1 /*
   2  * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2012, 2019 SAP SE. 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 "jvm.h"
  32 #include "classfile/classLoader.hpp"
  33 #include "classfile/systemDictionary.hpp"
  34 #include "classfile/vmSymbols.hpp"
  35 #include "code/icBuffer.hpp"
  36 #include "code/vtableStubs.hpp"
  37 #include "compiler/compileBroker.hpp"
  38 #include "interpreter/interpreter.hpp"
  39 #include "logging/log.hpp"
  40 #include "logging/logStream.hpp"
  41 #include "libo4.hpp"
  42 #include "libperfstat_aix.hpp"
  43 #include "libodm_aix.hpp"
  44 #include "loadlib_aix.hpp"
  45 #include "memory/allocation.inline.hpp"
  46 #include "memory/filemap.hpp"
  47 #include "misc_aix.hpp"
  48 #include "oops/oop.inline.hpp"
  49 #include "os_aix.inline.hpp"
  50 #include "os_share_aix.hpp"
  51 #include "porting_aix.hpp"
  52 #include "prims/jniFastGetField.hpp"
  53 #include "prims/jvm_misc.hpp"
  54 #include "runtime/arguments.hpp"
  55 #include "runtime/atomic.hpp"
  56 #include "runtime/extendedPC.hpp"
  57 #include "runtime/globals.hpp"
  58 #include "runtime/interfaceSupport.inline.hpp"
  59 #include "runtime/java.hpp"
  60 #include "runtime/javaCalls.hpp"
  61 #include "runtime/mutexLocker.hpp"
  62 #include "runtime/objectMonitor.hpp"
  63 #include "runtime/orderAccess.hpp"
  64 #include "runtime/os.hpp"
  65 #include "runtime/osThread.hpp"
  66 #include "runtime/perfMemory.hpp"
  67 #include "runtime/sharedRuntime.hpp"
  68 #include "runtime/statSampler.hpp"
  69 #include "runtime/stubRoutines.hpp"
  70 #include "runtime/thread.inline.hpp"
  71 #include "runtime/threadCritical.hpp"
  72 #include "runtime/timer.hpp"
  73 #include "runtime/vm_version.hpp"
  74 #include "services/attachListener.hpp"
  75 #include "services/runtimeService.hpp"
  76 #include "utilities/align.hpp"
  77 #include "utilities/decoder.hpp"
  78 #include "utilities/defaultStream.hpp"
  79 #include "utilities/events.hpp"
  80 #include "utilities/growableArray.hpp"
  81 #include "utilities/vmError.hpp"
  82 
  83 // put OS-includes here (sorted alphabetically)
  84 #include <errno.h>
  85 #include <fcntl.h>
  86 #include <inttypes.h>
  87 #include <poll.h>
  88 #include <procinfo.h>
  89 #include <pthread.h>
  90 #include <pwd.h>
  91 #include <semaphore.h>
  92 #include <signal.h>
  93 #include <stdint.h>
  94 #include <stdio.h>
  95 #include <string.h>
  96 #include <unistd.h>
  97 #include <sys/ioctl.h>
  98 #include <sys/ipc.h>
  99 #include <sys/mman.h>
 100 #include <sys/resource.h>
 101 #include <sys/select.h>
 102 #include <sys/shm.h>
 103 #include <sys/socket.h>
 104 #include <sys/stat.h>
 105 #include <sys/sysinfo.h>
 106 #include <sys/systemcfg.h>
 107 #include <sys/time.h>
 108 #include <sys/times.h>
 109 #include <sys/types.h>
 110 #include <sys/utsname.h>
 111 #include <sys/vminfo.h>
 112 #include <sys/wait.h>
 113 
 114 // Missing prototypes for various system APIs.
 115 extern "C"
 116 int mread_real_time(timebasestruct_t *t, size_t size_of_timebasestruct_t);
 117 
 118 #if !defined(_AIXVERSION_610)
 119 extern "C" int getthrds64(pid_t, struct thrdentry64*, int, tid64_t*, int);
 120 extern "C" int getprocs64(procentry64*, int, fdsinfo*, int, pid_t*, int);
 121 extern "C" int getargs(procsinfo*, int, char*, int);
 122 #endif
 123 
 124 #define MAX_PATH (2 * K)
 125 
 126 // for timer info max values which include all bits
 127 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 128 // for multipage initialization error analysis (in 'g_multipage_error')
 129 #define ERROR_MP_OS_TOO_OLD                          100
 130 #define ERROR_MP_EXTSHM_ACTIVE                       101
 131 #define ERROR_MP_VMGETINFO_FAILED                    102
 132 #define ERROR_MP_VMGETINFO_CLAIMS_NO_SUPPORT_FOR_64K 103
 133 
 134 // excerpts from systemcfg.h that might be missing on older os levels
 135 #ifndef PV_5_Compat
 136   #define PV_5_Compat 0x0F8000   /* Power PC 5 */
 137 #endif
 138 #ifndef PV_6
 139   #define PV_6 0x100000          /* Power PC 6 */
 140 #endif
 141 #ifndef PV_6_1
 142   #define PV_6_1 0x100001        /* Power PC 6 DD1.x */
 143 #endif
 144 #ifndef PV_6_Compat
 145   #define PV_6_Compat 0x108000   /* Power PC 6 */
 146 #endif
 147 #ifndef PV_7
 148   #define PV_7 0x200000          /* Power PC 7 */
 149 #endif
 150 #ifndef PV_7_Compat
 151   #define PV_7_Compat 0x208000   /* Power PC 7 */
 152 #endif
 153 #ifndef PV_8
 154   #define PV_8 0x300000          /* Power PC 8 */
 155 #endif
 156 #ifndef PV_8_Compat
 157   #define PV_8_Compat 0x308000   /* Power PC 8 */
 158 #endif
 159 
 160 static address resolve_function_descriptor_to_code_pointer(address p);
 161 
 162 static void vmembk_print_on(outputStream* os);
 163 
 164 ////////////////////////////////////////////////////////////////////////////////
 165 // global variables (for a description see os_aix.hpp)
 166 
 167 julong    os::Aix::_physical_memory = 0;
 168 
 169 pthread_t os::Aix::_main_thread = ((pthread_t)0);
 170 int       os::Aix::_page_size = -1;
 171 
 172 // -1 = uninitialized, 0 if AIX, 1 if OS/400 pase
 173 int       os::Aix::_on_pase = -1;
 174 
 175 // 0 = uninitialized, otherwise 32 bit number:
 176 //  0xVVRRTTSS
 177 //  VV - major version
 178 //  RR - minor version
 179 //  TT - tech level, if known, 0 otherwise
 180 //  SS - service pack, if known, 0 otherwise
 181 uint32_t  os::Aix::_os_version = 0;
 182 
 183 // -1 = uninitialized, 0 - no, 1 - yes
 184 int       os::Aix::_xpg_sus_mode = -1;
 185 
 186 // -1 = uninitialized, 0 - no, 1 - yes
 187 int       os::Aix::_extshm = -1;
 188 
 189 ////////////////////////////////////////////////////////////////////////////////
 190 // local variables
 191 
 192 static volatile jlong max_real_time = 0;
 193 static jlong    initial_time_count = 0;
 194 static int      clock_tics_per_sec = 100;
 195 static sigset_t check_signal_done;         // For diagnostics to print a message once (see run_periodic_checks)
 196 static bool     check_signals      = true;
 197 static int      SR_signum          = SIGUSR2; // Signal used to suspend/resume a thread (must be > SIGSEGV, see 4355769)
 198 static sigset_t SR_sigset;
 199 
 200 // Process break recorded at startup.
 201 static address g_brk_at_startup = NULL;
 202 
 203 // This describes the state of multipage support of the underlying
 204 // OS. Note that this is of no interest to the outsize world and
 205 // therefore should not be defined in AIX class.
 206 //
 207 // AIX supports four different page sizes - 4K, 64K, 16MB, 16GB. The
 208 // latter two (16M "large" resp. 16G "huge" pages) require special
 209 // setup and are normally not available.
 210 //
 211 // AIX supports multiple page sizes per process, for:
 212 //  - Stack (of the primordial thread, so not relevant for us)
 213 //  - Data - data, bss, heap, for us also pthread stacks
 214 //  - Text - text code
 215 //  - shared memory
 216 //
 217 // Default page sizes can be set via linker options (-bdatapsize, -bstacksize, ...)
 218 // and via environment variable LDR_CNTRL (DATAPSIZE, STACKPSIZE, ...).
 219 //
 220 // For shared memory, page size can be set dynamically via
 221 // shmctl(). Different shared memory regions can have different page
 222 // sizes.
 223 //
 224 // More information can be found at AIBM info center:
 225 //   http://publib.boulder.ibm.com/infocenter/aix/v6r1/index.jsp?topic=/com.ibm.aix.prftungd/doc/prftungd/multiple_page_size_app_support.htm
 226 //
 227 static struct {
 228   size_t pagesize;            // sysconf _SC_PAGESIZE (4K)
 229   size_t datapsize;           // default data page size (LDR_CNTRL DATAPSIZE)
 230   size_t shmpsize;            // default shared memory page size (LDR_CNTRL SHMPSIZE)
 231   size_t pthr_stack_pagesize; // stack page size of pthread threads
 232   size_t textpsize;           // default text page size (LDR_CNTRL STACKPSIZE)
 233   bool can_use_64K_pages;     // True if we can alloc 64K pages dynamically with Sys V shm.
 234   bool can_use_16M_pages;     // True if we can alloc 16M pages dynamically with Sys V shm.
 235   int error;                  // Error describing if something went wrong at multipage init.
 236 } g_multipage_support = {
 237   (size_t) -1,
 238   (size_t) -1,
 239   (size_t) -1,
 240   (size_t) -1,
 241   (size_t) -1,
 242   false, false,
 243   0
 244 };
 245 
 246 // We must not accidentally allocate memory close to the BRK - even if
 247 // that would work - because then we prevent the BRK segment from
 248 // growing which may result in a malloc OOM even though there is
 249 // enough memory. The problem only arises if we shmat() or mmap() at
 250 // a specific wish address, e.g. to place the heap in a
 251 // compressed-oops-friendly way.
 252 static bool is_close_to_brk(address a) {
 253   assert0(g_brk_at_startup != NULL);
 254   if (a >= g_brk_at_startup &&
 255       a < (g_brk_at_startup + MaxExpectedDataSegmentSize)) {
 256     return true;
 257   }
 258   return false;
 259 }
 260 
 261 julong os::available_memory() {
 262   return Aix::available_memory();
 263 }
 264 
 265 julong os::Aix::available_memory() {
 266   // Avoid expensive API call here, as returned value will always be null.
 267   if (os::Aix::on_pase()) {
 268     return 0x0LL;
 269   }
 270   os::Aix::meminfo_t mi;
 271   if (os::Aix::get_meminfo(&mi)) {
 272     return mi.real_free;
 273   } else {
 274     return ULONG_MAX;
 275   }
 276 }
 277 
 278 julong os::physical_memory() {
 279   return Aix::physical_memory();
 280 }
 281 
 282 // Return true if user is running as root.
 283 
 284 bool os::have_special_privileges() {
 285   static bool init = false;
 286   static bool privileges = false;
 287   if (!init) {
 288     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 289     init = true;
 290   }
 291   return privileges;
 292 }
 293 
 294 // Helper function, emulates disclaim64 using multiple 32bit disclaims
 295 // because we cannot use disclaim64() on AS/400 and old AIX releases.
 296 static bool my_disclaim64(char* addr, size_t size) {
 297 
 298   if (size == 0) {
 299     return true;
 300   }
 301 
 302   // Maximum size 32bit disclaim() accepts. (Theoretically 4GB, but I just do not trust that.)
 303   const unsigned int maxDisclaimSize = 0x40000000;
 304 
 305   const unsigned int numFullDisclaimsNeeded = (size / maxDisclaimSize);
 306   const unsigned int lastDisclaimSize = (size % maxDisclaimSize);
 307 
 308   char* p = addr;
 309 
 310   for (int i = 0; i < numFullDisclaimsNeeded; i ++) {
 311     if (::disclaim(p, maxDisclaimSize, DISCLAIM_ZEROMEM) != 0) {
 312       trcVerbose("Cannot disclaim %p - %p (errno %d)\n", p, p + maxDisclaimSize, errno);
 313       return false;
 314     }
 315     p += maxDisclaimSize;
 316   }
 317 
 318   if (lastDisclaimSize > 0) {
 319     if (::disclaim(p, lastDisclaimSize, DISCLAIM_ZEROMEM) != 0) {
 320       trcVerbose("Cannot disclaim %p - %p (errno %d)\n", p, p + lastDisclaimSize, errno);
 321       return false;
 322     }
 323   }
 324 
 325   return true;
 326 }
 327 
 328 // Cpu architecture string
 329 #if defined(PPC32)
 330 static char cpu_arch[] = "ppc";
 331 #elif defined(PPC64)
 332 static char cpu_arch[] = "ppc64";
 333 #else
 334 #error Add appropriate cpu_arch setting
 335 #endif
 336 
 337 // Wrap the function "vmgetinfo" which is not available on older OS releases.
 338 static int checked_vmgetinfo(void *out, int command, int arg) {
 339   if (os::Aix::on_pase() && os::Aix::os_version_short() < 0x0601) {
 340     guarantee(false, "cannot call vmgetinfo on AS/400 older than V6R1");
 341   }
 342   return ::vmgetinfo(out, command, arg);
 343 }
 344 
 345 // Given an address, returns the size of the page backing that address.
 346 size_t os::Aix::query_pagesize(void* addr) {
 347 
 348   if (os::Aix::on_pase() && os::Aix::os_version_short() < 0x0601) {
 349     // AS/400 older than V6R1: no vmgetinfo here, default to 4K
 350     return 4*K;
 351   }
 352 
 353   vm_page_info pi;
 354   pi.addr = (uint64_t)addr;
 355   if (checked_vmgetinfo(&pi, VM_PAGE_INFO, sizeof(pi)) == 0) {
 356     return pi.pagesize;
 357   } else {
 358     assert(false, "vmgetinfo failed to retrieve page size");
 359     return 4*K;
 360   }
 361 }
 362 
 363 void os::Aix::initialize_system_info() {
 364 
 365   // Get the number of online(logical) cpus instead of configured.
 366   os::_processor_count = sysconf(_SC_NPROCESSORS_ONLN);
 367   assert(_processor_count > 0, "_processor_count must be > 0");
 368 
 369   // Retrieve total physical storage.
 370   os::Aix::meminfo_t mi;
 371   if (!os::Aix::get_meminfo(&mi)) {
 372     assert(false, "os::Aix::get_meminfo failed.");
 373   }
 374   _physical_memory = (julong) mi.real_total;
 375 }
 376 
 377 // Helper function for tracing page sizes.
 378 static const char* describe_pagesize(size_t pagesize) {
 379   switch (pagesize) {
 380     case 4*K : return "4K";
 381     case 64*K: return "64K";
 382     case 16*M: return "16M";
 383     case 16*G: return "16G";
 384     default:
 385       assert(false, "surprise");
 386       return "??";
 387   }
 388 }
 389 
 390 // Probe OS for multipage support.
 391 // Will fill the global g_multipage_support structure.
 392 // Must be called before calling os::large_page_init().
 393 static void query_multipage_support() {
 394 
 395   guarantee(g_multipage_support.pagesize == -1,
 396             "do not call twice");
 397 
 398   g_multipage_support.pagesize = ::sysconf(_SC_PAGESIZE);
 399 
 400   // This really would surprise me.
 401   assert(g_multipage_support.pagesize == 4*K, "surprise!");
 402 
 403   // Query default data page size (default page size for C-Heap, pthread stacks and .bss).
 404   // Default data page size is defined either by linker options (-bdatapsize)
 405   // or by environment variable LDR_CNTRL (suboption DATAPSIZE). If none is given,
 406   // default should be 4K.
 407   {
 408     void* p = ::malloc(16*M);
 409     g_multipage_support.datapsize = os::Aix::query_pagesize(p);
 410     ::free(p);
 411   }
 412 
 413   // Query default shm page size (LDR_CNTRL SHMPSIZE).
 414   // Note that this is pure curiosity. We do not rely on default page size but set
 415   // our own page size after allocated.
 416   {
 417     const int shmid = ::shmget(IPC_PRIVATE, 1, IPC_CREAT | S_IRUSR | S_IWUSR);
 418     guarantee(shmid != -1, "shmget failed");
 419     void* p = ::shmat(shmid, NULL, 0);
 420     ::shmctl(shmid, IPC_RMID, NULL);
 421     guarantee(p != (void*) -1, "shmat failed");
 422     g_multipage_support.shmpsize = os::Aix::query_pagesize(p);
 423     ::shmdt(p);
 424   }
 425 
 426   // Before querying the stack page size, make sure we are not running as primordial
 427   // thread (because primordial thread's stack may have different page size than
 428   // pthread thread stacks). Running a VM on the primordial thread won't work for a
 429   // number of reasons so we may just as well guarantee it here.
 430   guarantee0(!os::is_primordial_thread());
 431 
 432   // Query pthread stack page size. Should be the same as data page size because
 433   // pthread stacks are allocated from C-Heap.
 434   {
 435     int dummy = 0;
 436     g_multipage_support.pthr_stack_pagesize = os::Aix::query_pagesize(&dummy);
 437   }
 438 
 439   // Query default text page size (LDR_CNTRL TEXTPSIZE).
 440   {
 441     address any_function =
 442       resolve_function_descriptor_to_code_pointer((address)describe_pagesize);
 443     g_multipage_support.textpsize = os::Aix::query_pagesize(any_function);
 444   }
 445 
 446   // Now probe for support of 64K pages and 16M pages.
 447 
 448   // Before OS/400 V6R1, there is no support for pages other than 4K.
 449   if (os::Aix::on_pase_V5R4_or_older()) {
 450     trcVerbose("OS/400 < V6R1 - no large page support.");
 451     g_multipage_support.error = ERROR_MP_OS_TOO_OLD;
 452     goto query_multipage_support_end;
 453   }
 454 
 455   // Now check which page sizes the OS claims it supports, and of those, which actually can be used.
 456   {
 457     const int MAX_PAGE_SIZES = 4;
 458     psize_t sizes[MAX_PAGE_SIZES];
 459     const int num_psizes = checked_vmgetinfo(sizes, VMINFO_GETPSIZES, MAX_PAGE_SIZES);
 460     if (num_psizes == -1) {
 461       trcVerbose("vmgetinfo(VMINFO_GETPSIZES) failed (errno: %d)", errno);
 462       trcVerbose("disabling multipage support.");
 463       g_multipage_support.error = ERROR_MP_VMGETINFO_FAILED;
 464       goto query_multipage_support_end;
 465     }
 466     guarantee(num_psizes > 0, "vmgetinfo(.., VMINFO_GETPSIZES, ...) failed.");
 467     assert(num_psizes <= MAX_PAGE_SIZES, "Surprise! more than 4 page sizes?");
 468     trcVerbose("vmgetinfo(.., VMINFO_GETPSIZES, ...) returns %d supported page sizes: ", num_psizes);
 469     for (int i = 0; i < num_psizes; i ++) {
 470       trcVerbose(" %s ", describe_pagesize(sizes[i]));
 471     }
 472 
 473     // Can we use 64K, 16M pages?
 474     for (int i = 0; i < num_psizes; i ++) {
 475       const size_t pagesize = sizes[i];
 476       if (pagesize != 64*K && pagesize != 16*M) {
 477         continue;
 478       }
 479       bool can_use = false;
 480       trcVerbose("Probing support for %s pages...", describe_pagesize(pagesize));
 481       const int shmid = ::shmget(IPC_PRIVATE, pagesize,
 482         IPC_CREAT | S_IRUSR | S_IWUSR);
 483       guarantee0(shmid != -1); // Should always work.
 484       // Try to set pagesize.
 485       struct shmid_ds shm_buf = { 0 };
 486       shm_buf.shm_pagesize = pagesize;
 487       if (::shmctl(shmid, SHM_PAGESIZE, &shm_buf) != 0) {
 488         const int en = errno;
 489         ::shmctl(shmid, IPC_RMID, NULL); // As early as possible!
 490         trcVerbose("shmctl(SHM_PAGESIZE) failed with errno=%n",
 491           errno);
 492       } else {
 493         // Attach and double check pageisze.
 494         void* p = ::shmat(shmid, NULL, 0);
 495         ::shmctl(shmid, IPC_RMID, NULL); // As early as possible!
 496         guarantee0(p != (void*) -1); // Should always work.
 497         const size_t real_pagesize = os::Aix::query_pagesize(p);
 498         if (real_pagesize != pagesize) {
 499           trcVerbose("real page size (0x%llX) differs.", real_pagesize);
 500         } else {
 501           can_use = true;
 502         }
 503         ::shmdt(p);
 504       }
 505       trcVerbose("Can use: %s", (can_use ? "yes" : "no"));
 506       if (pagesize == 64*K) {
 507         g_multipage_support.can_use_64K_pages = can_use;
 508       } else if (pagesize == 16*M) {
 509         g_multipage_support.can_use_16M_pages = can_use;
 510       }
 511     }
 512 
 513   } // end: check which pages can be used for shared memory
 514 
 515 query_multipage_support_end:
 516 
 517   trcVerbose("base page size (sysconf _SC_PAGESIZE): %s",
 518       describe_pagesize(g_multipage_support.pagesize));
 519   trcVerbose("Data page size (C-Heap, bss, etc): %s",
 520       describe_pagesize(g_multipage_support.datapsize));
 521   trcVerbose("Text page size: %s",
 522       describe_pagesize(g_multipage_support.textpsize));
 523   trcVerbose("Thread stack page size (pthread): %s",
 524       describe_pagesize(g_multipage_support.pthr_stack_pagesize));
 525   trcVerbose("Default shared memory page size: %s",
 526       describe_pagesize(g_multipage_support.shmpsize));
 527   trcVerbose("Can use 64K pages dynamically with shared meory: %s",
 528       (g_multipage_support.can_use_64K_pages ? "yes" :"no"));
 529   trcVerbose("Can use 16M pages dynamically with shared memory: %s",
 530       (g_multipage_support.can_use_16M_pages ? "yes" :"no"));
 531   trcVerbose("Multipage error details: %d",
 532       g_multipage_support.error);
 533 
 534   // sanity checks
 535   assert0(g_multipage_support.pagesize == 4*K);
 536   assert0(g_multipage_support.datapsize == 4*K || g_multipage_support.datapsize == 64*K);
 537   assert0(g_multipage_support.textpsize == 4*K || g_multipage_support.textpsize == 64*K);
 538   assert0(g_multipage_support.pthr_stack_pagesize == g_multipage_support.datapsize);
 539   assert0(g_multipage_support.shmpsize == 4*K || g_multipage_support.shmpsize == 64*K);
 540 
 541 }
 542 
 543 void os::init_system_properties_values() {
 544 
 545 #ifndef OVERRIDE_LIBPATH
 546   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 547 #else
 548   #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
 549 #endif
 550 #define EXTENSIONS_DIR  "/lib/ext"
 551 
 552   // Buffer that fits several sprintfs.
 553   // Note that the space for the trailing null is provided
 554   // by the nulls included by the sizeof operator.
 555   const size_t bufsize =
 556     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 557          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR)); // extensions dir
 558   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 559 
 560   // sysclasspath, java_home, dll_dir
 561   {
 562     char *pslash;
 563     os::jvm_path(buf, bufsize);
 564 
 565     // Found the full path to libjvm.so.
 566     // Now cut the path to <java_home>/jre if we can.
 567     pslash = strrchr(buf, '/');
 568     if (pslash != NULL) {
 569       *pslash = '\0';            // Get rid of /libjvm.so.
 570     }
 571     pslash = strrchr(buf, '/');
 572     if (pslash != NULL) {
 573       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 574     }
 575     Arguments::set_dll_dir(buf);
 576 
 577     if (pslash != NULL) {
 578       pslash = strrchr(buf, '/');
 579       if (pslash != NULL) {
 580         *pslash = '\0';        // Get rid of /lib.
 581       }
 582     }
 583     Arguments::set_java_home(buf);
 584     if (!set_boot_path('/', ':')) {
 585       vm_exit_during_initialization("Failed setting boot class path.", NULL);
 586     }
 587   }
 588 
 589   // Where to look for native libraries.
 590 
 591   // On Aix we get the user setting of LIBPATH.
 592   // Eventually, all the library path setting will be done here.
 593   // Get the user setting of LIBPATH.
 594   const char *v = ::getenv("LIBPATH");
 595   const char *v_colon = ":";
 596   if (v == NULL) { v = ""; v_colon = ""; }
 597 
 598   // Concatenate user and invariant part of ld_library_path.
 599   // That's +1 for the colon and +1 for the trailing '\0'.
 600   char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, strlen(v) + 1 + sizeof(DEFAULT_LIBPATH) + 1, mtInternal);
 601   sprintf(ld_library_path, "%s%s" DEFAULT_LIBPATH, v, v_colon);
 602   Arguments::set_library_path(ld_library_path);
 603   FREE_C_HEAP_ARRAY(char, ld_library_path);
 604 
 605   // Extensions directories.
 606   sprintf(buf, "%s" EXTENSIONS_DIR, Arguments::get_java_home());
 607   Arguments::set_ext_dirs(buf);
 608 
 609   FREE_C_HEAP_ARRAY(char, buf);
 610 
 611 #undef DEFAULT_LIBPATH
 612 #undef EXTENSIONS_DIR
 613 }
 614 
 615 ////////////////////////////////////////////////////////////////////////////////
 616 // breakpoint support
 617 
 618 void os::breakpoint() {
 619   BREAKPOINT;
 620 }
 621 
 622 extern "C" void breakpoint() {
 623   // use debugger to set breakpoint here
 624 }
 625 
 626 ////////////////////////////////////////////////////////////////////////////////
 627 // signal support
 628 
 629 debug_only(static bool signal_sets_initialized = false);
 630 static sigset_t unblocked_sigs, vm_sigs;
 631 
 632 void os::Aix::signal_sets_init() {
 633   // Should also have an assertion stating we are still single-threaded.
 634   assert(!signal_sets_initialized, "Already initialized");
 635   // Fill in signals that are necessarily unblocked for all threads in
 636   // the VM. Currently, we unblock the following signals:
 637   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 638   //                         by -Xrs (=ReduceSignalUsage));
 639   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 640   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 641   // the dispositions or masks wrt these signals.
 642   // Programs embedding the VM that want to use the above signals for their
 643   // own purposes must, at this time, use the "-Xrs" option to prevent
 644   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 645   // (See bug 4345157, and other related bugs).
 646   // In reality, though, unblocking these signals is really a nop, since
 647   // these signals are not blocked by default.
 648   sigemptyset(&unblocked_sigs);
 649   sigaddset(&unblocked_sigs, SIGILL);
 650   sigaddset(&unblocked_sigs, SIGSEGV);
 651   sigaddset(&unblocked_sigs, SIGBUS);
 652   sigaddset(&unblocked_sigs, SIGFPE);
 653   sigaddset(&unblocked_sigs, SIGTRAP);
 654   sigaddset(&unblocked_sigs, SR_signum);
 655 
 656   if (!ReduceSignalUsage) {
 657    if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 658      sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 659    }
 660    if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 661      sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 662    }
 663    if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 664      sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 665    }
 666   }
 667   // Fill in signals that are blocked by all but the VM thread.
 668   sigemptyset(&vm_sigs);
 669   if (!ReduceSignalUsage)
 670     sigaddset(&vm_sigs, BREAK_SIGNAL);
 671   debug_only(signal_sets_initialized = true);
 672 }
 673 
 674 // These are signals that are unblocked while a thread is running Java.
 675 // (For some reason, they get blocked by default.)
 676 sigset_t* os::Aix::unblocked_signals() {
 677   assert(signal_sets_initialized, "Not initialized");
 678   return &unblocked_sigs;
 679 }
 680 
 681 // These are the signals that are blocked while a (non-VM) thread is
 682 // running Java. Only the VM thread handles these signals.
 683 sigset_t* os::Aix::vm_signals() {
 684   assert(signal_sets_initialized, "Not initialized");
 685   return &vm_sigs;
 686 }
 687 
 688 void os::Aix::hotspot_sigmask(Thread* thread) {
 689 
 690   //Save caller's signal mask before setting VM signal mask
 691   sigset_t caller_sigmask;
 692   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 693 
 694   OSThread* osthread = thread->osthread();
 695   osthread->set_caller_sigmask(caller_sigmask);
 696 
 697   pthread_sigmask(SIG_UNBLOCK, os::Aix::unblocked_signals(), NULL);
 698 
 699   if (!ReduceSignalUsage) {
 700     if (thread->is_VM_thread()) {
 701       // Only the VM thread handles BREAK_SIGNAL ...
 702       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 703     } else {
 704       // ... all other threads block BREAK_SIGNAL
 705       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 706     }
 707   }
 708 }
 709 
 710 // retrieve memory information.
 711 // Returns false if something went wrong;
 712 // content of pmi undefined in this case.
 713 bool os::Aix::get_meminfo(meminfo_t* pmi) {
 714 
 715   assert(pmi, "get_meminfo: invalid parameter");
 716 
 717   memset(pmi, 0, sizeof(meminfo_t));
 718 
 719   if (os::Aix::on_pase()) {
 720     // On PASE, use the libo4 porting library.
 721 
 722     unsigned long long virt_total = 0;
 723     unsigned long long real_total = 0;
 724     unsigned long long real_free = 0;
 725     unsigned long long pgsp_total = 0;
 726     unsigned long long pgsp_free = 0;
 727     if (libo4::get_memory_info(&virt_total, &real_total, &real_free, &pgsp_total, &pgsp_free)) {
 728       pmi->virt_total = virt_total;
 729       pmi->real_total = real_total;
 730       pmi->real_free = real_free;
 731       pmi->pgsp_total = pgsp_total;
 732       pmi->pgsp_free = pgsp_free;
 733       return true;
 734     }
 735     return false;
 736 
 737   } else {
 738 
 739     // On AIX, I use the (dynamically loaded) perfstat library to retrieve memory statistics
 740     // See:
 741     // http://publib.boulder.ibm.com/infocenter/systems/index.jsp
 742     //        ?topic=/com.ibm.aix.basetechref/doc/basetrf1/perfstat_memtot.htm
 743     // http://publib.boulder.ibm.com/infocenter/systems/index.jsp
 744     //        ?topic=/com.ibm.aix.files/doc/aixfiles/libperfstat.h.htm
 745 
 746     perfstat_memory_total_t psmt;
 747     memset (&psmt, '\0', sizeof(psmt));
 748     const int rc = libperfstat::perfstat_memory_total(NULL, &psmt, sizeof(psmt), 1);
 749     if (rc == -1) {
 750       trcVerbose("perfstat_memory_total() failed (errno=%d)", errno);
 751       assert(0, "perfstat_memory_total() failed");
 752       return false;
 753     }
 754 
 755     assert(rc == 1, "perfstat_memory_total() - weird return code");
 756 
 757     // excerpt from
 758     // http://publib.boulder.ibm.com/infocenter/systems/index.jsp
 759     //        ?topic=/com.ibm.aix.files/doc/aixfiles/libperfstat.h.htm
 760     // The fields of perfstat_memory_total_t:
 761     // u_longlong_t virt_total         Total virtual memory (in 4 KB pages).
 762     // u_longlong_t real_total         Total real memory (in 4 KB pages).
 763     // u_longlong_t real_free          Free real memory (in 4 KB pages).
 764     // u_longlong_t pgsp_total         Total paging space (in 4 KB pages).
 765     // u_longlong_t pgsp_free          Free paging space (in 4 KB pages).
 766 
 767     pmi->virt_total = psmt.virt_total * 4096;
 768     pmi->real_total = psmt.real_total * 4096;
 769     pmi->real_free = psmt.real_free * 4096;
 770     pmi->pgsp_total = psmt.pgsp_total * 4096;
 771     pmi->pgsp_free = psmt.pgsp_free * 4096;
 772 
 773     return true;
 774 
 775   }
 776 } // end os::Aix::get_meminfo
 777 
 778 //////////////////////////////////////////////////////////////////////////////
 779 // create new thread
 780 
 781 // Thread start routine for all newly created threads
 782 static void *thread_native_entry(Thread *thread) {
 783 
 784   thread->record_stack_base_and_size();
 785 
 786   const pthread_t pthread_id = ::pthread_self();
 787   const tid_t kernel_thread_id = ::thread_self();
 788 
 789   LogTarget(Info, os, thread) lt;
 790   if (lt.is_enabled()) {
 791     address low_address = thread->stack_end();
 792     address high_address = thread->stack_base();
 793     lt.print("Thread is alive (tid: " UINTX_FORMAT ", kernel thread id: " UINTX_FORMAT
 794              ", stack [" PTR_FORMAT " - " PTR_FORMAT " (" SIZE_FORMAT "k using %uk pages)).",
 795              os::current_thread_id(), (uintx) kernel_thread_id, low_address, high_address,
 796              (high_address - low_address) / K, os::Aix::query_pagesize(low_address) / K);
 797   }
 798 
 799   // Normally, pthread stacks on AIX live in the data segment (are allocated with malloc()
 800   // by the pthread library). In rare cases, this may not be the case, e.g. when third-party
 801   // tools hook pthread_create(). In this case, we may run into problems establishing
 802   // guard pages on those stacks, because the stacks may reside in memory which is not
 803   // protectable (shmated).
 804   if (thread->stack_base() > ::sbrk(0)) {
 805     log_warning(os, thread)("Thread stack not in data segment.");
 806   }
 807 
 808   // Try to randomize the cache line index of hot stack frames.
 809   // This helps when threads of the same stack traces evict each other's
 810   // cache lines. The threads can be either from the same JVM instance, or
 811   // from different JVM instances. The benefit is especially true for
 812   // processors with hyperthreading technology.
 813 
 814   static int counter = 0;
 815   int pid = os::current_process_id();
 816   alloca(((pid ^ counter++) & 7) * 128);
 817 
 818   thread->initialize_thread_current();
 819 
 820   OSThread* osthread = thread->osthread();
 821 
 822   // Thread_id is pthread id.
 823   osthread->set_thread_id(pthread_id);
 824 
 825   // .. but keep kernel thread id too for diagnostics
 826   osthread->set_kernel_thread_id(kernel_thread_id);
 827 
 828   // Initialize signal mask for this thread.
 829   os::Aix::hotspot_sigmask(thread);
 830 
 831   // Initialize floating point control register.
 832   os::Aix::init_thread_fpu_state();
 833 
 834   assert(osthread->get_state() == RUNNABLE, "invalid os thread state");
 835 
 836   // Call one more level start routine.
 837   thread->call_run();
 838 
 839   // Note: at this point the thread object may already have deleted itself.
 840   // Prevent dereferencing it from here on out.
 841   thread = NULL;
 842 
 843   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", kernel thread id: " UINTX_FORMAT ").",
 844     os::current_thread_id(), (uintx) kernel_thread_id);
 845 
 846   return 0;
 847 }
 848 
 849 bool os::create_thread(Thread* thread, ThreadType thr_type,
 850                        size_t req_stack_size) {
 851 
 852   assert(thread->osthread() == NULL, "caller responsible");
 853 
 854   // Allocate the OSThread object.
 855   OSThread* osthread = new OSThread(NULL, NULL);
 856   if (osthread == NULL) {
 857     return false;
 858   }
 859 
 860   // Set the correct thread state.
 861   osthread->set_thread_type(thr_type);
 862 
 863   // Initial state is ALLOCATED but not INITIALIZED
 864   osthread->set_state(ALLOCATED);
 865 
 866   thread->set_osthread(osthread);
 867 
 868   // Init thread attributes.
 869   pthread_attr_t attr;
 870   pthread_attr_init(&attr);
 871   guarantee(pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED) == 0, "???");
 872 
 873   // Make sure we run in 1:1 kernel-user-thread mode.
 874   if (os::Aix::on_aix()) {
 875     guarantee(pthread_attr_setscope(&attr, PTHREAD_SCOPE_SYSTEM) == 0, "???");
 876     guarantee(pthread_attr_setinheritsched(&attr, PTHREAD_EXPLICIT_SCHED) == 0, "???");
 877   }
 878 
 879   // Start in suspended state, and in os::thread_start, wake the thread up.
 880   guarantee(pthread_attr_setsuspendstate_np(&attr, PTHREAD_CREATE_SUSPENDED_NP) == 0, "???");
 881 
 882   // Calculate stack size if it's not specified by caller.
 883   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 884 
 885   // JDK-8187028: It was observed that on some configurations (4K backed thread stacks)
 886   // the real thread stack size may be smaller than the requested stack size, by as much as 64K.
 887   // This very much looks like a pthread lib error. As a workaround, increase the stack size
 888   // by 64K for small thread stacks (arbitrarily choosen to be < 4MB)
 889   if (stack_size < 4096 * K) {
 890     stack_size += 64 * K;
 891   }
 892 
 893   // On Aix, pthread_attr_setstacksize fails with huge values and leaves the
 894   // thread size in attr unchanged. If this is the minimal stack size as set
 895   // by pthread_attr_init this leads to crashes after thread creation. E.g. the
 896   // guard pages might not fit on the tiny stack created.
 897   int ret = pthread_attr_setstacksize(&attr, stack_size);
 898   if (ret != 0) {
 899     log_warning(os, thread)("The %sthread stack size specified is invalid: " SIZE_FORMAT "k",
 900                             (thr_type == compiler_thread) ? "compiler " : ((thr_type == java_thread) ? "" : "VM "),
 901                             stack_size / K);
 902     thread->set_osthread(NULL);
 903     delete osthread;
 904     return false;
 905   }
 906 
 907   // Save some cycles and a page by disabling OS guard pages where we have our own
 908   // VM guard pages (in java threads). For other threads, keep system default guard
 909   // pages in place.
 910   if (thr_type == java_thread || thr_type == compiler_thread) {
 911     ret = pthread_attr_setguardsize(&attr, 0);
 912   }
 913 
 914   pthread_t tid = 0;
 915   if (ret == 0) {
 916     ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 917   }
 918 
 919   if (ret == 0) {
 920     char buf[64];
 921     log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 922       (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 923   } else {
 924     char buf[64];
 925     log_warning(os, thread)("Failed to start thread - pthread_create failed (%d=%s) for attributes: %s.",
 926       ret, os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 927     // Log some OS information which might explain why creating the thread failed.
 928     log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads());
 929     LogStream st(Log(os, thread)::info());
 930     os::Posix::print_rlimit_info(&st);
 931     os::print_memory_info(&st);
 932   }
 933 
 934   pthread_attr_destroy(&attr);
 935 
 936   if (ret != 0) {
 937     // Need to clean up stuff we've allocated so far.
 938     thread->set_osthread(NULL);
 939     delete osthread;
 940     return false;
 941   }
 942 
 943   // OSThread::thread_id is the pthread id.
 944   osthread->set_thread_id(tid);
 945 
 946   return true;
 947 }
 948 
 949 /////////////////////////////////////////////////////////////////////////////
 950 // attach existing thread
 951 
 952 // bootstrap the main thread
 953 bool os::create_main_thread(JavaThread* thread) {
 954   assert(os::Aix::_main_thread == pthread_self(), "should be called inside main thread");
 955   return create_attached_thread(thread);
 956 }
 957 
 958 bool os::create_attached_thread(JavaThread* thread) {
 959 #ifdef ASSERT
 960     thread->verify_not_published();
 961 #endif
 962 
 963   // Allocate the OSThread object
 964   OSThread* osthread = new OSThread(NULL, NULL);
 965 
 966   if (osthread == NULL) {
 967     return false;
 968   }
 969 
 970   const pthread_t pthread_id = ::pthread_self();
 971   const tid_t kernel_thread_id = ::thread_self();
 972 
 973   // OSThread::thread_id is the pthread id.
 974   osthread->set_thread_id(pthread_id);
 975 
 976   // .. but keep kernel thread id too for diagnostics
 977   osthread->set_kernel_thread_id(kernel_thread_id);
 978 
 979   // initialize floating point control register
 980   os::Aix::init_thread_fpu_state();
 981 
 982   // Initial thread state is RUNNABLE
 983   osthread->set_state(RUNNABLE);
 984 
 985   thread->set_osthread(osthread);
 986 
 987   if (UseNUMA) {
 988     int lgrp_id = os::numa_get_group_id();
 989     if (lgrp_id != -1) {
 990       thread->set_lgrp_id(lgrp_id);
 991     }
 992   }
 993 
 994   // initialize signal mask for this thread
 995   // and save the caller's signal mask
 996   os::Aix::hotspot_sigmask(thread);
 997 
 998   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", kernel thread id: " UINTX_FORMAT ").",
 999     os::current_thread_id(), (uintx) kernel_thread_id);
1000 
1001   return true;
1002 }
1003 
1004 void os::pd_start_thread(Thread* thread) {
1005   int status = pthread_continue_np(thread->osthread()->pthread_id());
1006   assert(status == 0, "thr_continue failed");
1007 }
1008 
1009 // Free OS resources related to the OSThread
1010 void os::free_thread(OSThread* osthread) {
1011   assert(osthread != NULL, "osthread not set");
1012 
1013   // We are told to free resources of the argument thread,
1014   // but we can only really operate on the current thread.
1015   assert(Thread::current()->osthread() == osthread,
1016          "os::free_thread but not current thread");
1017 
1018   // Restore caller's signal mask
1019   sigset_t sigmask = osthread->caller_sigmask();
1020   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1021 
1022   delete osthread;
1023 }
1024 
1025 ////////////////////////////////////////////////////////////////////////////////
1026 // time support
1027 
1028 // Time since start-up in seconds to a fine granularity.
1029 // Used by VMSelfDestructTimer and the MemProfiler.
1030 double os::elapsedTime() {
1031   return (double)(os::elapsed_counter()) * 0.000001;
1032 }
1033 
1034 jlong os::elapsed_counter() {
1035   timeval time;
1036   int status = gettimeofday(&time, NULL);
1037   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1038 }
1039 
1040 jlong os::elapsed_frequency() {
1041   return (1000 * 1000);
1042 }
1043 
1044 bool os::supports_vtime() { return true; }
1045 bool os::enable_vtime()   { return false; }
1046 bool os::vtime_enabled()  { return false; }
1047 
1048 double os::elapsedVTime() {
1049   struct rusage usage;
1050   int retval = getrusage(RUSAGE_THREAD, &usage);
1051   if (retval == 0) {
1052     return usage.ru_utime.tv_sec + usage.ru_stime.tv_sec + (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000.0 * 1000);
1053   } else {
1054     // better than nothing, but not much
1055     return elapsedTime();
1056   }
1057 }
1058 
1059 jlong os::javaTimeMillis() {
1060   timeval time;
1061   int status = gettimeofday(&time, NULL);
1062   assert(status != -1, "aix error at gettimeofday()");
1063   return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1064 }
1065 
1066 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1067   timeval time;
1068   int status = gettimeofday(&time, NULL);
1069   assert(status != -1, "aix error at gettimeofday()");
1070   seconds = jlong(time.tv_sec);
1071   nanos = jlong(time.tv_usec) * 1000;
1072 }
1073 
1074 // We use mread_real_time here.
1075 // On AIX: If the CPU has a time register, the result will be RTC_POWER and
1076 // it has to be converted to real time. AIX documentations suggests to do
1077 // this unconditionally, so we do it.
1078 //
1079 // See: https://www.ibm.com/support/knowledgecenter/ssw_aix_61/com.ibm.aix.basetrf2/read_real_time.htm
1080 //
1081 // On PASE: mread_real_time will always return RTC_POWER_PC data, so no
1082 // conversion is necessary. However, mread_real_time will not return
1083 // monotonic results but merely matches read_real_time. So we need a tweak
1084 // to ensure monotonic results.
1085 //
1086 // For PASE no public documentation exists, just word by IBM
1087 jlong os::javaTimeNanos() {
1088   timebasestruct_t time;
1089   int rc = mread_real_time(&time, TIMEBASE_SZ);
1090   if (os::Aix::on_pase()) {
1091     assert(rc == RTC_POWER, "expected time format RTC_POWER from mread_real_time in PASE");
1092     jlong now = jlong(time.tb_high) * NANOSECS_PER_SEC + jlong(time.tb_low);
1093     jlong prev = max_real_time;
1094     if (now <= prev) {
1095       return prev;   // same or retrograde time;
1096     }
1097     jlong obsv = Atomic::cmpxchg(now, &max_real_time, prev);
1098     assert(obsv >= prev, "invariant");   // Monotonicity
1099     // If the CAS succeeded then we're done and return "now".
1100     // If the CAS failed and the observed value "obsv" is >= now then
1101     // we should return "obsv".  If the CAS failed and now > obsv > prv then
1102     // some other thread raced this thread and installed a new value, in which case
1103     // we could either (a) retry the entire operation, (b) retry trying to install now
1104     // or (c) just return obsv.  We use (c).   No loop is required although in some cases
1105     // we might discard a higher "now" value in deference to a slightly lower but freshly
1106     // installed obsv value.   That's entirely benign -- it admits no new orderings compared
1107     // to (a) or (b) -- and greatly reduces coherence traffic.
1108     // We might also condition (c) on the magnitude of the delta between obsv and now.
1109     // Avoiding excessive CAS operations to hot RW locations is critical.
1110     // See https://blogs.oracle.com/dave/entry/cas_and_cache_trivia_invalidate
1111     return (prev == obsv) ? now : obsv;
1112   } else {
1113     if (rc != RTC_POWER) {
1114       rc = time_base_to_time(&time, TIMEBASE_SZ);
1115       assert(rc != -1, "error calling time_base_to_time()");
1116     }
1117     return jlong(time.tb_high) * NANOSECS_PER_SEC + jlong(time.tb_low);
1118   }
1119 }
1120 
1121 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1122   info_ptr->max_value = ALL_64_BITS;
1123   // mread_real_time() is monotonic (see 'os::javaTimeNanos()')
1124   info_ptr->may_skip_backward = false;
1125   info_ptr->may_skip_forward = false;
1126   info_ptr->kind = JVMTI_TIMER_ELAPSED;    // elapsed not CPU time
1127 }
1128 
1129 // Return the real, user, and system times in seconds from an
1130 // arbitrary fixed point in the past.
1131 bool os::getTimesSecs(double* process_real_time,
1132                       double* process_user_time,
1133                       double* process_system_time) {
1134   struct tms ticks;
1135   clock_t real_ticks = times(&ticks);
1136 
1137   if (real_ticks == (clock_t) (-1)) {
1138     return false;
1139   } else {
1140     double ticks_per_second = (double) clock_tics_per_sec;
1141     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1142     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1143     *process_real_time = ((double) real_ticks) / ticks_per_second;
1144 
1145     return true;
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 // Note: os::abort() might be called very early during initialization, or
1189 // called from signal handler. Before adding something to os::abort(), make
1190 // sure it is async-safe and can handle partially initialized VM.
1191 void os::abort(bool dump_core, void* siginfo, const void* context) {
1192   os::shutdown();
1193   if (dump_core) {
1194 #ifndef PRODUCT
1195     fdStream out(defaultStream::output_fd());
1196     out.print_raw("Current thread is ");
1197     char buf[16];
1198     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1199     out.print_raw_cr(buf);
1200     out.print_raw_cr("Dumping core ...");
1201 #endif
1202     ::abort(); // dump core
1203   }
1204 
1205   ::exit(1);
1206 }
1207 
1208 // Die immediately, no exit hook, no abort hook, no cleanup.
1209 void os::die() {
1210   ::abort();
1211 }
1212 
1213 intx os::current_thread_id() {
1214   return (intx)pthread_self();
1215 }
1216 
1217 int os::current_process_id() {
1218   return getpid();
1219 }
1220 
1221 // DLL functions
1222 
1223 const char* os::dll_file_extension() { return ".so"; }
1224 
1225 // This must be hard coded because it's the system's temporary
1226 // directory not the java application's temp directory, ala java.io.tmpdir.
1227 const char* os::get_temp_directory() { return "/tmp"; }
1228 
1229 // Check if addr is inside libjvm.so.
1230 bool os::address_is_in_vm(address addr) {
1231 
1232   // Input could be a real pc or a function pointer literal. The latter
1233   // would be a function descriptor residing in the data segment of a module.
1234   loaded_module_t lm;
1235   if (LoadedLibraries::find_for_text_address(addr, &lm) != NULL) {
1236     return lm.is_in_vm;
1237   } else if (LoadedLibraries::find_for_data_address(addr, &lm) != NULL) {
1238     return lm.is_in_vm;
1239   } else {
1240     return false;
1241   }
1242 
1243 }
1244 
1245 // Resolve an AIX function descriptor literal to a code pointer.
1246 // If the input is a valid code pointer to a text segment of a loaded module,
1247 //   it is returned unchanged.
1248 // If the input is a valid AIX function descriptor, it is resolved to the
1249 //   code entry point.
1250 // If the input is neither a valid function descriptor nor a valid code pointer,
1251 //   NULL is returned.
1252 static address resolve_function_descriptor_to_code_pointer(address p) {
1253 
1254   if (LoadedLibraries::find_for_text_address(p, NULL) != NULL) {
1255     // It is a real code pointer.
1256     return p;
1257   } else if (LoadedLibraries::find_for_data_address(p, NULL) != NULL) {
1258     // Pointer to data segment, potential function descriptor.
1259     address code_entry = (address)(((FunctionDescriptor*)p)->entry());
1260     if (LoadedLibraries::find_for_text_address(code_entry, NULL) != NULL) {
1261       // It is a function descriptor.
1262       return code_entry;
1263     }
1264   }
1265 
1266   return NULL;
1267 }
1268 
1269 bool os::dll_address_to_function_name(address addr, char *buf,
1270                                       int buflen, int *offset,
1271                                       bool demangle) {
1272   if (offset) {
1273     *offset = -1;
1274   }
1275   // Buf is not optional, but offset is optional.
1276   assert(buf != NULL, "sanity check");
1277   buf[0] = '\0';
1278 
1279   // Resolve function ptr literals first.
1280   addr = resolve_function_descriptor_to_code_pointer(addr);
1281   if (!addr) {
1282     return false;
1283   }
1284 
1285   return AixSymbols::get_function_name(addr, buf, buflen, offset, NULL, demangle);
1286 }
1287 
1288 bool os::dll_address_to_library_name(address addr, char* buf,
1289                                      int buflen, int* offset) {
1290   if (offset) {
1291     *offset = -1;
1292   }
1293   // Buf is not optional, but offset is optional.
1294   assert(buf != NULL, "sanity check");
1295   buf[0] = '\0';
1296 
1297   // Resolve function ptr literals first.
1298   addr = resolve_function_descriptor_to_code_pointer(addr);
1299   if (!addr) {
1300     return false;
1301   }
1302 
1303   return AixSymbols::get_module_name(addr, buf, buflen);
1304 }
1305 
1306 // Loads .dll/.so and in case of error it checks if .dll/.so was built
1307 // for the same architecture as Hotspot is running on.
1308 void *os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1309 
1310   if (ebuf && ebuflen > 0) {
1311     ebuf[0] = '\0';
1312     ebuf[ebuflen - 1] = '\0';
1313   }
1314 
1315   if (!filename || strlen(filename) == 0) {
1316     ::strncpy(ebuf, "dll_load: empty filename specified", ebuflen - 1);
1317     return NULL;
1318   }
1319 
1320   // RTLD_LAZY is currently not implemented. The dl is loaded immediately with all its dependants.
1321   void * result= ::dlopen(filename, RTLD_LAZY);
1322   if (result != NULL) {
1323     // Reload dll cache. Don't do this in signal handling.
1324     LoadedLibraries::reload();
1325     return result;
1326   } else {
1327     // error analysis when dlopen fails
1328     const char* const error_report = ::dlerror();
1329     if (error_report && ebuf && ebuflen > 0) {
1330       snprintf(ebuf, ebuflen - 1, "%s, LIBPATH=%s, LD_LIBRARY_PATH=%s : %s",
1331                filename, ::getenv("LIBPATH"), ::getenv("LD_LIBRARY_PATH"), error_report);
1332     }
1333   }
1334   return NULL;
1335 }
1336 
1337 void* os::dll_lookup(void* handle, const char* name) {
1338   void* res = dlsym(handle, name);
1339   return res;
1340 }
1341 
1342 void* os::get_default_process_handle() {
1343   return (void*)::dlopen(NULL, RTLD_LAZY);
1344 }
1345 
1346 void os::print_dll_info(outputStream *st) {
1347   st->print_cr("Dynamic libraries:");
1348   LoadedLibraries::print(st);
1349 }
1350 
1351 void os::get_summary_os_info(char* buf, size_t buflen) {
1352   // There might be something more readable than uname results for AIX.
1353   struct utsname name;
1354   uname(&name);
1355   snprintf(buf, buflen, "%s %s", name.release, name.version);
1356 }
1357 
1358 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1359   // Not yet implemented.
1360   return 0;
1361 }
1362 
1363 void os::print_os_info_brief(outputStream* st) {
1364   uint32_t ver = os::Aix::os_version();
1365   st->print_cr("AIX kernel version %u.%u.%u.%u",
1366                (ver >> 24) & 0xFF, (ver >> 16) & 0xFF, (ver >> 8) & 0xFF, ver & 0xFF);
1367 
1368   os::Posix::print_uname_info(st);
1369 
1370   // Linux uses print_libversion_info(st); here.
1371 }
1372 
1373 void os::print_os_info(outputStream* st) {
1374   st->print("OS:");
1375 
1376   st->print("uname:");
1377   struct utsname name;
1378   uname(&name);
1379   st->print(name.sysname); st->print(" ");
1380   st->print(name.nodename); st->print(" ");
1381   st->print(name.release); st->print(" ");
1382   st->print(name.version); st->print(" ");
1383   st->print(name.machine);
1384   st->cr();
1385 
1386   uint32_t ver = os::Aix::os_version();
1387   st->print_cr("AIX kernel version %u.%u.%u.%u",
1388                (ver >> 24) & 0xFF, (ver >> 16) & 0xFF, (ver >> 8) & 0xFF, ver & 0xFF);
1389 
1390   os::Posix::print_rlimit_info(st);
1391 
1392   // _SC_THREAD_THREADS_MAX is the maximum number of threads within a process.
1393   long tmax = sysconf(_SC_THREAD_THREADS_MAX);
1394   st->print_cr("maximum #threads within a process:%ld", tmax);
1395 
1396   // load average
1397   st->print("load average:");
1398   double loadavg[3] = {-1.L, -1.L, -1.L};
1399   os::loadavg(loadavg, 3);
1400   st->print_cr("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
1401 
1402   // print wpar info
1403   libperfstat::wparinfo_t wi;
1404   if (libperfstat::get_wparinfo(&wi)) {
1405     st->print_cr("wpar info");
1406     st->print_cr("name: %s", wi.name);
1407     st->print_cr("id:   %d", wi.wpar_id);
1408     st->print_cr("type: %s", (wi.app_wpar ? "application" : "system"));
1409   }
1410 
1411   VM_Version::print_platform_virtualization_info(st);
1412 }
1413 
1414 void os::print_memory_info(outputStream* st) {
1415 
1416   st->print_cr("Memory:");
1417 
1418   st->print_cr("  Base page size (sysconf _SC_PAGESIZE):  %s",
1419     describe_pagesize(g_multipage_support.pagesize));
1420   st->print_cr("  Data page size (C-Heap, bss, etc):      %s",
1421     describe_pagesize(g_multipage_support.datapsize));
1422   st->print_cr("  Text page size:                         %s",
1423     describe_pagesize(g_multipage_support.textpsize));
1424   st->print_cr("  Thread stack page size (pthread):       %s",
1425     describe_pagesize(g_multipage_support.pthr_stack_pagesize));
1426   st->print_cr("  Default shared memory page size:        %s",
1427     describe_pagesize(g_multipage_support.shmpsize));
1428   st->print_cr("  Can use 64K pages dynamically with shared meory:  %s",
1429     (g_multipage_support.can_use_64K_pages ? "yes" :"no"));
1430   st->print_cr("  Can use 16M pages dynamically with shared memory: %s",
1431     (g_multipage_support.can_use_16M_pages ? "yes" :"no"));
1432   st->print_cr("  Multipage error: %d",
1433     g_multipage_support.error);
1434   st->cr();
1435   st->print_cr("  os::vm_page_size:       %s", describe_pagesize(os::vm_page_size()));
1436 
1437   // print out LDR_CNTRL because it affects the default page sizes
1438   const char* const ldr_cntrl = ::getenv("LDR_CNTRL");
1439   st->print_cr("  LDR_CNTRL=%s.", ldr_cntrl ? ldr_cntrl : "<unset>");
1440 
1441   // Print out EXTSHM because it is an unsupported setting.
1442   const char* const extshm = ::getenv("EXTSHM");
1443   st->print_cr("  EXTSHM=%s.", extshm ? extshm : "<unset>");
1444   if ( (strcmp(extshm, "on") == 0) || (strcmp(extshm, "ON") == 0) ) {
1445     st->print_cr("  *** Unsupported! Please remove EXTSHM from your environment! ***");
1446   }
1447 
1448   // Print out AIXTHREAD_GUARDPAGES because it affects the size of pthread stacks.
1449   const char* const aixthread_guardpages = ::getenv("AIXTHREAD_GUARDPAGES");
1450   st->print_cr("  AIXTHREAD_GUARDPAGES=%s.",
1451       aixthread_guardpages ? aixthread_guardpages : "<unset>");
1452   st->cr();
1453 
1454   os::Aix::meminfo_t mi;
1455   if (os::Aix::get_meminfo(&mi)) {
1456     if (os::Aix::on_aix()) {
1457       st->print_cr("physical total : " SIZE_FORMAT, mi.real_total);
1458       st->print_cr("physical free  : " SIZE_FORMAT, mi.real_free);
1459       st->print_cr("swap total     : " SIZE_FORMAT, mi.pgsp_total);
1460       st->print_cr("swap free      : " SIZE_FORMAT, mi.pgsp_free);
1461     } else {
1462       // PASE - Numbers are result of QWCRSSTS; they mean:
1463       // real_total: Sum of all system pools
1464       // real_free: always 0
1465       // pgsp_total: we take the size of the system ASP
1466       // pgsp_free: size of system ASP times percentage of system ASP unused
1467       st->print_cr("physical total     : " SIZE_FORMAT, mi.real_total);
1468       st->print_cr("system asp total   : " SIZE_FORMAT, mi.pgsp_total);
1469       st->print_cr("%% system asp used : %.2f",
1470         mi.pgsp_total ? (100.0f * (mi.pgsp_total - mi.pgsp_free) / mi.pgsp_total) : -1.0f);
1471     }
1472   }
1473   st->cr();
1474 
1475   // Print program break.
1476   st->print_cr("Program break at VM startup: " PTR_FORMAT ".", p2i(g_brk_at_startup));
1477   address brk_now = (address)::sbrk(0);
1478   if (brk_now != (address)-1) {
1479     st->print_cr("Program break now          : " PTR_FORMAT " (distance: " SIZE_FORMAT "k).",
1480                  p2i(brk_now), (size_t)((brk_now - g_brk_at_startup) / K));
1481   }
1482   st->print_cr("MaxExpectedDataSegmentSize    : " SIZE_FORMAT "k.", MaxExpectedDataSegmentSize / K);
1483   st->cr();
1484 
1485   // Print segments allocated with os::reserve_memory.
1486   st->print_cr("internal virtual memory regions used by vm:");
1487   vmembk_print_on(st);
1488 }
1489 
1490 // Get a string for the cpuinfo that is a summary of the cpu type
1491 void os::get_summary_cpu_info(char* buf, size_t buflen) {
1492   // read _system_configuration.version
1493   switch (_system_configuration.version) {
1494   case PV_8:
1495     strncpy(buf, "Power PC 8", buflen);
1496     break;
1497   case PV_7:
1498     strncpy(buf, "Power PC 7", buflen);
1499     break;
1500   case PV_6_1:
1501     strncpy(buf, "Power PC 6 DD1.x", buflen);
1502     break;
1503   case PV_6:
1504     strncpy(buf, "Power PC 6", buflen);
1505     break;
1506   case PV_5:
1507     strncpy(buf, "Power PC 5", buflen);
1508     break;
1509   case PV_5_2:
1510     strncpy(buf, "Power PC 5_2", buflen);
1511     break;
1512   case PV_5_3:
1513     strncpy(buf, "Power PC 5_3", buflen);
1514     break;
1515   case PV_5_Compat:
1516     strncpy(buf, "PV_5_Compat", buflen);
1517     break;
1518   case PV_6_Compat:
1519     strncpy(buf, "PV_6_Compat", buflen);
1520     break;
1521   case PV_7_Compat:
1522     strncpy(buf, "PV_7_Compat", buflen);
1523     break;
1524   case PV_8_Compat:
1525     strncpy(buf, "PV_8_Compat", buflen);
1526     break;
1527   default:
1528     strncpy(buf, "unknown", buflen);
1529   }
1530 }
1531 
1532 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
1533   // Nothing to do beyond of what os::print_cpu_info() does.
1534 }
1535 
1536 static void print_signal_handler(outputStream* st, int sig,
1537                                  char* buf, size_t buflen);
1538 
1539 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
1540   st->print_cr("Signal Handlers:");
1541   print_signal_handler(st, SIGSEGV, buf, buflen);
1542   print_signal_handler(st, SIGBUS , buf, buflen);
1543   print_signal_handler(st, SIGFPE , buf, buflen);
1544   print_signal_handler(st, SIGPIPE, buf, buflen);
1545   print_signal_handler(st, SIGXFSZ, buf, buflen);
1546   print_signal_handler(st, SIGILL , buf, buflen);
1547   print_signal_handler(st, SR_signum, buf, buflen);
1548   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
1549   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
1550   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
1551   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
1552   print_signal_handler(st, SIGTRAP, buf, buflen);
1553   // We also want to know if someone else adds a SIGDANGER handler because
1554   // that will interfere with OOM killling.
1555   print_signal_handler(st, SIGDANGER, buf, buflen);
1556 }
1557 
1558 static char saved_jvm_path[MAXPATHLEN] = {0};
1559 
1560 // Find the full path to the current module, libjvm.so.
1561 void os::jvm_path(char *buf, jint buflen) {
1562   // Error checking.
1563   if (buflen < MAXPATHLEN) {
1564     assert(false, "must use a large-enough buffer");
1565     buf[0] = '\0';
1566     return;
1567   }
1568   // Lazy resolve the path to current module.
1569   if (saved_jvm_path[0] != 0) {
1570     strcpy(buf, saved_jvm_path);
1571     return;
1572   }
1573 
1574   Dl_info dlinfo;
1575   int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
1576   assert(ret != 0, "cannot locate libjvm");
1577   char* rp = os::Posix::realpath((char *)dlinfo.dli_fname, buf, buflen);
1578   assert(rp != NULL, "error in realpath(): maybe the 'path' argument is too long?");
1579 
1580   if (Arguments::sun_java_launcher_is_altjvm()) {
1581     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
1582     // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
1583     // If "/jre/lib/" appears at the right place in the string, then
1584     // assume we are installed in a JDK and we're done. Otherwise, check
1585     // for a JAVA_HOME environment variable and fix up the path so it
1586     // looks like libjvm.so is installed there (append a fake suffix
1587     // hotspot/libjvm.so).
1588     const char *p = buf + strlen(buf) - 1;
1589     for (int count = 0; p > buf && count < 4; ++count) {
1590       for (--p; p > buf && *p != '/'; --p)
1591         /* empty */ ;
1592     }
1593 
1594     if (strncmp(p, "/jre/lib/", 9) != 0) {
1595       // Look for JAVA_HOME in the environment.
1596       char* java_home_var = ::getenv("JAVA_HOME");
1597       if (java_home_var != NULL && java_home_var[0] != 0) {
1598         char* jrelib_p;
1599         int len;
1600 
1601         // Check the current module name "libjvm.so".
1602         p = strrchr(buf, '/');
1603         if (p == NULL) {
1604           return;
1605         }
1606         assert(strstr(p, "/libjvm") == p, "invalid library name");
1607 
1608         rp = os::Posix::realpath(java_home_var, buf, buflen);
1609         if (rp == NULL) {
1610           return;
1611         }
1612 
1613         // determine if this is a legacy image or modules image
1614         // modules image doesn't have "jre" subdirectory
1615         len = strlen(buf);
1616         assert(len < buflen, "Ran out of buffer room");
1617         jrelib_p = buf + len;
1618         snprintf(jrelib_p, buflen-len, "/jre/lib");
1619         if (0 != access(buf, F_OK)) {
1620           snprintf(jrelib_p, buflen-len, "/lib");
1621         }
1622 
1623         if (0 == access(buf, F_OK)) {
1624           // Use current module name "libjvm.so"
1625           len = strlen(buf);
1626           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
1627         } else {
1628           // Go back to path of .so
1629           rp = os::Posix::realpath((char *)dlinfo.dli_fname, buf, buflen);
1630           if (rp == NULL) {
1631             return;
1632           }
1633         }
1634       }
1635     }
1636   }
1637 
1638   strncpy(saved_jvm_path, buf, sizeof(saved_jvm_path));
1639   saved_jvm_path[sizeof(saved_jvm_path) - 1] = '\0';
1640 }
1641 
1642 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
1643   // no prefix required, not even "_"
1644 }
1645 
1646 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
1647   // no suffix required
1648 }
1649 
1650 ////////////////////////////////////////////////////////////////////////////////
1651 // sun.misc.Signal support
1652 
1653 static volatile jint sigint_count = 0;
1654 
1655 static void
1656 UserHandler(int sig, void *siginfo, void *context) {
1657   // 4511530 - sem_post is serialized and handled by the manager thread. When
1658   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
1659   // don't want to flood the manager thread with sem_post requests.
1660   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
1661     return;
1662 
1663   // Ctrl-C is pressed during error reporting, likely because the error
1664   // handler fails to abort. Let VM die immediately.
1665   if (sig == SIGINT && VMError::is_error_reported()) {
1666     os::die();
1667   }
1668 
1669   os::signal_notify(sig);
1670 }
1671 
1672 void* os::user_handler() {
1673   return CAST_FROM_FN_PTR(void*, UserHandler);
1674 }
1675 
1676 extern "C" {
1677   typedef void (*sa_handler_t)(int);
1678   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
1679 }
1680 
1681 void* os::signal(int signal_number, void* handler) {
1682   struct sigaction sigAct, oldSigAct;
1683 
1684   sigfillset(&(sigAct.sa_mask));
1685 
1686   // Do not block out synchronous signals in the signal handler.
1687   // Blocking synchronous signals only makes sense if you can really
1688   // be sure that those signals won't happen during signal handling,
1689   // when the blocking applies. Normal signal handlers are lean and
1690   // do not cause signals. But our signal handlers tend to be "risky"
1691   // - secondary SIGSEGV, SIGILL, SIGBUS' may and do happen.
1692   // On AIX, PASE there was a case where a SIGSEGV happened, followed
1693   // by a SIGILL, which was blocked due to the signal mask. The process
1694   // just hung forever. Better to crash from a secondary signal than to hang.
1695   sigdelset(&(sigAct.sa_mask), SIGSEGV);
1696   sigdelset(&(sigAct.sa_mask), SIGBUS);
1697   sigdelset(&(sigAct.sa_mask), SIGILL);
1698   sigdelset(&(sigAct.sa_mask), SIGFPE);
1699   sigdelset(&(sigAct.sa_mask), SIGTRAP);
1700 
1701   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
1702 
1703   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
1704 
1705   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
1706     // -1 means registration failed
1707     return (void *)-1;
1708   }
1709 
1710   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
1711 }
1712 
1713 void os::signal_raise(int signal_number) {
1714   ::raise(signal_number);
1715 }
1716 
1717 //
1718 // The following code is moved from os.cpp for making this
1719 // code platform specific, which it is by its very nature.
1720 //
1721 
1722 // Will be modified when max signal is changed to be dynamic
1723 int os::sigexitnum_pd() {
1724   return NSIG;
1725 }
1726 
1727 // a counter for each possible signal value
1728 static volatile jint pending_signals[NSIG+1] = { 0 };
1729 
1730 // Wrapper functions for: sem_init(), sem_post(), sem_wait()
1731 // On AIX, we use sem_init(), sem_post(), sem_wait()
1732 // On Pase, we need to use msem_lock() and msem_unlock(), because Posix Semaphores
1733 // do not seem to work at all on PASE (unimplemented, will cause SIGILL).
1734 // Note that just using msem_.. APIs for both PASE and AIX is not an option either, as
1735 // on AIX, msem_..() calls are suspected of causing problems.
1736 static sem_t sig_sem;
1737 static msemaphore* p_sig_msem = 0;
1738 
1739 static void local_sem_init() {
1740   if (os::Aix::on_aix()) {
1741     int rc = ::sem_init(&sig_sem, 0, 0);
1742     guarantee(rc != -1, "sem_init failed");
1743   } else {
1744     // Memory semaphores must live in shared mem.
1745     guarantee0(p_sig_msem == NULL);
1746     p_sig_msem = (msemaphore*)os::reserve_memory(sizeof(msemaphore), NULL);
1747     guarantee(p_sig_msem, "Cannot allocate memory for memory semaphore");
1748     guarantee(::msem_init(p_sig_msem, 0) == p_sig_msem, "msem_init failed");
1749   }
1750 }
1751 
1752 static void local_sem_post() {
1753   static bool warn_only_once = false;
1754   if (os::Aix::on_aix()) {
1755     int rc = ::sem_post(&sig_sem);
1756     if (rc == -1 && !warn_only_once) {
1757       trcVerbose("sem_post failed (errno = %d, %s)", errno, os::errno_name(errno));
1758       warn_only_once = true;
1759     }
1760   } else {
1761     guarantee0(p_sig_msem != NULL);
1762     int rc = ::msem_unlock(p_sig_msem, 0);
1763     if (rc == -1 && !warn_only_once) {
1764       trcVerbose("msem_unlock failed (errno = %d, %s)", errno, os::errno_name(errno));
1765       warn_only_once = true;
1766     }
1767   }
1768 }
1769 
1770 static void local_sem_wait() {
1771   static bool warn_only_once = false;
1772   if (os::Aix::on_aix()) {
1773     int rc = ::sem_wait(&sig_sem);
1774     if (rc == -1 && !warn_only_once) {
1775       trcVerbose("sem_wait failed (errno = %d, %s)", errno, os::errno_name(errno));
1776       warn_only_once = true;
1777     }
1778   } else {
1779     guarantee0(p_sig_msem != NULL); // must init before use
1780     int rc = ::msem_lock(p_sig_msem, 0);
1781     if (rc == -1 && !warn_only_once) {
1782       trcVerbose("msem_lock failed (errno = %d, %s)", errno, os::errno_name(errno));
1783       warn_only_once = true;
1784     }
1785   }
1786 }
1787 
1788 static void jdk_misc_signal_init() {
1789   // Initialize signal structures
1790   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
1791 
1792   // Initialize signal semaphore
1793   local_sem_init();
1794 }
1795 
1796 void os::signal_notify(int sig) {
1797   Atomic::inc(&pending_signals[sig]);
1798   local_sem_post();
1799 }
1800 
1801 static int check_pending_signals() {
1802   Atomic::store(0, &sigint_count);
1803   for (;;) {
1804     for (int i = 0; i < NSIG + 1; i++) {
1805       jint n = pending_signals[i];
1806       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
1807         return i;
1808       }
1809     }
1810     JavaThread *thread = JavaThread::current();
1811     ThreadBlockInVM tbivm(thread);
1812 
1813     bool threadIsSuspended;
1814     do {
1815       thread->set_suspend_equivalent();
1816       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
1817 
1818       local_sem_wait();
1819 
1820       // were we externally suspended while we were waiting?
1821       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
1822       if (threadIsSuspended) {
1823         //
1824         // The semaphore has been incremented, but while we were waiting
1825         // another thread suspended us. We don't want to continue running
1826         // while suspended because that would surprise the thread that
1827         // suspended us.
1828         //
1829 
1830         local_sem_post();
1831 
1832         thread->java_suspend_self();
1833       }
1834     } while (threadIsSuspended);
1835   }
1836 }
1837 
1838 int os::signal_wait() {
1839   return check_pending_signals();
1840 }
1841 
1842 ////////////////////////////////////////////////////////////////////////////////
1843 // Virtual Memory
1844 
1845 // We need to keep small simple bookkeeping for os::reserve_memory and friends.
1846 
1847 #define VMEM_MAPPED  1
1848 #define VMEM_SHMATED 2
1849 
1850 struct vmembk_t {
1851   int type;         // 1 - mmap, 2 - shmat
1852   char* addr;
1853   size_t size;      // Real size, may be larger than usersize.
1854   size_t pagesize;  // page size of area
1855   vmembk_t* next;
1856 
1857   bool contains_addr(char* p) const {
1858     return p >= addr && p < (addr + size);
1859   }
1860 
1861   bool contains_range(char* p, size_t s) const {
1862     return contains_addr(p) && contains_addr(p + s - 1);
1863   }
1864 
1865   void print_on(outputStream* os) const {
1866     os->print("[" PTR_FORMAT " - " PTR_FORMAT "] (" UINTX_FORMAT
1867       " bytes, %d %s pages), %s",
1868       addr, addr + size - 1, size, size / pagesize, describe_pagesize(pagesize),
1869       (type == VMEM_SHMATED ? "shmat" : "mmap")
1870     );
1871   }
1872 
1873   // Check that range is a sub range of memory block (or equal to memory block);
1874   // also check that range is fully page aligned to the page size if the block.
1875   void assert_is_valid_subrange(char* p, size_t s) const {
1876     if (!contains_range(p, s)) {
1877       trcVerbose("[" PTR_FORMAT " - " PTR_FORMAT "] is not a sub "
1878               "range of [" PTR_FORMAT " - " PTR_FORMAT "].",
1879               p, p + s, addr, addr + size);
1880       guarantee0(false);
1881     }
1882     if (!is_aligned_to(p, pagesize) || !is_aligned_to(p + s, pagesize)) {
1883       trcVerbose("range [" PTR_FORMAT " - " PTR_FORMAT "] is not"
1884               " aligned to pagesize (%lu)", p, p + s, (unsigned long) pagesize);
1885       guarantee0(false);
1886     }
1887   }
1888 };
1889 
1890 static struct {
1891   vmembk_t* first;
1892   MiscUtils::CritSect cs;
1893 } vmem;
1894 
1895 static void vmembk_add(char* addr, size_t size, size_t pagesize, int type) {
1896   vmembk_t* p = (vmembk_t*) ::malloc(sizeof(vmembk_t));
1897   assert0(p);
1898   if (p) {
1899     MiscUtils::AutoCritSect lck(&vmem.cs);
1900     p->addr = addr; p->size = size;
1901     p->pagesize = pagesize;
1902     p->type = type;
1903     p->next = vmem.first;
1904     vmem.first = p;
1905   }
1906 }
1907 
1908 static vmembk_t* vmembk_find(char* addr) {
1909   MiscUtils::AutoCritSect lck(&vmem.cs);
1910   for (vmembk_t* p = vmem.first; p; p = p->next) {
1911     if (p->addr <= addr && (p->addr + p->size) > addr) {
1912       return p;
1913     }
1914   }
1915   return NULL;
1916 }
1917 
1918 static void vmembk_remove(vmembk_t* p0) {
1919   MiscUtils::AutoCritSect lck(&vmem.cs);
1920   assert0(p0);
1921   assert0(vmem.first); // List should not be empty.
1922   for (vmembk_t** pp = &(vmem.first); *pp; pp = &((*pp)->next)) {
1923     if (*pp == p0) {
1924       *pp = p0->next;
1925       ::free(p0);
1926       return;
1927     }
1928   }
1929   assert0(false); // Not found?
1930 }
1931 
1932 static void vmembk_print_on(outputStream* os) {
1933   MiscUtils::AutoCritSect lck(&vmem.cs);
1934   for (vmembk_t* vmi = vmem.first; vmi; vmi = vmi->next) {
1935     vmi->print_on(os);
1936     os->cr();
1937   }
1938 }
1939 
1940 // Reserve and attach a section of System V memory.
1941 // If <requested_addr> is not NULL, function will attempt to attach the memory at the given
1942 // address. Failing that, it will attach the memory anywhere.
1943 // If <requested_addr> is NULL, function will attach the memory anywhere.
1944 //
1945 // <alignment_hint> is being ignored by this function. It is very probable however that the
1946 // alignment requirements are met anyway, because shmat() attaches at 256M boundaries.
1947 // Should this be not enogh, we can put more work into it.
1948 static char* reserve_shmated_memory (
1949   size_t bytes,
1950   char* requested_addr,
1951   size_t alignment_hint) {
1952 
1953   trcVerbose("reserve_shmated_memory " UINTX_FORMAT " bytes, wishaddress "
1954     PTR_FORMAT ", alignment_hint " UINTX_FORMAT "...",
1955     bytes, requested_addr, alignment_hint);
1956 
1957   // Either give me wish address or wish alignment but not both.
1958   assert0(!(requested_addr != NULL && alignment_hint != 0));
1959 
1960   // We must prevent anyone from attaching too close to the
1961   // BRK because that may cause malloc OOM.
1962   if (requested_addr != NULL && is_close_to_brk((address)requested_addr)) {
1963     trcVerbose("Wish address " PTR_FORMAT " is too close to the BRK segment. "
1964       "Will attach anywhere.", requested_addr);
1965     // Act like the OS refused to attach there.
1966     requested_addr = NULL;
1967   }
1968 
1969   // For old AS/400's (V5R4 and older) we should not even be here - System V shared memory is not
1970   // really supported (max size 4GB), so reserve_mmapped_memory should have been used instead.
1971   if (os::Aix::on_pase_V5R4_or_older()) {
1972     ShouldNotReachHere();
1973   }
1974 
1975   // Align size of shm up to 64K to avoid errors if we later try to change the page size.
1976   const size_t size = align_up(bytes, 64*K);
1977 
1978   // Reserve the shared segment.
1979   int shmid = shmget(IPC_PRIVATE, size, IPC_CREAT | S_IRUSR | S_IWUSR);
1980   if (shmid == -1) {
1981     trcVerbose("shmget(.., " UINTX_FORMAT ", ..) failed (errno: %d).", size, errno);
1982     return NULL;
1983   }
1984 
1985   // Important note:
1986   // It is very important that we, upon leaving this function, do not leave a shm segment alive.
1987   // We must right after attaching it remove it from the system. System V shm segments are global and
1988   // survive the process.
1989   // So, from here on: Do not assert, do not return, until we have called shmctl(IPC_RMID) (A).
1990 
1991   struct shmid_ds shmbuf;
1992   memset(&shmbuf, 0, sizeof(shmbuf));
1993   shmbuf.shm_pagesize = 64*K;
1994   if (shmctl(shmid, SHM_PAGESIZE, &shmbuf) != 0) {
1995     trcVerbose("Failed to set page size (need " UINTX_FORMAT " 64K pages) - shmctl failed with %d.",
1996                size / (64*K), errno);
1997     // I want to know if this ever happens.
1998     assert(false, "failed to set page size for shmat");
1999   }
2000 
2001   // Now attach the shared segment.
2002   // Note that I attach with SHM_RND - which means that the requested address is rounded down, if
2003   // needed, to the next lowest segment boundary. Otherwise the attach would fail if the address
2004   // were not a segment boundary.
2005   char* const addr = (char*) shmat(shmid, requested_addr, SHM_RND);
2006   const int errno_shmat = errno;
2007 
2008   // (A) Right after shmat and before handing shmat errors delete the shm segment.
2009   if (::shmctl(shmid, IPC_RMID, NULL) == -1) {
2010     trcVerbose("shmctl(%u, IPC_RMID) failed (%d)\n", shmid, errno);
2011     assert(false, "failed to remove shared memory segment!");
2012   }
2013 
2014   // Handle shmat error. If we failed to attach, just return.
2015   if (addr == (char*)-1) {
2016     trcVerbose("Failed to attach segment at " PTR_FORMAT " (%d).", requested_addr, errno_shmat);
2017     return NULL;
2018   }
2019 
2020   // Just for info: query the real page size. In case setting the page size did not
2021   // work (see above), the system may have given us something other then 4K (LDR_CNTRL).
2022   const size_t real_pagesize = os::Aix::query_pagesize(addr);
2023   if (real_pagesize != shmbuf.shm_pagesize) {
2024     trcVerbose("pagesize is, surprisingly, %h.", real_pagesize);
2025   }
2026 
2027   if (addr) {
2028     trcVerbose("shm-allocated " PTR_FORMAT " .. " PTR_FORMAT " (" UINTX_FORMAT " bytes, " UINTX_FORMAT " %s pages)",
2029       addr, addr + size - 1, size, size/real_pagesize, describe_pagesize(real_pagesize));
2030   } else {
2031     if (requested_addr != NULL) {
2032       trcVerbose("failed to shm-allocate " UINTX_FORMAT " bytes at with address " PTR_FORMAT ".", size, requested_addr);
2033     } else {
2034       trcVerbose("failed to shm-allocate " UINTX_FORMAT " bytes at any address.", size);
2035     }
2036   }
2037 
2038   // book-keeping
2039   vmembk_add(addr, size, real_pagesize, VMEM_SHMATED);
2040   assert0(is_aligned_to(addr, os::vm_page_size()));
2041 
2042   return addr;
2043 }
2044 
2045 static bool release_shmated_memory(char* addr, size_t size) {
2046 
2047   trcVerbose("release_shmated_memory [" PTR_FORMAT " - " PTR_FORMAT "].",
2048     addr, addr + size - 1);
2049 
2050   bool rc = false;
2051 
2052   // TODO: is there a way to verify shm size without doing bookkeeping?
2053   if (::shmdt(addr) != 0) {
2054     trcVerbose("error (%d).", errno);
2055   } else {
2056     trcVerbose("ok.");
2057     rc = true;
2058   }
2059   return rc;
2060 }
2061 
2062 static bool uncommit_shmated_memory(char* addr, size_t size) {
2063   trcVerbose("uncommit_shmated_memory [" PTR_FORMAT " - " PTR_FORMAT "].",
2064     addr, addr + size - 1);
2065 
2066   const bool rc = my_disclaim64(addr, size);
2067 
2068   if (!rc) {
2069     trcVerbose("my_disclaim64(" PTR_FORMAT ", " UINTX_FORMAT ") failed.\n", addr, size);
2070     return false;
2071   }
2072   return true;
2073 }
2074 
2075 ////////////////////////////////  mmap-based routines /////////////////////////////////
2076 
2077 // Reserve memory via mmap.
2078 // If <requested_addr> is given, an attempt is made to attach at the given address.
2079 // Failing that, memory is allocated at any address.
2080 // If <alignment_hint> is given and <requested_addr> is NULL, an attempt is made to
2081 // allocate at an address aligned with the given alignment. Failing that, memory
2082 // is aligned anywhere.
2083 static char* reserve_mmaped_memory(size_t bytes, char* requested_addr, size_t alignment_hint) {
2084   trcVerbose("reserve_mmaped_memory " UINTX_FORMAT " bytes, wishaddress " PTR_FORMAT ", "
2085     "alignment_hint " UINTX_FORMAT "...",
2086     bytes, requested_addr, alignment_hint);
2087 
2088   // If a wish address is given, but not aligned to 4K page boundary, mmap will fail.
2089   if (requested_addr && !is_aligned_to(requested_addr, os::vm_page_size()) != 0) {
2090     trcVerbose("Wish address " PTR_FORMAT " not aligned to page boundary.", requested_addr);
2091     return NULL;
2092   }
2093 
2094   // We must prevent anyone from attaching too close to the
2095   // BRK because that may cause malloc OOM.
2096   if (requested_addr != NULL && is_close_to_brk((address)requested_addr)) {
2097     trcVerbose("Wish address " PTR_FORMAT " is too close to the BRK segment. "
2098       "Will attach anywhere.", requested_addr);
2099     // Act like the OS refused to attach there.
2100     requested_addr = NULL;
2101   }
2102 
2103   // Specify one or the other but not both.
2104   assert0(!(requested_addr != NULL && alignment_hint > 0));
2105 
2106   // In 64K mode, we claim the global page size (os::vm_page_size())
2107   // is 64K. This is one of the few points where that illusion may
2108   // break, because mmap() will always return memory aligned to 4K. So
2109   // we must ensure we only ever return memory aligned to 64k.
2110   if (alignment_hint) {
2111     alignment_hint = lcm(alignment_hint, os::vm_page_size());
2112   } else {
2113     alignment_hint = os::vm_page_size();
2114   }
2115 
2116   // Size shall always be a multiple of os::vm_page_size (esp. in 64K mode).
2117   const size_t size = align_up(bytes, os::vm_page_size());
2118 
2119   // alignment: Allocate memory large enough to include an aligned range of the right size and
2120   // cut off the leading and trailing waste pages.
2121   assert0(alignment_hint != 0 && is_aligned_to(alignment_hint, os::vm_page_size())); // see above
2122   const size_t extra_size = size + alignment_hint;
2123 
2124   // Note: MAP_SHARED (instead of MAP_PRIVATE) needed to be able to
2125   // later use msync(MS_INVALIDATE) (see os::uncommit_memory).
2126   int flags = MAP_ANONYMOUS | MAP_SHARED;
2127 
2128   // MAP_FIXED is needed to enforce requested_addr - manpage is vague about what
2129   // it means if wishaddress is given but MAP_FIXED is not set.
2130   //
2131   // Important! Behaviour differs depending on whether SPEC1170 mode is active or not.
2132   // SPEC1170 mode active: behaviour like POSIX, MAP_FIXED will clobber existing mappings.
2133   // SPEC1170 mode not active: behaviour, unlike POSIX, is that no existing mappings will
2134   // get clobbered.
2135   if (requested_addr != NULL) {
2136     if (!os::Aix::xpg_sus_mode()) {  // not SPEC1170 Behaviour
2137       flags |= MAP_FIXED;
2138     }
2139   }
2140 
2141   char* addr = (char*)::mmap(requested_addr, extra_size,
2142       PROT_READ|PROT_WRITE|PROT_EXEC, flags, -1, 0);
2143 
2144   if (addr == MAP_FAILED) {
2145     trcVerbose("mmap(" PTR_FORMAT ", " UINTX_FORMAT ", ..) failed (%d)", requested_addr, size, errno);
2146     return NULL;
2147   }
2148 
2149   // Handle alignment.
2150   char* const addr_aligned = align_up(addr, alignment_hint);
2151   const size_t waste_pre = addr_aligned - addr;
2152   char* const addr_aligned_end = addr_aligned + size;
2153   const size_t waste_post = extra_size - waste_pre - size;
2154   if (waste_pre > 0) {
2155     ::munmap(addr, waste_pre);
2156   }
2157   if (waste_post > 0) {
2158     ::munmap(addr_aligned_end, waste_post);
2159   }
2160   addr = addr_aligned;
2161 
2162   if (addr) {
2163     trcVerbose("mmap-allocated " PTR_FORMAT " .. " PTR_FORMAT " (" UINTX_FORMAT " bytes)",
2164       addr, addr + bytes, bytes);
2165   } else {
2166     if (requested_addr != NULL) {
2167       trcVerbose("failed to mmap-allocate " UINTX_FORMAT " bytes at wish address " PTR_FORMAT ".", bytes, requested_addr);
2168     } else {
2169       trcVerbose("failed to mmap-allocate " UINTX_FORMAT " bytes at any address.", bytes);
2170     }
2171   }
2172 
2173   // bookkeeping
2174   vmembk_add(addr, size, 4*K, VMEM_MAPPED);
2175 
2176   // Test alignment, see above.
2177   assert0(is_aligned_to(addr, os::vm_page_size()));
2178 
2179   return addr;
2180 }
2181 
2182 static bool release_mmaped_memory(char* addr, size_t size) {
2183   assert0(is_aligned_to(addr, os::vm_page_size()));
2184   assert0(is_aligned_to(size, os::vm_page_size()));
2185 
2186   trcVerbose("release_mmaped_memory [" PTR_FORMAT " - " PTR_FORMAT "].",
2187     addr, addr + size - 1);
2188   bool rc = false;
2189 
2190   if (::munmap(addr, size) != 0) {
2191     trcVerbose("failed (%d)\n", errno);
2192     rc = false;
2193   } else {
2194     trcVerbose("ok.");
2195     rc = true;
2196   }
2197 
2198   return rc;
2199 }
2200 
2201 static bool uncommit_mmaped_memory(char* addr, size_t size) {
2202 
2203   assert0(is_aligned_to(addr, os::vm_page_size()));
2204   assert0(is_aligned_to(size, os::vm_page_size()));
2205 
2206   trcVerbose("uncommit_mmaped_memory [" PTR_FORMAT " - " PTR_FORMAT "].",
2207     addr, addr + size - 1);
2208   bool rc = false;
2209 
2210   // Uncommit mmap memory with msync MS_INVALIDATE.
2211   if (::msync(addr, size, MS_INVALIDATE) != 0) {
2212     trcVerbose("failed (%d)\n", errno);
2213     rc = false;
2214   } else {
2215     trcVerbose("ok.");
2216     rc = true;
2217   }
2218 
2219   return rc;
2220 }
2221 
2222 int os::vm_page_size() {
2223   // Seems redundant as all get out.
2224   assert(os::Aix::page_size() != -1, "must call os::init");
2225   return os::Aix::page_size();
2226 }
2227 
2228 // Aix allocates memory by pages.
2229 int os::vm_allocation_granularity() {
2230   assert(os::Aix::page_size() != -1, "must call os::init");
2231   return os::Aix::page_size();
2232 }
2233 
2234 #ifdef PRODUCT
2235 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2236                                     int err) {
2237   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2238           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2239           os::errno_name(err), err);
2240 }
2241 #endif
2242 
2243 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2244                                   const char* mesg) {
2245   assert(mesg != NULL, "mesg must be specified");
2246   if (!pd_commit_memory(addr, size, exec)) {
2247     // Add extra info in product mode for vm_exit_out_of_memory():
2248     PRODUCT_ONLY(warn_fail_commit_memory(addr, size, exec, errno);)
2249     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2250   }
2251 }
2252 
2253 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2254 
2255   assert(is_aligned_to(addr, os::vm_page_size()),
2256     "addr " PTR_FORMAT " not aligned to vm_page_size (" PTR_FORMAT ")",
2257     p2i(addr), os::vm_page_size());
2258   assert(is_aligned_to(size, os::vm_page_size()),
2259     "size " PTR_FORMAT " not aligned to vm_page_size (" PTR_FORMAT ")",
2260     size, os::vm_page_size());
2261 
2262   vmembk_t* const vmi = vmembk_find(addr);
2263   guarantee0(vmi);
2264   vmi->assert_is_valid_subrange(addr, size);
2265 
2266   trcVerbose("commit_memory [" PTR_FORMAT " - " PTR_FORMAT "].", addr, addr + size - 1);
2267 
2268   if (UseExplicitCommit) {
2269     // AIX commits memory on touch. So, touch all pages to be committed.
2270     for (char* p = addr; p < (addr + size); p += 4*K) {
2271       *p = '\0';
2272     }
2273   }
2274 
2275   return true;
2276 }
2277 
2278 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, bool exec) {
2279   return pd_commit_memory(addr, size, exec);
2280 }
2281 
2282 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2283                                   size_t alignment_hint, bool exec,
2284                                   const char* mesg) {
2285   // Alignment_hint is ignored on this OS.
2286   pd_commit_memory_or_exit(addr, size, exec, mesg);
2287 }
2288 
2289 bool os::pd_uncommit_memory(char* addr, size_t size) {
2290   assert(is_aligned_to(addr, os::vm_page_size()),
2291     "addr " PTR_FORMAT " not aligned to vm_page_size (" PTR_FORMAT ")",
2292     p2i(addr), os::vm_page_size());
2293   assert(is_aligned_to(size, os::vm_page_size()),
2294     "size " PTR_FORMAT " not aligned to vm_page_size (" PTR_FORMAT ")",
2295     size, os::vm_page_size());
2296 
2297   // Dynamically do different things for mmap/shmat.
2298   const vmembk_t* const vmi = vmembk_find(addr);
2299   guarantee0(vmi);
2300   vmi->assert_is_valid_subrange(addr, size);
2301 
2302   if (vmi->type == VMEM_SHMATED) {
2303     return uncommit_shmated_memory(addr, size);
2304   } else {
2305     return uncommit_mmaped_memory(addr, size);
2306   }
2307 }
2308 
2309 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2310   // Do not call this; no need to commit stack pages on AIX.
2311   ShouldNotReachHere();
2312   return true;
2313 }
2314 
2315 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2316   // Do not call this; no need to commit stack pages on AIX.
2317   ShouldNotReachHere();
2318   return true;
2319 }
2320 
2321 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2322 }
2323 
2324 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2325 }
2326 
2327 void os::numa_make_global(char *addr, size_t bytes) {
2328 }
2329 
2330 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2331 }
2332 
2333 bool os::numa_topology_changed() {
2334   return false;
2335 }
2336 
2337 size_t os::numa_get_groups_num() {
2338   return 1;
2339 }
2340 
2341 int os::numa_get_group_id() {
2342   return 0;
2343 }
2344 
2345 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2346   if (size > 0) {
2347     ids[0] = 0;
2348     return 1;
2349   }
2350   return 0;
2351 }
2352 
2353 bool os::get_page_info(char *start, page_info* info) {
2354   return false;
2355 }
2356 
2357 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2358   return end;
2359 }
2360 
2361 // Reserves and attaches a shared memory segment.
2362 // Will assert if a wish address is given and could not be obtained.
2363 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, size_t alignment_hint) {
2364 
2365   // All other Unices do a mmap(MAP_FIXED) if the addr is given,
2366   // thereby clobbering old mappings at that place. That is probably
2367   // not intended, never used and almost certainly an error were it
2368   // ever be used this way (to try attaching at a specified address
2369   // without clobbering old mappings an alternate API exists,
2370   // os::attempt_reserve_memory_at()).
2371   // Instead of mimicking the dangerous coding of the other platforms, here I
2372   // just ignore the request address (release) or assert(debug).
2373   assert0(requested_addr == NULL);
2374 
2375   // Always round to os::vm_page_size(), which may be larger than 4K.
2376   bytes = align_up(bytes, os::vm_page_size());
2377   const size_t alignment_hint0 =
2378     alignment_hint ? align_up(alignment_hint, os::vm_page_size()) : 0;
2379 
2380   // In 4K mode always use mmap.
2381   // In 64K mode allocate small sizes with mmap, large ones with 64K shmatted.
2382   if (os::vm_page_size() == 4*K) {
2383     return reserve_mmaped_memory(bytes, requested_addr, alignment_hint);
2384   } else {
2385     if (bytes >= Use64KPagesThreshold) {
2386       return reserve_shmated_memory(bytes, requested_addr, alignment_hint);
2387     } else {
2388       return reserve_mmaped_memory(bytes, requested_addr, alignment_hint);
2389     }
2390   }
2391 }
2392 
2393 bool os::pd_release_memory(char* addr, size_t size) {
2394 
2395   // Dynamically do different things for mmap/shmat.
2396   vmembk_t* const vmi = vmembk_find(addr);
2397   guarantee0(vmi);
2398 
2399   // Always round to os::vm_page_size(), which may be larger than 4K.
2400   size = align_up(size, os::vm_page_size());
2401   addr = align_up(addr, os::vm_page_size());
2402 
2403   bool rc = false;
2404   bool remove_bookkeeping = false;
2405   if (vmi->type == VMEM_SHMATED) {
2406     // For shmatted memory, we do:
2407     // - If user wants to release the whole range, release the memory (shmdt).
2408     // - If user only wants to release a partial range, uncommit (disclaim) that
2409     //   range. That way, at least, we do not use memory anymore (bust still page
2410     //   table space).
2411     vmi->assert_is_valid_subrange(addr, size);
2412     if (addr == vmi->addr && size == vmi->size) {
2413       rc = release_shmated_memory(addr, size);
2414       remove_bookkeeping = true;
2415     } else {
2416       rc = uncommit_shmated_memory(addr, size);
2417     }
2418   } else {
2419     // User may unmap partial regions but region has to be fully contained.
2420 #ifdef ASSERT
2421     vmi->assert_is_valid_subrange(addr, size);
2422 #endif
2423     rc = release_mmaped_memory(addr, size);
2424     remove_bookkeeping = true;
2425   }
2426 
2427   // update bookkeeping
2428   if (rc && remove_bookkeeping) {
2429     vmembk_remove(vmi);
2430   }
2431 
2432   return rc;
2433 }
2434 
2435 static bool checked_mprotect(char* addr, size_t size, int prot) {
2436 
2437   // Little problem here: if SPEC1170 behaviour is off, mprotect() on AIX will
2438   // not tell me if protection failed when trying to protect an un-protectable range.
2439   //
2440   // This means if the memory was allocated using shmget/shmat, protection wont work
2441   // but mprotect will still return 0:
2442   //
2443   // See http://publib.boulder.ibm.com/infocenter/pseries/v5r3/index.jsp?topic=/com.ibm.aix.basetechref/doc/basetrf1/mprotect.htm
2444 
2445   Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(addr), p2i(addr+size), prot);
2446   bool rc = ::mprotect(addr, size, prot) == 0 ? true : false;
2447 
2448   if (!rc) {
2449     const char* const s_errno = os::errno_name(errno);
2450     warning("mprotect(" PTR_FORMAT "-" PTR_FORMAT ", 0x%X) failed (%s).", addr, addr + size, prot, s_errno);
2451     return false;
2452   }
2453 
2454   // mprotect success check
2455   //
2456   // Mprotect said it changed the protection but can I believe it?
2457   //
2458   // To be sure I need to check the protection afterwards. Try to
2459   // read from protected memory and check whether that causes a segfault.
2460   //
2461   if (!os::Aix::xpg_sus_mode()) {
2462 
2463     if (CanUseSafeFetch32()) {
2464 
2465       const bool read_protected =
2466         (SafeFetch32((int*)addr, 0x12345678) == 0x12345678 &&
2467          SafeFetch32((int*)addr, 0x76543210) == 0x76543210) ? true : false;
2468 
2469       if (prot & PROT_READ) {
2470         rc = !read_protected;
2471       } else {
2472         rc = read_protected;
2473       }
2474 
2475       if (!rc) {
2476         if (os::Aix::on_pase()) {
2477           // There is an issue on older PASE systems where mprotect() will return success but the
2478           // memory will not be protected.
2479           // This has nothing to do with the problem of using mproect() on SPEC1170 incompatible
2480           // machines; we only see it rarely, when using mprotect() to protect the guard page of
2481           // a stack. It is an OS error.
2482           //
2483           // A valid strategy is just to try again. This usually works. :-/
2484 
2485           ::usleep(1000);
2486           Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(addr), p2i(addr+size), prot);
2487           if (::mprotect(addr, size, prot) == 0) {
2488             const bool read_protected_2 =
2489               (SafeFetch32((int*)addr, 0x12345678) == 0x12345678 &&
2490               SafeFetch32((int*)addr, 0x76543210) == 0x76543210) ? true : false;
2491             rc = true;
2492           }
2493         }
2494       }
2495     }
2496   }
2497 
2498   assert(rc == true, "mprotect failed.");
2499 
2500   return rc;
2501 }
2502 
2503 // Set protections specified
2504 bool os::protect_memory(char* addr, size_t size, ProtType prot, bool is_committed) {
2505   unsigned int p = 0;
2506   switch (prot) {
2507   case MEM_PROT_NONE: p = PROT_NONE; break;
2508   case MEM_PROT_READ: p = PROT_READ; break;
2509   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2510   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2511   default:
2512     ShouldNotReachHere();
2513   }
2514   // is_committed is unused.
2515   return checked_mprotect(addr, size, p);
2516 }
2517 
2518 bool os::guard_memory(char* addr, size_t size) {
2519   return checked_mprotect(addr, size, PROT_NONE);
2520 }
2521 
2522 bool os::unguard_memory(char* addr, size_t size) {
2523   return checked_mprotect(addr, size, PROT_READ|PROT_WRITE|PROT_EXEC);
2524 }
2525 
2526 // Large page support
2527 
2528 static size_t _large_page_size = 0;
2529 
2530 // Enable large page support if OS allows that.
2531 void os::large_page_init() {
2532   return; // Nothing to do. See query_multipage_support and friends.
2533 }
2534 
2535 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
2536   // reserve_memory_special() is used to allocate large paged memory. On AIX, we implement
2537   // 64k paged memory reservation using the normal memory allocation paths (os::reserve_memory()),
2538   // so this is not needed.
2539   assert(false, "should not be called on AIX");
2540   return NULL;
2541 }
2542 
2543 bool os::release_memory_special(char* base, size_t bytes) {
2544   // Detaching the SHM segment will also delete it, see reserve_memory_special().
2545   Unimplemented();
2546   return false;
2547 }
2548 
2549 size_t os::large_page_size() {
2550   return _large_page_size;
2551 }
2552 
2553 bool os::can_commit_large_page_memory() {
2554   // Does not matter, we do not support huge pages.
2555   return false;
2556 }
2557 
2558 bool os::can_execute_large_page_memory() {
2559   // Does not matter, we do not support huge pages.
2560   return false;
2561 }
2562 
2563 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
2564   assert(file_desc >= 0, "file_desc is not valid");
2565   char* result = NULL;
2566 
2567   // Always round to os::vm_page_size(), which may be larger than 4K.
2568   bytes = align_up(bytes, os::vm_page_size());
2569   result = reserve_mmaped_memory(bytes, requested_addr, 0);
2570 
2571   if (result != NULL) {
2572     if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
2573       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
2574     }
2575   }
2576   return result;
2577 }
2578 
2579 // Reserve memory at an arbitrary address, only if that area is
2580 // available (and not reserved for something else).
2581 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2582   char* addr = NULL;
2583 
2584   // Always round to os::vm_page_size(), which may be larger than 4K.
2585   bytes = align_up(bytes, os::vm_page_size());
2586 
2587   // In 4K mode always use mmap.
2588   // In 64K mode allocate small sizes with mmap, large ones with 64K shmatted.
2589   if (os::vm_page_size() == 4*K) {
2590     return reserve_mmaped_memory(bytes, requested_addr, 0);
2591   } else {
2592     if (bytes >= Use64KPagesThreshold) {
2593       return reserve_shmated_memory(bytes, requested_addr, 0);
2594     } else {
2595       return reserve_mmaped_memory(bytes, requested_addr, 0);
2596     }
2597   }
2598 
2599   return addr;
2600 }
2601 
2602 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2603 void os::infinite_sleep() {
2604   while (true) {    // sleep forever ...
2605     ::sleep(100);   // ... 100 seconds at a time
2606   }
2607 }
2608 
2609 // Used to convert frequent JVM_Yield() to nops
2610 bool os::dont_yield() {
2611   return DontYieldALot;
2612 }
2613 
2614 void os::naked_yield() {
2615   sched_yield();
2616 }
2617 
2618 ////////////////////////////////////////////////////////////////////////////////
2619 // thread priority support
2620 
2621 // From AIX manpage to pthread_setschedparam
2622 // (see: http://publib.boulder.ibm.com/infocenter/pseries/v5r3/index.jsp?
2623 //    topic=/com.ibm.aix.basetechref/doc/basetrf1/pthread_setschedparam.htm):
2624 //
2625 // "If schedpolicy is SCHED_OTHER, then sched_priority must be in the
2626 // range from 40 to 80, where 40 is the least favored priority and 80
2627 // is the most favored."
2628 //
2629 // (Actually, I doubt this even has an impact on AIX, as we do kernel
2630 // scheduling there; however, this still leaves iSeries.)
2631 //
2632 // We use the same values for AIX and PASE.
2633 int os::java_to_os_priority[CriticalPriority + 1] = {
2634   54,             // 0 Entry should never be used
2635 
2636   55,             // 1 MinPriority
2637   55,             // 2
2638   56,             // 3
2639 
2640   56,             // 4
2641   57,             // 5 NormPriority
2642   57,             // 6
2643 
2644   58,             // 7
2645   58,             // 8
2646   59,             // 9 NearMaxPriority
2647 
2648   60,             // 10 MaxPriority
2649 
2650   60              // 11 CriticalPriority
2651 };
2652 
2653 OSReturn os::set_native_priority(Thread* thread, int newpri) {
2654   if (!UseThreadPriorities) return OS_OK;
2655   pthread_t thr = thread->osthread()->pthread_id();
2656   int policy = SCHED_OTHER;
2657   struct sched_param param;
2658   param.sched_priority = newpri;
2659   int ret = pthread_setschedparam(thr, policy, &param);
2660 
2661   if (ret != 0) {
2662     trcVerbose("Could not change priority for thread %d to %d (error %d, %s)",
2663         (int)thr, newpri, ret, os::errno_name(ret));
2664   }
2665   return (ret == 0) ? OS_OK : OS_ERR;
2666 }
2667 
2668 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
2669   if (!UseThreadPriorities) {
2670     *priority_ptr = java_to_os_priority[NormPriority];
2671     return OS_OK;
2672   }
2673   pthread_t thr = thread->osthread()->pthread_id();
2674   int policy = SCHED_OTHER;
2675   struct sched_param param;
2676   int ret = pthread_getschedparam(thr, &policy, &param);
2677   *priority_ptr = param.sched_priority;
2678 
2679   return (ret == 0) ? OS_OK : OS_ERR;
2680 }
2681 
2682 ////////////////////////////////////////////////////////////////////////////////
2683 // suspend/resume support
2684 
2685 //  The low-level signal-based suspend/resume support is a remnant from the
2686 //  old VM-suspension that used to be for java-suspension, safepoints etc,
2687 //  within hotspot. Currently used by JFR's OSThreadSampler
2688 //
2689 //  The remaining code is greatly simplified from the more general suspension
2690 //  code that used to be used.
2691 //
2692 //  The protocol is quite simple:
2693 //  - suspend:
2694 //      - sends a signal to the target thread
2695 //      - polls the suspend state of the osthread using a yield loop
2696 //      - target thread signal handler (SR_handler) sets suspend state
2697 //        and blocks in sigsuspend until continued
2698 //  - resume:
2699 //      - sets target osthread state to continue
2700 //      - sends signal to end the sigsuspend loop in the SR_handler
2701 //
2702 //  Note that the SR_lock plays no role in this suspend/resume protocol,
2703 //  but is checked for NULL in SR_handler as a thread termination indicator.
2704 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
2705 //
2706 //  Note that resume_clear_context() and suspend_save_context() are needed
2707 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
2708 //  which in part is used by:
2709 //    - Forte Analyzer: AsyncGetCallTrace()
2710 //    - StackBanging: get_frame_at_stack_banging_point()
2711 
2712 static void resume_clear_context(OSThread *osthread) {
2713   osthread->set_ucontext(NULL);
2714   osthread->set_siginfo(NULL);
2715 }
2716 
2717 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
2718   osthread->set_ucontext(context);
2719   osthread->set_siginfo(siginfo);
2720 }
2721 
2722 //
2723 // Handler function invoked when a thread's execution is suspended or
2724 // resumed. We have to be careful that only async-safe functions are
2725 // called here (Note: most pthread functions are not async safe and
2726 // should be avoided.)
2727 //
2728 // Note: sigwait() is a more natural fit than sigsuspend() from an
2729 // interface point of view, but sigwait() prevents the signal hander
2730 // from being run. libpthread would get very confused by not having
2731 // its signal handlers run and prevents sigwait()'s use with the
2732 // mutex granting granting signal.
2733 //
2734 // Currently only ever called on the VMThread and JavaThreads (PC sampling).
2735 //
2736 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
2737   // Save and restore errno to avoid confusing native code with EINTR
2738   // after sigsuspend.
2739   int old_errno = errno;
2740 
2741   Thread* thread = Thread::current_or_null_safe();
2742   assert(thread != NULL, "Missing current thread in SR_handler");
2743 
2744   // On some systems we have seen signal delivery get "stuck" until the signal
2745   // mask is changed as part of thread termination. Check that the current thread
2746   // has not already terminated (via SR_lock()) - else the following assertion
2747   // will fail because the thread is no longer a JavaThread as the ~JavaThread
2748   // destructor has completed.
2749 
2750   if (thread->SR_lock() == NULL) {
2751     return;
2752   }
2753 
2754   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
2755 
2756   OSThread* osthread = thread->osthread();
2757 
2758   os::SuspendResume::State current = osthread->sr.state();
2759   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
2760     suspend_save_context(osthread, siginfo, context);
2761 
2762     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
2763     os::SuspendResume::State state = osthread->sr.suspended();
2764     if (state == os::SuspendResume::SR_SUSPENDED) {
2765       sigset_t suspend_set;  // signals for sigsuspend()
2766 
2767       // get current set of blocked signals and unblock resume signal
2768       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
2769       sigdelset(&suspend_set, SR_signum);
2770 
2771       // wait here until we are resumed
2772       while (1) {
2773         sigsuspend(&suspend_set);
2774 
2775         os::SuspendResume::State result = osthread->sr.running();
2776         if (result == os::SuspendResume::SR_RUNNING) {
2777           break;
2778         }
2779       }
2780 
2781     } else if (state == os::SuspendResume::SR_RUNNING) {
2782       // request was cancelled, continue
2783     } else {
2784       ShouldNotReachHere();
2785     }
2786 
2787     resume_clear_context(osthread);
2788   } else if (current == os::SuspendResume::SR_RUNNING) {
2789     // request was cancelled, continue
2790   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
2791     // ignore
2792   } else {
2793     ShouldNotReachHere();
2794   }
2795 
2796   errno = old_errno;
2797 }
2798 
2799 static int SR_initialize() {
2800   struct sigaction act;
2801   char *s;
2802   // Get signal number to use for suspend/resume
2803   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
2804     int sig = ::strtol(s, 0, 10);
2805     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
2806         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
2807       SR_signum = sig;
2808     } else {
2809       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
2810               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
2811     }
2812   }
2813 
2814   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
2815         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
2816 
2817   sigemptyset(&SR_sigset);
2818   sigaddset(&SR_sigset, SR_signum);
2819 
2820   // Set up signal handler for suspend/resume.
2821   act.sa_flags = SA_RESTART|SA_SIGINFO;
2822   act.sa_handler = (void (*)(int)) SR_handler;
2823 
2824   // SR_signum is blocked by default.
2825   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
2826 
2827   if (sigaction(SR_signum, &act, 0) == -1) {
2828     return -1;
2829   }
2830 
2831   // Save signal flag
2832   os::Aix::set_our_sigflags(SR_signum, act.sa_flags);
2833   return 0;
2834 }
2835 
2836 static int SR_finalize() {
2837   return 0;
2838 }
2839 
2840 static int sr_notify(OSThread* osthread) {
2841   int status = pthread_kill(osthread->pthread_id(), SR_signum);
2842   assert_status(status == 0, status, "pthread_kill");
2843   return status;
2844 }
2845 
2846 // "Randomly" selected value for how long we want to spin
2847 // before bailing out on suspending a thread, also how often
2848 // we send a signal to a thread we want to resume
2849 static const int RANDOMLY_LARGE_INTEGER = 1000000;
2850 static const int RANDOMLY_LARGE_INTEGER2 = 100;
2851 
2852 // returns true on success and false on error - really an error is fatal
2853 // but this seems the normal response to library errors
2854 static bool do_suspend(OSThread* osthread) {
2855   assert(osthread->sr.is_running(), "thread should be running");
2856   // mark as suspended and send signal
2857 
2858   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
2859     // failed to switch, state wasn't running?
2860     ShouldNotReachHere();
2861     return false;
2862   }
2863 
2864   if (sr_notify(osthread) != 0) {
2865     // try to cancel, switch to running
2866 
2867     os::SuspendResume::State result = osthread->sr.cancel_suspend();
2868     if (result == os::SuspendResume::SR_RUNNING) {
2869       // cancelled
2870       return false;
2871     } else if (result == os::SuspendResume::SR_SUSPENDED) {
2872       // somehow managed to suspend
2873       return true;
2874     } else {
2875       ShouldNotReachHere();
2876       return false;
2877     }
2878   }
2879 
2880   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
2881 
2882   for (int n = 0; !osthread->sr.is_suspended(); n++) {
2883     for (int i = 0; i < RANDOMLY_LARGE_INTEGER2 && !osthread->sr.is_suspended(); i++) {
2884       os::naked_yield();
2885     }
2886 
2887     // timeout, try to cancel the request
2888     if (n >= RANDOMLY_LARGE_INTEGER) {
2889       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
2890       if (cancelled == os::SuspendResume::SR_RUNNING) {
2891         return false;
2892       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
2893         return true;
2894       } else {
2895         ShouldNotReachHere();
2896         return false;
2897       }
2898     }
2899   }
2900 
2901   guarantee(osthread->sr.is_suspended(), "Must be suspended");
2902   return true;
2903 }
2904 
2905 static void do_resume(OSThread* osthread) {
2906   //assert(osthread->sr.is_suspended(), "thread should be suspended");
2907 
2908   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
2909     // failed to switch to WAKEUP_REQUEST
2910     ShouldNotReachHere();
2911     return;
2912   }
2913 
2914   while (!osthread->sr.is_running()) {
2915     if (sr_notify(osthread) == 0) {
2916       for (int n = 0; n < RANDOMLY_LARGE_INTEGER && !osthread->sr.is_running(); n++) {
2917         for (int i = 0; i < 100 && !osthread->sr.is_running(); i++) {
2918           os::naked_yield();
2919         }
2920       }
2921     } else {
2922       ShouldNotReachHere();
2923     }
2924   }
2925 
2926   guarantee(osthread->sr.is_running(), "Must be running!");
2927 }
2928 
2929 ///////////////////////////////////////////////////////////////////////////////////
2930 // signal handling (except suspend/resume)
2931 
2932 // This routine may be used by user applications as a "hook" to catch signals.
2933 // The user-defined signal handler must pass unrecognized signals to this
2934 // routine, and if it returns true (non-zero), then the signal handler must
2935 // return immediately. If the flag "abort_if_unrecognized" is true, then this
2936 // routine will never retun false (zero), but instead will execute a VM panic
2937 // routine kill the process.
2938 //
2939 // If this routine returns false, it is OK to call it again. This allows
2940 // the user-defined signal handler to perform checks either before or after
2941 // the VM performs its own checks. Naturally, the user code would be making
2942 // a serious error if it tried to handle an exception (such as a null check
2943 // or breakpoint) that the VM was generating for its own correct operation.
2944 //
2945 // This routine may recognize any of the following kinds of signals:
2946 //   SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
2947 // It should be consulted by handlers for any of those signals.
2948 //
2949 // The caller of this routine must pass in the three arguments supplied
2950 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
2951 // field of the structure passed to sigaction(). This routine assumes that
2952 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
2953 //
2954 // Note that the VM will print warnings if it detects conflicting signal
2955 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
2956 //
2957 extern "C" JNIEXPORT int
2958 JVM_handle_aix_signal(int signo, siginfo_t* siginfo, void* ucontext, int abort_if_unrecognized);
2959 
2960 // Set thread signal mask (for some reason on AIX sigthreadmask() seems
2961 // to be the thing to call; documentation is not terribly clear about whether
2962 // pthread_sigmask also works, and if it does, whether it does the same.
2963 bool set_thread_signal_mask(int how, const sigset_t* set, sigset_t* oset) {
2964   const int rc = ::pthread_sigmask(how, set, oset);
2965   // return value semantics differ slightly for error case:
2966   // pthread_sigmask returns error number, sigthreadmask -1 and sets global errno
2967   // (so, pthread_sigmask is more theadsafe for error handling)
2968   // But success is always 0.
2969   return rc == 0 ? true : false;
2970 }
2971 
2972 // Function to unblock all signals which are, according
2973 // to POSIX, typical program error signals. If they happen while being blocked,
2974 // they typically will bring down the process immediately.
2975 bool unblock_program_error_signals() {
2976   sigset_t set;
2977   ::sigemptyset(&set);
2978   ::sigaddset(&set, SIGILL);
2979   ::sigaddset(&set, SIGBUS);
2980   ::sigaddset(&set, SIGFPE);
2981   ::sigaddset(&set, SIGSEGV);
2982   return set_thread_signal_mask(SIG_UNBLOCK, &set, NULL);
2983 }
2984 
2985 // Renamed from 'signalHandler' to avoid collision with other shared libs.
2986 static void javaSignalHandler(int sig, siginfo_t* info, void* uc) {
2987   assert(info != NULL && uc != NULL, "it must be old kernel");
2988 
2989   // Never leave program error signals blocked;
2990   // on all our platforms they would bring down the process immediately when
2991   // getting raised while being blocked.
2992   unblock_program_error_signals();
2993 
2994   int orig_errno = errno;  // Preserve errno value over signal handler.
2995   JVM_handle_aix_signal(sig, info, uc, true);
2996   errno = orig_errno;
2997 }
2998 
2999 // This boolean allows users to forward their own non-matching signals
3000 // to JVM_handle_aix_signal, harmlessly.
3001 bool os::Aix::signal_handlers_are_installed = false;
3002 
3003 // For signal-chaining
3004 bool os::Aix::libjsig_is_loaded = false;
3005 typedef struct sigaction *(*get_signal_t)(int);
3006 get_signal_t os::Aix::get_signal_action = NULL;
3007 
3008 struct sigaction* os::Aix::get_chained_signal_action(int sig) {
3009   struct sigaction *actp = NULL;
3010 
3011   if (libjsig_is_loaded) {
3012     // Retrieve the old signal handler from libjsig
3013     actp = (*get_signal_action)(sig);
3014   }
3015   if (actp == NULL) {
3016     // Retrieve the preinstalled signal handler from jvm
3017     actp = os::Posix::get_preinstalled_handler(sig);
3018   }
3019 
3020   return actp;
3021 }
3022 
3023 static bool call_chained_handler(struct sigaction *actp, int sig,
3024                                  siginfo_t *siginfo, void *context) {
3025   // Call the old signal handler
3026   if (actp->sa_handler == SIG_DFL) {
3027     // It's more reasonable to let jvm treat it as an unexpected exception
3028     // instead of taking the default action.
3029     return false;
3030   } else if (actp->sa_handler != SIG_IGN) {
3031     if ((actp->sa_flags & SA_NODEFER) == 0) {
3032       // automaticlly block the signal
3033       sigaddset(&(actp->sa_mask), sig);
3034     }
3035 
3036     sa_handler_t hand = NULL;
3037     sa_sigaction_t sa = NULL;
3038     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3039     // retrieve the chained handler
3040     if (siginfo_flag_set) {
3041       sa = actp->sa_sigaction;
3042     } else {
3043       hand = actp->sa_handler;
3044     }
3045 
3046     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3047       actp->sa_handler = SIG_DFL;
3048     }
3049 
3050     // try to honor the signal mask
3051     sigset_t oset;
3052     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3053 
3054     // call into the chained handler
3055     if (siginfo_flag_set) {
3056       (*sa)(sig, siginfo, context);
3057     } else {
3058       (*hand)(sig);
3059     }
3060 
3061     // restore the signal mask
3062     pthread_sigmask(SIG_SETMASK, &oset, 0);
3063   }
3064   // Tell jvm's signal handler the signal is taken care of.
3065   return true;
3066 }
3067 
3068 bool os::Aix::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3069   bool chained = false;
3070   // signal-chaining
3071   if (UseSignalChaining) {
3072     struct sigaction *actp = get_chained_signal_action(sig);
3073     if (actp != NULL) {
3074       chained = call_chained_handler(actp, sig, siginfo, context);
3075     }
3076   }
3077   return chained;
3078 }
3079 
3080 // for diagnostic
3081 int sigflags[NSIG];
3082 
3083 int os::Aix::get_our_sigflags(int sig) {
3084   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
3085   return sigflags[sig];
3086 }
3087 
3088 void os::Aix::set_our_sigflags(int sig, int flags) {
3089   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
3090   if (sig > 0 && sig < NSIG) {
3091     sigflags[sig] = flags;
3092   }
3093 }
3094 
3095 void os::Aix::set_signal_handler(int sig, bool set_installed) {
3096   // Check for overwrite.
3097   struct sigaction oldAct;
3098   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3099 
3100   void* oldhand = oldAct.sa_sigaction
3101     ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3102     : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3103   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3104       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3105       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)javaSignalHandler)) {
3106     if (AllowUserSignalHandlers || !set_installed) {
3107       // Do not overwrite; user takes responsibility to forward to us.
3108       return;
3109     } else if (UseSignalChaining) {
3110       // save the old handler in jvm
3111       os::Posix::save_preinstalled_handler(sig, oldAct);
3112       // libjsig also interposes the sigaction() call below and saves the
3113       // old sigaction on it own.
3114     } else {
3115       fatal("Encountered unexpected pre-existing sigaction handler "
3116             "%#lx for signal %d.", (long)oldhand, sig);
3117     }
3118   }
3119 
3120   struct sigaction sigAct;
3121   sigfillset(&(sigAct.sa_mask));
3122   if (!set_installed) {
3123     sigAct.sa_handler = SIG_DFL;
3124     sigAct.sa_flags = SA_RESTART;
3125   } else {
3126     sigAct.sa_sigaction = javaSignalHandler;
3127     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3128   }
3129   // Save flags, which are set by ours
3130   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
3131   sigflags[sig] = sigAct.sa_flags;
3132 
3133   int ret = sigaction(sig, &sigAct, &oldAct);
3134   assert(ret == 0, "check");
3135 
3136   void* oldhand2 = oldAct.sa_sigaction
3137                  ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3138                  : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3139   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3140 }
3141 
3142 // install signal handlers for signals that HotSpot needs to
3143 // handle in order to support Java-level exception handling.
3144 void os::Aix::install_signal_handlers() {
3145   if (!signal_handlers_are_installed) {
3146     signal_handlers_are_installed = true;
3147 
3148     // signal-chaining
3149     typedef void (*signal_setting_t)();
3150     signal_setting_t begin_signal_setting = NULL;
3151     signal_setting_t end_signal_setting = NULL;
3152     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3153                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3154     if (begin_signal_setting != NULL) {
3155       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3156                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3157       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3158                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3159       libjsig_is_loaded = true;
3160       assert(UseSignalChaining, "should enable signal-chaining");
3161     }
3162     if (libjsig_is_loaded) {
3163       // Tell libjsig jvm is setting signal handlers.
3164       (*begin_signal_setting)();
3165     }
3166 
3167     set_signal_handler(SIGSEGV, true);
3168     set_signal_handler(SIGPIPE, true);
3169     set_signal_handler(SIGBUS, true);
3170     set_signal_handler(SIGILL, true);
3171     set_signal_handler(SIGFPE, true);
3172     set_signal_handler(SIGTRAP, true);
3173     set_signal_handler(SIGXFSZ, true);
3174 
3175     if (libjsig_is_loaded) {
3176       // Tell libjsig jvm finishes setting signal handlers.
3177       (*end_signal_setting)();
3178     }
3179 
3180     // We don't activate signal checker if libjsig is in place, we trust ourselves
3181     // and if UserSignalHandler is installed all bets are off.
3182     // Log that signal checking is off only if -verbose:jni is specified.
3183     if (CheckJNICalls) {
3184       if (libjsig_is_loaded) {
3185         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3186         check_signals = false;
3187       }
3188       if (AllowUserSignalHandlers) {
3189         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3190         check_signals = false;
3191       }
3192       // Need to initialize check_signal_done.
3193       ::sigemptyset(&check_signal_done);
3194     }
3195   }
3196 }
3197 
3198 static const char* get_signal_handler_name(address handler,
3199                                            char* buf, int buflen) {
3200   int offset;
3201   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3202   if (found) {
3203     // skip directory names
3204     const char *p1, *p2;
3205     p1 = buf;
3206     size_t len = strlen(os::file_separator());
3207     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3208     // The way os::dll_address_to_library_name is implemented on Aix
3209     // right now, it always returns -1 for the offset which is not
3210     // terribly informative.
3211     // Will fix that. For now, omit the offset.
3212     jio_snprintf(buf, buflen, "%s", p1);
3213   } else {
3214     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3215   }
3216   return buf;
3217 }
3218 
3219 static void print_signal_handler(outputStream* st, int sig,
3220                                  char* buf, size_t buflen) {
3221   struct sigaction sa;
3222   sigaction(sig, NULL, &sa);
3223 
3224   st->print("%s: ", os::exception_name(sig, buf, buflen));
3225 
3226   address handler = (sa.sa_flags & SA_SIGINFO)
3227     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3228     : CAST_FROM_FN_PTR(address, sa.sa_handler);
3229 
3230   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3231     st->print("SIG_DFL");
3232   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3233     st->print("SIG_IGN");
3234   } else {
3235     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3236   }
3237 
3238   // Print readable mask.
3239   st->print(", sa_mask[0]=");
3240   os::Posix::print_signal_set_short(st, &sa.sa_mask);
3241 
3242   address rh = VMError::get_resetted_sighandler(sig);
3243   // May be, handler was resetted by VMError?
3244   if (rh != NULL) {
3245     handler = rh;
3246     sa.sa_flags = VMError::get_resetted_sigflags(sig);
3247   }
3248 
3249   // Print textual representation of sa_flags.
3250   st->print(", sa_flags=");
3251   os::Posix::print_sa_flags(st, sa.sa_flags);
3252 
3253   // Check: is it our handler?
3254   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)javaSignalHandler) ||
3255       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3256     // It is our signal handler.
3257     // Check for flags, reset system-used one!
3258     if ((int)sa.sa_flags != os::Aix::get_our_sigflags(sig)) {
3259       st->print(", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3260                 os::Aix::get_our_sigflags(sig));
3261     }
3262   }
3263   st->cr();
3264 }
3265 
3266 #define DO_SIGNAL_CHECK(sig) \
3267   if (!sigismember(&check_signal_done, sig)) \
3268     os::Aix::check_signal_handler(sig)
3269 
3270 // This method is a periodic task to check for misbehaving JNI applications
3271 // under CheckJNI, we can add any periodic checks here
3272 
3273 void os::run_periodic_checks() {
3274 
3275   if (check_signals == false) return;
3276 
3277   // SEGV and BUS if overridden could potentially prevent
3278   // generation of hs*.log in the event of a crash, debugging
3279   // such a case can be very challenging, so we absolutely
3280   // check the following for a good measure:
3281   DO_SIGNAL_CHECK(SIGSEGV);
3282   DO_SIGNAL_CHECK(SIGILL);
3283   DO_SIGNAL_CHECK(SIGFPE);
3284   DO_SIGNAL_CHECK(SIGBUS);
3285   DO_SIGNAL_CHECK(SIGPIPE);
3286   DO_SIGNAL_CHECK(SIGXFSZ);
3287   if (UseSIGTRAP) {
3288     DO_SIGNAL_CHECK(SIGTRAP);
3289   }
3290 
3291   // ReduceSignalUsage allows the user to override these handlers
3292   // see comments at the very top and jvm_md.h
3293   if (!ReduceSignalUsage) {
3294     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3295     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3296     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3297     DO_SIGNAL_CHECK(BREAK_SIGNAL);
3298   }
3299 
3300   DO_SIGNAL_CHECK(SR_signum);
3301 }
3302 
3303 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3304 
3305 static os_sigaction_t os_sigaction = NULL;
3306 
3307 void os::Aix::check_signal_handler(int sig) {
3308   char buf[O_BUFLEN];
3309   address jvmHandler = NULL;
3310 
3311   struct sigaction act;
3312   if (os_sigaction == NULL) {
3313     // only trust the default sigaction, in case it has been interposed
3314     os_sigaction = CAST_TO_FN_PTR(os_sigaction_t, dlsym(RTLD_DEFAULT, "sigaction"));
3315     if (os_sigaction == NULL) return;
3316   }
3317 
3318   os_sigaction(sig, (struct sigaction*)NULL, &act);
3319 
3320   address thisHandler = (act.sa_flags & SA_SIGINFO)
3321     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3322     : CAST_FROM_FN_PTR(address, act.sa_handler);
3323 
3324   switch(sig) {
3325   case SIGSEGV:
3326   case SIGBUS:
3327   case SIGFPE:
3328   case SIGPIPE:
3329   case SIGILL:
3330   case SIGXFSZ:
3331     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)javaSignalHandler);
3332     break;
3333 
3334   case SHUTDOWN1_SIGNAL:
3335   case SHUTDOWN2_SIGNAL:
3336   case SHUTDOWN3_SIGNAL:
3337   case BREAK_SIGNAL:
3338     jvmHandler = (address)user_handler();
3339     break;
3340 
3341   default:
3342     if (sig == SR_signum) {
3343       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3344     } else {
3345       return;
3346     }
3347     break;
3348   }
3349 
3350   if (thisHandler != jvmHandler) {
3351     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3352     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3353     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3354     // No need to check this sig any longer
3355     sigaddset(&check_signal_done, sig);
3356     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
3357     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
3358       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
3359                     exception_name(sig, buf, O_BUFLEN));
3360     }
3361   } else if (os::Aix::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Aix::get_our_sigflags(sig)) {
3362     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3363     tty->print("expected:");
3364     os::Posix::print_sa_flags(tty, os::Aix::get_our_sigflags(sig));
3365     tty->cr();
3366     tty->print("  found:");
3367     os::Posix::print_sa_flags(tty, act.sa_flags);
3368     tty->cr();
3369     // No need to check this sig any longer
3370     sigaddset(&check_signal_done, sig);
3371   }
3372 
3373   // Dump all the signal
3374   if (sigismember(&check_signal_done, sig)) {
3375     print_signal_handlers(tty, buf, O_BUFLEN);
3376   }
3377 }
3378 
3379 // To install functions for atexit system call
3380 extern "C" {
3381   static void perfMemory_exit_helper() {
3382     perfMemory_exit();
3383   }
3384 }
3385 
3386 // This is called _before_ the most of global arguments have been parsed.
3387 void os::init(void) {
3388   // This is basic, we want to know if that ever changes.
3389   // (Shared memory boundary is supposed to be a 256M aligned.)
3390   assert(SHMLBA == ((uint64_t)0x10000000ULL)/*256M*/, "unexpected");
3391 
3392   // Record process break at startup.
3393   g_brk_at_startup = (address) ::sbrk(0);
3394   assert(g_brk_at_startup != (address) -1, "sbrk failed");
3395 
3396   // First off, we need to know whether we run on AIX or PASE, and
3397   // the OS level we run on.
3398   os::Aix::initialize_os_info();
3399 
3400   // Scan environment (SPEC1170 behaviour, etc).
3401   os::Aix::scan_environment();
3402 
3403   // Probe multipage support.
3404   query_multipage_support();
3405 
3406   // Act like we only have one page size by eliminating corner cases which
3407   // we did not support very well anyway.
3408   // We have two input conditions:
3409   // 1) Data segment page size. This is controlled by linker setting (datapsize) on the
3410   //    launcher, and/or by LDR_CNTRL environment variable. The latter overrules the linker
3411   //    setting.
3412   //    Data segment page size is important for us because it defines the thread stack page
3413   //    size, which is needed for guard page handling, stack banging etc.
3414   // 2) The ability to allocate 64k pages dynamically. If this is a given, java heap can
3415   //    and should be allocated with 64k pages.
3416   //
3417   // So, we do the following:
3418   // LDR_CNTRL    can_use_64K_pages_dynamically       what we do                      remarks
3419   // 4K           no                                  4K                              old systems (aix 5.2, as/400 v5r4) or new systems with AME activated
3420   // 4k           yes                                 64k (treat 4k stacks as 64k)    different loader than java and standard settings
3421   // 64k          no              --- AIX 5.2 ? ---
3422   // 64k          yes                                 64k                             new systems and standard java loader (we set datapsize=64k when linking)
3423 
3424   // We explicitly leave no option to change page size, because only upgrading would work,
3425   // not downgrading (if stack page size is 64k you cannot pretend its 4k).
3426 
3427   if (g_multipage_support.datapsize == 4*K) {
3428     // datapsize = 4K. Data segment, thread stacks are 4K paged.
3429     if (g_multipage_support.can_use_64K_pages) {
3430       // .. but we are able to use 64K pages dynamically.
3431       // This would be typical for java launchers which are not linked
3432       // with datapsize=64K (like, any other launcher but our own).
3433       //
3434       // In this case it would be smart to allocate the java heap with 64K
3435       // to get the performance benefit, and to fake 64k pages for the
3436       // data segment (when dealing with thread stacks).
3437       //
3438       // However, leave a possibility to downgrade to 4K, using
3439       // -XX:-Use64KPages.
3440       if (Use64KPages) {
3441         trcVerbose("64K page mode (faked for data segment)");
3442         Aix::_page_size = 64*K;
3443       } else {
3444         trcVerbose("4K page mode (Use64KPages=off)");
3445         Aix::_page_size = 4*K;
3446       }
3447     } else {
3448       // .. and not able to allocate 64k pages dynamically. Here, just
3449       // fall back to 4K paged mode and use mmap for everything.
3450       trcVerbose("4K page mode");
3451       Aix::_page_size = 4*K;
3452       FLAG_SET_ERGO(Use64KPages, false);
3453     }
3454   } else {
3455     // datapsize = 64k. Data segment, thread stacks are 64k paged.
3456     // This normally means that we can allocate 64k pages dynamically.
3457     // (There is one special case where this may be false: EXTSHM=on.
3458     // but we decided to not support that mode).
3459     assert0(g_multipage_support.can_use_64K_pages);
3460     Aix::_page_size = 64*K;
3461     trcVerbose("64K page mode");
3462     FLAG_SET_ERGO(Use64KPages, true);
3463   }
3464 
3465   // For now UseLargePages is just ignored.
3466   FLAG_SET_ERGO(UseLargePages, false);
3467   _page_sizes[0] = 0;
3468 
3469   // debug trace
3470   trcVerbose("os::vm_page_size %s", describe_pagesize(os::vm_page_size()));
3471 
3472   // Next, we need to initialize libo4 and libperfstat libraries.
3473   if (os::Aix::on_pase()) {
3474     os::Aix::initialize_libo4();
3475   } else {
3476     os::Aix::initialize_libperfstat();
3477   }
3478 
3479   // Reset the perfstat information provided by ODM.
3480   if (os::Aix::on_aix()) {
3481     libperfstat::perfstat_reset();
3482   }
3483 
3484   // Now initialze basic system properties. Note that for some of the values we
3485   // need libperfstat etc.
3486   os::Aix::initialize_system_info();
3487 
3488   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3489 
3490   init_random(1234567);
3491 
3492   // _main_thread points to the thread that created/loaded the JVM.
3493   Aix::_main_thread = pthread_self();
3494 
3495   initial_time_count = os::elapsed_counter();
3496 
3497   os::Posix::init();
3498 }
3499 
3500 // This is called _after_ the global arguments have been parsed.
3501 jint os::init_2(void) {
3502 
3503   // This could be set after os::Posix::init() but all platforms
3504   // have to set it the same so we have to mirror Solaris.
3505   DEBUG_ONLY(os::set_mutex_init_done();)
3506 
3507   os::Posix::init_2();
3508 
3509   if (os::Aix::on_pase()) {
3510     trcVerbose("Running on PASE.");
3511   } else {
3512     trcVerbose("Running on AIX (not PASE).");
3513   }
3514 
3515   trcVerbose("processor count: %d", os::_processor_count);
3516   trcVerbose("physical memory: %lu", Aix::_physical_memory);
3517 
3518   // Initially build up the loaded dll map.
3519   LoadedLibraries::reload();
3520   if (Verbose) {
3521     trcVerbose("Loaded Libraries: ");
3522     LoadedLibraries::print(tty);
3523   }
3524 
3525   // initialize suspend/resume support - must do this before signal_sets_init()
3526   if (SR_initialize() != 0) {
3527     perror("SR_initialize failed");
3528     return JNI_ERR;
3529   }
3530 
3531   Aix::signal_sets_init();
3532   Aix::install_signal_handlers();
3533   // Initialize data for jdk.internal.misc.Signal
3534   if (!ReduceSignalUsage) {
3535     jdk_misc_signal_init();
3536   }
3537 
3538   // Check and sets minimum stack sizes against command line options
3539   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
3540     return JNI_ERR;
3541   }
3542 
3543   if (UseNUMA) {
3544     UseNUMA = false;
3545     warning("NUMA optimizations are not available on this OS.");
3546   }
3547 
3548   if (MaxFDLimit) {
3549     // Set the number of file descriptors to max. print out error
3550     // if getrlimit/setrlimit fails but continue regardless.
3551     struct rlimit nbr_files;
3552     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3553     if (status != 0) {
3554       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
3555     } else {
3556       nbr_files.rlim_cur = nbr_files.rlim_max;
3557       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3558       if (status != 0) {
3559         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
3560       }
3561     }
3562   }
3563 
3564   if (PerfAllowAtExitRegistration) {
3565     // Only register atexit functions if PerfAllowAtExitRegistration is set.
3566     // At exit functions can be delayed until process exit time, which
3567     // can be problematic for embedded VM situations. Embedded VMs should
3568     // call DestroyJavaVM() to assure that VM resources are released.
3569 
3570     // Note: perfMemory_exit_helper atexit function may be removed in
3571     // the future if the appropriate cleanup code can be added to the
3572     // VM_Exit VMOperation's doit method.
3573     if (atexit(perfMemory_exit_helper) != 0) {
3574       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
3575     }
3576   }
3577 
3578   return JNI_OK;
3579 }
3580 
3581 // Mark the polling page as unreadable
3582 void os::make_polling_page_unreadable(void) {
3583   if (!guard_memory((char*)_polling_page, Aix::page_size())) {
3584     fatal("Could not disable polling page");
3585   }
3586 };
3587 
3588 // Mark the polling page as readable
3589 void os::make_polling_page_readable(void) {
3590   // Changed according to os_linux.cpp.
3591   if (!checked_mprotect((char *)_polling_page, Aix::page_size(), PROT_READ)) {
3592     fatal("Could not enable polling page at " PTR_FORMAT, _polling_page);
3593   }
3594 };
3595 
3596 int os::active_processor_count() {
3597   // User has overridden the number of active processors
3598   if (ActiveProcessorCount > 0) {
3599     log_trace(os)("active_processor_count: "
3600                   "active processor count set by user : %d",
3601                   ActiveProcessorCount);
3602     return ActiveProcessorCount;
3603   }
3604 
3605   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
3606   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
3607   return online_cpus;
3608 }
3609 
3610 void os::set_native_thread_name(const char *name) {
3611   // Not yet implemented.
3612   return;
3613 }
3614 
3615 bool os::distribute_processes(uint length, uint* distribution) {
3616   // Not yet implemented.
3617   return false;
3618 }
3619 
3620 bool os::bind_to_processor(uint processor_id) {
3621   // Not yet implemented.
3622   return false;
3623 }
3624 
3625 void os::SuspendedThreadTask::internal_do_task() {
3626   if (do_suspend(_thread->osthread())) {
3627     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
3628     do_task(context);
3629     do_resume(_thread->osthread());
3630   }
3631 }
3632 
3633 ////////////////////////////////////////////////////////////////////////////////
3634 // debug support
3635 
3636 bool os::find(address addr, outputStream* st) {
3637 
3638   st->print(PTR_FORMAT ": ", addr);
3639 
3640   loaded_module_t lm;
3641   if (LoadedLibraries::find_for_text_address(addr, &lm) != NULL ||
3642       LoadedLibraries::find_for_data_address(addr, &lm) != NULL) {
3643     st->print_cr("%s", lm.path);
3644     return true;
3645   }
3646 
3647   return false;
3648 }
3649 
3650 ////////////////////////////////////////////////////////////////////////////////
3651 // misc
3652 
3653 // This does not do anything on Aix. This is basically a hook for being
3654 // able to use structured exception handling (thread-local exception filters)
3655 // on, e.g., Win32.
3656 void
3657 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
3658                          JavaCallArguments* args, Thread* thread) {
3659   f(value, method, args, thread);
3660 }
3661 
3662 void os::print_statistics() {
3663 }
3664 
3665 bool os::message_box(const char* title, const char* message) {
3666   int i;
3667   fdStream err(defaultStream::error_fd());
3668   for (i = 0; i < 78; i++) err.print_raw("=");
3669   err.cr();
3670   err.print_raw_cr(title);
3671   for (i = 0; i < 78; i++) err.print_raw("-");
3672   err.cr();
3673   err.print_raw_cr(message);
3674   for (i = 0; i < 78; i++) err.print_raw("=");
3675   err.cr();
3676 
3677   char buf[16];
3678   // Prevent process from exiting upon "read error" without consuming all CPU
3679   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3680 
3681   return buf[0] == 'y' || buf[0] == 'Y';
3682 }
3683 
3684 // Is a (classpath) directory empty?
3685 bool os::dir_is_empty(const char* path) {
3686   DIR *dir = NULL;
3687   struct dirent *ptr;
3688 
3689   dir = opendir(path);
3690   if (dir == NULL) return true;
3691 
3692   /* Scan the directory */
3693   bool result = true;
3694   while (result && (ptr = readdir(dir)) != NULL) {
3695     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
3696       result = false;
3697     }
3698   }
3699   closedir(dir);
3700   return result;
3701 }
3702 
3703 // This code originates from JDK's sysOpen and open64_w
3704 // from src/solaris/hpi/src/system_md.c
3705 
3706 int os::open(const char *path, int oflag, int mode) {
3707 
3708   if (strlen(path) > MAX_PATH - 1) {
3709     errno = ENAMETOOLONG;
3710     return -1;
3711   }
3712   int fd;
3713 
3714   fd = ::open64(path, oflag, mode);
3715   if (fd == -1) return -1;
3716 
3717   // If the open succeeded, the file might still be a directory.
3718   {
3719     struct stat64 buf64;
3720     int ret = ::fstat64(fd, &buf64);
3721     int st_mode = buf64.st_mode;
3722 
3723     if (ret != -1) {
3724       if ((st_mode & S_IFMT) == S_IFDIR) {
3725         errno = EISDIR;
3726         ::close(fd);
3727         return -1;
3728       }
3729     } else {
3730       ::close(fd);
3731       return -1;
3732     }
3733   }
3734 
3735   // All file descriptors that are opened in the JVM and not
3736   // specifically destined for a subprocess should have the
3737   // close-on-exec flag set. If we don't set it, then careless 3rd
3738   // party native code might fork and exec without closing all
3739   // appropriate file descriptors (e.g. as we do in closeDescriptors in
3740   // UNIXProcess.c), and this in turn might:
3741   //
3742   // - cause end-of-file to fail to be detected on some file
3743   //   descriptors, resulting in mysterious hangs, or
3744   //
3745   // - might cause an fopen in the subprocess to fail on a system
3746   //   suffering from bug 1085341.
3747   //
3748   // (Yes, the default setting of the close-on-exec flag is a Unix
3749   // design flaw.)
3750   //
3751   // See:
3752   // 1085341: 32-bit stdio routines should support file descriptors >255
3753   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
3754   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
3755 #ifdef FD_CLOEXEC
3756   {
3757     int flags = ::fcntl(fd, F_GETFD);
3758     if (flags != -1)
3759       ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
3760   }
3761 #endif
3762 
3763   return fd;
3764 }
3765 
3766 // create binary file, rewriting existing file if required
3767 int os::create_binary_file(const char* path, bool rewrite_existing) {
3768   int oflags = O_WRONLY | O_CREAT;
3769   if (!rewrite_existing) {
3770     oflags |= O_EXCL;
3771   }
3772   return ::open64(path, oflags, S_IREAD | S_IWRITE);
3773 }
3774 
3775 // return current position of file pointer
3776 jlong os::current_file_offset(int fd) {
3777   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
3778 }
3779 
3780 // move file pointer to the specified offset
3781 jlong os::seek_to_file_offset(int fd, jlong offset) {
3782   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
3783 }
3784 
3785 // This code originates from JDK's sysAvailable
3786 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
3787 
3788 int os::available(int fd, jlong *bytes) {
3789   jlong cur, end;
3790   int mode;
3791   struct stat64 buf64;
3792 
3793   if (::fstat64(fd, &buf64) >= 0) {
3794     mode = buf64.st_mode;
3795     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
3796       int n;
3797       if (::ioctl(fd, FIONREAD, &n) >= 0) {
3798         *bytes = n;
3799         return 1;
3800       }
3801     }
3802   }
3803   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
3804     return 0;
3805   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
3806     return 0;
3807   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
3808     return 0;
3809   }
3810   *bytes = end - cur;
3811   return 1;
3812 }
3813 
3814 // Map a block of memory.
3815 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
3816                         char *addr, size_t bytes, bool read_only,
3817                         bool allow_exec) {
3818   int prot;
3819   int flags = MAP_PRIVATE;
3820 
3821   if (read_only) {
3822     prot = PROT_READ;
3823     flags = MAP_SHARED;
3824   } else {
3825     prot = PROT_READ | PROT_WRITE;
3826     flags = MAP_PRIVATE;
3827   }
3828 
3829   if (allow_exec) {
3830     prot |= PROT_EXEC;
3831   }
3832 
3833   if (addr != NULL) {
3834     flags |= MAP_FIXED;
3835   }
3836 
3837   // Allow anonymous mappings if 'fd' is -1.
3838   if (fd == -1) {
3839     flags |= MAP_ANONYMOUS;
3840   }
3841 
3842   char* mapped_address = (char*)::mmap(addr, (size_t)bytes, prot, flags,
3843                                      fd, file_offset);
3844   if (mapped_address == MAP_FAILED) {
3845     return NULL;
3846   }
3847   return mapped_address;
3848 }
3849 
3850 // Remap a block of memory.
3851 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
3852                           char *addr, size_t bytes, bool read_only,
3853                           bool allow_exec) {
3854   // same as map_memory() on this OS
3855   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
3856                         allow_exec);
3857 }
3858 
3859 // Unmap a block of memory.
3860 bool os::pd_unmap_memory(char* addr, size_t bytes) {
3861   return munmap(addr, bytes) == 0;
3862 }
3863 
3864 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
3865 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
3866 // of a thread.
3867 //
3868 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
3869 // the fast estimate available on the platform.
3870 
3871 jlong os::current_thread_cpu_time() {
3872   // return user + sys since the cost is the same
3873   const jlong n = os::thread_cpu_time(Thread::current(), true /* user + sys */);
3874   assert(n >= 0, "negative CPU time");
3875   return n;
3876 }
3877 
3878 jlong os::thread_cpu_time(Thread* thread) {
3879   // consistent with what current_thread_cpu_time() returns
3880   const jlong n = os::thread_cpu_time(thread, true /* user + sys */);
3881   assert(n >= 0, "negative CPU time");
3882   return n;
3883 }
3884 
3885 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
3886   const jlong n = os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
3887   assert(n >= 0, "negative CPU time");
3888   return n;
3889 }
3890 
3891 static bool thread_cpu_time_unchecked(Thread* thread, jlong* p_sys_time, jlong* p_user_time) {
3892   bool error = false;
3893 
3894   jlong sys_time = 0;
3895   jlong user_time = 0;
3896 
3897   // Reimplemented using getthrds64().
3898   //
3899   // Works like this:
3900   // For the thread in question, get the kernel thread id. Then get the
3901   // kernel thread statistics using that id.
3902   //
3903   // This only works of course when no pthread scheduling is used,
3904   // i.e. there is a 1:1 relationship to kernel threads.
3905   // On AIX, see AIXTHREAD_SCOPE variable.
3906 
3907   pthread_t pthtid = thread->osthread()->pthread_id();
3908 
3909   // retrieve kernel thread id for the pthread:
3910   tid64_t tid = 0;
3911   struct __pthrdsinfo pinfo;
3912   // I just love those otherworldly IBM APIs which force me to hand down
3913   // dummy buffers for stuff I dont care for...
3914   char dummy[1];
3915   int dummy_size = sizeof(dummy);
3916   if (pthread_getthrds_np(&pthtid, PTHRDSINFO_QUERY_TID, &pinfo, sizeof(pinfo),
3917                           dummy, &dummy_size) == 0) {
3918     tid = pinfo.__pi_tid;
3919   } else {
3920     tty->print_cr("pthread_getthrds_np failed.");
3921     error = true;
3922   }
3923 
3924   // retrieve kernel timing info for that kernel thread
3925   if (!error) {
3926     struct thrdentry64 thrdentry;
3927     if (getthrds64(getpid(), &thrdentry, sizeof(thrdentry), &tid, 1) == 1) {
3928       sys_time = thrdentry.ti_ru.ru_stime.tv_sec * 1000000000LL + thrdentry.ti_ru.ru_stime.tv_usec * 1000LL;
3929       user_time = thrdentry.ti_ru.ru_utime.tv_sec * 1000000000LL + thrdentry.ti_ru.ru_utime.tv_usec * 1000LL;
3930     } else {
3931       tty->print_cr("pthread_getthrds_np failed.");
3932       error = true;
3933     }
3934   }
3935 
3936   if (p_sys_time) {
3937     *p_sys_time = sys_time;
3938   }
3939 
3940   if (p_user_time) {
3941     *p_user_time = user_time;
3942   }
3943 
3944   if (error) {
3945     return false;
3946   }
3947 
3948   return true;
3949 }
3950 
3951 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
3952   jlong sys_time;
3953   jlong user_time;
3954 
3955   if (!thread_cpu_time_unchecked(thread, &sys_time, &user_time)) {
3956     return -1;
3957   }
3958 
3959   return user_sys_cpu_time ? sys_time + user_time : user_time;
3960 }
3961 
3962 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
3963   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
3964   info_ptr->may_skip_backward = false;     // elapsed time not wall time
3965   info_ptr->may_skip_forward = false;      // elapsed time not wall time
3966   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
3967 }
3968 
3969 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
3970   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
3971   info_ptr->may_skip_backward = false;     // elapsed time not wall time
3972   info_ptr->may_skip_forward = false;      // elapsed time not wall time
3973   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
3974 }
3975 
3976 bool os::is_thread_cpu_time_supported() {
3977   return true;
3978 }
3979 
3980 // System loadavg support. Returns -1 if load average cannot be obtained.
3981 // For now just return the system wide load average (no processor sets).
3982 int os::loadavg(double values[], int nelem) {
3983 
3984   guarantee(nelem >= 0 && nelem <= 3, "argument error");
3985   guarantee(values, "argument error");
3986 
3987   if (os::Aix::on_pase()) {
3988 
3989     // AS/400 PASE: use libo4 porting library
3990     double v[3] = { 0.0, 0.0, 0.0 };
3991 
3992     if (libo4::get_load_avg(v, v + 1, v + 2)) {
3993       for (int i = 0; i < nelem; i ++) {
3994         values[i] = v[i];
3995       }
3996       return nelem;
3997     } else {
3998       return -1;
3999     }
4000 
4001   } else {
4002 
4003     // AIX: use libperfstat
4004     libperfstat::cpuinfo_t ci;
4005     if (libperfstat::get_cpuinfo(&ci)) {
4006       for (int i = 0; i < nelem; i++) {
4007         values[i] = ci.loadavg[i];
4008       }
4009     } else {
4010       return -1;
4011     }
4012     return nelem;
4013   }
4014 }
4015 
4016 void os::pause() {
4017   char filename[MAX_PATH];
4018   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4019     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4020   } else {
4021     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4022   }
4023 
4024   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4025   if (fd != -1) {
4026     struct stat buf;
4027     ::close(fd);
4028     while (::stat(filename, &buf) == 0) {
4029       (void)::poll(NULL, 0, 100);
4030     }
4031   } else {
4032     trcVerbose("Could not open pause file '%s', continuing immediately.", filename);
4033   }
4034 }
4035 
4036 bool os::is_primordial_thread(void) {
4037   if (pthread_self() == (pthread_t)1) {
4038     return true;
4039   } else {
4040     return false;
4041   }
4042 }
4043 
4044 // OS recognitions (PASE/AIX, OS level) call this before calling any
4045 // one of Aix::on_pase(), Aix::os_version() static
4046 void os::Aix::initialize_os_info() {
4047 
4048   assert(_on_pase == -1 && _os_version == 0, "already called.");
4049 
4050   struct utsname uts;
4051   memset(&uts, 0, sizeof(uts));
4052   strcpy(uts.sysname, "?");
4053   if (::uname(&uts) == -1) {
4054     trcVerbose("uname failed (%d)", errno);
4055     guarantee(0, "Could not determine whether we run on AIX or PASE");
4056   } else {
4057     trcVerbose("uname says: sysname \"%s\" version \"%s\" release \"%s\" "
4058                "node \"%s\" machine \"%s\"\n",
4059                uts.sysname, uts.version, uts.release, uts.nodename, uts.machine);
4060     const int major = atoi(uts.version);
4061     assert(major > 0, "invalid OS version");
4062     const int minor = atoi(uts.release);
4063     assert(minor > 0, "invalid OS release");
4064     _os_version = (major << 24) | (minor << 16);
4065     char ver_str[20] = {0};
4066     char *name_str = "unknown OS";
4067     if (strcmp(uts.sysname, "OS400") == 0) {
4068       // We run on AS/400 PASE. We do not support versions older than V5R4M0.
4069       _on_pase = 1;
4070       if (os_version_short() < 0x0504) {
4071         trcVerbose("OS/400 releases older than V5R4M0 not supported.");
4072         assert(false, "OS/400 release too old.");
4073       }
4074       name_str = "OS/400 (pase)";
4075       jio_snprintf(ver_str, sizeof(ver_str), "%u.%u", major, minor);
4076     } else if (strcmp(uts.sysname, "AIX") == 0) {
4077       // We run on AIX. We do not support versions older than AIX 5.3.
4078       _on_pase = 0;
4079       // Determine detailed AIX version: Version, Release, Modification, Fix Level.
4080       odmWrapper::determine_os_kernel_version(&_os_version);
4081       if (os_version_short() < 0x0503) {
4082         trcVerbose("AIX release older than AIX 5.3 not supported.");
4083         assert(false, "AIX release too old.");
4084       }
4085       name_str = "AIX";
4086       jio_snprintf(ver_str, sizeof(ver_str), "%u.%u.%u.%u",
4087                    major, minor, (_os_version >> 8) & 0xFF, _os_version & 0xFF);
4088     } else {
4089       assert(false, name_str);
4090     }
4091     trcVerbose("We run on %s %s", name_str, ver_str);
4092   }
4093 
4094   guarantee(_on_pase != -1 && _os_version, "Could not determine AIX/OS400 release");
4095 } // end: os::Aix::initialize_os_info()
4096 
4097 // Scan environment for important settings which might effect the VM.
4098 // Trace out settings. Warn about invalid settings and/or correct them.
4099 //
4100 // Must run after os::Aix::initialue_os_info().
4101 void os::Aix::scan_environment() {
4102 
4103   char* p;
4104   int rc;
4105 
4106   // Warn explicity if EXTSHM=ON is used. That switch changes how
4107   // System V shared memory behaves. One effect is that page size of
4108   // shared memory cannot be change dynamically, effectivly preventing
4109   // large pages from working.
4110   // This switch was needed on AIX 32bit, but on AIX 64bit the general
4111   // recommendation is (in OSS notes) to switch it off.
4112   p = ::getenv("EXTSHM");
4113   trcVerbose("EXTSHM=%s.", p ? p : "<unset>");
4114   if (p && strcasecmp(p, "ON") == 0) {
4115     _extshm = 1;
4116     trcVerbose("*** Unsupported mode! Please remove EXTSHM from your environment! ***");
4117     if (!AllowExtshm) {
4118       // We allow under certain conditions the user to continue. However, we want this
4119       // to be a fatal error by default. On certain AIX systems, leaving EXTSHM=ON means
4120       // that the VM is not able to allocate 64k pages for the heap.
4121       // We do not want to run with reduced performance.
4122       vm_exit_during_initialization("EXTSHM is ON. Please remove EXTSHM from your environment.");
4123     }
4124   } else {
4125     _extshm = 0;
4126   }
4127 
4128   // SPEC1170 behaviour: will change the behaviour of a number of POSIX APIs.
4129   // Not tested, not supported.
4130   //
4131   // Note that it might be worth the trouble to test and to require it, if only to
4132   // get useful return codes for mprotect.
4133   //
4134   // Note: Setting XPG_SUS_ENV in the process is too late. Must be set earlier (before
4135   // exec() ? before loading the libjvm ? ....)
4136   p = ::getenv("XPG_SUS_ENV");
4137   trcVerbose("XPG_SUS_ENV=%s.", p ? p : "<unset>");
4138   if (p && strcmp(p, "ON") == 0) {
4139     _xpg_sus_mode = 1;
4140     trcVerbose("Unsupported setting: XPG_SUS_ENV=ON");
4141     // This is not supported. Worst of all, it changes behaviour of mmap MAP_FIXED to
4142     // clobber address ranges. If we ever want to support that, we have to do some
4143     // testing first.
4144     guarantee(false, "XPG_SUS_ENV=ON not supported");
4145   } else {
4146     _xpg_sus_mode = 0;
4147   }
4148 
4149   if (os::Aix::on_pase()) {
4150     p = ::getenv("QIBM_MULTI_THREADED");
4151     trcVerbose("QIBM_MULTI_THREADED=%s.", p ? p : "<unset>");
4152   }
4153 
4154   p = ::getenv("LDR_CNTRL");
4155   trcVerbose("LDR_CNTRL=%s.", p ? p : "<unset>");
4156   if (os::Aix::on_pase() && os::Aix::os_version_short() == 0x0701) {
4157     if (p && ::strstr(p, "TEXTPSIZE")) {
4158       trcVerbose("*** WARNING - LDR_CNTRL contains TEXTPSIZE. "
4159         "you may experience hangs or crashes on OS/400 V7R1.");
4160     }
4161   }
4162 
4163   p = ::getenv("AIXTHREAD_GUARDPAGES");
4164   trcVerbose("AIXTHREAD_GUARDPAGES=%s.", p ? p : "<unset>");
4165 
4166 } // end: os::Aix::scan_environment()
4167 
4168 // PASE: initialize the libo4 library (PASE porting library).
4169 void os::Aix::initialize_libo4() {
4170   guarantee(os::Aix::on_pase(), "OS/400 only.");
4171   if (!libo4::init()) {
4172     trcVerbose("libo4 initialization failed.");
4173     assert(false, "libo4 initialization failed");
4174   } else {
4175     trcVerbose("libo4 initialized.");
4176   }
4177 }
4178 
4179 // AIX: initialize the libperfstat library.
4180 void os::Aix::initialize_libperfstat() {
4181   assert(os::Aix::on_aix(), "AIX only");
4182   if (!libperfstat::init()) {
4183     trcVerbose("libperfstat initialization failed.");
4184     assert(false, "libperfstat initialization failed");
4185   } else {
4186     trcVerbose("libperfstat initialized.");
4187   }
4188 }
4189 
4190 /////////////////////////////////////////////////////////////////////////////
4191 // thread stack
4192 
4193 // Get the current stack base from the OS (actually, the pthread library).
4194 // Note: usually not page aligned.
4195 address os::current_stack_base() {
4196   AixMisc::stackbounds_t bounds;
4197   bool rc = AixMisc::query_stack_bounds_for_current_thread(&bounds);
4198   guarantee(rc, "Unable to retrieve stack bounds.");
4199   return bounds.base;
4200 }
4201 
4202 // Get the current stack size from the OS (actually, the pthread library).
4203 // Returned size is such that (base - size) is always aligned to page size.
4204 size_t os::current_stack_size() {
4205   AixMisc::stackbounds_t bounds;
4206   bool rc = AixMisc::query_stack_bounds_for_current_thread(&bounds);
4207   guarantee(rc, "Unable to retrieve stack bounds.");
4208   // Align the returned stack size such that the stack low address
4209   // is aligned to page size (Note: base is usually not and we do not care).
4210   // We need to do this because caller code will assume stack low address is
4211   // page aligned and will place guard pages without checking.
4212   address low = bounds.base - bounds.size;
4213   address low_aligned = (address)align_up(low, os::vm_page_size());
4214   size_t s = bounds.base - low_aligned;
4215   return s;
4216 }
4217 
4218 extern char** environ;
4219 
4220 // Run the specified command in a separate process. Return its exit value,
4221 // or -1 on failure (e.g. can't fork a new process).
4222 // Unlike system(), this function can be called from signal handler. It
4223 // doesn't block SIGINT et al.
4224 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
4225   char * argv[4] = {"sh", "-c", cmd, NULL};
4226 
4227   pid_t pid = fork();
4228 
4229   if (pid < 0) {
4230     // fork failed
4231     return -1;
4232 
4233   } else if (pid == 0) {
4234     // child process
4235 
4236     // Try to be consistent with system(), which uses "/usr/bin/sh" on AIX.
4237     execve("/usr/bin/sh", argv, environ);
4238 
4239     // execve failed
4240     _exit(-1);
4241 
4242   } else {
4243     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
4244     // care about the actual exit code, for now.
4245 
4246     int status;
4247 
4248     // Wait for the child process to exit. This returns immediately if
4249     // the child has already exited. */
4250     while (waitpid(pid, &status, 0) < 0) {
4251       switch (errno) {
4252         case ECHILD: return 0;
4253         case EINTR: break;
4254         default: return -1;
4255       }
4256     }
4257 
4258     if (WIFEXITED(status)) {
4259       // The child exited normally; get its exit code.
4260       return WEXITSTATUS(status);
4261     } else if (WIFSIGNALED(status)) {
4262       // The child exited because of a signal.
4263       // The best value to return is 0x80 + signal number,
4264       // because that is what all Unix shells do, and because
4265       // it allows callers to distinguish between process exit and
4266       // process death by signal.
4267       return 0x80 + WTERMSIG(status);
4268     } else {
4269       // Unknown exit code; pass it through.
4270       return status;
4271     }
4272   }
4273   return -1;
4274 }
4275 
4276 // Get the default path to the core file
4277 // Returns the length of the string
4278 int os::get_core_path(char* buffer, size_t bufferSize) {
4279   const char* p = get_current_directory(buffer, bufferSize);
4280 
4281   if (p == NULL) {
4282     assert(p != NULL, "failed to get current directory");
4283     return 0;
4284   }
4285 
4286   jio_snprintf(buffer, bufferSize, "%s/core or core.%d",
4287                                                p, current_process_id());
4288 
4289   return strlen(buffer);
4290 }
4291 
4292 #ifndef PRODUCT
4293 void TestReserveMemorySpecial_test() {
4294   // No tests available for this platform
4295 }
4296 #endif
4297 
4298 bool os::start_debugging(char *buf, int buflen) {
4299   int len = (int)strlen(buf);
4300   char *p = &buf[len];
4301 
4302   jio_snprintf(p, buflen -len,
4303                  "\n\n"
4304                  "Do you want to debug the problem?\n\n"
4305                  "To debug, run 'dbx -a %d'; then switch to thread tid " INTX_FORMAT ", k-tid " INTX_FORMAT "\n"
4306                  "Enter 'yes' to launch dbx automatically (PATH must include dbx)\n"
4307                  "Otherwise, press RETURN to abort...",
4308                  os::current_process_id(),
4309                  os::current_thread_id(), thread_self());
4310 
4311   bool yes = os::message_box("Unexpected Error", buf);
4312 
4313   if (yes) {
4314     // yes, user asked VM to launch debugger
4315     jio_snprintf(buf, buflen, "dbx -a %d", os::current_process_id());
4316 
4317     os::fork_and_exec(buf);
4318     yes = false;
4319   }
4320   return yes;
4321 }
4322 
4323 static inline time_t get_mtime(const char* filename) {
4324   struct stat st;
4325   int ret = os::stat(filename, &st);
4326   assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
4327   return st.st_mtime;
4328 }
4329 
4330 int os::compare_file_modified_times(const char* file1, const char* file2) {
4331   time_t t1 = get_mtime(file1);
4332   time_t t2 = get_mtime(file2);
4333   return t1 - t2;
4334 }