1 /* 2 * Copyright (c) 1997, 2017, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2012, 2017, 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 "prims/jvm.h" 32 #include "runtime/java.hpp" 33 #include "runtime/os.hpp" 34 #include "runtime/stubCodeGenerator.hpp" 35 #include "utilities/align.hpp" 36 #include "utilities/defaultStream.hpp" 37 #include "utilities/globalDefinitions.hpp" 38 #include "vm_version_ppc.hpp" 39 40 # include <sys/sysinfo.h> 41 42 bool VM_Version::_is_determine_features_test_running = false; 43 uint64_t VM_Version::_dscr_val = 0; 44 45 #define MSG(flag) \ 46 if (flag && !FLAG_IS_DEFAULT(flag)) \ 47 jio_fprintf(defaultStream::error_stream(), \ 48 "warning: -XX:+" #flag " requires -XX:+UseSIGTRAP\n" \ 49 " -XX:+" #flag " will be disabled!\n"); 50 51 void VM_Version::initialize() { 52 53 // Test which instructions are supported and measure cache line size. 54 determine_features(); 55 56 // If PowerArchitecturePPC64 hasn't been specified explicitly determine from features. 57 if (FLAG_IS_DEFAULT(PowerArchitecturePPC64)) { 58 if (VM_Version::has_lqarx()) { 59 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 8); 60 } else if (VM_Version::has_popcntw()) { 61 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 7); 62 } else if (VM_Version::has_cmpb()) { 63 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 6); 64 } else if (VM_Version::has_popcntb()) { 65 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 5); 66 } else { 67 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 0); 68 } 69 } 70 71 bool PowerArchitecturePPC64_ok = false; 72 switch (PowerArchitecturePPC64) { 73 case 8: if (!VM_Version::has_lqarx() ) break; 74 case 7: if (!VM_Version::has_popcntw()) break; 75 case 6: if (!VM_Version::has_cmpb() ) break; 76 case 5: if (!VM_Version::has_popcntb()) break; 77 case 0: PowerArchitecturePPC64_ok = true; break; 78 default: break; 79 } 80 guarantee(PowerArchitecturePPC64_ok, "PowerArchitecturePPC64 cannot be set to " 81 UINTX_FORMAT " on this machine", PowerArchitecturePPC64); 82 83 // Power 8: Configure Data Stream Control Register. 84 if (PowerArchitecturePPC64 >= 8 && has_mfdscr()) { 85 config_dscr(); 86 } 87 88 if (!UseSIGTRAP) { 89 MSG(TrapBasedICMissChecks); 90 MSG(TrapBasedNotEntrantChecks); 91 MSG(TrapBasedNullChecks); 92 FLAG_SET_ERGO(bool, TrapBasedNotEntrantChecks, false); 93 FLAG_SET_ERGO(bool, TrapBasedNullChecks, false); 94 FLAG_SET_ERGO(bool, TrapBasedICMissChecks, false); 95 } 96 97 #ifdef COMPILER2 98 if (!UseSIGTRAP) { 99 MSG(TrapBasedRangeChecks); 100 FLAG_SET_ERGO(bool, TrapBasedRangeChecks, false); 101 } 102 103 // On Power6 test for section size. 104 if (PowerArchitecturePPC64 == 6) { 105 determine_section_size(); 106 // TODO: PPC port } else { 107 // TODO: PPC port PdScheduling::power6SectorSize = 0x20; 108 } 109 110 if (PowerArchitecturePPC64 >= 8) { 111 if (FLAG_IS_DEFAULT(SuperwordUseVSX)) { 112 FLAG_SET_ERGO(bool, SuperwordUseVSX, true); 113 } 114 } else { 115 if (SuperwordUseVSX) { 116 warning("SuperwordUseVSX specified, but needs at least Power8."); 117 FLAG_SET_DEFAULT(SuperwordUseVSX, false); 118 } 119 } 120 MaxVectorSize = SuperwordUseVSX ? 16 : 8; 121 #endif 122 123 // Create and print feature-string. 124 char buf[(num_features+1) * 16]; // Max 16 chars per feature. 125 jio_snprintf(buf, sizeof(buf), 126 "ppc64%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s", 127 (has_fsqrt() ? " fsqrt" : ""), 128 (has_isel() ? " isel" : ""), 129 (has_lxarxeh() ? " lxarxeh" : ""), 130 (has_cmpb() ? " cmpb" : ""), 131 //(has_mftgpr()? " mftgpr" : ""), 132 (has_popcntb() ? " popcntb" : ""), 133 (has_popcntw() ? " popcntw" : ""), 134 (has_fcfids() ? " fcfids" : ""), 135 (has_vand() ? " vand" : ""), 136 (has_lqarx() ? " lqarx" : ""), 137 (has_vcipher() ? " aes" : ""), 138 (has_vpmsumb() ? " vpmsumb" : ""), 139 (has_tcheck() ? " tcheck" : ""), 140 (has_mfdscr() ? " mfdscr" : ""), 141 (has_vsx() ? " vsx" : ""), 142 (has_ldbrx() ? " ldbrx" : ""), 143 (has_stdbrx() ? " stdbrx" : ""), 144 (has_vshasig() ? " sha" : "") 145 // Make sure number of %s matches num_features! 146 ); 147 _features_string = os::strdup(buf); 148 if (Verbose) { 149 print_features(); 150 } 151 152 // PPC64 supports 8-byte compare-exchange operations (see 153 // Atomic::cmpxchg and StubGenerator::generate_atomic_cmpxchg_ptr) 154 // and 'atomic long memory ops' (see Unsafe_GetLongVolatile). 155 _supports_cx8 = true; 156 157 // Used by C1. 158 _supports_atomic_getset4 = true; 159 _supports_atomic_getadd4 = true; 160 _supports_atomic_getset8 = true; 161 _supports_atomic_getadd8 = true; 162 163 UseSSE = 0; // Only on x86 and x64 164 165 intx cache_line_size = L1_data_cache_line_size(); 166 167 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) AllocatePrefetchStyle = 1; 168 169 if (AllocatePrefetchStyle == 4) { 170 AllocatePrefetchStepSize = cache_line_size; // Need exact value. 171 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 12; // Use larger blocks by default. 172 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 2*cache_line_size; // Default is not defined? 173 } else { 174 if (cache_line_size > AllocatePrefetchStepSize) AllocatePrefetchStepSize = cache_line_size; 175 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 3; // Optimistic value. 176 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 3*cache_line_size; // Default is not defined? 177 } 178 179 assert(AllocatePrefetchLines > 0, "invalid value"); 180 if (AllocatePrefetchLines < 1) { // Set valid value in product VM. 181 AllocatePrefetchLines = 1; // Conservative value. 182 } 183 184 if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size) { 185 AllocatePrefetchStyle = 1; // Fall back if inappropriate. 186 } 187 188 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive"); 189 190 // If defined(VM_LITTLE_ENDIAN) and running on Power8 or newer hardware, 191 // the implementation uses the vector instructions available with Power8. 192 // In all other cases, the implementation uses only generally available instructions. 193 if (!UseCRC32Intrinsics) { 194 if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) { 195 FLAG_SET_DEFAULT(UseCRC32Intrinsics, true); 196 } 197 } 198 199 // Implementation does not use any of the vector instructions available with Power8. 200 // Their exploitation is still pending (aka "work in progress"). 201 if (!UseCRC32CIntrinsics) { 202 if (FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) { 203 FLAG_SET_DEFAULT(UseCRC32CIntrinsics, true); 204 } 205 } 206 207 // TODO: Provide implementation. 208 if (UseAdler32Intrinsics) { 209 warning("Adler32Intrinsics not available on this CPU."); 210 FLAG_SET_DEFAULT(UseAdler32Intrinsics, false); 211 } 212 213 // The AES intrinsic stubs require AES instruction support. 214 #if defined(VM_LITTLE_ENDIAN) 215 if (has_vcipher()) { 216 if (FLAG_IS_DEFAULT(UseAES)) { 217 UseAES = true; 218 } 219 } else if (UseAES) { 220 if (!FLAG_IS_DEFAULT(UseAES)) 221 warning("AES instructions are not available on this CPU"); 222 FLAG_SET_DEFAULT(UseAES, false); 223 } 224 225 if (UseAES && has_vcipher()) { 226 if (FLAG_IS_DEFAULT(UseAESIntrinsics)) { 227 UseAESIntrinsics = true; 228 } 229 } else if (UseAESIntrinsics) { 230 if (!FLAG_IS_DEFAULT(UseAESIntrinsics)) 231 warning("AES intrinsics are not available on this CPU"); 232 FLAG_SET_DEFAULT(UseAESIntrinsics, false); 233 } 234 235 #else 236 if (UseAES) { 237 warning("AES instructions are not available on this CPU"); 238 FLAG_SET_DEFAULT(UseAES, false); 239 } 240 if (UseAESIntrinsics) { 241 if (!FLAG_IS_DEFAULT(UseAESIntrinsics)) 242 warning("AES intrinsics are not available on this CPU"); 243 FLAG_SET_DEFAULT(UseAESIntrinsics, false); 244 } 245 #endif 246 247 if (UseAESCTRIntrinsics) { 248 warning("AES/CTR intrinsics are not available on this CPU"); 249 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false); 250 } 251 252 if (UseGHASHIntrinsics) { 253 warning("GHASH intrinsics are not available on this CPU"); 254 FLAG_SET_DEFAULT(UseGHASHIntrinsics, false); 255 } 256 257 if (FLAG_IS_DEFAULT(UseFMA)) { 258 FLAG_SET_DEFAULT(UseFMA, true); 259 } 260 261 if (has_vshasig()) { 262 if (FLAG_IS_DEFAULT(UseSHA)) { 263 UseSHA = true; 264 } 265 } else if (UseSHA) { 266 if (!FLAG_IS_DEFAULT(UseSHA)) 267 warning("SHA instructions are not available on this CPU"); 268 FLAG_SET_DEFAULT(UseSHA, false); 269 } 270 271 if (UseSHA1Intrinsics) { 272 warning("Intrinsics for SHA-1 crypto hash functions not available on this CPU."); 273 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false); 274 } 275 276 if (UseSHA && has_vshasig()) { 277 if (FLAG_IS_DEFAULT(UseSHA256Intrinsics)) { 278 FLAG_SET_DEFAULT(UseSHA256Intrinsics, true); 279 } 280 } else if (UseSHA256Intrinsics) { 281 warning("Intrinsics for SHA-224 and SHA-256 crypto hash functions not available on this CPU."); 282 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false); 283 } 284 285 if (UseSHA && has_vshasig()) { 286 if (FLAG_IS_DEFAULT(UseSHA512Intrinsics)) { 287 FLAG_SET_DEFAULT(UseSHA512Intrinsics, true); 288 } 289 } else if (UseSHA512Intrinsics) { 290 warning("Intrinsics for SHA-384 and SHA-512 crypto hash functions not available on this CPU."); 291 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false); 292 } 293 294 if (!(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics)) { 295 FLAG_SET_DEFAULT(UseSHA, false); 296 } 297 298 if (FLAG_IS_DEFAULT(UseSquareToLenIntrinsic)) { 299 UseSquareToLenIntrinsic = true; 300 } 301 if (FLAG_IS_DEFAULT(UseMulAddIntrinsic)) { 302 UseMulAddIntrinsic = true; 303 } 304 if (FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) { 305 UseMultiplyToLenIntrinsic = true; 306 } 307 if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) { 308 UseMontgomeryMultiplyIntrinsic = true; 309 } 310 if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) { 311 UseMontgomerySquareIntrinsic = true; 312 } 313 314 if (UseVectorizedMismatchIntrinsic) { 315 warning("UseVectorizedMismatchIntrinsic specified, but not available on this CPU."); 316 FLAG_SET_DEFAULT(UseVectorizedMismatchIntrinsic, false); 317 } 318 319 320 // Adjust RTM (Restricted Transactional Memory) flags. 321 if (UseRTMLocking) { 322 // If CPU or OS are too old: 323 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 324 // setting during arguments processing. See use_biased_locking(). 325 // VM_Version_init() is executed after UseBiasedLocking is used 326 // in Thread::allocate(). 327 if (!has_tcheck()) { 328 vm_exit_during_initialization("RTM instructions are not available on this CPU"); 329 } 330 bool os_too_old = true; 331 #ifdef AIX 332 // Actually, this is supported since AIX 7.1.. Unfortunately, this first 333 // contained bugs, so that it can only be enabled after AIX 7.1.3.30. 334 // The Java property os.version, which is used in RTM tests to decide 335 // whether the feature is available, only knows major and minor versions. 336 // We don't want to change this property, as user code might depend on it. 337 // So the tests can not check on subversion 3.30, and we only enable RTM 338 // with AIX 7.2. 339 if (os::Aix::os_version() >= 0x07020000) { // At least AIX 7.2. 340 os_too_old = false; 341 } 342 #endif 343 #ifdef LINUX 344 // At least Linux kernel 4.2, as the problematic behavior of syscalls 345 // being called in the middle of a transaction has been addressed. 346 // Please, refer to commit b4b56f9ecab40f3b4ef53e130c9f6663be491894 347 // in Linux kernel source tree: https://goo.gl/Kc5i7A 348 if (os::Linux::os_version_is_known()) { 349 if (os::Linux::os_version() >= 0x040200) 350 os_too_old = false; 351 } else { 352 vm_exit_during_initialization("RTM can not be enabled: kernel version is unknown."); 353 } 354 #endif 355 if (os_too_old) { 356 vm_exit_during_initialization("RTM is not supported on this OS version."); 357 } 358 } 359 360 if (UseRTMLocking) { 361 #if INCLUDE_RTM_OPT 362 if (!UnlockExperimentalVMOptions) { 363 vm_exit_during_initialization("UseRTMLocking is only available as experimental option on this platform. " 364 "It must be enabled via -XX:+UnlockExperimentalVMOptions flag."); 365 } else { 366 warning("UseRTMLocking is only available as experimental option on this platform."); 367 } 368 if (!FLAG_IS_CMDLINE(UseRTMLocking)) { 369 // RTM locking should be used only for applications with 370 // high lock contention. For now we do not use it by default. 371 vm_exit_during_initialization("UseRTMLocking flag should be only set on command line"); 372 } 373 #else 374 // Only C2 does RTM locking optimization. 375 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 376 // setting during arguments processing. See use_biased_locking(). 377 vm_exit_during_initialization("RTM locking optimization is not supported in this VM"); 378 #endif 379 } else { // !UseRTMLocking 380 if (UseRTMForStackLocks) { 381 if (!FLAG_IS_DEFAULT(UseRTMForStackLocks)) { 382 warning("UseRTMForStackLocks flag should be off when UseRTMLocking flag is off"); 383 } 384 FLAG_SET_DEFAULT(UseRTMForStackLocks, false); 385 } 386 if (UseRTMDeopt) { 387 FLAG_SET_DEFAULT(UseRTMDeopt, false); 388 } 389 if (PrintPreciseRTMLockingStatistics) { 390 FLAG_SET_DEFAULT(PrintPreciseRTMLockingStatistics, false); 391 } 392 } 393 394 // This machine allows unaligned memory accesses 395 if (FLAG_IS_DEFAULT(UseUnalignedAccesses)) { 396 FLAG_SET_DEFAULT(UseUnalignedAccesses, true); 397 } 398 } 399 400 bool VM_Version::use_biased_locking() { 401 #if INCLUDE_RTM_OPT 402 // RTM locking is most useful when there is high lock contention and 403 // low data contention. With high lock contention the lock is usually 404 // inflated and biased locking is not suitable for that case. 405 // RTM locking code requires that biased locking is off. 406 // Note: we can't switch off UseBiasedLocking in get_processor_features() 407 // because it is used by Thread::allocate() which is called before 408 // VM_Version::initialize(). 409 if (UseRTMLocking && UseBiasedLocking) { 410 if (FLAG_IS_DEFAULT(UseBiasedLocking)) { 411 FLAG_SET_DEFAULT(UseBiasedLocking, false); 412 } else { 413 warning("Biased locking is not supported with RTM locking; ignoring UseBiasedLocking flag." ); 414 UseBiasedLocking = false; 415 } 416 } 417 #endif 418 return UseBiasedLocking; 419 } 420 421 void VM_Version::print_features() { 422 tty->print_cr("Version: %s L1_data_cache_line_size=%d", features_string(), L1_data_cache_line_size()); 423 } 424 425 #ifdef COMPILER2 426 // Determine section size on power6: If section size is 8 instructions, 427 // there should be a difference between the two testloops of ~15 %. If 428 // no difference is detected the section is assumed to be 32 instructions. 429 void VM_Version::determine_section_size() { 430 431 int unroll = 80; 432 433 const int code_size = (2* unroll * 32 + 100)*BytesPerInstWord; 434 435 // Allocate space for the code. 436 ResourceMark rm; 437 CodeBuffer cb("detect_section_size", code_size, 0); 438 MacroAssembler* a = new MacroAssembler(&cb); 439 440 uint32_t *code = (uint32_t *)a->pc(); 441 // Emit code. 442 void (*test1)() = (void(*)())(void *)a->function_entry(); 443 444 Label l1; 445 446 a->li(R4, 1); 447 a->sldi(R4, R4, 28); 448 a->b(l1); 449 a->align(CodeEntryAlignment); 450 451 a->bind(l1); 452 453 for (int i = 0; i < unroll; i++) { 454 // Schleife 1 455 // ------- sector 0 ------------ 456 // ;; 0 457 a->nop(); // 1 458 a->fpnop0(); // 2 459 a->fpnop1(); // 3 460 a->addi(R4,R4, -1); // 4 461 462 // ;; 1 463 a->nop(); // 5 464 a->fmr(F6, F6); // 6 465 a->fmr(F7, F7); // 7 466 a->endgroup(); // 8 467 // ------- sector 8 ------------ 468 469 // ;; 2 470 a->nop(); // 9 471 a->nop(); // 10 472 a->fmr(F8, F8); // 11 473 a->fmr(F9, F9); // 12 474 475 // ;; 3 476 a->nop(); // 13 477 a->fmr(F10, F10); // 14 478 a->fmr(F11, F11); // 15 479 a->endgroup(); // 16 480 // -------- sector 16 ------------- 481 482 // ;; 4 483 a->nop(); // 17 484 a->nop(); // 18 485 a->fmr(F15, F15); // 19 486 a->fmr(F16, F16); // 20 487 488 // ;; 5 489 a->nop(); // 21 490 a->fmr(F17, F17); // 22 491 a->fmr(F18, F18); // 23 492 a->endgroup(); // 24 493 // ------- sector 24 ------------ 494 495 // ;; 6 496 a->nop(); // 25 497 a->nop(); // 26 498 a->fmr(F19, F19); // 27 499 a->fmr(F20, F20); // 28 500 501 // ;; 7 502 a->nop(); // 29 503 a->fmr(F21, F21); // 30 504 a->fmr(F22, F22); // 31 505 a->brnop0(); // 32 506 507 // ------- sector 32 ------------ 508 } 509 510 // ;; 8 511 a->cmpdi(CCR0, R4, unroll); // 33 512 a->bge(CCR0, l1); // 34 513 a->blr(); 514 515 // Emit code. 516 void (*test2)() = (void(*)())(void *)a->function_entry(); 517 // uint32_t *code = (uint32_t *)a->pc(); 518 519 Label l2; 520 521 a->li(R4, 1); 522 a->sldi(R4, R4, 28); 523 a->b(l2); 524 a->align(CodeEntryAlignment); 525 526 a->bind(l2); 527 528 for (int i = 0; i < unroll; i++) { 529 // Schleife 2 530 // ------- sector 0 ------------ 531 // ;; 0 532 a->brnop0(); // 1 533 a->nop(); // 2 534 //a->cmpdi(CCR0, R4, unroll); 535 a->fpnop0(); // 3 536 a->fpnop1(); // 4 537 a->addi(R4,R4, -1); // 5 538 539 // ;; 1 540 541 a->nop(); // 6 542 a->fmr(F6, F6); // 7 543 a->fmr(F7, F7); // 8 544 // ------- sector 8 --------------- 545 546 // ;; 2 547 a->endgroup(); // 9 548 549 // ;; 3 550 a->nop(); // 10 551 a->nop(); // 11 552 a->fmr(F8, F8); // 12 553 554 // ;; 4 555 a->fmr(F9, F9); // 13 556 a->nop(); // 14 557 a->fmr(F10, F10); // 15 558 559 // ;; 5 560 a->fmr(F11, F11); // 16 561 // -------- sector 16 ------------- 562 563 // ;; 6 564 a->endgroup(); // 17 565 566 // ;; 7 567 a->nop(); // 18 568 a->nop(); // 19 569 a->fmr(F15, F15); // 20 570 571 // ;; 8 572 a->fmr(F16, F16); // 21 573 a->nop(); // 22 574 a->fmr(F17, F17); // 23 575 576 // ;; 9 577 a->fmr(F18, F18); // 24 578 // -------- sector 24 ------------- 579 580 // ;; 10 581 a->endgroup(); // 25 582 583 // ;; 11 584 a->nop(); // 26 585 a->nop(); // 27 586 a->fmr(F19, F19); // 28 587 588 // ;; 12 589 a->fmr(F20, F20); // 29 590 a->nop(); // 30 591 a->fmr(F21, F21); // 31 592 593 // ;; 13 594 a->fmr(F22, F22); // 32 595 } 596 597 // -------- sector 32 ------------- 598 // ;; 14 599 a->cmpdi(CCR0, R4, unroll); // 33 600 a->bge(CCR0, l2); // 34 601 602 a->blr(); 603 uint32_t *code_end = (uint32_t *)a->pc(); 604 a->flush(); 605 606 double loop1_seconds,loop2_seconds, rel_diff; 607 uint64_t start1, stop1; 608 609 start1 = os::current_thread_cpu_time(false); 610 (*test1)(); 611 stop1 = os::current_thread_cpu_time(false); 612 loop1_seconds = (stop1- start1) / (1000 *1000 *1000.0); 613 614 615 start1 = os::current_thread_cpu_time(false); 616 (*test2)(); 617 stop1 = os::current_thread_cpu_time(false); 618 619 loop2_seconds = (stop1 - start1) / (1000 *1000 *1000.0); 620 621 rel_diff = (loop2_seconds - loop1_seconds) / loop1_seconds *100; 622 623 if (PrintAssembly) { 624 ttyLocker ttyl; 625 tty->print_cr("Decoding section size detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 626 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 627 tty->print_cr("Time loop1 :%f", loop1_seconds); 628 tty->print_cr("Time loop2 :%f", loop2_seconds); 629 tty->print_cr("(time2 - time1) / time1 = %f %%", rel_diff); 630 631 if (rel_diff > 12.0) { 632 tty->print_cr("Section Size 8 Instructions"); 633 } else{ 634 tty->print_cr("Section Size 32 Instructions or Power5"); 635 } 636 } 637 638 #if 0 // TODO: PPC port 639 // Set sector size (if not set explicitly). 640 if (FLAG_IS_DEFAULT(Power6SectorSize128PPC64)) { 641 if (rel_diff > 12.0) { 642 PdScheduling::power6SectorSize = 0x20; 643 } else { 644 PdScheduling::power6SectorSize = 0x80; 645 } 646 } else if (Power6SectorSize128PPC64) { 647 PdScheduling::power6SectorSize = 0x80; 648 } else { 649 PdScheduling::power6SectorSize = 0x20; 650 } 651 #endif 652 if (UsePower6SchedulerPPC64) Unimplemented(); 653 } 654 #endif // COMPILER2 655 656 void VM_Version::determine_features() { 657 #if defined(ABI_ELFv2) 658 // 1 InstWord per call for the blr instruction. 659 const int code_size = (num_features+1+2*1)*BytesPerInstWord; 660 #else 661 // 7 InstWords for each call (function descriptor + blr instruction). 662 const int code_size = (num_features+1+2*7)*BytesPerInstWord; 663 #endif 664 int features = 0; 665 666 // create test area 667 enum { BUFFER_SIZE = 2*4*K }; // Needs to be >=2* max cache line size (cache line size can't exceed min page size). 668 char test_area[BUFFER_SIZE]; 669 char *mid_of_test_area = &test_area[BUFFER_SIZE>>1]; 670 671 // Allocate space for the code. 672 ResourceMark rm; 673 CodeBuffer cb("detect_cpu_features", code_size, 0); 674 MacroAssembler* a = new MacroAssembler(&cb); 675 676 // Must be set to true so we can generate the test code. 677 _features = VM_Version::all_features_m; 678 679 // Emit code. 680 void (*test)(address addr, uint64_t offset)=(void(*)(address addr, uint64_t offset))(void *)a->function_entry(); 681 uint32_t *code = (uint32_t *)a->pc(); 682 // Don't use R0 in ldarx. 683 // Keep R3_ARG1 unmodified, it contains &field (see below). 684 // Keep R4_ARG2 unmodified, it contains offset = 0 (see below). 685 a->fsqrt(F3, F4); // code[0] -> fsqrt_m 686 a->fsqrts(F3, F4); // code[1] -> fsqrts_m 687 a->isel(R7, R5, R6, 0); // code[2] -> isel_m 688 a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[3] -> lxarx_m 689 a->cmpb(R7, R5, R6); // code[4] -> cmpb 690 a->popcntb(R7, R5); // code[5] -> popcntb 691 a->popcntw(R7, R5); // code[6] -> popcntw 692 a->fcfids(F3, F4); // code[7] -> fcfids 693 a->vand(VR0, VR0, VR0); // code[8] -> vand 694 // arg0 of lqarx must be an even register, (arg1 + arg2) must be a multiple of 16 695 a->lqarx_unchecked(R6, R3_ARG1, R4_ARG2, 1); // code[9] -> lqarx_m 696 a->vcipher(VR0, VR1, VR2); // code[10] -> vcipher 697 a->vpmsumb(VR0, VR1, VR2); // code[11] -> vpmsumb 698 a->tcheck(0); // code[12] -> tcheck 699 a->mfdscr(R0); // code[13] -> mfdscr 700 a->lxvd2x(VSR0, R3_ARG1); // code[14] -> vsx 701 a->ldbrx(R7, R3_ARG1, R4_ARG2); // code[15] -> ldbrx 702 a->stdbrx(R7, R3_ARG1, R4_ARG2); // code[16] -> stdbrx 703 a->vshasigmaw(VR0, VR1, 1, 0xF); // code[17] -> vshasig 704 a->blr(); 705 706 // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it. 707 void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->function_entry(); 708 a->dcbz(R3_ARG1); // R3_ARG1 = addr 709 a->blr(); 710 711 uint32_t *code_end = (uint32_t *)a->pc(); 712 a->flush(); 713 _features = VM_Version::unknown_m; 714 715 // Print the detection code. 716 if (PrintAssembly) { 717 ttyLocker ttyl; 718 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 719 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 720 } 721 722 // Measure cache line size. 723 memset(test_area, 0xFF, BUFFER_SIZE); // Fill test area with 0xFF. 724 (*zero_cacheline_func_ptr)(mid_of_test_area); // Call function which executes dcbz to the middle. 725 int count = 0; // count zeroed bytes 726 for (int i = 0; i < BUFFER_SIZE; i++) if (test_area[i] == 0) count++; 727 guarantee(is_power_of_2(count), "cache line size needs to be a power of 2"); 728 _L1_data_cache_line_size = count; 729 730 // Execute code. Illegal instructions will be replaced by 0 in the signal handler. 731 VM_Version::_is_determine_features_test_running = true; 732 // We must align the first argument to 16 bytes because of the lqarx check. 733 (*test)(align_up((address)mid_of_test_area, 16), 0); 734 VM_Version::_is_determine_features_test_running = false; 735 736 // determine which instructions are legal. 737 int feature_cntr = 0; 738 if (code[feature_cntr++]) features |= fsqrt_m; 739 if (code[feature_cntr++]) features |= fsqrts_m; 740 if (code[feature_cntr++]) features |= isel_m; 741 if (code[feature_cntr++]) features |= lxarxeh_m; 742 if (code[feature_cntr++]) features |= cmpb_m; 743 if (code[feature_cntr++]) features |= popcntb_m; 744 if (code[feature_cntr++]) features |= popcntw_m; 745 if (code[feature_cntr++]) features |= fcfids_m; 746 if (code[feature_cntr++]) features |= vand_m; 747 if (code[feature_cntr++]) features |= lqarx_m; 748 if (code[feature_cntr++]) features |= vcipher_m; 749 if (code[feature_cntr++]) features |= vpmsumb_m; 750 if (code[feature_cntr++]) features |= tcheck_m; 751 if (code[feature_cntr++]) features |= mfdscr_m; 752 if (code[feature_cntr++]) features |= vsx_m; 753 if (code[feature_cntr++]) features |= ldbrx_m; 754 if (code[feature_cntr++]) features |= stdbrx_m; 755 if (code[feature_cntr++]) features |= vshasig_m; 756 757 // Print the detection code. 758 if (PrintAssembly) { 759 ttyLocker ttyl; 760 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " after execution:", p2i(code)); 761 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 762 } 763 764 _features = features; 765 } 766 767 // Power 8: Configure Data Stream Control Register. 768 void VM_Version::config_dscr() { 769 // 7 InstWords for each call (function descriptor + blr instruction). 770 const int code_size = (2+2*7)*BytesPerInstWord; 771 772 // Allocate space for the code. 773 ResourceMark rm; 774 CodeBuffer cb("config_dscr", code_size, 0); 775 MacroAssembler* a = new MacroAssembler(&cb); 776 777 // Emit code. 778 uint64_t (*get_dscr)() = (uint64_t(*)())(void *)a->function_entry(); 779 uint32_t *code = (uint32_t *)a->pc(); 780 a->mfdscr(R3); 781 a->blr(); 782 783 void (*set_dscr)(long) = (void(*)(long))(void *)a->function_entry(); 784 a->mtdscr(R3); 785 a->blr(); 786 787 uint32_t *code_end = (uint32_t *)a->pc(); 788 a->flush(); 789 790 // Print the detection code. 791 if (PrintAssembly) { 792 ttyLocker ttyl; 793 tty->print_cr("Decoding dscr configuration stub at " INTPTR_FORMAT " before execution:", p2i(code)); 794 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 795 } 796 797 // Apply the configuration if needed. 798 _dscr_val = (*get_dscr)(); 799 if (Verbose) { 800 tty->print_cr("dscr value was 0x%lx" , _dscr_val); 801 } 802 bool change_requested = false; 803 if (DSCR_PPC64 != (uintx)-1) { 804 _dscr_val = DSCR_PPC64; 805 change_requested = true; 806 } 807 if (DSCR_DPFD_PPC64 <= 7) { 808 uint64_t mask = 0x7; 809 if ((_dscr_val & mask) != DSCR_DPFD_PPC64) { 810 _dscr_val = (_dscr_val & ~mask) | (DSCR_DPFD_PPC64); 811 change_requested = true; 812 } 813 } 814 if (DSCR_URG_PPC64 <= 7) { 815 uint64_t mask = 0x7 << 6; 816 if ((_dscr_val & mask) != DSCR_DPFD_PPC64 << 6) { 817 _dscr_val = (_dscr_val & ~mask) | (DSCR_URG_PPC64 << 6); 818 change_requested = true; 819 } 820 } 821 if (change_requested) { 822 (*set_dscr)(_dscr_val); 823 if (Verbose) { 824 tty->print_cr("dscr was set to 0x%lx" , (*get_dscr)()); 825 } 826 } 827 } 828 829 static uint64_t saved_features = 0; 830 831 void VM_Version::allow_all() { 832 saved_features = _features; 833 _features = all_features_m; 834 } 835 836 void VM_Version::revert() { 837 _features = saved_features; 838 }