1 /*
   2  * Copyright (c) 1997, 2016, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2012, 2016 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 #include "precompiled.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "asm/macroAssembler.inline.hpp"
  29 #include "compiler/disassembler.hpp"
  30 #include "memory/resourceArea.hpp"
  31 #include "runtime/java.hpp"
  32 #include "runtime/os.hpp"
  33 #include "runtime/stubCodeGenerator.hpp"
  34 #include "utilities/defaultStream.hpp"
  35 #include "utilities/globalDefinitions.hpp"
  36 #include "vm_version_ppc.hpp"
  37 
  38 # include <sys/sysinfo.h>
  39 
  40 bool VM_Version::_is_determine_features_test_running = false;
  41 
  42 
  43 #define MSG(flag)   \
  44   if (flag && !FLAG_IS_DEFAULT(flag))                                  \
  45       jio_fprintf(defaultStream::error_stream(),                       \
  46                   "warning: -XX:+" #flag " requires -XX:+UseSIGTRAP\n" \
  47                   "         -XX:+" #flag " will be disabled!\n");
  48 
  49 void VM_Version::initialize() {
  50 
  51   // Test which instructions are supported and measure cache line size.
  52   determine_features();
  53 
  54   // If PowerArchitecturePPC64 hasn't been specified explicitly determine from features.
  55   if (FLAG_IS_DEFAULT(PowerArchitecturePPC64)) {
  56     if (VM_Version::has_lqarx()) {
  57       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 8);
  58     } else if (VM_Version::has_popcntw()) {
  59       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 7);
  60     } else if (VM_Version::has_cmpb()) {
  61       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 6);
  62     } else if (VM_Version::has_popcntb()) {
  63       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 5);
  64     } else {
  65       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 0);
  66     }
  67   }
  68 
  69   bool PowerArchitecturePPC64_ok = false;
  70   switch (PowerArchitecturePPC64) {
  71     case 8: if (!VM_Version::has_lqarx()  ) break;
  72     case 7: if (!VM_Version::has_popcntw()) break;
  73     case 6: if (!VM_Version::has_cmpb()   ) break;
  74     case 5: if (!VM_Version::has_popcntb()) break;
  75     case 0: PowerArchitecturePPC64_ok = true; break;
  76     default: break;
  77   }
  78   guarantee(PowerArchitecturePPC64_ok, "PowerArchitecturePPC64 cannot be set to "
  79             UINTX_FORMAT " on this machine", PowerArchitecturePPC64);
  80 
  81   // Power 8: Configure Data Stream Control Register.
  82   if (has_mfdscr()) {
  83     config_dscr();
  84   }
  85 
  86   if (!UseSIGTRAP) {
  87     MSG(TrapBasedICMissChecks);
  88     MSG(TrapBasedNotEntrantChecks);
  89     MSG(TrapBasedNullChecks);
  90     FLAG_SET_ERGO(bool, TrapBasedNotEntrantChecks, false);
  91     FLAG_SET_ERGO(bool, TrapBasedNullChecks,       false);
  92     FLAG_SET_ERGO(bool, TrapBasedICMissChecks,     false);
  93   }
  94 
  95 #ifdef COMPILER2
  96   if (!UseSIGTRAP) {
  97     MSG(TrapBasedRangeChecks);
  98     FLAG_SET_ERGO(bool, TrapBasedRangeChecks, false);
  99   }
 100 
 101   // On Power6 test for section size.
 102   if (PowerArchitecturePPC64 == 6) {
 103     determine_section_size();
 104   // TODO: PPC port } else {
 105   // TODO: PPC port PdScheduling::power6SectorSize = 0x20;
 106   }
 107 
 108   MaxVectorSize = 8;
 109 #endif
 110 
 111   // Create and print feature-string.
 112   char buf[(num_features+1) * 16]; // Max 16 chars per feature.
 113   jio_snprintf(buf, sizeof(buf),
 114                "ppc64%s%s%s%s%s%s%s%s%s%s%s%s%s",
 115                (has_fsqrt()   ? " fsqrt"   : ""),
 116                (has_isel()    ? " isel"    : ""),
 117                (has_lxarxeh() ? " lxarxeh" : ""),
 118                (has_cmpb()    ? " cmpb"    : ""),
 119                //(has_mftgpr()? " mftgpr"  : ""),
 120                (has_popcntb() ? " popcntb" : ""),
 121                (has_popcntw() ? " popcntw" : ""),
 122                (has_fcfids()  ? " fcfids"  : ""),
 123                (has_vand()    ? " vand"    : ""),
 124                (has_lqarx()   ? " lqarx"   : ""),
 125                (has_vcipher() ? " vcipher" : ""),
 126                (has_vpmsumb() ? " vpmsumb" : ""),
 127                (has_tcheck()  ? " tcheck"  : ""),
 128                (has_mfdscr()  ? " mfdscr"  : "")
 129                // Make sure number of %s matches num_features!
 130               );
 131   _features_string = os::strdup(buf);
 132   if (Verbose) {
 133     print_features();
 134   }
 135 
 136   // PPC64 supports 8-byte compare-exchange operations (see
 137   // Atomic::cmpxchg and StubGenerator::generate_atomic_cmpxchg_ptr)
 138   // and 'atomic long memory ops' (see Unsafe_GetLongVolatile).
 139   _supports_cx8 = true;
 140 
 141   // Used by C1.
 142   _supports_atomic_getset4 = true;
 143   _supports_atomic_getadd4 = true;
 144   _supports_atomic_getset8 = true;
 145   _supports_atomic_getadd8 = true;
 146 
 147   UseSSE = 0; // Only on x86 and x64
 148 
 149   intx cache_line_size = L1_data_cache_line_size();
 150 
 151   if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) AllocatePrefetchStyle = 1;
 152 
 153   if (AllocatePrefetchStyle == 4) {
 154     AllocatePrefetchStepSize = cache_line_size; // Need exact value.
 155     if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 12; // Use larger blocks by default.
 156     if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 2*cache_line_size; // Default is not defined?
 157   } else {
 158     if (cache_line_size > AllocatePrefetchStepSize) AllocatePrefetchStepSize = cache_line_size;
 159     if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 3; // Optimistic value.
 160     if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 3*cache_line_size; // Default is not defined?
 161   }
 162 
 163   assert(AllocatePrefetchLines > 0, "invalid value");
 164   if (AllocatePrefetchLines < 1) { // Set valid value in product VM.
 165     AllocatePrefetchLines = 1; // Conservative value.
 166   }
 167 
 168   if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size) {
 169     AllocatePrefetchStyle = 1; // Fall back if inappropriate.
 170   }
 171 
 172   assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive");
 173 
 174   // Implementation does not use any of the vector instructions
 175   // available with Power8. Their exploitation is still pending.
 176   if (!UseCRC32Intrinsics) {
 177     if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) {
 178       FLAG_SET_DEFAULT(UseCRC32Intrinsics, true);
 179     }
 180   }
 181 
 182   if (UseCRC32CIntrinsics) {
 183     if (!FLAG_IS_DEFAULT(UseCRC32CIntrinsics))
 184       warning("CRC32C intrinsics are not available on this CPU");
 185     FLAG_SET_DEFAULT(UseCRC32CIntrinsics, false);
 186   }
 187 
 188   // The AES intrinsic stubs require AES instruction support.
 189   if (UseAES) {
 190     warning("AES instructions are not available on this CPU");
 191     FLAG_SET_DEFAULT(UseAES, false);
 192   }
 193   if (UseAESIntrinsics) {
 194     if (!FLAG_IS_DEFAULT(UseAESIntrinsics))
 195       warning("AES intrinsics are not available on this CPU");
 196     FLAG_SET_DEFAULT(UseAESIntrinsics, false);
 197   }
 198 
 199   if (UseAESCTRIntrinsics) {
 200     warning("AES/CTR intrinsics are not available on this CPU");
 201     FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false);
 202   }
 203 
 204   if (UseGHASHIntrinsics) {
 205     warning("GHASH intrinsics are not available on this CPU");
 206     FLAG_SET_DEFAULT(UseGHASHIntrinsics, false);
 207   }
 208 
 209   if (UseSHA) {
 210     warning("SHA instructions are not available on this CPU");
 211     FLAG_SET_DEFAULT(UseSHA, false);
 212   }
 213   if (UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics) {
 214     warning("SHA intrinsics are not available on this CPU");
 215     FLAG_SET_DEFAULT(UseSHA1Intrinsics, false);
 216     FLAG_SET_DEFAULT(UseSHA256Intrinsics, false);
 217     FLAG_SET_DEFAULT(UseSHA512Intrinsics, false);
 218   }
 219 
 220   if (UseAdler32Intrinsics) {
 221     warning("Adler32Intrinsics not available on this CPU.");
 222     FLAG_SET_DEFAULT(UseAdler32Intrinsics, false);
 223   }
 224 
 225   if (FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) {
 226     UseMultiplyToLenIntrinsic = true;
 227   }
 228   if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) {
 229     UseMontgomeryMultiplyIntrinsic = true;
 230   }
 231   if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) {
 232     UseMontgomerySquareIntrinsic = true;
 233   }
 234 
 235   if (UseVectorizedMismatchIntrinsic) {
 236     warning("UseVectorizedMismatchIntrinsic specified, but not available on this CPU.");
 237     FLAG_SET_DEFAULT(UseVectorizedMismatchIntrinsic, false);
 238   }
 239 
 240 
 241   // Adjust RTM (Restricted Transactional Memory) flags.
 242   if (UseRTMLocking) {
 243     // If CPU or OS are too old:
 244     // Can't continue because UseRTMLocking affects UseBiasedLocking flag
 245     // setting during arguments processing. See use_biased_locking().
 246     // VM_Version_init() is executed after UseBiasedLocking is used
 247     // in Thread::allocate().
 248     if (!has_tcheck()) {
 249       vm_exit_during_initialization("RTM instructions are not available on this CPU");
 250     }
 251     bool os_too_old = true;
 252 #ifdef AIX
 253     if (os::Aix::os_version() >= 0x0701031e) { // at least AIX 7.1.3.30
 254       os_too_old = false;
 255     }
 256 #endif
 257 #ifdef linux
 258     // At least Linux kernel 4.2, as the problematic behavior of syscalls
 259     // being called in the middle of a transaction has been addressed.
 260     // Please, refer to commit b4b56f9ecab40f3b4ef53e130c9f6663be491894
 261     // in Linux kernel source tree: https://goo.gl/Kc5i7A
 262     if (os::Linux::os_version_is_known()) {
 263       if (os::Linux::os_version() >= 0x040200)
 264         os_too_old = false;
 265     } else {
 266       vm_exit_during_initialization("RTM can not be enabled: kernel version is unknown.");
 267     }
 268 #endif
 269     if (os_too_old) {
 270       vm_exit_during_initialization("RTM is not supported on this OS version.");
 271     }
 272   }
 273 
 274   if (UseRTMLocking) {
 275 #if INCLUDE_RTM_OPT
 276     if (!UnlockExperimentalVMOptions) {
 277       vm_exit_during_initialization("UseRTMLocking is only available as experimental option on this platform. "
 278                                     "It must be enabled via -XX:+UnlockExperimentalVMOptions flag.");
 279     } else {
 280       warning("UseRTMLocking is only available as experimental option on this platform.");
 281     }
 282     if (!FLAG_IS_CMDLINE(UseRTMLocking)) {
 283       // RTM locking should be used only for applications with
 284       // high lock contention. For now we do not use it by default.
 285       vm_exit_during_initialization("UseRTMLocking flag should be only set on command line");
 286     }
 287     if (!is_power_of_2(RTMTotalCountIncrRate)) {
 288       warning("RTMTotalCountIncrRate must be a power of 2, resetting it to 64");
 289       FLAG_SET_DEFAULT(RTMTotalCountIncrRate, 64);
 290     }
 291     if (RTMAbortRatio < 0 || RTMAbortRatio > 100) {
 292       warning("RTMAbortRatio must be in the range 0 to 100, resetting it to 50");
 293       FLAG_SET_DEFAULT(RTMAbortRatio, 50);
 294     }
 295     guarantee(RTMSpinLoopCount > 0, "unsupported");
 296 #else
 297     // Only C2 does RTM locking optimization.
 298     // Can't continue because UseRTMLocking affects UseBiasedLocking flag
 299     // setting during arguments processing. See use_biased_locking().
 300     vm_exit_during_initialization("RTM locking optimization is not supported in this VM");
 301 #endif
 302   } else { // !UseRTMLocking
 303     if (UseRTMForStackLocks) {
 304       if (!FLAG_IS_DEFAULT(UseRTMForStackLocks)) {
 305         warning("UseRTMForStackLocks flag should be off when UseRTMLocking flag is off");
 306       }
 307       FLAG_SET_DEFAULT(UseRTMForStackLocks, false);
 308     }
 309     if (UseRTMDeopt) {
 310       FLAG_SET_DEFAULT(UseRTMDeopt, false);
 311     }
 312     if (PrintPreciseRTMLockingStatistics) {
 313       FLAG_SET_DEFAULT(PrintPreciseRTMLockingStatistics, false);
 314     }
 315   }
 316 
 317   // This machine allows unaligned memory accesses
 318   if (FLAG_IS_DEFAULT(UseUnalignedAccesses)) {
 319     FLAG_SET_DEFAULT(UseUnalignedAccesses, true);
 320   }
 321 }
 322 
 323 bool VM_Version::use_biased_locking() {
 324 #if INCLUDE_RTM_OPT
 325   // RTM locking is most useful when there is high lock contention and
 326   // low data contention. With high lock contention the lock is usually
 327   // inflated and biased locking is not suitable for that case.
 328   // RTM locking code requires that biased locking is off.
 329   // Note: we can't switch off UseBiasedLocking in get_processor_features()
 330   // because it is used by Thread::allocate() which is called before
 331   // VM_Version::initialize().
 332   if (UseRTMLocking && UseBiasedLocking) {
 333     if (FLAG_IS_DEFAULT(UseBiasedLocking)) {
 334       FLAG_SET_DEFAULT(UseBiasedLocking, false);
 335     } else {
 336       warning("Biased locking is not supported with RTM locking; ignoring UseBiasedLocking flag." );
 337       UseBiasedLocking = false;
 338     }
 339   }
 340 #endif
 341   return UseBiasedLocking;
 342 }
 343 
 344 void VM_Version::print_features() {
 345   tty->print_cr("Version: %s L1_data_cache_line_size=%d", features_string(), L1_data_cache_line_size());
 346 }
 347 
 348 #ifdef COMPILER2
 349 // Determine section size on power6: If section size is 8 instructions,
 350 // there should be a difference between the two testloops of ~15 %. If
 351 // no difference is detected the section is assumed to be 32 instructions.
 352 void VM_Version::determine_section_size() {
 353 
 354   int unroll = 80;
 355 
 356   const int code_size = (2* unroll * 32 + 100)*BytesPerInstWord;
 357 
 358   // Allocate space for the code.
 359   ResourceMark rm;
 360   CodeBuffer cb("detect_section_size", code_size, 0);
 361   MacroAssembler* a = new MacroAssembler(&cb);
 362 
 363   uint32_t *code = (uint32_t *)a->pc();
 364   // Emit code.
 365   void (*test1)() = (void(*)())(void *)a->function_entry();
 366 
 367   Label l1;
 368 
 369   a->li(R4, 1);
 370   a->sldi(R4, R4, 28);
 371   a->b(l1);
 372   a->align(CodeEntryAlignment);
 373 
 374   a->bind(l1);
 375 
 376   for (int i = 0; i < unroll; i++) {
 377     // Schleife 1
 378     // ------- sector 0 ------------
 379     // ;; 0
 380     a->nop();                   // 1
 381     a->fpnop0();                // 2
 382     a->fpnop1();                // 3
 383     a->addi(R4,R4, -1); // 4
 384 
 385     // ;;  1
 386     a->nop();                   // 5
 387     a->fmr(F6, F6);             // 6
 388     a->fmr(F7, F7);             // 7
 389     a->endgroup();              // 8
 390     // ------- sector 8 ------------
 391 
 392     // ;;  2
 393     a->nop();                   // 9
 394     a->nop();                   // 10
 395     a->fmr(F8, F8);             // 11
 396     a->fmr(F9, F9);             // 12
 397 
 398     // ;;  3
 399     a->nop();                   // 13
 400     a->fmr(F10, F10);           // 14
 401     a->fmr(F11, F11);           // 15
 402     a->endgroup();              // 16
 403     // -------- sector 16 -------------
 404 
 405     // ;;  4
 406     a->nop();                   // 17
 407     a->nop();                   // 18
 408     a->fmr(F15, F15);           // 19
 409     a->fmr(F16, F16);           // 20
 410 
 411     // ;;  5
 412     a->nop();                   // 21
 413     a->fmr(F17, F17);           // 22
 414     a->fmr(F18, F18);           // 23
 415     a->endgroup();              // 24
 416     // ------- sector 24  ------------
 417 
 418     // ;;  6
 419     a->nop();                   // 25
 420     a->nop();                   // 26
 421     a->fmr(F19, F19);           // 27
 422     a->fmr(F20, F20);           // 28
 423 
 424     // ;;  7
 425     a->nop();                   // 29
 426     a->fmr(F21, F21);           // 30
 427     a->fmr(F22, F22);           // 31
 428     a->brnop0();                // 32
 429 
 430     // ------- sector 32 ------------
 431   }
 432 
 433   // ;; 8
 434   a->cmpdi(CCR0, R4, unroll);   // 33
 435   a->bge(CCR0, l1);             // 34
 436   a->blr();
 437 
 438   // Emit code.
 439   void (*test2)() = (void(*)())(void *)a->function_entry();
 440   // uint32_t *code = (uint32_t *)a->pc();
 441 
 442   Label l2;
 443 
 444   a->li(R4, 1);
 445   a->sldi(R4, R4, 28);
 446   a->b(l2);
 447   a->align(CodeEntryAlignment);
 448 
 449   a->bind(l2);
 450 
 451   for (int i = 0; i < unroll; i++) {
 452     // Schleife 2
 453     // ------- sector 0 ------------
 454     // ;; 0
 455     a->brnop0();                  // 1
 456     a->nop();                     // 2
 457     //a->cmpdi(CCR0, R4, unroll);
 458     a->fpnop0();                  // 3
 459     a->fpnop1();                  // 4
 460     a->addi(R4,R4, -1);           // 5
 461 
 462     // ;; 1
 463 
 464     a->nop();                     // 6
 465     a->fmr(F6, F6);               // 7
 466     a->fmr(F7, F7);               // 8
 467     // ------- sector 8 ---------------
 468 
 469     // ;; 2
 470     a->endgroup();                // 9
 471 
 472     // ;; 3
 473     a->nop();                     // 10
 474     a->nop();                     // 11
 475     a->fmr(F8, F8);               // 12
 476 
 477     // ;; 4
 478     a->fmr(F9, F9);               // 13
 479     a->nop();                     // 14
 480     a->fmr(F10, F10);             // 15
 481 
 482     // ;; 5
 483     a->fmr(F11, F11);             // 16
 484     // -------- sector 16 -------------
 485 
 486     // ;; 6
 487     a->endgroup();                // 17
 488 
 489     // ;; 7
 490     a->nop();                     // 18
 491     a->nop();                     // 19
 492     a->fmr(F15, F15);             // 20
 493 
 494     // ;; 8
 495     a->fmr(F16, F16);             // 21
 496     a->nop();                     // 22
 497     a->fmr(F17, F17);             // 23
 498 
 499     // ;; 9
 500     a->fmr(F18, F18);             // 24
 501     // -------- sector 24 -------------
 502 
 503     // ;; 10
 504     a->endgroup();                // 25
 505 
 506     // ;; 11
 507     a->nop();                     // 26
 508     a->nop();                     // 27
 509     a->fmr(F19, F19);             // 28
 510 
 511     // ;; 12
 512     a->fmr(F20, F20);             // 29
 513     a->nop();                     // 30
 514     a->fmr(F21, F21);             // 31
 515 
 516     // ;; 13
 517     a->fmr(F22, F22);             // 32
 518   }
 519 
 520   // -------- sector 32 -------------
 521   // ;; 14
 522   a->cmpdi(CCR0, R4, unroll); // 33
 523   a->bge(CCR0, l2);           // 34
 524 
 525   a->blr();
 526   uint32_t *code_end = (uint32_t *)a->pc();
 527   a->flush();
 528 
 529   double loop1_seconds,loop2_seconds, rel_diff;
 530   uint64_t start1, stop1;
 531 
 532   start1 = os::current_thread_cpu_time(false);
 533   (*test1)();
 534   stop1 = os::current_thread_cpu_time(false);
 535   loop1_seconds = (stop1- start1) / (1000 *1000 *1000.0);
 536 
 537 
 538   start1 = os::current_thread_cpu_time(false);
 539   (*test2)();
 540   stop1 = os::current_thread_cpu_time(false);
 541 
 542   loop2_seconds = (stop1 - start1) / (1000 *1000 *1000.0);
 543 
 544   rel_diff = (loop2_seconds - loop1_seconds) / loop1_seconds *100;
 545 
 546   if (PrintAssembly) {
 547     ttyLocker ttyl;
 548     tty->print_cr("Decoding section size detection stub at " INTPTR_FORMAT " before execution:", p2i(code));
 549     Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
 550     tty->print_cr("Time loop1 :%f", loop1_seconds);
 551     tty->print_cr("Time loop2 :%f", loop2_seconds);
 552     tty->print_cr("(time2 - time1) / time1 = %f %%", rel_diff);
 553 
 554     if (rel_diff > 12.0) {
 555       tty->print_cr("Section Size 8 Instructions");
 556     } else{
 557       tty->print_cr("Section Size 32 Instructions or Power5");
 558     }
 559   }
 560 
 561 #if 0 // TODO: PPC port
 562   // Set sector size (if not set explicitly).
 563   if (FLAG_IS_DEFAULT(Power6SectorSize128PPC64)) {
 564     if (rel_diff > 12.0) {
 565       PdScheduling::power6SectorSize = 0x20;
 566     } else {
 567       PdScheduling::power6SectorSize = 0x80;
 568     }
 569   } else if (Power6SectorSize128PPC64) {
 570     PdScheduling::power6SectorSize = 0x80;
 571   } else {
 572     PdScheduling::power6SectorSize = 0x20;
 573   }
 574 #endif
 575   if (UsePower6SchedulerPPC64) Unimplemented();
 576 }
 577 #endif // COMPILER2
 578 
 579 void VM_Version::determine_features() {
 580 #if defined(ABI_ELFv2)
 581   // 1 InstWord per call for the blr instruction.
 582   const int code_size = (num_features+1+2*1)*BytesPerInstWord;
 583 #else
 584   // 7 InstWords for each call (function descriptor + blr instruction).
 585   const int code_size = (num_features+1+2*7)*BytesPerInstWord;
 586 #endif
 587   int features = 0;
 588 
 589   // create test area
 590   enum { BUFFER_SIZE = 2*4*K }; // Needs to be >=2* max cache line size (cache line size can't exceed min page size).
 591   char test_area[BUFFER_SIZE];
 592   char *mid_of_test_area = &test_area[BUFFER_SIZE>>1];
 593 
 594   // Allocate space for the code.
 595   ResourceMark rm;
 596   CodeBuffer cb("detect_cpu_features", code_size, 0);
 597   MacroAssembler* a = new MacroAssembler(&cb);
 598 
 599   // Must be set to true so we can generate the test code.
 600   _features = VM_Version::all_features_m;
 601 
 602   // Emit code.
 603   void (*test)(address addr, uint64_t offset)=(void(*)(address addr, uint64_t offset))(void *)a->function_entry();
 604   uint32_t *code = (uint32_t *)a->pc();
 605   // Don't use R0 in ldarx.
 606   // Keep R3_ARG1 unmodified, it contains &field (see below).
 607   // Keep R4_ARG2 unmodified, it contains offset = 0 (see below).
 608   a->fsqrt(F3, F4);                            // code[0]  -> fsqrt_m
 609   a->fsqrts(F3, F4);                           // code[1]  -> fsqrts_m
 610   a->isel(R7, R5, R6, 0);                      // code[2]  -> isel_m
 611   a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[3]  -> lxarx_m
 612   a->cmpb(R7, R5, R6);                         // code[4]  -> cmpb
 613   a->popcntb(R7, R5);                          // code[5]  -> popcntb
 614   a->popcntw(R7, R5);                          // code[6]  -> popcntw
 615   a->fcfids(F3, F4);                           // code[7]  -> fcfids
 616   a->vand(VR0, VR0, VR0);                      // code[8]  -> vand
 617   // arg0 of lqarx must be an even register, (arg1 + arg2) must be a multiple of 16
 618   a->lqarx_unchecked(R6, R3_ARG1, R4_ARG2, 1); // code[9]  -> lqarx_m
 619   a->vcipher(VR0, VR1, VR2);                   // code[10] -> vcipher
 620   a->vpmsumb(VR0, VR1, VR2);                   // code[11] -> vpmsumb
 621   a->tcheck(0);                                // code[12] -> tcheck
 622   a->mfdscr(R0);                               // code[13] -> mfdscr
 623   a->blr();
 624 
 625   // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it.
 626   void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->function_entry();
 627   a->dcbz(R3_ARG1); // R3_ARG1 = addr
 628   a->blr();
 629 
 630   uint32_t *code_end = (uint32_t *)a->pc();
 631   a->flush();
 632   _features = VM_Version::unknown_m;
 633 
 634   // Print the detection code.
 635   if (PrintAssembly) {
 636     ttyLocker ttyl;
 637     tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " before execution:", p2i(code));
 638     Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
 639   }
 640 
 641   // Measure cache line size.
 642   memset(test_area, 0xFF, BUFFER_SIZE); // Fill test area with 0xFF.
 643   (*zero_cacheline_func_ptr)(mid_of_test_area); // Call function which executes dcbz to the middle.
 644   int count = 0; // count zeroed bytes
 645   for (int i = 0; i < BUFFER_SIZE; i++) if (test_area[i] == 0) count++;
 646   guarantee(is_power_of_2(count), "cache line size needs to be a power of 2");
 647   _L1_data_cache_line_size = count;
 648 
 649   // Execute code. Illegal instructions will be replaced by 0 in the signal handler.
 650   VM_Version::_is_determine_features_test_running = true;
 651   // We must align the first argument to 16 bytes because of the lqarx check.
 652   (*test)((address)align_size_up((intptr_t)mid_of_test_area, 16), (uint64_t)0);
 653   VM_Version::_is_determine_features_test_running = false;
 654 
 655   // determine which instructions are legal.
 656   int feature_cntr = 0;
 657   if (code[feature_cntr++]) features |= fsqrt_m;
 658   if (code[feature_cntr++]) features |= fsqrts_m;
 659   if (code[feature_cntr++]) features |= isel_m;
 660   if (code[feature_cntr++]) features |= lxarxeh_m;
 661   if (code[feature_cntr++]) features |= cmpb_m;
 662   if (code[feature_cntr++]) features |= popcntb_m;
 663   if (code[feature_cntr++]) features |= popcntw_m;
 664   if (code[feature_cntr++]) features |= fcfids_m;
 665   if (code[feature_cntr++]) features |= vand_m;
 666   if (code[feature_cntr++]) features |= lqarx_m;
 667   if (code[feature_cntr++]) features |= vcipher_m;
 668   if (code[feature_cntr++]) features |= vpmsumb_m;
 669   if (code[feature_cntr++]) features |= tcheck_m;
 670   if (code[feature_cntr++]) features |= mfdscr_m;
 671 
 672   // Print the detection code.
 673   if (PrintAssembly) {
 674     ttyLocker ttyl;
 675     tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " after execution:", p2i(code));
 676     Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
 677   }
 678 
 679   _features = features;
 680 }
 681 
 682 // Power 8: Configure Data Stream Control Register.
 683 void VM_Version::config_dscr() {
 684   // 7 InstWords for each call (function descriptor + blr instruction).
 685   const int code_size = (2+2*7)*BytesPerInstWord;
 686 
 687   // Allocate space for the code.
 688   ResourceMark rm;
 689   CodeBuffer cb("config_dscr", code_size, 0);
 690   MacroAssembler* a = new MacroAssembler(&cb);
 691 
 692   // Emit code.
 693   uint64_t (*get_dscr)() = (uint64_t(*)())(void *)a->function_entry();
 694   uint32_t *code = (uint32_t *)a->pc();
 695   a->mfdscr(R3);
 696   a->blr();
 697 
 698   void (*set_dscr)(long) = (void(*)(long))(void *)a->function_entry();
 699   a->mtdscr(R3);
 700   a->blr();
 701 
 702   uint32_t *code_end = (uint32_t *)a->pc();
 703   a->flush();
 704 
 705   // Print the detection code.
 706   if (PrintAssembly) {
 707     ttyLocker ttyl;
 708     tty->print_cr("Decoding dscr configuration stub at " INTPTR_FORMAT " before execution:", p2i(code));
 709     Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
 710   }
 711 
 712   // Apply the configuration if needed.
 713   uint64_t dscr_val = (*get_dscr)();
 714   if (Verbose) {
 715     tty->print_cr("dscr value was 0x%lx" , dscr_val);
 716   }
 717   bool change_requested = false;
 718   if (DSCR_PPC64 != (uintx)-1) {
 719     dscr_val = DSCR_PPC64;
 720     change_requested = true;
 721   }
 722   if (DSCR_DPFD_PPC64 <= 7) {
 723     uint64_t mask = 0x7;
 724     if ((dscr_val & mask) != DSCR_DPFD_PPC64) {
 725       dscr_val = (dscr_val & ~mask) | (DSCR_DPFD_PPC64);
 726       change_requested = true;
 727     }
 728   }
 729   if (DSCR_URG_PPC64 <= 7) {
 730     uint64_t mask = 0x7 << 6;
 731     if ((dscr_val & mask) != DSCR_DPFD_PPC64 << 6) {
 732       dscr_val = (dscr_val & ~mask) | (DSCR_URG_PPC64 << 6);
 733       change_requested = true;
 734     }
 735   }
 736   if (change_requested) {
 737     (*set_dscr)(dscr_val);
 738     if (Verbose) {
 739       tty->print_cr("dscr was set to 0x%lx" , (*get_dscr)());
 740     }
 741   }
 742 }
 743 
 744 static uint64_t saved_features = 0;
 745 
 746 void VM_Version::allow_all() {
 747   saved_features = _features;
 748   _features      = all_features_m;
 749 }
 750 
 751 void VM_Version::revert() {
 752   _features = saved_features;
 753 }