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_darn()) { 67 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 9); 68 } else if (VM_Version::has_lqarx()) { 69 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 8); 70 } else if (VM_Version::has_popcntw()) { 71 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 7); 72 } else if (VM_Version::has_cmpb()) { 73 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 6); 74 } else if (VM_Version::has_popcntb()) { 75 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 5); 76 } else { 77 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 0); 78 } 79 } 80 81 bool PowerArchitecturePPC64_ok = false; 82 switch (PowerArchitecturePPC64) { 83 case 9: if (!VM_Version::has_darn() ) break; 84 case 8: if (!VM_Version::has_lqarx() ) break; 85 case 7: if (!VM_Version::has_popcntw()) break; 86 case 6: if (!VM_Version::has_cmpb() ) break; 87 case 5: if (!VM_Version::has_popcntb()) break; 88 case 0: PowerArchitecturePPC64_ok = true; break; 89 default: break; 90 } 91 guarantee(PowerArchitecturePPC64_ok, "PowerArchitecturePPC64 cannot be set to " 92 UINTX_FORMAT " on this machine", PowerArchitecturePPC64); 93 94 // Power 8: Configure Data Stream Control Register. 95 if (PowerArchitecturePPC64 >= 8 && has_mfdscr()) { 96 config_dscr(); 97 } 98 99 if (!UseSIGTRAP) { 100 MSG(TrapBasedICMissChecks); 101 MSG(TrapBasedNotEntrantChecks); 102 MSG(TrapBasedNullChecks); 103 FLAG_SET_ERGO(bool, TrapBasedNotEntrantChecks, false); 104 FLAG_SET_ERGO(bool, TrapBasedNullChecks, false); 105 FLAG_SET_ERGO(bool, TrapBasedICMissChecks, false); 106 } 107 108 #ifdef COMPILER2 109 if (!UseSIGTRAP) { 110 MSG(TrapBasedRangeChecks); 111 FLAG_SET_ERGO(bool, TrapBasedRangeChecks, false); 112 } 113 114 // On Power6 test for section size. 115 if (PowerArchitecturePPC64 == 6) { 116 determine_section_size(); 117 // TODO: PPC port } else { 118 // TODO: PPC port PdScheduling::power6SectorSize = 0x20; 119 } 120 121 if (PowerArchitecturePPC64 >= 8) { 122 if (FLAG_IS_DEFAULT(SuperwordUseVSX)) { 123 FLAG_SET_ERGO(bool, SuperwordUseVSX, true); 124 } 125 } else { 126 if (SuperwordUseVSX) { 127 warning("SuperwordUseVSX specified, but needs at least Power8."); 128 FLAG_SET_DEFAULT(SuperwordUseVSX, false); 129 } 130 } 131 MaxVectorSize = SuperwordUseVSX ? 16 : 8; 132 #endif 133 134 // Create and print feature-string. 135 char buf[(num_features+1) * 16]; // Max 16 chars per feature. 136 jio_snprintf(buf, sizeof(buf), 137 "ppc64%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s", 138 (has_fsqrt() ? " fsqrt" : ""), 139 (has_isel() ? " isel" : ""), 140 (has_lxarxeh() ? " lxarxeh" : ""), 141 (has_cmpb() ? " cmpb" : ""), 142 (has_popcntb() ? " popcntb" : ""), 143 (has_popcntw() ? " popcntw" : ""), 144 (has_fcfids() ? " fcfids" : ""), 145 (has_vand() ? " vand" : ""), 146 (has_lqarx() ? " lqarx" : ""), 147 (has_vcipher() ? " aes" : ""), 148 (has_vpmsumb() ? " vpmsumb" : ""), 149 (has_mfdscr() ? " mfdscr" : ""), 150 (has_vsx() ? " vsx" : ""), 151 (has_ldbrx() ? " ldbrx" : ""), 152 (has_stdbrx() ? " stdbrx" : ""), 153 (has_vshasig() ? " sha" : ""), 154 (has_tm() ? " rtm" : ""), 155 (has_darn() ? " darn" : "") 156 // Make sure number of %s matches num_features! 157 ); 158 _features_string = os::strdup(buf); 159 if (Verbose) { 160 print_features(); 161 } 162 163 // PPC64 supports 8-byte compare-exchange operations (see Atomic::cmpxchg) 164 // and 'atomic long memory ops' (see Unsafe_GetLongVolatile). 165 _supports_cx8 = true; 166 167 // Used by C1. 168 _supports_atomic_getset4 = true; 169 _supports_atomic_getadd4 = true; 170 _supports_atomic_getset8 = true; 171 _supports_atomic_getadd8 = true; 172 173 UseSSE = 0; // Only on x86 and x64 174 175 intx cache_line_size = L1_data_cache_line_size(); 176 177 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) AllocatePrefetchStyle = 1; 178 179 if (AllocatePrefetchStyle == 4) { 180 AllocatePrefetchStepSize = cache_line_size; // Need exact value. 181 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 12; // Use larger blocks by default. 182 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 2*cache_line_size; // Default is not defined? 183 } else { 184 if (cache_line_size > AllocatePrefetchStepSize) AllocatePrefetchStepSize = cache_line_size; 185 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 3; // Optimistic value. 186 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 3*cache_line_size; // Default is not defined? 187 } 188 189 assert(AllocatePrefetchLines > 0, "invalid value"); 190 if (AllocatePrefetchLines < 1) { // Set valid value in product VM. 191 AllocatePrefetchLines = 1; // Conservative value. 192 } 193 194 if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size) { 195 AllocatePrefetchStyle = 1; // Fall back if inappropriate. 196 } 197 198 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive"); 199 200 // If running on Power8 or newer hardware, the implementation uses the available vector instructions. 201 // In all other cases, the implementation uses only generally available instructions. 202 if (!UseCRC32Intrinsics) { 203 if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) { 204 FLAG_SET_DEFAULT(UseCRC32Intrinsics, true); 205 } 206 } 207 208 // Implementation does not use any of the vector instructions available with Power8. 209 // Their exploitation is still pending (aka "work in progress"). 210 if (!UseCRC32CIntrinsics) { 211 if (FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) { 212 FLAG_SET_DEFAULT(UseCRC32CIntrinsics, true); 213 } 214 } 215 216 // TODO: Provide implementation. 217 if (UseAdler32Intrinsics) { 218 warning("Adler32Intrinsics not available on this CPU."); 219 FLAG_SET_DEFAULT(UseAdler32Intrinsics, false); 220 } 221 222 // The AES intrinsic stubs require AES instruction support. 223 if (has_vcipher()) { 224 if (FLAG_IS_DEFAULT(UseAES)) { 225 UseAES = true; 226 } 227 } else if (UseAES) { 228 if (!FLAG_IS_DEFAULT(UseAES)) 229 warning("AES instructions are not available on this CPU"); 230 FLAG_SET_DEFAULT(UseAES, false); 231 } 232 233 if (UseAES && has_vcipher()) { 234 if (FLAG_IS_DEFAULT(UseAESIntrinsics)) { 235 UseAESIntrinsics = true; 236 } 237 } else if (UseAESIntrinsics) { 238 if (!FLAG_IS_DEFAULT(UseAESIntrinsics)) 239 warning("AES intrinsics are not available on this CPU"); 240 FLAG_SET_DEFAULT(UseAESIntrinsics, false); 241 } 242 243 if (UseAESCTRIntrinsics) { 244 warning("AES/CTR intrinsics are not available on this CPU"); 245 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false); 246 } 247 248 if (UseGHASHIntrinsics) { 249 warning("GHASH intrinsics are not available on this CPU"); 250 FLAG_SET_DEFAULT(UseGHASHIntrinsics, false); 251 } 252 253 if (FLAG_IS_DEFAULT(UseFMA)) { 254 FLAG_SET_DEFAULT(UseFMA, true); 255 } 256 257 if (has_vshasig()) { 258 if (FLAG_IS_DEFAULT(UseSHA)) { 259 UseSHA = true; 260 } 261 } else if (UseSHA) { 262 if (!FLAG_IS_DEFAULT(UseSHA)) 263 warning("SHA instructions are not available on this CPU"); 264 FLAG_SET_DEFAULT(UseSHA, false); 265 } 266 267 if (UseSHA1Intrinsics) { 268 warning("Intrinsics for SHA-1 crypto hash functions not available on this CPU."); 269 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false); 270 } 271 272 if (UseSHA && has_vshasig()) { 273 if (FLAG_IS_DEFAULT(UseSHA256Intrinsics)) { 274 FLAG_SET_DEFAULT(UseSHA256Intrinsics, true); 275 } 276 } else if (UseSHA256Intrinsics) { 277 warning("Intrinsics for SHA-224 and SHA-256 crypto hash functions not available on this CPU."); 278 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false); 279 } 280 281 if (UseSHA && has_vshasig()) { 282 if (FLAG_IS_DEFAULT(UseSHA512Intrinsics)) { 283 FLAG_SET_DEFAULT(UseSHA512Intrinsics, true); 284 } 285 } else if (UseSHA512Intrinsics) { 286 warning("Intrinsics for SHA-384 and SHA-512 crypto hash functions not available on this CPU."); 287 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false); 288 } 289 290 if (!(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics)) { 291 FLAG_SET_DEFAULT(UseSHA, false); 292 } 293 294 if (FLAG_IS_DEFAULT(UseSquareToLenIntrinsic)) { 295 UseSquareToLenIntrinsic = true; 296 } 297 if (FLAG_IS_DEFAULT(UseMulAddIntrinsic)) { 298 UseMulAddIntrinsic = true; 299 } 300 if (FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) { 301 UseMultiplyToLenIntrinsic = true; 302 } 303 if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) { 304 UseMontgomeryMultiplyIntrinsic = true; 305 } 306 if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) { 307 UseMontgomerySquareIntrinsic = true; 308 } 309 310 if (UseVectorizedMismatchIntrinsic) { 311 warning("UseVectorizedMismatchIntrinsic specified, but not available on this CPU."); 312 FLAG_SET_DEFAULT(UseVectorizedMismatchIntrinsic, false); 313 } 314 315 316 // Adjust RTM (Restricted Transactional Memory) flags. 317 if (UseRTMLocking) { 318 // If CPU or OS do not support TM: 319 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 320 // setting during arguments processing. See use_biased_locking(). 321 // VM_Version_init() is executed after UseBiasedLocking is used 322 // in Thread::allocate(). 323 if (PowerArchitecturePPC64 < 8) { 324 vm_exit_during_initialization("RTM instructions are not available on this CPU."); 325 } 326 327 if (!has_tm()) { 328 vm_exit_during_initialization("RTM is not supported on this OS version."); 329 } 330 } 331 332 if (UseRTMLocking) { 333 #if INCLUDE_RTM_OPT 334 if (!FLAG_IS_CMDLINE(UseRTMLocking)) { 335 // RTM locking should be used only for applications with 336 // high lock contention. For now we do not use it by default. 337 vm_exit_during_initialization("UseRTMLocking flag should be only set on command line"); 338 } 339 #else 340 // Only C2 does RTM locking optimization. 341 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 342 // setting during arguments processing. See use_biased_locking(). 343 vm_exit_during_initialization("RTM locking optimization is not supported in this VM"); 344 #endif 345 } else { // !UseRTMLocking 346 if (UseRTMForStackLocks) { 347 if (!FLAG_IS_DEFAULT(UseRTMForStackLocks)) { 348 warning("UseRTMForStackLocks flag should be off when UseRTMLocking flag is off"); 349 } 350 FLAG_SET_DEFAULT(UseRTMForStackLocks, false); 351 } 352 if (UseRTMDeopt) { 353 FLAG_SET_DEFAULT(UseRTMDeopt, false); 354 } 355 if (PrintPreciseRTMLockingStatistics) { 356 FLAG_SET_DEFAULT(PrintPreciseRTMLockingStatistics, false); 357 } 358 } 359 360 // This machine allows unaligned memory accesses 361 if (FLAG_IS_DEFAULT(UseUnalignedAccesses)) { 362 FLAG_SET_DEFAULT(UseUnalignedAccesses, true); 363 } 364 } 365 366 bool VM_Version::use_biased_locking() { 367 #if INCLUDE_RTM_OPT 368 // RTM locking is most useful when there is high lock contention and 369 // low data contention. With high lock contention the lock is usually 370 // inflated and biased locking is not suitable for that case. 371 // RTM locking code requires that biased locking is off. 372 // Note: we can't switch off UseBiasedLocking in get_processor_features() 373 // because it is used by Thread::allocate() which is called before 374 // VM_Version::initialize(). 375 if (UseRTMLocking && UseBiasedLocking) { 376 if (FLAG_IS_DEFAULT(UseBiasedLocking)) { 377 FLAG_SET_DEFAULT(UseBiasedLocking, false); 378 } else { 379 warning("Biased locking is not supported with RTM locking; ignoring UseBiasedLocking flag." ); 380 UseBiasedLocking = false; 381 } 382 } 383 #endif 384 return UseBiasedLocking; 385 } 386 387 void VM_Version::print_features() { 388 tty->print_cr("Version: %s L1_data_cache_line_size=%d", features_string(), L1_data_cache_line_size()); 389 } 390 391 #ifdef COMPILER2 392 // Determine section size on power6: If section size is 8 instructions, 393 // there should be a difference between the two testloops of ~15 %. If 394 // no difference is detected the section is assumed to be 32 instructions. 395 void VM_Version::determine_section_size() { 396 397 int unroll = 80; 398 399 const int code_size = (2* unroll * 32 + 100)*BytesPerInstWord; 400 401 // Allocate space for the code. 402 ResourceMark rm; 403 CodeBuffer cb("detect_section_size", code_size, 0); 404 MacroAssembler* a = new MacroAssembler(&cb); 405 406 uint32_t *code = (uint32_t *)a->pc(); 407 // Emit code. 408 void (*test1)() = (void(*)())(void *)a->function_entry(); 409 410 Label l1; 411 412 a->li(R4, 1); 413 a->sldi(R4, R4, 28); 414 a->b(l1); 415 a->align(CodeEntryAlignment); 416 417 a->bind(l1); 418 419 for (int i = 0; i < unroll; i++) { 420 // Schleife 1 421 // ------- sector 0 ------------ 422 // ;; 0 423 a->nop(); // 1 424 a->fpnop0(); // 2 425 a->fpnop1(); // 3 426 a->addi(R4,R4, -1); // 4 427 428 // ;; 1 429 a->nop(); // 5 430 a->fmr(F6, F6); // 6 431 a->fmr(F7, F7); // 7 432 a->endgroup(); // 8 433 // ------- sector 8 ------------ 434 435 // ;; 2 436 a->nop(); // 9 437 a->nop(); // 10 438 a->fmr(F8, F8); // 11 439 a->fmr(F9, F9); // 12 440 441 // ;; 3 442 a->nop(); // 13 443 a->fmr(F10, F10); // 14 444 a->fmr(F11, F11); // 15 445 a->endgroup(); // 16 446 // -------- sector 16 ------------- 447 448 // ;; 4 449 a->nop(); // 17 450 a->nop(); // 18 451 a->fmr(F15, F15); // 19 452 a->fmr(F16, F16); // 20 453 454 // ;; 5 455 a->nop(); // 21 456 a->fmr(F17, F17); // 22 457 a->fmr(F18, F18); // 23 458 a->endgroup(); // 24 459 // ------- sector 24 ------------ 460 461 // ;; 6 462 a->nop(); // 25 463 a->nop(); // 26 464 a->fmr(F19, F19); // 27 465 a->fmr(F20, F20); // 28 466 467 // ;; 7 468 a->nop(); // 29 469 a->fmr(F21, F21); // 30 470 a->fmr(F22, F22); // 31 471 a->brnop0(); // 32 472 473 // ------- sector 32 ------------ 474 } 475 476 // ;; 8 477 a->cmpdi(CCR0, R4, unroll); // 33 478 a->bge(CCR0, l1); // 34 479 a->blr(); 480 481 // Emit code. 482 void (*test2)() = (void(*)())(void *)a->function_entry(); 483 // uint32_t *code = (uint32_t *)a->pc(); 484 485 Label l2; 486 487 a->li(R4, 1); 488 a->sldi(R4, R4, 28); 489 a->b(l2); 490 a->align(CodeEntryAlignment); 491 492 a->bind(l2); 493 494 for (int i = 0; i < unroll; i++) { 495 // Schleife 2 496 // ------- sector 0 ------------ 497 // ;; 0 498 a->brnop0(); // 1 499 a->nop(); // 2 500 //a->cmpdi(CCR0, R4, unroll); 501 a->fpnop0(); // 3 502 a->fpnop1(); // 4 503 a->addi(R4,R4, -1); // 5 504 505 // ;; 1 506 507 a->nop(); // 6 508 a->fmr(F6, F6); // 7 509 a->fmr(F7, F7); // 8 510 // ------- sector 8 --------------- 511 512 // ;; 2 513 a->endgroup(); // 9 514 515 // ;; 3 516 a->nop(); // 10 517 a->nop(); // 11 518 a->fmr(F8, F8); // 12 519 520 // ;; 4 521 a->fmr(F9, F9); // 13 522 a->nop(); // 14 523 a->fmr(F10, F10); // 15 524 525 // ;; 5 526 a->fmr(F11, F11); // 16 527 // -------- sector 16 ------------- 528 529 // ;; 6 530 a->endgroup(); // 17 531 532 // ;; 7 533 a->nop(); // 18 534 a->nop(); // 19 535 a->fmr(F15, F15); // 20 536 537 // ;; 8 538 a->fmr(F16, F16); // 21 539 a->nop(); // 22 540 a->fmr(F17, F17); // 23 541 542 // ;; 9 543 a->fmr(F18, F18); // 24 544 // -------- sector 24 ------------- 545 546 // ;; 10 547 a->endgroup(); // 25 548 549 // ;; 11 550 a->nop(); // 26 551 a->nop(); // 27 552 a->fmr(F19, F19); // 28 553 554 // ;; 12 555 a->fmr(F20, F20); // 29 556 a->nop(); // 30 557 a->fmr(F21, F21); // 31 558 559 // ;; 13 560 a->fmr(F22, F22); // 32 561 } 562 563 // -------- sector 32 ------------- 564 // ;; 14 565 a->cmpdi(CCR0, R4, unroll); // 33 566 a->bge(CCR0, l2); // 34 567 568 a->blr(); 569 uint32_t *code_end = (uint32_t *)a->pc(); 570 a->flush(); 571 572 double loop1_seconds,loop2_seconds, rel_diff; 573 uint64_t start1, stop1; 574 575 start1 = os::current_thread_cpu_time(false); 576 (*test1)(); 577 stop1 = os::current_thread_cpu_time(false); 578 loop1_seconds = (stop1- start1) / (1000 *1000 *1000.0); 579 580 581 start1 = os::current_thread_cpu_time(false); 582 (*test2)(); 583 stop1 = os::current_thread_cpu_time(false); 584 585 loop2_seconds = (stop1 - start1) / (1000 *1000 *1000.0); 586 587 rel_diff = (loop2_seconds - loop1_seconds) / loop1_seconds *100; 588 589 if (PrintAssembly) { 590 ttyLocker ttyl; 591 tty->print_cr("Decoding section size detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 592 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 593 tty->print_cr("Time loop1 :%f", loop1_seconds); 594 tty->print_cr("Time loop2 :%f", loop2_seconds); 595 tty->print_cr("(time2 - time1) / time1 = %f %%", rel_diff); 596 597 if (rel_diff > 12.0) { 598 tty->print_cr("Section Size 8 Instructions"); 599 } else{ 600 tty->print_cr("Section Size 32 Instructions or Power5"); 601 } 602 } 603 604 #if 0 // TODO: PPC port 605 // Set sector size (if not set explicitly). 606 if (FLAG_IS_DEFAULT(Power6SectorSize128PPC64)) { 607 if (rel_diff > 12.0) { 608 PdScheduling::power6SectorSize = 0x20; 609 } else { 610 PdScheduling::power6SectorSize = 0x80; 611 } 612 } else if (Power6SectorSize128PPC64) { 613 PdScheduling::power6SectorSize = 0x80; 614 } else { 615 PdScheduling::power6SectorSize = 0x20; 616 } 617 #endif 618 if (UsePower6SchedulerPPC64) Unimplemented(); 619 } 620 #endif // COMPILER2 621 622 void VM_Version::determine_features() { 623 #if defined(ABI_ELFv2) 624 // 1 InstWord per call for the blr instruction. 625 const int code_size = (num_features+1+2*1)*BytesPerInstWord; 626 #else 627 // 7 InstWords for each call (function descriptor + blr instruction). 628 const int code_size = (num_features+1+2*7)*BytesPerInstWord; 629 #endif 630 int features = 0; 631 632 // create test area 633 enum { BUFFER_SIZE = 2*4*K }; // Needs to be >=2* max cache line size (cache line size can't exceed min page size). 634 char test_area[BUFFER_SIZE]; 635 char *mid_of_test_area = &test_area[BUFFER_SIZE>>1]; 636 637 // Allocate space for the code. 638 ResourceMark rm; 639 CodeBuffer cb("detect_cpu_features", code_size, 0); 640 MacroAssembler* a = new MacroAssembler(&cb); 641 642 // Must be set to true so we can generate the test code. 643 _features = VM_Version::all_features_m; 644 645 // Emit code. 646 void (*test)(address addr, uint64_t offset)=(void(*)(address addr, uint64_t offset))(void *)a->function_entry(); 647 uint32_t *code = (uint32_t *)a->pc(); 648 // Don't use R0 in ldarx. 649 // Keep R3_ARG1 unmodified, it contains &field (see below). 650 // Keep R4_ARG2 unmodified, it contains offset = 0 (see below). 651 a->fsqrt(F3, F4); // code[0] -> fsqrt_m 652 a->fsqrts(F3, F4); // code[1] -> fsqrts_m 653 a->isel(R7, R5, R6, 0); // code[2] -> isel_m 654 a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[3] -> lxarx_m 655 a->cmpb(R7, R5, R6); // code[4] -> cmpb 656 a->popcntb(R7, R5); // code[5] -> popcntb 657 a->popcntw(R7, R5); // code[6] -> popcntw 658 a->fcfids(F3, F4); // code[7] -> fcfids 659 a->vand(VR0, VR0, VR0); // code[8] -> vand 660 // arg0 of lqarx must be an even register, (arg1 + arg2) must be a multiple of 16 661 a->lqarx_unchecked(R6, R3_ARG1, R4_ARG2, 1); // code[9] -> lqarx_m 662 a->vcipher(VR0, VR1, VR2); // code[10] -> vcipher 663 a->vpmsumb(VR0, VR1, VR2); // code[11] -> vpmsumb 664 a->mfdscr(R0); // code[12] -> mfdscr 665 a->lxvd2x(VSR0, R3_ARG1); // code[13] -> vsx 666 a->ldbrx(R7, R3_ARG1, R4_ARG2); // code[14] -> ldbrx 667 a->stdbrx(R7, R3_ARG1, R4_ARG2); // code[15] -> stdbrx 668 a->vshasigmaw(VR0, VR1, 1, 0xF); // code[16] -> vshasig 669 // rtm is determined by OS 670 a->darn(R7); // code[17] -> darn 671 a->blr(); 672 673 // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it. 674 void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->function_entry(); 675 a->dcbz(R3_ARG1); // R3_ARG1 = addr 676 a->blr(); 677 678 uint32_t *code_end = (uint32_t *)a->pc(); 679 a->flush(); 680 _features = VM_Version::unknown_m; 681 682 // Print the detection code. 683 if (PrintAssembly) { 684 ttyLocker ttyl; 685 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 686 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 687 } 688 689 // Measure cache line size. 690 memset(test_area, 0xFF, BUFFER_SIZE); // Fill test area with 0xFF. 691 (*zero_cacheline_func_ptr)(mid_of_test_area); // Call function which executes dcbz to the middle. 692 int count = 0; // count zeroed bytes 693 for (int i = 0; i < BUFFER_SIZE; i++) if (test_area[i] == 0) count++; 694 guarantee(is_power_of_2(count), "cache line size needs to be a power of 2"); 695 _L1_data_cache_line_size = count; 696 697 // Execute code. Illegal instructions will be replaced by 0 in the signal handler. 698 VM_Version::_is_determine_features_test_running = true; 699 // We must align the first argument to 16 bytes because of the lqarx check. 700 (*test)(align_up((address)mid_of_test_area, 16), 0); 701 VM_Version::_is_determine_features_test_running = false; 702 703 // determine which instructions are legal. 704 int feature_cntr = 0; 705 if (code[feature_cntr++]) features |= fsqrt_m; 706 if (code[feature_cntr++]) features |= fsqrts_m; 707 if (code[feature_cntr++]) features |= isel_m; 708 if (code[feature_cntr++]) features |= lxarxeh_m; 709 if (code[feature_cntr++]) features |= cmpb_m; 710 if (code[feature_cntr++]) features |= popcntb_m; 711 if (code[feature_cntr++]) features |= popcntw_m; 712 if (code[feature_cntr++]) features |= fcfids_m; 713 if (code[feature_cntr++]) features |= vand_m; 714 if (code[feature_cntr++]) features |= lqarx_m; 715 if (code[feature_cntr++]) features |= vcipher_m; 716 if (code[feature_cntr++]) features |= vpmsumb_m; 717 if (code[feature_cntr++]) features |= mfdscr_m; 718 if (code[feature_cntr++]) features |= vsx_m; 719 if (code[feature_cntr++]) features |= ldbrx_m; 720 if (code[feature_cntr++]) features |= stdbrx_m; 721 if (code[feature_cntr++]) features |= vshasig_m; 722 // feature rtm_m is determined by OS 723 if (code[feature_cntr++]) features |= darn_m; 724 725 // Print the detection code. 726 if (PrintAssembly) { 727 ttyLocker ttyl; 728 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " after execution:", p2i(code)); 729 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 730 } 731 732 _features = features; 733 734 #ifdef AIX 735 // To enable it on AIX it's necessary POWER8 or above and at least AIX 7.2. 736 // Actually, this is supported since AIX 7.1.. Unfortunately, this first 737 // contained bugs, so that it can only be enabled after AIX 7.1.3.30. 738 // The Java property os.version, which is used in RTM tests to decide 739 // whether the feature is available, only knows major and minor versions. 740 // We don't want to change this property, as user code might depend on it. 741 // So the tests can not check on subversion 3.30, and we only enable RTM 742 // with AIX 7.2. 743 if (has_lqarx()) { // POWER8 or above 744 if (os::Aix::os_version() >= 0x07020000) { // At least AIX 7.2. 745 _features |= rtm_m; 746 } 747 } 748 #endif 749 #if defined(LINUX) && defined(VM_LITTLE_ENDIAN) 750 unsigned long auxv = getauxval(AT_HWCAP2); 751 752 if (auxv & PPC_FEATURE2_HTM_NOSC) { 753 if (auxv & PPC_FEATURE2_HAS_HTM) { 754 // TM on POWER8 and POWER9 in compat mode (VM) is supported by the JVM. 755 // TM on POWER9 DD2.1 NV (baremetal) is not supported by the JVM (TM on 756 // POWER9 DD2.1 NV has a few issues that need a couple of firmware 757 // and kernel workarounds, so there is a new mode only supported 758 // on non-virtualized P9 machines called HTM with no Suspend Mode). 759 // TM on POWER9 D2.2+ NV is not supported at all by Linux. 760 _features |= rtm_m; 761 } 762 } 763 #endif 764 } 765 766 // Power 8: Configure Data Stream Control Register. 767 void VM_Version::config_dscr() { 768 // 7 InstWords for each call (function descriptor + blr instruction). 769 const int code_size = (2+2*7)*BytesPerInstWord; 770 771 // Allocate space for the code. 772 ResourceMark rm; 773 CodeBuffer cb("config_dscr", code_size, 0); 774 MacroAssembler* a = new MacroAssembler(&cb); 775 776 // Emit code. 777 uint64_t (*get_dscr)() = (uint64_t(*)())(void *)a->function_entry(); 778 uint32_t *code = (uint32_t *)a->pc(); 779 a->mfdscr(R3); 780 a->blr(); 781 782 void (*set_dscr)(long) = (void(*)(long))(void *)a->function_entry(); 783 a->mtdscr(R3); 784 a->blr(); 785 786 uint32_t *code_end = (uint32_t *)a->pc(); 787 a->flush(); 788 789 // Print the detection code. 790 if (PrintAssembly) { 791 ttyLocker ttyl; 792 tty->print_cr("Decoding dscr configuration stub at " INTPTR_FORMAT " before execution:", p2i(code)); 793 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 794 } 795 796 // Apply the configuration if needed. 797 _dscr_val = (*get_dscr)(); 798 if (Verbose) { 799 tty->print_cr("dscr value was 0x%lx" , _dscr_val); 800 } 801 bool change_requested = false; 802 if (DSCR_PPC64 != (uintx)-1) { 803 _dscr_val = DSCR_PPC64; 804 change_requested = true; 805 } 806 if (DSCR_DPFD_PPC64 <= 7) { 807 uint64_t mask = 0x7; 808 if ((_dscr_val & mask) != DSCR_DPFD_PPC64) { 809 _dscr_val = (_dscr_val & ~mask) | (DSCR_DPFD_PPC64); 810 change_requested = true; 811 } 812 } 813 if (DSCR_URG_PPC64 <= 7) { 814 uint64_t mask = 0x7 << 6; 815 if ((_dscr_val & mask) != DSCR_DPFD_PPC64 << 6) { 816 _dscr_val = (_dscr_val & ~mask) | (DSCR_URG_PPC64 << 6); 817 change_requested = true; 818 } 819 } 820 if (change_requested) { 821 (*set_dscr)(_dscr_val); 822 if (Verbose) { 823 tty->print_cr("dscr was set to 0x%lx" , (*get_dscr)()); 824 } 825 } 826 } 827 828 static uint64_t saved_features = 0; 829 830 void VM_Version::allow_all() { 831 saved_features = _features; 832 _features = all_features_m; 833 } 834 835 void VM_Version::revert() { 836 _features = saved_features; 837 }