1 /* 2 * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2012, 2018, 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 "jvm.h" 28 #include "asm/assembler.inline.hpp" 29 #include "asm/macroAssembler.inline.hpp" 30 #include "compiler/disassembler.hpp" 31 #include "memory/resourceArea.hpp" 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 #if defined(LINUX) && defined(VM_LITTLE_ENDIAN) 43 #include <sys/auxv.h> 44 45 #ifndef PPC_FEATURE2_HTM_NOSC 46 #define PPC_FEATURE2_HTM_NOSC (1 << 24) 47 #endif 48 #endif 49 50 bool VM_Version::_is_determine_features_test_running = false; 51 uint64_t VM_Version::_dscr_val = 0; 52 53 #define MSG(flag) \ 54 if (flag && !FLAG_IS_DEFAULT(flag)) \ 55 jio_fprintf(defaultStream::error_stream(), \ 56 "warning: -XX:+" #flag " requires -XX:+UseSIGTRAP\n" \ 57 " -XX:+" #flag " will be disabled!\n"); 58 59 void VM_Version::initialize() { 60 61 // Test which instructions are supported and measure cache line size. 62 determine_features(); 63 64 // If PowerArchitecturePPC64 hasn't been specified explicitly determine from features. 65 if (FLAG_IS_DEFAULT(PowerArchitecturePPC64)) { 66 if (VM_Version::has_lqarx()) { 67 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 8); 68 } else if (VM_Version::has_popcntw()) { 69 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 7); 70 } else if (VM_Version::has_cmpb()) { 71 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 6); 72 } else if (VM_Version::has_popcntb()) { 73 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 5); 74 } else { 75 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 0); 76 } 77 } 78 79 bool PowerArchitecturePPC64_ok = false; 80 switch (PowerArchitecturePPC64) { 81 case 8: if (!VM_Version::has_lqarx() ) break; 82 case 7: if (!VM_Version::has_popcntw()) break; 83 case 6: if (!VM_Version::has_cmpb() ) break; 84 case 5: if (!VM_Version::has_popcntb()) break; 85 case 0: PowerArchitecturePPC64_ok = true; break; 86 default: break; 87 } 88 guarantee(PowerArchitecturePPC64_ok, "PowerArchitecturePPC64 cannot be set to " 89 UINTX_FORMAT " on this machine", PowerArchitecturePPC64); 90 91 // Power 8: Configure Data Stream Control Register. 92 if (PowerArchitecturePPC64 >= 8 && has_mfdscr()) { 93 config_dscr(); 94 } 95 96 if (!UseSIGTRAP) { 97 MSG(TrapBasedICMissChecks); 98 MSG(TrapBasedNotEntrantChecks); 99 MSG(TrapBasedNullChecks); 100 FLAG_SET_ERGO(bool, TrapBasedNotEntrantChecks, false); 101 FLAG_SET_ERGO(bool, TrapBasedNullChecks, false); 102 FLAG_SET_ERGO(bool, TrapBasedICMissChecks, false); 103 } 104 105 #ifdef COMPILER2 106 if (!UseSIGTRAP) { 107 MSG(TrapBasedRangeChecks); 108 FLAG_SET_ERGO(bool, TrapBasedRangeChecks, false); 109 } 110 111 // On Power6 test for section size. 112 if (PowerArchitecturePPC64 == 6) { 113 determine_section_size(); 114 // TODO: PPC port } else { 115 // TODO: PPC port PdScheduling::power6SectorSize = 0x20; 116 } 117 118 if (PowerArchitecturePPC64 >= 8) { 119 if (FLAG_IS_DEFAULT(SuperwordUseVSX)) { 120 FLAG_SET_ERGO(bool, SuperwordUseVSX, true); 121 } 122 } else { 123 if (SuperwordUseVSX) { 124 warning("SuperwordUseVSX specified, but needs at least Power8."); 125 FLAG_SET_DEFAULT(SuperwordUseVSX, false); 126 } 127 } 128 MaxVectorSize = SuperwordUseVSX ? 16 : 8; 129 #endif 130 131 // Create and print feature-string. 132 char buf[(num_features+1) * 16]; // Max 16 chars per feature. 133 jio_snprintf(buf, sizeof(buf), 134 "ppc64%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s", 135 (has_fsqrt() ? " fsqrt" : ""), 136 (has_isel() ? " isel" : ""), 137 (has_lxarxeh() ? " lxarxeh" : ""), 138 (has_cmpb() ? " cmpb" : ""), 139 //(has_mftgpr()? " mftgpr" : ""), 140 (has_popcntb() ? " popcntb" : ""), 141 (has_popcntw() ? " popcntw" : ""), 142 (has_fcfids() ? " fcfids" : ""), 143 (has_vand() ? " vand" : ""), 144 (has_lqarx() ? " lqarx" : ""), 145 (has_vcipher() ? " aes" : ""), 146 (has_vpmsumb() ? " vpmsumb" : ""), 147 (has_mfdscr() ? " mfdscr" : ""), 148 (has_vsx() ? " vsx" : ""), 149 (has_ldbrx() ? " ldbrx" : ""), 150 (has_stdbrx() ? " stdbrx" : ""), 151 (has_vshasig() ? " sha" : ""), 152 (has_tm() ? " rtm" : "") 153 // Make sure number of %s matches num_features! 154 ); 155 _features_string = os::strdup(buf); 156 if (Verbose) { 157 print_features(); 158 } 159 160 // PPC64 supports 8-byte compare-exchange operations (see Atomic::cmpxchg) 161 // and 'atomic long memory ops' (see Unsafe_GetLongVolatile). 162 _supports_cx8 = true; 163 164 // Used by C1. 165 _supports_atomic_getset4 = true; 166 _supports_atomic_getadd4 = true; 167 _supports_atomic_getset8 = true; 168 _supports_atomic_getadd8 = true; 169 170 UseSSE = 0; // Only on x86 and x64 171 172 intx cache_line_size = L1_data_cache_line_size(); 173 174 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) AllocatePrefetchStyle = 1; 175 176 if (AllocatePrefetchStyle == 4) { 177 AllocatePrefetchStepSize = cache_line_size; // Need exact value. 178 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 12; // Use larger blocks by default. 179 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 2*cache_line_size; // Default is not defined? 180 } else { 181 if (cache_line_size > AllocatePrefetchStepSize) AllocatePrefetchStepSize = cache_line_size; 182 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 3; // Optimistic value. 183 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 3*cache_line_size; // Default is not defined? 184 } 185 186 assert(AllocatePrefetchLines > 0, "invalid value"); 187 if (AllocatePrefetchLines < 1) { // Set valid value in product VM. 188 AllocatePrefetchLines = 1; // Conservative value. 189 } 190 191 if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size) { 192 AllocatePrefetchStyle = 1; // Fall back if inappropriate. 193 } 194 195 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive"); 196 197 // If running on Power8 or newer hardware, the implementation uses the available vector instructions. 198 // In all other cases, the implementation uses only generally available instructions. 199 if (!UseCRC32Intrinsics) { 200 if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) { 201 FLAG_SET_DEFAULT(UseCRC32Intrinsics, true); 202 } 203 } 204 205 // Implementation does not use any of the vector instructions available with Power8. 206 // Their exploitation is still pending (aka "work in progress"). 207 if (!UseCRC32CIntrinsics) { 208 if (FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) { 209 FLAG_SET_DEFAULT(UseCRC32CIntrinsics, true); 210 } 211 } 212 213 // TODO: Provide implementation. 214 if (UseAdler32Intrinsics) { 215 warning("Adler32Intrinsics not available on this CPU."); 216 FLAG_SET_DEFAULT(UseAdler32Intrinsics, false); 217 } 218 219 // The AES intrinsic stubs require AES instruction support. 220 if (has_vcipher()) { 221 if (FLAG_IS_DEFAULT(UseAES)) { 222 UseAES = true; 223 } 224 } else if (UseAES) { 225 if (!FLAG_IS_DEFAULT(UseAES)) 226 warning("AES instructions are not available on this CPU"); 227 FLAG_SET_DEFAULT(UseAES, false); 228 } 229 230 if (UseAES && has_vcipher()) { 231 if (FLAG_IS_DEFAULT(UseAESIntrinsics)) { 232 UseAESIntrinsics = true; 233 } 234 } else if (UseAESIntrinsics) { 235 if (!FLAG_IS_DEFAULT(UseAESIntrinsics)) 236 warning("AES intrinsics are not available on this CPU"); 237 FLAG_SET_DEFAULT(UseAESIntrinsics, false); 238 } 239 240 if (UseAESCTRIntrinsics) { 241 warning("AES/CTR intrinsics are not available on this CPU"); 242 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false); 243 } 244 245 if (UseGHASHIntrinsics) { 246 warning("GHASH intrinsics are not available on this CPU"); 247 FLAG_SET_DEFAULT(UseGHASHIntrinsics, false); 248 } 249 250 if (FLAG_IS_DEFAULT(UseFMA)) { 251 FLAG_SET_DEFAULT(UseFMA, true); 252 } 253 254 if (has_vshasig()) { 255 if (FLAG_IS_DEFAULT(UseSHA)) { 256 UseSHA = true; 257 } 258 } else if (UseSHA) { 259 if (!FLAG_IS_DEFAULT(UseSHA)) 260 warning("SHA instructions are not available on this CPU"); 261 FLAG_SET_DEFAULT(UseSHA, false); 262 } 263 264 if (UseSHA1Intrinsics) { 265 warning("Intrinsics for SHA-1 crypto hash functions not available on this CPU."); 266 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false); 267 } 268 269 if (UseSHA && has_vshasig()) { 270 if (FLAG_IS_DEFAULT(UseSHA256Intrinsics)) { 271 FLAG_SET_DEFAULT(UseSHA256Intrinsics, true); 272 } 273 } else if (UseSHA256Intrinsics) { 274 warning("Intrinsics for SHA-224 and SHA-256 crypto hash functions not available on this CPU."); 275 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false); 276 } 277 278 if (UseSHA && has_vshasig()) { 279 if (FLAG_IS_DEFAULT(UseSHA512Intrinsics)) { 280 FLAG_SET_DEFAULT(UseSHA512Intrinsics, true); 281 } 282 } else if (UseSHA512Intrinsics) { 283 warning("Intrinsics for SHA-384 and SHA-512 crypto hash functions not available on this CPU."); 284 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false); 285 } 286 287 if (!(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics)) { 288 FLAG_SET_DEFAULT(UseSHA, false); 289 } 290 291 if (FLAG_IS_DEFAULT(UseSquareToLenIntrinsic)) { 292 UseSquareToLenIntrinsic = true; 293 } 294 if (FLAG_IS_DEFAULT(UseMulAddIntrinsic)) { 295 UseMulAddIntrinsic = true; 296 } 297 if (FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) { 298 UseMultiplyToLenIntrinsic = true; 299 } 300 if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) { 301 UseMontgomeryMultiplyIntrinsic = true; 302 } 303 if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) { 304 UseMontgomerySquareIntrinsic = true; 305 } 306 307 if (UseVectorizedMismatchIntrinsic) { 308 warning("UseVectorizedMismatchIntrinsic specified, but not available on this CPU."); 309 FLAG_SET_DEFAULT(UseVectorizedMismatchIntrinsic, false); 310 } 311 312 313 // Adjust RTM (Restricted Transactional Memory) flags. 314 if (UseRTMLocking) { 315 // If CPU or OS do not support TM: 316 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 317 // setting during arguments processing. See use_biased_locking(). 318 // VM_Version_init() is executed after UseBiasedLocking is used 319 // in Thread::allocate(). 320 if (PowerArchitecturePPC64 < 8) { 321 vm_exit_during_initialization("RTM instructions are not available on this CPU."); 322 } 323 324 if (!has_tm()) { 325 vm_exit_during_initialization("RTM is not supported on this OS version."); 326 } 327 } 328 329 if (UseRTMLocking) { 330 #if INCLUDE_RTM_OPT 331 if (!UnlockExperimentalVMOptions) { 332 vm_exit_during_initialization("UseRTMLocking is only available as experimental option on this platform. " 333 "It must be enabled via -XX:+UnlockExperimentalVMOptions flag."); 334 } else { 335 warning("UseRTMLocking is only available as experimental option on this platform."); 336 } 337 if (!FLAG_IS_CMDLINE(UseRTMLocking)) { 338 // RTM locking should be used only for applications with 339 // high lock contention. For now we do not use it by default. 340 vm_exit_during_initialization("UseRTMLocking flag should be only set on command line"); 341 } 342 #else 343 // Only C2 does RTM locking optimization. 344 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 345 // setting during arguments processing. See use_biased_locking(). 346 vm_exit_during_initialization("RTM locking optimization is not supported in this VM"); 347 #endif 348 } else { // !UseRTMLocking 349 if (UseRTMForStackLocks) { 350 if (!FLAG_IS_DEFAULT(UseRTMForStackLocks)) { 351 warning("UseRTMForStackLocks flag should be off when UseRTMLocking flag is off"); 352 } 353 FLAG_SET_DEFAULT(UseRTMForStackLocks, false); 354 } 355 if (UseRTMDeopt) { 356 FLAG_SET_DEFAULT(UseRTMDeopt, false); 357 } 358 if (PrintPreciseRTMLockingStatistics) { 359 FLAG_SET_DEFAULT(PrintPreciseRTMLockingStatistics, false); 360 } 361 } 362 363 // This machine allows unaligned memory accesses 364 if (FLAG_IS_DEFAULT(UseUnalignedAccesses)) { 365 FLAG_SET_DEFAULT(UseUnalignedAccesses, true); 366 } 367 } 368 369 bool VM_Version::use_biased_locking() { 370 #if INCLUDE_RTM_OPT 371 // RTM locking is most useful when there is high lock contention and 372 // low data contention. With high lock contention the lock is usually 373 // inflated and biased locking is not suitable for that case. 374 // RTM locking code requires that biased locking is off. 375 // Note: we can't switch off UseBiasedLocking in get_processor_features() 376 // because it is used by Thread::allocate() which is called before 377 // VM_Version::initialize(). 378 if (UseRTMLocking && UseBiasedLocking) { 379 if (FLAG_IS_DEFAULT(UseBiasedLocking)) { 380 FLAG_SET_DEFAULT(UseBiasedLocking, false); 381 } else { 382 warning("Biased locking is not supported with RTM locking; ignoring UseBiasedLocking flag." ); 383 UseBiasedLocking = false; 384 } 385 } 386 #endif 387 return UseBiasedLocking; 388 } 389 390 void VM_Version::print_features() { 391 tty->print_cr("Version: %s L1_data_cache_line_size=%d", features_string(), L1_data_cache_line_size()); 392 } 393 394 #ifdef COMPILER2 395 // Determine section size on power6: If section size is 8 instructions, 396 // there should be a difference between the two testloops of ~15 %. If 397 // no difference is detected the section is assumed to be 32 instructions. 398 void VM_Version::determine_section_size() { 399 400 int unroll = 80; 401 402 const int code_size = (2* unroll * 32 + 100)*BytesPerInstWord; 403 404 // Allocate space for the code. 405 ResourceMark rm; 406 CodeBuffer cb("detect_section_size", code_size, 0); 407 MacroAssembler* a = new MacroAssembler(&cb); 408 409 uint32_t *code = (uint32_t *)a->pc(); 410 // Emit code. 411 void (*test1)() = (void(*)())(void *)a->function_entry(); 412 413 Label l1; 414 415 a->li(R4, 1); 416 a->sldi(R4, R4, 28); 417 a->b(l1); 418 a->align(CodeEntryAlignment); 419 420 a->bind(l1); 421 422 for (int i = 0; i < unroll; i++) { 423 // Schleife 1 424 // ------- sector 0 ------------ 425 // ;; 0 426 a->nop(); // 1 427 a->fpnop0(); // 2 428 a->fpnop1(); // 3 429 a->addi(R4,R4, -1); // 4 430 431 // ;; 1 432 a->nop(); // 5 433 a->fmr(F6, F6); // 6 434 a->fmr(F7, F7); // 7 435 a->endgroup(); // 8 436 // ------- sector 8 ------------ 437 438 // ;; 2 439 a->nop(); // 9 440 a->nop(); // 10 441 a->fmr(F8, F8); // 11 442 a->fmr(F9, F9); // 12 443 444 // ;; 3 445 a->nop(); // 13 446 a->fmr(F10, F10); // 14 447 a->fmr(F11, F11); // 15 448 a->endgroup(); // 16 449 // -------- sector 16 ------------- 450 451 // ;; 4 452 a->nop(); // 17 453 a->nop(); // 18 454 a->fmr(F15, F15); // 19 455 a->fmr(F16, F16); // 20 456 457 // ;; 5 458 a->nop(); // 21 459 a->fmr(F17, F17); // 22 460 a->fmr(F18, F18); // 23 461 a->endgroup(); // 24 462 // ------- sector 24 ------------ 463 464 // ;; 6 465 a->nop(); // 25 466 a->nop(); // 26 467 a->fmr(F19, F19); // 27 468 a->fmr(F20, F20); // 28 469 470 // ;; 7 471 a->nop(); // 29 472 a->fmr(F21, F21); // 30 473 a->fmr(F22, F22); // 31 474 a->brnop0(); // 32 475 476 // ------- sector 32 ------------ 477 } 478 479 // ;; 8 480 a->cmpdi(CCR0, R4, unroll); // 33 481 a->bge(CCR0, l1); // 34 482 a->blr(); 483 484 // Emit code. 485 void (*test2)() = (void(*)())(void *)a->function_entry(); 486 // uint32_t *code = (uint32_t *)a->pc(); 487 488 Label l2; 489 490 a->li(R4, 1); 491 a->sldi(R4, R4, 28); 492 a->b(l2); 493 a->align(CodeEntryAlignment); 494 495 a->bind(l2); 496 497 for (int i = 0; i < unroll; i++) { 498 // Schleife 2 499 // ------- sector 0 ------------ 500 // ;; 0 501 a->brnop0(); // 1 502 a->nop(); // 2 503 //a->cmpdi(CCR0, R4, unroll); 504 a->fpnop0(); // 3 505 a->fpnop1(); // 4 506 a->addi(R4,R4, -1); // 5 507 508 // ;; 1 509 510 a->nop(); // 6 511 a->fmr(F6, F6); // 7 512 a->fmr(F7, F7); // 8 513 // ------- sector 8 --------------- 514 515 // ;; 2 516 a->endgroup(); // 9 517 518 // ;; 3 519 a->nop(); // 10 520 a->nop(); // 11 521 a->fmr(F8, F8); // 12 522 523 // ;; 4 524 a->fmr(F9, F9); // 13 525 a->nop(); // 14 526 a->fmr(F10, F10); // 15 527 528 // ;; 5 529 a->fmr(F11, F11); // 16 530 // -------- sector 16 ------------- 531 532 // ;; 6 533 a->endgroup(); // 17 534 535 // ;; 7 536 a->nop(); // 18 537 a->nop(); // 19 538 a->fmr(F15, F15); // 20 539 540 // ;; 8 541 a->fmr(F16, F16); // 21 542 a->nop(); // 22 543 a->fmr(F17, F17); // 23 544 545 // ;; 9 546 a->fmr(F18, F18); // 24 547 // -------- sector 24 ------------- 548 549 // ;; 10 550 a->endgroup(); // 25 551 552 // ;; 11 553 a->nop(); // 26 554 a->nop(); // 27 555 a->fmr(F19, F19); // 28 556 557 // ;; 12 558 a->fmr(F20, F20); // 29 559 a->nop(); // 30 560 a->fmr(F21, F21); // 31 561 562 // ;; 13 563 a->fmr(F22, F22); // 32 564 } 565 566 // -------- sector 32 ------------- 567 // ;; 14 568 a->cmpdi(CCR0, R4, unroll); // 33 569 a->bge(CCR0, l2); // 34 570 571 a->blr(); 572 uint32_t *code_end = (uint32_t *)a->pc(); 573 a->flush(); 574 575 double loop1_seconds,loop2_seconds, rel_diff; 576 uint64_t start1, stop1; 577 578 start1 = os::current_thread_cpu_time(false); 579 (*test1)(); 580 stop1 = os::current_thread_cpu_time(false); 581 loop1_seconds = (stop1- start1) / (1000 *1000 *1000.0); 582 583 584 start1 = os::current_thread_cpu_time(false); 585 (*test2)(); 586 stop1 = os::current_thread_cpu_time(false); 587 588 loop2_seconds = (stop1 - start1) / (1000 *1000 *1000.0); 589 590 rel_diff = (loop2_seconds - loop1_seconds) / loop1_seconds *100; 591 592 if (PrintAssembly) { 593 ttyLocker ttyl; 594 tty->print_cr("Decoding section size detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 595 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 596 tty->print_cr("Time loop1 :%f", loop1_seconds); 597 tty->print_cr("Time loop2 :%f", loop2_seconds); 598 tty->print_cr("(time2 - time1) / time1 = %f %%", rel_diff); 599 600 if (rel_diff > 12.0) { 601 tty->print_cr("Section Size 8 Instructions"); 602 } else{ 603 tty->print_cr("Section Size 32 Instructions or Power5"); 604 } 605 } 606 607 #if 0 // TODO: PPC port 608 // Set sector size (if not set explicitly). 609 if (FLAG_IS_DEFAULT(Power6SectorSize128PPC64)) { 610 if (rel_diff > 12.0) { 611 PdScheduling::power6SectorSize = 0x20; 612 } else { 613 PdScheduling::power6SectorSize = 0x80; 614 } 615 } else if (Power6SectorSize128PPC64) { 616 PdScheduling::power6SectorSize = 0x80; 617 } else { 618 PdScheduling::power6SectorSize = 0x20; 619 } 620 #endif 621 if (UsePower6SchedulerPPC64) Unimplemented(); 622 } 623 #endif // COMPILER2 624 625 void VM_Version::determine_features() { 626 #if defined(ABI_ELFv2) 627 // 1 InstWord per call for the blr instruction. 628 const int code_size = (num_features+1+2*1)*BytesPerInstWord; 629 #else 630 // 7 InstWords for each call (function descriptor + blr instruction). 631 const int code_size = (num_features+1+2*7)*BytesPerInstWord; 632 #endif 633 int features = 0; 634 635 // create test area 636 enum { BUFFER_SIZE = 2*4*K }; // Needs to be >=2* max cache line size (cache line size can't exceed min page size). 637 char test_area[BUFFER_SIZE]; 638 char *mid_of_test_area = &test_area[BUFFER_SIZE>>1]; 639 640 // Allocate space for the code. 641 ResourceMark rm; 642 CodeBuffer cb("detect_cpu_features", code_size, 0); 643 MacroAssembler* a = new MacroAssembler(&cb); 644 645 // Must be set to true so we can generate the test code. 646 _features = VM_Version::all_features_m; 647 648 // Emit code. 649 void (*test)(address addr, uint64_t offset)=(void(*)(address addr, uint64_t offset))(void *)a->function_entry(); 650 uint32_t *code = (uint32_t *)a->pc(); 651 // Don't use R0 in ldarx. 652 // Keep R3_ARG1 unmodified, it contains &field (see below). 653 // Keep R4_ARG2 unmodified, it contains offset = 0 (see below). 654 a->fsqrt(F3, F4); // code[0] -> fsqrt_m 655 a->fsqrts(F3, F4); // code[1] -> fsqrts_m 656 a->isel(R7, R5, R6, 0); // code[2] -> isel_m 657 a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[3] -> lxarx_m 658 a->cmpb(R7, R5, R6); // code[4] -> cmpb 659 a->popcntb(R7, R5); // code[5] -> popcntb 660 a->popcntw(R7, R5); // code[6] -> popcntw 661 a->fcfids(F3, F4); // code[7] -> fcfids 662 a->vand(VR0, VR0, VR0); // code[8] -> vand 663 // arg0 of lqarx must be an even register, (arg1 + arg2) must be a multiple of 16 664 a->lqarx_unchecked(R6, R3_ARG1, R4_ARG2, 1); // code[9] -> lqarx_m 665 a->vcipher(VR0, VR1, VR2); // code[10] -> vcipher 666 a->vpmsumb(VR0, VR1, VR2); // code[11] -> vpmsumb 667 a->mfdscr(R0); // code[12] -> mfdscr 668 a->lxvd2x(VSR0, R3_ARG1); // code[13] -> vsx 669 a->ldbrx(R7, R3_ARG1, R4_ARG2); // code[14] -> ldbrx 670 a->stdbrx(R7, R3_ARG1, R4_ARG2); // code[15] -> stdbrx 671 a->vshasigmaw(VR0, VR1, 1, 0xF); // code[16] -> vshasig 672 a->blr(); 673 674 // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it. 675 void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->function_entry(); 676 a->dcbz(R3_ARG1); // R3_ARG1 = addr 677 a->blr(); 678 679 uint32_t *code_end = (uint32_t *)a->pc(); 680 a->flush(); 681 _features = VM_Version::unknown_m; 682 683 // Print the detection code. 684 if (PrintAssembly) { 685 ttyLocker ttyl; 686 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 687 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 688 } 689 690 // Measure cache line size. 691 memset(test_area, 0xFF, BUFFER_SIZE); // Fill test area with 0xFF. 692 (*zero_cacheline_func_ptr)(mid_of_test_area); // Call function which executes dcbz to the middle. 693 int count = 0; // count zeroed bytes 694 for (int i = 0; i < BUFFER_SIZE; i++) if (test_area[i] == 0) count++; 695 guarantee(is_power_of_2(count), "cache line size needs to be a power of 2"); 696 _L1_data_cache_line_size = count; 697 698 // Execute code. Illegal instructions will be replaced by 0 in the signal handler. 699 VM_Version::_is_determine_features_test_running = true; 700 // We must align the first argument to 16 bytes because of the lqarx check. 701 (*test)(align_up((address)mid_of_test_area, 16), 0); 702 VM_Version::_is_determine_features_test_running = false; 703 704 // determine which instructions are legal. 705 int feature_cntr = 0; 706 if (code[feature_cntr++]) features |= fsqrt_m; 707 if (code[feature_cntr++]) features |= fsqrts_m; 708 if (code[feature_cntr++]) features |= isel_m; 709 if (code[feature_cntr++]) features |= lxarxeh_m; 710 if (code[feature_cntr++]) features |= cmpb_m; 711 if (code[feature_cntr++]) features |= popcntb_m; 712 if (code[feature_cntr++]) features |= popcntw_m; 713 if (code[feature_cntr++]) features |= fcfids_m; 714 if (code[feature_cntr++]) features |= vand_m; 715 if (code[feature_cntr++]) features |= lqarx_m; 716 if (code[feature_cntr++]) features |= vcipher_m; 717 if (code[feature_cntr++]) features |= vpmsumb_m; 718 if (code[feature_cntr++]) features |= mfdscr_m; 719 if (code[feature_cntr++]) features |= vsx_m; 720 if (code[feature_cntr++]) features |= ldbrx_m; 721 if (code[feature_cntr++]) features |= stdbrx_m; 722 if (code[feature_cntr++]) features |= vshasig_m; 723 724 // Print the detection code. 725 if (PrintAssembly) { 726 ttyLocker ttyl; 727 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " after execution:", p2i(code)); 728 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 729 } 730 731 #ifdef AIX 732 // Actually, this is supported since AIX 7.1.. Unfortunately, this first 733 // contained bugs, so that it can only be enabled after AIX 7.1.3.30. 734 // The Java property os.version, which is used in RTM tests to decide 735 // whether the feature is available, only knows major and minor versions. 736 // We don't want to change this property, as user code might depend on it. 737 // So the tests can not check on subversion 3.30, and we only enable RTM 738 // with AIX 7.2. 739 if (os::Aix::os_version() >= 0x07020000) { // At least AIX 7.2. 740 features |= rtm_m; 741 } 742 #endif 743 #if defined(LINUX) && defined(VM_LITTLE_ENDIAN) 744 unsigned long auxv = getauxval(AT_HWCAP2); 745 746 if (auxv & PPC_FEATURE2_HTM_NOSC) { 747 if (auxv & PPC_FEATURE2_HAS_HTM) { 748 // TM on POWER8 and POWER9 in compat mode (VM) is supported by the JVM. 749 // TM on POWER9 DD2.1 NV (baremetal) is not supported by the JVM (TM on 750 // POWER9 DD2.1 NV has a few issues that need a couple of firmware 751 // and kernel workarounds, so there is a new mode only supported 752 // on non-virtualized P9 machines called HTM with no Suspend Mode). 753 // TM on POWER9 D2.2+ NV is not supported at all by Linux. 754 features |= rtm_m; 755 } 756 } 757 #endif 758 759 _features = features; 760 } 761 762 // Power 8: Configure Data Stream Control Register. 763 void VM_Version::config_dscr() { 764 // 7 InstWords for each call (function descriptor + blr instruction). 765 const int code_size = (2+2*7)*BytesPerInstWord; 766 767 // Allocate space for the code. 768 ResourceMark rm; 769 CodeBuffer cb("config_dscr", code_size, 0); 770 MacroAssembler* a = new MacroAssembler(&cb); 771 772 // Emit code. 773 uint64_t (*get_dscr)() = (uint64_t(*)())(void *)a->function_entry(); 774 uint32_t *code = (uint32_t *)a->pc(); 775 a->mfdscr(R3); 776 a->blr(); 777 778 void (*set_dscr)(long) = (void(*)(long))(void *)a->function_entry(); 779 a->mtdscr(R3); 780 a->blr(); 781 782 uint32_t *code_end = (uint32_t *)a->pc(); 783 a->flush(); 784 785 // Print the detection code. 786 if (PrintAssembly) { 787 ttyLocker ttyl; 788 tty->print_cr("Decoding dscr configuration stub at " INTPTR_FORMAT " before execution:", p2i(code)); 789 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 790 } 791 792 // Apply the configuration if needed. 793 _dscr_val = (*get_dscr)(); 794 if (Verbose) { 795 tty->print_cr("dscr value was 0x%lx" , _dscr_val); 796 } 797 bool change_requested = false; 798 if (DSCR_PPC64 != (uintx)-1) { 799 _dscr_val = DSCR_PPC64; 800 change_requested = true; 801 } 802 if (DSCR_DPFD_PPC64 <= 7) { 803 uint64_t mask = 0x7; 804 if ((_dscr_val & mask) != DSCR_DPFD_PPC64) { 805 _dscr_val = (_dscr_val & ~mask) | (DSCR_DPFD_PPC64); 806 change_requested = true; 807 } 808 } 809 if (DSCR_URG_PPC64 <= 7) { 810 uint64_t mask = 0x7 << 6; 811 if ((_dscr_val & mask) != DSCR_DPFD_PPC64 << 6) { 812 _dscr_val = (_dscr_val & ~mask) | (DSCR_URG_PPC64 << 6); 813 change_requested = true; 814 } 815 } 816 if (change_requested) { 817 (*set_dscr)(_dscr_val); 818 if (Verbose) { 819 tty->print_cr("dscr was set to 0x%lx" , (*get_dscr)()); 820 } 821 } 822 } 823 824 static uint64_t saved_features = 0; 825 826 void VM_Version::allow_all() { 827 saved_features = _features; 828 _features = all_features_m; 829 } 830 831 void VM_Version::revert() { 832 _features = saved_features; 833 }