4096     if (flags != -1)
4097       ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4098   }
4099 #endif
4100 
4101   if (o_delete != 0) {
4102     ::unlink(path);
4103   }
4104   return fd;
4105 }
4106 
4107 
4108 // create binary file, rewriting existing file if required
4109 int os::create_binary_file(const char* path, bool rewrite_existing) {
4110   int oflags = O_WRONLY | O_CREAT;
4111   if (!rewrite_existing) {
4112     oflags |= O_EXCL;
4113   }
4114   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4115 }
4116 
4117 // return current position of file pointer
4118 jlong os::current_file_offset(int fd) {
4119   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4120 }
4121 
4122 // move file pointer to the specified offset
4123 jlong os::seek_to_file_offset(int fd, jlong offset) {
4124   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4125 }
4126 
4127 // This code originates from JDK's sysAvailable
4128 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4129 
4130 int os::available(int fd, jlong *bytes) {
4131   jlong cur, end;
4132   int mode;
4133   struct stat64 buf64;
4134 
4135   if (::fstat64(fd, &buf64) >= 0) {
4136     mode = buf64.st_mode;
4137     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4138       // XXX: is the following call interruptible? If so, this might
4139       // need to go through the INTERRUPT_IO() wrapper as for other
4140       // blocking, interruptible calls in this file.
4141       int n;
4142       if (::ioctl(fd, FIONREAD, &n) >= 0) {
4143         *bytes = n;
4144         return 1;
4145       }
4146     }
4147   }
4148   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4149     return 0;
4150   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4151     return 0;
4152   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4153     return 0;
4154   }
4155   *bytes = end - cur;
4156   return 1;
4157 }
4158 
4159 int os::socket_available(int fd, jint *pbytes) {
4160   // Linux doc says EINTR not returned, unlike Solaris
4161   int ret = ::ioctl(fd, FIONREAD, pbytes);
4162 
4163   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4164   // is expected to return 0 on failure and 1 on success to the jdk.
4165   return (ret < 0) ? 0 : 1;
4166 }
4167 
4168 // Map a block of memory.
4169 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4170                         char *addr, size_t bytes, bool read_only,
4171                         bool allow_exec) {
4172   Unimplemented();
4173   return NULL;
4174 }
4175 
4176 
4177 // Remap a block of memory.
4178 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4179                           char *addr, size_t bytes, bool read_only,
4180                           bool allow_exec) {
4181   // same as map_memory() on this OS
4182   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4183                         allow_exec);
4184 }
4185 
4186 // Unmap a block of memory.
4187 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4188   return munmap(addr, bytes) == 0;
4189 }
4190 
4191 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4192 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4193 // of a thread.
4194 //
4195 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4196 // the fast estimate available on the platform.
4197 
4198 jlong os::current_thread_cpu_time() {
4199   // return user + sys since the cost is the same
4200   const jlong n = os::thread_cpu_time(Thread::current(), true /* user + sys */);
4201   assert(n >= 0, "negative CPU time");
4202   return n;
4203 }
4204 
4205 jlong os::thread_cpu_time(Thread* thread) {
4206   // consistent with what current_thread_cpu_time() returns
4207   const jlong n = os::thread_cpu_time(thread, true /* user + sys */);
4208   assert(n >= 0, "negative CPU time");
4209   return n;
4210 }
4211 
4212 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4213   const jlong n = os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
4214   assert(n >= 0, "negative CPU time");
4215   return n;
4216 }
4217 
4218 static bool thread_cpu_time_unchecked(Thread* thread, jlong* p_sys_time, jlong* p_user_time) {
4219   bool error = false;
4220 
4221   jlong sys_time = 0;
4222   jlong user_time = 0;
4223 
4224   // reimplemented using getthrds64().
4225   //
4226   // goes like this:
4227   // For the thread in question, get the kernel thread id. Then get the
4228   // kernel thread statistics using that id.
4229   //
4230   // This only works of course when no pthread scheduling is used,
4231   // ie there is a 1:1 relationship to kernel threads.
4232   // On AIX, see AIXTHREAD_SCOPE variable.
4233 
4234   pthread_t pthtid = thread->osthread()->pthread_id();
4235 
4236   // retrieve kernel thread id for the pthread:
4237   tid64_t tid = 0;
4238   struct __pthrdsinfo pinfo;
4239   // I just love those otherworldly IBM APIs which force me to hand down
4240   // dummy buffers for stuff I dont care for...
4241   char dummy[1];
4242   int dummy_size = sizeof(dummy);
4243   if (pthread_getthrds_np(&pthtid, PTHRDSINFO_QUERY_TID, &pinfo, sizeof(pinfo),
4244                           dummy, &dummy_size) == 0) {
4245     tid = pinfo.__pi_tid;
4246   } else {
4247     tty->print_cr("pthread_getthrds_np failed.");
4248     error = true;
4249   }
4250 
4251   // retrieve kernel timing info for that kernel thread
4252   if (!error) {
4253     struct thrdentry64 thrdentry;
4254     if (getthrds64(getpid(), &thrdentry, sizeof(thrdentry), &tid, 1) == 1) {
4255       sys_time = thrdentry.ti_ru.ru_stime.tv_sec * 1000000000LL + thrdentry.ti_ru.ru_stime.tv_usec * 1000LL;
4256       user_time = thrdentry.ti_ru.ru_utime.tv_sec * 1000000000LL + thrdentry.ti_ru.ru_utime.tv_usec * 1000LL;
4257     } else {
4258       tty->print_cr("pthread_getthrds_np failed.");
4259       error = true;
4260     }
4261   }
4262 
4263   if (p_sys_time) {
4264     *p_sys_time = sys_time;
4265   }
4266 
4267   if (p_user_time) {
4268     *p_user_time = user_time;
4269   }
4270 
4271   if (error) {
4272     return false;
4273   }
4274 
4275   return true;
4276 }
4277 
4278 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4279   jlong sys_time;
4280   jlong user_time;
4281 
4282   if (!thread_cpu_time_unchecked(thread, &sys_time, &user_time)) {
4283     return -1;
4284   }
4285 
4286   return user_sys_cpu_time ? sys_time + user_time : user_time;
4287 }
4288 
4289 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4290   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4291   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4292   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4293   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4294 }
4295 
4296 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4297   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4298   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4299   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4300   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4301 }
4302 
4303 bool os::is_thread_cpu_time_supported() {
4304   return true;
4305 }
4306 
4307 // System loadavg support. Returns -1 if load average cannot be obtained.
4308 // For now just return the system wide load average (no processor sets).
4309 int os::loadavg(double values[], int nelem) {
4310 
4311   // Implemented using libperfstat on AIX.
4312 
4313   guarantee(nelem >= 0 && nelem <= 3, "argument error");
4314   guarantee(values, "argument error");
4315 
4316   if (os::Aix::on_pase()) {
4317     Unimplemented();
4318     return -1;
4319   } else {
4320     // AIX: use libperfstat
4321     //
4322     // See also:
4323     // http://publib.boulder.ibm.com/infocenter/pseries/v5r3/index.jsp?topic=/com.ibm.aix.basetechref/doc/basetrf1/perfstat_cputot.htm
4324     // /usr/include/libperfstat.h:
4325 
4326     // Use the already AIX version independent get_cpuinfo.
4327     os::Aix::cpuinfo_t ci;
4328     if (os::Aix::get_cpuinfo(&ci)) {
4329       for (int i = 0; i < nelem; i++) {
4330         values[i] = ci.loadavg[i];
4331       }
4332     } else {
4333       return -1;
4334     }
4335     return nelem;
4336   }
4337 }
4338 
4339 void os::pause() {
4340   char filename[MAX_PATH];
4341   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4342     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4343   } else {
4344     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4345   }
4346 
4347   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4348   if (fd != -1) {
4349     struct stat buf;
4350     ::close(fd);
4351     while (::stat(filename, &buf) == 0) {
4352       (void)::poll(NULL, 0, 100);
4353     }
4354   } else {
4355     jio_fprintf(stderr,
4356       "Could not open pause file '%s', continuing immediately.\n", filename);
4357   }
4358 }
4359 
4360 bool os::Aix::is_primordial_thread() {
4361   if (pthread_self() == (pthread_t)1) {
4362     return true;
4363   } else {
4364     return false;
4365   }
4366 }
4367 
4368 // OS recognitions (PASE/AIX, OS level) call this before calling any
4369 // one of Aix::on_pase(), Aix::os_version() static
4370 void os::Aix::initialize_os_info() {
4371 
4372   assert(_on_pase == -1 && _os_version == -1, "already called.");
4373 
4374   struct utsname uts;
4375   memset(&uts, 0, sizeof(uts));
4376   strcpy(uts.sysname, "?");
4377   if (::uname(&uts) == -1) {
4378     fprintf(stderr, "uname failed (%d)\n", errno);
4379     guarantee(0, "Could not determine whether we run on AIX or PASE");
4380   } else {
4381     if (Verbose) {
4382       fprintf(stderr,"uname says: sysname \"%s\" version \"%s\" release \"%s\" "
4383               "node \"%s\" machine \"%s\"\n",
4384               uts.sysname, uts.version, uts.release, uts.nodename, uts.machine);
4385     }
4386     const int major = atoi(uts.version);
4387     assert(major > 0, "invalid OS version");
4388     const int minor = atoi(uts.release);
4389     assert(minor > 0, "invalid OS release");
4390     _os_version = (major << 8) | minor;
4391     if (strcmp(uts.sysname, "OS400") == 0) {
4392       Unimplemented();
4393     } else if (strcmp(uts.sysname, "AIX") == 0) {
4394       // We run on AIX. We do not support versions older than AIX 5.3.
4395       _on_pase = 0;
4396       if (_os_version < 0x0503) {
4397         fprintf(stderr, "AIX release older than AIX 5.3 not supported.\n");
4398         assert(false, "AIX release too old.");
4399       } else {
4400         if (Verbose) {
4401           fprintf(stderr, "We run on AIX %d.%d\n", major, minor);
4402         }
4403       }
4404     } else {
4405       assert(false, "unknown OS");
4406     }
4407   }
4408 
4409   guarantee(_on_pase != -1 && _os_version, "Could not determine AIX/OS400 release");
4410 
4411 } // end: os::Aix::initialize_os_info()
4412 
4413 // Scan environment for important settings which might effect the VM.
4414 // Trace out settings. Warn about invalid settings and/or correct them.
4415 //
4416 // Must run after os::Aix::initialue_os_info().
4417 void os::Aix::scan_environment() {
4418 
4419   char* p;
4420   int rc;
4421 
4422   // Warn explicity if EXTSHM=ON is used. That switch changes how
4423   // System V shared memory behaves. One effect is that page size of
4424   // shared memory cannot be change dynamically, effectivly preventing
4425   // large pages from working.
4426   // This switch was needed on AIX 32bit, but on AIX 64bit the general
4427   // recommendation is (in OSS notes) to switch it off.
4428   p = ::getenv("EXTSHM");
4429   if (Verbose) {
4430     fprintf(stderr, "EXTSHM=%s.\n", p ? p : "<unset>");
4431   }
4432   if (p && strcmp(p, "ON") == 0) {
4433     fprintf(stderr, "Unsupported setting: EXTSHM=ON. Large Page support will be disabled.\n");
4434     _extshm = 1;
4435   } else {
4436     _extshm = 0;
4437   }
4438 
4439   // SPEC1170 behaviour: will change the behaviour of a number of POSIX APIs.
4440   // Not tested, not supported.
4441   //
4442   // Note that it might be worth the trouble to test and to require it, if only to
4443   // get useful return codes for mprotect.
4444   //
4445   // Note: Setting XPG_SUS_ENV in the process is too late. Must be set earlier (before
4446   // exec() ? before loading the libjvm ? ....)
4447   p = ::getenv("XPG_SUS_ENV");
4448   if (Verbose) {
4449     fprintf(stderr, "XPG_SUS_ENV=%s.\n", p ? p : "<unset>");
4450   }
4451   if (p && strcmp(p, "ON") == 0) {
4452     _xpg_sus_mode = 1;
4453     fprintf(stderr, "Unsupported setting: XPG_SUS_ENV=ON\n");
4454     // This is not supported. Worst of all, it changes behaviour of mmap MAP_FIXED to
4455     // clobber address ranges. If we ever want to support that, we have to do some
4456     // testing first.
4457     guarantee(false, "XPG_SUS_ENV=ON not supported");
4458   } else {
4459     _xpg_sus_mode = 0;
4460   }
4461 
4462   // Switch off AIX internal (pthread) guard pages. This has
4463   // immediate effect for any pthread_create calls which follow.
4464   p = ::getenv("AIXTHREAD_GUARDPAGES");
4465   if (Verbose) {
4466     fprintf(stderr, "AIXTHREAD_GUARDPAGES=%s.\n", p ? p : "<unset>");
4467     fprintf(stderr, "setting AIXTHREAD_GUARDPAGES=0.\n");
4468   }
4469   rc = ::putenv("AIXTHREAD_GUARDPAGES=0");
4470   guarantee(rc == 0, "");
4471 
4472 } // end: os::Aix::scan_environment()
4473 
4474 // PASE: initialize the libo4 library (AS400 PASE porting library).
4475 void os::Aix::initialize_libo4() {
4476   Unimplemented();
4477 }
4478 
4479 // AIX: initialize the libperfstat library (we load this dynamically
4480 // because it is only available on AIX.
4481 void os::Aix::initialize_libperfstat() {
4482 
4483   assert(os::Aix::on_aix(), "AIX only");
4484 
4485   if (!libperfstat::init()) {
4486     fprintf(stderr, "libperfstat initialization failed.\n");
4487     assert(false, "libperfstat initialization failed");
4488   } else {
4489     if (Verbose) {
4490       fprintf(stderr, "libperfstat initialized.\n");
4491     }
4492   }
4493 } // end: os::Aix::initialize_libperfstat
4494 
4495 /////////////////////////////////////////////////////////////////////////////
4496 // thread stack
4497 
4498 // function to query the current stack size using pthread_getthrds_np
4499 //
4500 // ! do not change anything here unless you know what you are doing !
4501 static void query_stack_dimensions(address* p_stack_base, size_t* p_stack_size) {
4502 
4503   // This only works when invoked on a pthread. As we agreed not to use
4504   // primordial threads anyway, I assert here
4505   guarantee(!os::Aix::is_primordial_thread(), "not allowed on the primordial thread");
4506 
4507   // information about this api can be found (a) in the pthread.h header and
4508   // (b) in http://publib.boulder.ibm.com/infocenter/pseries/v5r3/index.jsp?topic=/com.ibm.aix.basetechref/doc/basetrf1/pthread_getthrds_np.htm
4509   //
4510   // The use of this API to find out the current stack is kind of undefined.
4511   // But after a lot of tries and asking IBM about it, I concluded that it is safe
4512   // enough for cases where I let the pthread library create its stacks. For cases
4513   // where I create an own stack and pass this to pthread_create, it seems not to
4514   // work (the returned stack size in that case is 0).
4515 
4516   pthread_t tid = pthread_self();
4517   struct __pthrdsinfo pinfo;
4518   char dummy[1]; // we only need this to satisfy the api and to not get E
4519   int dummy_size = sizeof(dummy);
4520 
4521   memset(&pinfo, 0, sizeof(pinfo));
4522 
4523   const int rc = pthread_getthrds_np (&tid, PTHRDSINFO_QUERY_ALL, &pinfo,
4524                                       sizeof(pinfo), dummy, &dummy_size);
4525 
4526   if (rc != 0) {
4527     fprintf(stderr, "pthread_getthrds_np failed (%d)\n", rc);
4528     guarantee(0, "pthread_getthrds_np failed");
4529   }
4530 
4531   guarantee(pinfo.__pi_stackend, "returned stack base invalid");
4532 
4533   // the following can happen when invoking pthread_getthrds_np on a pthread running on a user provided stack
4534   // (when handing down a stack to pthread create, see pthread_attr_setstackaddr).
4535   // Not sure what to do here - I feel inclined to forbid this use case completely.
4536   guarantee(pinfo.__pi_stacksize, "returned stack size invalid");
4537 
4538   // On AIX, stacks are not necessarily page aligned so round the base and size accordingly
4539   if (p_stack_base) {
4540     (*p_stack_base) = (address) align_size_up((intptr_t)pinfo.__pi_stackend, os::Aix::stack_page_size());
4541   }
4542 
4543   if (p_stack_size) {
4544     (*p_stack_size) = pinfo.__pi_stacksize - os::Aix::stack_page_size();
4545   }
4546 
4547 #ifndef PRODUCT
4548   if (Verbose) {
4549     fprintf(stderr,
4550             "query_stack_dimensions() -> real stack_base=" INTPTR_FORMAT ", real stack_addr=" INTPTR_FORMAT
4551             ", real stack_size=" INTPTR_FORMAT
4552             ", stack_base=" INTPTR_FORMAT ", stack_size=" INTPTR_FORMAT "\n",
4553             (intptr_t)pinfo.__pi_stackend, (intptr_t)pinfo.__pi_stackaddr, pinfo.__pi_stacksize,
4554             (intptr_t)align_size_up((intptr_t)pinfo.__pi_stackend, os::Aix::stack_page_size()),
4555             pinfo.__pi_stacksize - os::Aix::stack_page_size());
4556   }
4557 #endif
4558 
4559 } // end query_stack_dimensions
4560 
4561 // get the current stack base from the OS (actually, the pthread library)
4562 address os::current_stack_base() {
4563   address p;
4564   query_stack_dimensions(&p, 0);
4565   return p;
4566 }
4567 
4568 // get the current stack size from the OS (actually, the pthread library)
4569 size_t os::current_stack_size() {
4570   size_t s;
4571   query_stack_dimensions(0, &s);
4572   return s;
4573 }
4574 
4575 // Refer to the comments in os_solaris.cpp park-unpark.
4576 //
4577 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4578 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4579 // For specifics regarding the bug see GLIBC BUGID 261237 :
4580 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4581 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4582 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4583 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4584 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4585 // and monitorenter when we're using 1-0 locking. All those operations may result in
4586 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4587 // of libpthread avoids the problem, but isn't practical.
4588 //
4589 // Possible remedies:
4590 //
4591 // 1.   Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4592 //      This is palliative and probabilistic, however. If the thread is preempted
4593 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
4594 //      than the minimum period may have passed, and the abstime may be stale (in the
4595 //      past) resultin in a hang. Using this technique reduces the odds of a hang
4596 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
4597 //
4598 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4599 //      of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4600 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4601 //      reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4602 //      thread.
4603 //
4604 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
4605 //      that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4606 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
4607 //      This also works well. In fact it avoids kernel-level scalability impediments
4608 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
4609 //      timers in a graceful fashion.
4610 //
4611 // 4.   When the abstime value is in the past it appears that control returns
4612 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4613 //      Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4614 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
4615 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
4616 //      It may be possible to avoid reinitialization by checking the return
4617 //      value from pthread_cond_timedwait(). In addition to reinitializing the
4618 //      condvar we must establish the invariant that cond_signal() is only called
4619 //      within critical sections protected by the adjunct mutex. This prevents
4620 //      cond_signal() from "seeing" a condvar that's in the midst of being
4621 //      reinitialized or that is corrupt. Sadly, this invariant obviates the
4622 //      desirable signal-after-unlock optimization that avoids futile context switching.
4623 //
4624 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
4625 //      structure when a condvar is used or initialized. cond_destroy() would
4626 //      release the helper structure. Our reinitialize-after-timedwait fix
4627 //      put excessive stress on malloc/free and locks protecting the c-heap.
4628 //
4629 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4630 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4631 // and only enabling the work-around for vulnerable environments.
4632 
4633 // utility to compute the abstime argument to timedwait:
4634 // millis is the relative timeout time
4635 // abstime will be the absolute timeout time
4636 // TODO: replace compute_abstime() with unpackTime()
4637 
4638 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4639   if (millis < 0) millis = 0;
4640   struct timeval now;
4641   int status = gettimeofday(&now, NULL);
4642   assert(status == 0, "gettimeofday");
4643   jlong seconds = millis / 1000;
4644   millis %= 1000;
4645   if (seconds > 50000000) { // see man cond_timedwait(3T)
4646     seconds = 50000000;
4647   }
4648   abstime->tv_sec = now.tv_sec  + seconds;
4649   long       usec = now.tv_usec + millis * 1000;
4650   if (usec >= 1000000) {
4651     abstime->tv_sec += 1;
4652     usec -= 1000000;
4653   }
4654   abstime->tv_nsec = usec * 1000;
4655   return abstime;
4656 }
4657 
4658 
4659 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4660 // Conceptually TryPark() should be equivalent to park(0).
4661 
4662 int os::PlatformEvent::TryPark() {
4663   for (;;) {
4664     const int v = _Event;
4665     guarantee ((v == 0) || (v == 1), "invariant");
4666     if (Atomic::cmpxchg (0, &_Event, v) == v) return v;
4667   }
4668 }
4669 
4670 void os::PlatformEvent::park() {       // AKA "down()"
4671   // Invariant: Only the thread associated with the Event/PlatformEvent
4672   // may call park().
4673   // TODO: assert that _Assoc != NULL or _Assoc == Self
4674   int v;
4675   for (;;) {
4676     v = _Event;
4677     if (Atomic::cmpxchg (v-1, &_Event, v) == v) break;
4678   }
4679   guarantee (v >= 0, "invariant");
4680   if (v == 0) {
4681     // Do this the hard way by blocking ...
4682     int status = pthread_mutex_lock(_mutex);
4683     assert_status(status == 0, status, "mutex_lock");
4684     guarantee (_nParked == 0, "invariant");
4685     ++ _nParked;
4686     while (_Event < 0) {
4687       status = pthread_cond_wait(_cond, _mutex);
4688       assert_status(status == 0 || status == ETIMEDOUT, status, "cond_timedwait");
4689     }
4690     -- _nParked;
4691 
4692     // In theory we could move the ST of 0 into _Event past the unlock(),
4693     // but then we'd need a MEMBAR after the ST.
4694     _Event = 0;
4695     status = pthread_mutex_unlock(_mutex);
4696     assert_status(status == 0, status, "mutex_unlock");
4697   }
4698   guarantee (_Event >= 0, "invariant");
4699 }
4700 
4701 int os::PlatformEvent::park(jlong millis) {
4702   guarantee (_nParked == 0, "invariant");
4703 
4704   int v;
4705   for (;;) {
4706     v = _Event;
4707     if (Atomic::cmpxchg (v-1, &_Event, v) == v) break;
4708   }
4709   guarantee (v >= 0, "invariant");
4710   if (v != 0) return OS_OK;
4711 
4712   // We do this the hard way, by blocking the thread.
4713   // Consider enforcing a minimum timeout value.
4714   struct timespec abst;
4715   compute_abstime(&abst, millis);
4716 
4717   int ret = OS_TIMEOUT;
4718   int status = pthread_mutex_lock(_mutex);
4719   assert_status(status == 0, status, "mutex_lock");
4720   guarantee (_nParked == 0, "invariant");
4721   ++_nParked;
4722 
4723   // Object.wait(timo) will return because of
4724   // (a) notification
4725   // (b) timeout
4726   // (c) thread.interrupt
4727   //
4728   // Thread.interrupt and object.notify{All} both call Event::set.
4729   // That is, we treat thread.interrupt as a special case of notification.
4730   // The underlying Solaris implementation, cond_timedwait, admits
4731   // spurious/premature wakeups, but the JLS/JVM spec prevents the
4732   // JVM from making those visible to Java code. As such, we must
4733   // filter out spurious wakeups. We assume all ETIME returns are valid.
4734   //
4735   // TODO: properly differentiate simultaneous notify+interrupt.
4736   // In that case, we should propagate the notify to another waiter.
4737 
4738   while (_Event < 0) {
4739     status = pthread_cond_timedwait(_cond, _mutex, &abst);
4740     assert_status(status == 0 || status == ETIMEDOUT,
4741           status, "cond_timedwait");
4742     if (!FilterSpuriousWakeups) break;         // previous semantics
4743     if (status == ETIMEDOUT) break;
4744     // We consume and ignore EINTR and spurious wakeups.
4745   }
4746   --_nParked;
4747   if (_Event >= 0) {
4748      ret = OS_OK;
4749   }
4750   _Event = 0;
4751   status = pthread_mutex_unlock(_mutex);
4752   assert_status(status == 0, status, "mutex_unlock");
4753   assert (_nParked == 0, "invariant");
4754   return ret;
4755 }
4756 
4757 void os::PlatformEvent::unpark() {
4758   int v, AnyWaiters;
4759   for (;;) {
4760     v = _Event;
4761     if (v > 0) {
4762       // The LD of _Event could have reordered or be satisfied
4763       // by a read-aside from this processor's write buffer.
4764       // To avoid problems execute a barrier and then
4765       // ratify the value.
4766       OrderAccess::fence();
4767       if (_Event == v) return;
4768       continue;
4769     }
4770     if (Atomic::cmpxchg (v+1, &_Event, v) == v) break;
4771   }
4772   if (v < 0) {
4773     // Wait for the thread associated with the event to vacate
4774     int status = pthread_mutex_lock(_mutex);
4775     assert_status(status == 0, status, "mutex_lock");
4776     AnyWaiters = _nParked;
4777 
4778     if (AnyWaiters != 0) {
4779       // We intentional signal *after* dropping the lock
4780       // to avoid a common class of futile wakeups.
4781       status = pthread_cond_signal(_cond);
4782       assert_status(status == 0, status, "cond_signal");
4783     }
4784     // Mutex should be locked for pthread_cond_signal(_cond).
4785     status = pthread_mutex_unlock(_mutex);
4786     assert_status(status == 0, status, "mutex_unlock");
4787   }
4788 
4789   // Note that we signal() _after dropping the lock for "immortal" Events.
4790   // This is safe and avoids a common class of futile wakeups. In rare
4791   // circumstances this can cause a thread to return prematurely from
4792   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4793   // simply re-test the condition and re-park itself.
4794 }
4795 
4796 
4797 // JSR166
4798 // -------------------------------------------------------
4799 
4800 //
4801 // The solaris and linux implementations of park/unpark are fairly
4802 // conservative for now, but can be improved. They currently use a
4803 // mutex/condvar pair, plus a a count.
4804 // Park decrements count if > 0, else does a condvar wait. Unpark
4805 // sets count to 1 and signals condvar. Only one thread ever waits
4806 // on the condvar. Contention seen when trying to park implies that someone
4807 // is unparking you, so don't wait. And spurious returns are fine, so there
4808 // is no need to track notifications.
4809 //
4810 
4811 #define MAX_SECS 100000000
4812 //
4813 // This code is common to linux and solaris and will be moved to a
4814 // common place in dolphin.
4815 //
4816 // The passed in time value is either a relative time in nanoseconds
4817 // or an absolute time in milliseconds. Either way it has to be unpacked
4818 // into suitable seconds and nanoseconds components and stored in the
4819 // given timespec structure.
4820 // Given time is a 64-bit value and the time_t used in the timespec is only
4821 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
4822 // overflow if times way in the future are given. Further on Solaris versions
4823 // prior to 10 there is a restriction (see cond_timedwait) that the specified
4824 // number of seconds, in abstime, is less than current_time + 100,000,000.
4825 // As it will be 28 years before "now + 100000000" will overflow we can
4826 // ignore overflow and just impose a hard-limit on seconds using the value
4827 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
4828 // years from "now".
4829 //
4830 
4831 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4832   assert (time > 0, "convertTime");
4833 
4834   struct timeval now;
4835   int status = gettimeofday(&now, NULL);
4836   assert(status == 0, "gettimeofday");
4837 
4838   time_t max_secs = now.tv_sec + MAX_SECS;
4839 
4840   if (isAbsolute) {
4841     jlong secs = time / 1000;
4842     if (secs > max_secs) {
4843       absTime->tv_sec = max_secs;
4844     }
4845     else {
4846       absTime->tv_sec = secs;
4847     }
4848     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4849   }
4850   else {
4851     jlong secs = time / NANOSECS_PER_SEC;
4852     if (secs >= MAX_SECS) {
4853       absTime->tv_sec = max_secs;
4854       absTime->tv_nsec = 0;
4855     }
4856     else {
4857       absTime->tv_sec = now.tv_sec + secs;
4858       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4859       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4860         absTime->tv_nsec -= NANOSECS_PER_SEC;
4861         ++absTime->tv_sec; // note: this must be <= max_secs
4862       }
4863     }
4864   }
4865   assert(absTime->tv_sec >= 0, "tv_sec < 0");
4866   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4867   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4868   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4869 }
4870 
4871 void Parker::park(bool isAbsolute, jlong time) {
4872   // Optional fast-path check:
4873   // Return immediately if a permit is available.
4874   if (_counter > 0) {
4875       _counter = 0;
4876       OrderAccess::fence();
4877       return;
4878   }
4879 
4880   Thread* thread = Thread::current();
4881   assert(thread->is_Java_thread(), "Must be JavaThread");
4882   JavaThread *jt = (JavaThread *)thread;
4883 
4884   // Optional optimization -- avoid state transitions if there's an interrupt pending.
4885   // Check interrupt before trying to wait
4886   if (Thread::is_interrupted(thread, false)) {
4887     return;
4888   }
4889 
4890   // Next, demultiplex/decode time arguments
4891   timespec absTime;
4892   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
4893     return;
4894   }
4895   if (time > 0) {
4896     unpackTime(&absTime, isAbsolute, time);
4897   }
4898 
4899 
4900   // Enter safepoint region
4901   // Beware of deadlocks such as 6317397.
4902   // The per-thread Parker:: mutex is a classic leaf-lock.
4903   // In particular a thread must never block on the Threads_lock while
4904   // holding the Parker:: mutex. If safepoints are pending both the
4905   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4906   ThreadBlockInVM tbivm(jt);
4907 
4908   // Don't wait if cannot get lock since interference arises from
4909   // unblocking. Also. check interrupt before trying wait
4910   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4911     return;
4912   }
4913 
4914   int status;
4915   if (_counter > 0) { // no wait needed
4916     _counter = 0;
4917     status = pthread_mutex_unlock(_mutex);
4918     assert (status == 0, "invariant");
4919     OrderAccess::fence();
4920     return;
4921   }
4922 
4923 #ifdef ASSERT
4924   // Don't catch signals while blocked; let the running threads have the signals.
4925   // (This allows a debugger to break into the running thread.)
4926   sigset_t oldsigs;
4927   sigset_t* allowdebug_blocked = os::Aix::allowdebug_blocked_signals();
4928   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4929 #endif
4930 
4931   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4932   jt->set_suspend_equivalent();
4933   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4934 
4935   if (time == 0) {
4936     status = pthread_cond_wait (_cond, _mutex);
4937   } else {
4938     status = pthread_cond_timedwait (_cond, _mutex, &absTime);
4939     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4940       pthread_cond_destroy (_cond);
4941       pthread_cond_init    (_cond, NULL);
4942     }
4943   }
4944   assert_status(status == 0 || status == EINTR ||
4945                 status == ETIME || status == ETIMEDOUT,
4946                 status, "cond_timedwait");
4947 
4948 #ifdef ASSERT
4949   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4950 #endif
4951 
4952   _counter = 0;
4953   status = pthread_mutex_unlock(_mutex);
4954   assert_status(status == 0, status, "invariant");
4955   // If externally suspended while waiting, re-suspend
4956   if (jt->handle_special_suspend_equivalent_condition()) {
4957     jt->java_suspend_self();
4958   }
4959 
4960   OrderAccess::fence();
4961 }
4962 
4963 void Parker::unpark() {
4964   int s, status;
4965   status = pthread_mutex_lock(_mutex);
4966   assert (status == 0, "invariant");
4967   s = _counter;
4968   _counter = 1;
4969   if (s < 1) {
4970     if (WorkAroundNPTLTimedWaitHang) {
4971       status = pthread_cond_signal (_cond);
4972       assert (status == 0, "invariant");
4973       status = pthread_mutex_unlock(_mutex);
4974       assert (status == 0, "invariant");
4975     } else {
4976       status = pthread_mutex_unlock(_mutex);
4977       assert (status == 0, "invariant");
4978       status = pthread_cond_signal (_cond);
4979       assert (status == 0, "invariant");
4980     }
4981   } else {
4982     pthread_mutex_unlock(_mutex);
4983     assert (status == 0, "invariant");
4984   }
4985 }
4986 
4987 
4988 extern char** environ;
4989 
4990 // Run the specified command in a separate process. Return its exit value,
4991 // or -1 on failure (e.g. can't fork a new process).
4992 // Unlike system(), this function can be called from signal handler. It
4993 // doesn't block SIGINT et al.
4994 int os::fork_and_exec(char* cmd) {
4995   char * argv[4] = {"sh", "-c", cmd, NULL};
4996 
4997   pid_t pid = fork();
4998 
4999   if (pid < 0) {
5000     // fork failed
5001     return -1;
5002 
5003   } else if (pid == 0) {
5004     // child process
5005 
5006     // try to be consistent with system(), which uses "/usr/bin/sh" on AIX
5007     execve("/usr/bin/sh", argv, environ);
5008 
5009     // execve failed
5010     _exit(-1);
5011 
5012   } else  {
5013     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5014     // care about the actual exit code, for now.
5015 
5016     int status;
5017 
5018     // Wait for the child process to exit.  This returns immediately if
5019     // the child has already exited. */
5020     while (waitpid(pid, &status, 0) < 0) {
5021         switch (errno) {
5022         case ECHILD: return 0;
5023         case EINTR: break;
5024         default: return -1;
5025         }
5026     }
5027 
5028     if (WIFEXITED(status)) {
5029        // The child exited normally; get its exit code.
5030        return WEXITSTATUS(status);
5031     } else if (WIFSIGNALED(status)) {
5032        // The child exited because of a signal
5033        // The best value to return is 0x80 + signal number,
5034        // because that is what all Unix shells do, and because
5035        // it allows callers to distinguish between process exit and
5036        // process death by signal.
5037        return 0x80 + WTERMSIG(status);
5038     } else {
5039        // Unknown exit code; pass it through
5040        return status;
5041     }
5042   }
5043   // Remove warning.
5044   return -1;
5045 }
5046 
5047 // is_headless_jre()
5048 //
5049 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5050 // in order to report if we are running in a headless jre.
5051 //
5052 // Since JDK8 xawt/libmawt.so is moved into the same directory
5053 // as libawt.so, and renamed libawt_xawt.so
5054 bool os::is_headless_jre() {
5055   struct stat statbuf;
5056   char buf[MAXPATHLEN];
5057   char libmawtpath[MAXPATHLEN];
5058   const char *xawtstr  = "/xawt/libmawt.so";
5059   const char *new_xawtstr = "/libawt_xawt.so";
5060 
5061   char *p;
5062 
5063   // Get path to libjvm.so
5064   os::jvm_path(buf, sizeof(buf));
5065 
5066   // Get rid of libjvm.so
5067   p = strrchr(buf, '/');
5068   if (p == NULL) return false;
5069   else *p = '\0';
5070 
5071   // Get rid of client or server
5072   p = strrchr(buf, '/');
5073   if (p == NULL) return false;
5074   else *p = '\0';
5075 
5076   // check xawt/libmawt.so
5077   strcpy(libmawtpath, buf);
5078   strcat(libmawtpath, xawtstr);
5079   if (::stat(libmawtpath, &statbuf) == 0) return false;
5080 
5081   // check libawt_xawt.so
5082   strcpy(libmawtpath, buf);
5083   strcat(libmawtpath, new_xawtstr);
5084   if (::stat(libmawtpath, &statbuf) == 0) return false;
5085 
5086   return true;
5087 }
5088 
5089 // Get the default path to the core file
5090 // Returns the length of the string
5091 int os::get_core_path(char* buffer, size_t bufferSize) {
5092   const char* p = get_current_directory(buffer, bufferSize);
5093 
5094   if (p == NULL) {
5095     assert(p != NULL, "failed to get current directory");
5096     return 0;
5097   }
5098 
5099   return strlen(buffer);
5100 }
5101 
5102 #ifndef PRODUCT
5103 void TestReserveMemorySpecial_test() {
5104   // No tests available for this platform
5105 }
5106 #endif