1 /* 2 * Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved. 3 * Copyright 2012, 2015 SAP AG. 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 int VM_Version::_features = VM_Version::unknown_m; 41 int VM_Version::_measured_cache_line_size = 32; // pessimistic init value 42 const char* VM_Version::_features_str = ""; 43 bool VM_Version::_is_determine_features_test_running = false; 44 45 46 #define MSG(flag) \ 47 if (flag && !FLAG_IS_DEFAULT(flag)) \ 48 jio_fprintf(defaultStream::error_stream(), \ 49 "warning: -XX:+" #flag " requires -XX:+UseSIGTRAP\n" \ 50 " -XX:+" #flag " will be disabled!\n"); 51 52 void VM_Version::initialize() { 53 54 // Test which instructions are supported and measure cache line size. 55 determine_features(); 56 57 // If PowerArchitecturePPC64 hasn't been specified explicitly determine from features. 58 if (FLAG_IS_DEFAULT(PowerArchitecturePPC64)) { 59 if (VM_Version::has_lqarx()) { 60 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 8); 61 } else if (VM_Version::has_popcntw()) { 62 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 7); 63 } else if (VM_Version::has_cmpb()) { 64 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 6); 65 } else if (VM_Version::has_popcntb()) { 66 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 5); 67 } else { 68 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 0); 69 } 70 } 71 guarantee(PowerArchitecturePPC64 == 0 || PowerArchitecturePPC64 == 5 || 72 PowerArchitecturePPC64 == 6 || PowerArchitecturePPC64 == 7 || 73 PowerArchitecturePPC64 == 8, 74 "PowerArchitecturePPC64 should be 0, 5, 6, 7, or 8"); 75 76 // Power 8: Configure Data Stream Control Register. 77 if (PowerArchitecturePPC64 >= 8) { 78 config_dscr(); 79 } 80 81 if (!UseSIGTRAP) { 82 MSG(TrapBasedICMissChecks); 83 MSG(TrapBasedNotEntrantChecks); 84 MSG(TrapBasedNullChecks); 85 FLAG_SET_ERGO(bool, TrapBasedNotEntrantChecks, false); 86 FLAG_SET_ERGO(bool, TrapBasedNullChecks, false); 87 FLAG_SET_ERGO(bool, TrapBasedICMissChecks, false); 88 } 89 90 #ifdef COMPILER2 91 if (!UseSIGTRAP) { 92 MSG(TrapBasedRangeChecks); 93 FLAG_SET_ERGO(bool, TrapBasedRangeChecks, false); 94 } 95 96 // On Power6 test for section size. 97 if (PowerArchitecturePPC64 == 6) { 98 determine_section_size(); 99 // TODO: PPC port } else { 100 // TODO: PPC port PdScheduling::power6SectorSize = 0x20; 101 } 102 103 MaxVectorSize = 8; 104 #endif 105 106 // Create and print feature-string. 107 char buf[(num_features+1) * 16]; // Max 16 chars per feature. 108 jio_snprintf(buf, sizeof(buf), 109 "ppc64%s%s%s%s%s%s%s%s%s%s%s%s", 110 (has_fsqrt() ? " fsqrt" : ""), 111 (has_isel() ? " isel" : ""), 112 (has_lxarxeh() ? " lxarxeh" : ""), 113 (has_cmpb() ? " cmpb" : ""), 114 //(has_mftgpr()? " mftgpr" : ""), 115 (has_popcntb() ? " popcntb" : ""), 116 (has_popcntw() ? " popcntw" : ""), 117 (has_fcfids() ? " fcfids" : ""), 118 (has_vand() ? " vand" : ""), 119 (has_lqarx() ? " lqarx" : ""), 120 (has_vcipher() ? " vcipher" : ""), 121 (has_vpmsumb() ? " vpmsumb" : ""), 122 (has_tcheck() ? " tcheck" : "") 123 // Make sure number of %s matches num_features! 124 ); 125 _features_str = os::strdup(buf); 126 if (Verbose) { 127 print_features(); 128 } 129 130 // PPC64 supports 8-byte compare-exchange operations (see 131 // Atomic::cmpxchg and StubGenerator::generate_atomic_cmpxchg_ptr) 132 // and 'atomic long memory ops' (see Unsafe_GetLongVolatile). 133 _supports_cx8 = true; 134 135 UseSSE = 0; // Only on x86 and x64 136 137 intx cache_line_size = _measured_cache_line_size; 138 139 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) AllocatePrefetchStyle = 1; 140 141 if (AllocatePrefetchStyle == 4) { 142 AllocatePrefetchStepSize = cache_line_size; // Need exact value. 143 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 12; // Use larger blocks by default. 144 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 2*cache_line_size; // Default is not defined? 145 } else { 146 if (cache_line_size > AllocatePrefetchStepSize) AllocatePrefetchStepSize = cache_line_size; 147 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 3; // Optimistic value. 148 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 3*cache_line_size; // Default is not defined? 149 } 150 151 assert(AllocatePrefetchLines > 0, "invalid value"); 152 if (AllocatePrefetchLines < 1) { // Set valid value in product VM. 153 AllocatePrefetchLines = 1; // Conservative value. 154 } 155 156 if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size) { 157 AllocatePrefetchStyle = 1; // Fall back if inappropriate. 158 } 159 160 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive"); 161 162 // Implementation does not use any of the vector instructions 163 // available with Power8. Their exploitation is still pending. 164 if (!UseCRC32Intrinsics) { 165 if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) { 166 FLAG_SET_DEFAULT(UseCRC32Intrinsics, true); 167 } 168 } 169 170 if (UseCRC32CIntrinsics) { 171 if (!FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) 172 warning("CRC32C intrinsics are not available on this CPU"); 173 FLAG_SET_DEFAULT(UseCRC32CIntrinsics, false); 174 } 175 176 // The AES intrinsic stubs require AES instruction support. 177 if (UseAES) { 178 warning("AES instructions are not available on this CPU"); 179 FLAG_SET_DEFAULT(UseAES, false); 180 } 181 if (UseAESIntrinsics) { 182 if (!FLAG_IS_DEFAULT(UseAESIntrinsics)) 183 warning("AES intrinsics are not available on this CPU"); 184 FLAG_SET_DEFAULT(UseAESIntrinsics, false); 185 } 186 187 if (UseGHASHIntrinsics) { 188 warning("GHASH intrinsics are not available on this CPU"); 189 FLAG_SET_DEFAULT(UseGHASHIntrinsics, false); 190 } 191 192 if (UseSHA) { 193 warning("SHA instructions are not available on this CPU"); 194 FLAG_SET_DEFAULT(UseSHA, false); 195 } 196 if (UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics) { 197 warning("SHA intrinsics are not available on this CPU"); 198 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false); 199 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false); 200 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false); 201 } 202 203 if (UseAdler32Intrinsics) { 204 warning("Adler32Intrinsics not available on this CPU."); 205 FLAG_SET_DEFAULT(UseAdler32Intrinsics, false); 206 } 207 208 if (FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) { 209 UseMultiplyToLenIntrinsic = true; 210 } 211 212 // Adjust RTM (Restricted Transactional Memory) flags. 213 if (!has_tcheck() && UseRTMLocking) { 214 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 215 // setting during arguments processing. See use_biased_locking(). 216 // VM_Version_init() is executed after UseBiasedLocking is used 217 // in Thread::allocate(). 218 vm_exit_during_initialization("RTM instructions are not available on this CPU"); 219 } 220 221 if (UseRTMLocking) { 222 #if INCLUDE_RTM_OPT 223 if (!UnlockExperimentalVMOptions) { 224 vm_exit_during_initialization("UseRTMLocking is only available as experimental option on this platform. " 225 "It must be enabled via -XX:+UnlockExperimentalVMOptions flag."); 226 } else { 227 warning("UseRTMLocking is only available as experimental option on this platform."); 228 } 229 if (!FLAG_IS_CMDLINE(UseRTMLocking)) { 230 // RTM locking should be used only for applications with 231 // high lock contention. For now we do not use it by default. 232 vm_exit_during_initialization("UseRTMLocking flag should be only set on command line"); 233 } 234 if (!is_power_of_2(RTMTotalCountIncrRate)) { 235 warning("RTMTotalCountIncrRate must be a power of 2, resetting it to 64"); 236 FLAG_SET_DEFAULT(RTMTotalCountIncrRate, 64); 237 } 238 if (RTMAbortRatio < 0 || RTMAbortRatio > 100) { 239 warning("RTMAbortRatio must be in the range 0 to 100, resetting it to 50"); 240 FLAG_SET_DEFAULT(RTMAbortRatio, 50); 241 } 242 guarantee(RTMSpinLoopCount > 0, "unsupported"); 243 #else 244 // Only C2 does RTM locking optimization. 245 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 246 // setting during arguments processing. See use_biased_locking(). 247 vm_exit_during_initialization("RTM locking optimization is not supported in this VM"); 248 #endif 249 } else { // !UseRTMLocking 250 if (UseRTMForStackLocks) { 251 if (!FLAG_IS_DEFAULT(UseRTMForStackLocks)) { 252 warning("UseRTMForStackLocks flag should be off when UseRTMLocking flag is off"); 253 } 254 FLAG_SET_DEFAULT(UseRTMForStackLocks, false); 255 } 256 if (UseRTMDeopt) { 257 FLAG_SET_DEFAULT(UseRTMDeopt, false); 258 } 259 if (PrintPreciseRTMLockingStatistics) { 260 FLAG_SET_DEFAULT(PrintPreciseRTMLockingStatistics, false); 261 } 262 } 263 264 // This machine does not allow unaligned memory accesses 265 if (UseUnalignedAccesses) { 266 if (!FLAG_IS_DEFAULT(UseUnalignedAccesses)) 267 warning("Unaligned memory access is not available on this CPU"); 268 FLAG_SET_DEFAULT(UseUnalignedAccesses, false); 269 } 270 } 271 272 bool VM_Version::use_biased_locking() { 273 #if INCLUDE_RTM_OPT 274 // RTM locking is most useful when there is high lock contention and 275 // low data contention. With high lock contention the lock is usually 276 // inflated and biased locking is not suitable for that case. 277 // RTM locking code requires that biased locking is off. 278 // Note: we can't switch off UseBiasedLocking in get_processor_features() 279 // because it is used by Thread::allocate() which is called before 280 // VM_Version::initialize(). 281 if (UseRTMLocking && UseBiasedLocking) { 282 if (FLAG_IS_DEFAULT(UseBiasedLocking)) { 283 FLAG_SET_DEFAULT(UseBiasedLocking, false); 284 } else { 285 warning("Biased locking is not supported with RTM locking; ignoring UseBiasedLocking flag." ); 286 UseBiasedLocking = false; 287 } 288 } 289 #endif 290 return UseBiasedLocking; 291 } 292 293 void VM_Version::print_features() { 294 tty->print_cr("Version: %s cache_line_size = %d", cpu_features(), (int) get_cache_line_size()); 295 } 296 297 #ifdef COMPILER2 298 // Determine section size on power6: If section size is 8 instructions, 299 // there should be a difference between the two testloops of ~15 %. If 300 // no difference is detected the section is assumed to be 32 instructions. 301 void VM_Version::determine_section_size() { 302 303 int unroll = 80; 304 305 const int code_size = (2* unroll * 32 + 100)*BytesPerInstWord; 306 307 // Allocate space for the code. 308 ResourceMark rm; 309 CodeBuffer cb("detect_section_size", code_size, 0); 310 MacroAssembler* a = new MacroAssembler(&cb); 311 312 uint32_t *code = (uint32_t *)a->pc(); 313 // Emit code. 314 void (*test1)() = (void(*)())(void *)a->function_entry(); 315 316 Label l1; 317 318 a->li(R4, 1); 319 a->sldi(R4, R4, 28); 320 a->b(l1); 321 a->align(CodeEntryAlignment); 322 323 a->bind(l1); 324 325 for (int i = 0; i < unroll; i++) { 326 // Schleife 1 327 // ------- sector 0 ------------ 328 // ;; 0 329 a->nop(); // 1 330 a->fpnop0(); // 2 331 a->fpnop1(); // 3 332 a->addi(R4,R4, -1); // 4 333 334 // ;; 1 335 a->nop(); // 5 336 a->fmr(F6, F6); // 6 337 a->fmr(F7, F7); // 7 338 a->endgroup(); // 8 339 // ------- sector 8 ------------ 340 341 // ;; 2 342 a->nop(); // 9 343 a->nop(); // 10 344 a->fmr(F8, F8); // 11 345 a->fmr(F9, F9); // 12 346 347 // ;; 3 348 a->nop(); // 13 349 a->fmr(F10, F10); // 14 350 a->fmr(F11, F11); // 15 351 a->endgroup(); // 16 352 // -------- sector 16 ------------- 353 354 // ;; 4 355 a->nop(); // 17 356 a->nop(); // 18 357 a->fmr(F15, F15); // 19 358 a->fmr(F16, F16); // 20 359 360 // ;; 5 361 a->nop(); // 21 362 a->fmr(F17, F17); // 22 363 a->fmr(F18, F18); // 23 364 a->endgroup(); // 24 365 // ------- sector 24 ------------ 366 367 // ;; 6 368 a->nop(); // 25 369 a->nop(); // 26 370 a->fmr(F19, F19); // 27 371 a->fmr(F20, F20); // 28 372 373 // ;; 7 374 a->nop(); // 29 375 a->fmr(F21, F21); // 30 376 a->fmr(F22, F22); // 31 377 a->brnop0(); // 32 378 379 // ------- sector 32 ------------ 380 } 381 382 // ;; 8 383 a->cmpdi(CCR0, R4, unroll); // 33 384 a->bge(CCR0, l1); // 34 385 a->blr(); 386 387 // Emit code. 388 void (*test2)() = (void(*)())(void *)a->function_entry(); 389 // uint32_t *code = (uint32_t *)a->pc(); 390 391 Label l2; 392 393 a->li(R4, 1); 394 a->sldi(R4, R4, 28); 395 a->b(l2); 396 a->align(CodeEntryAlignment); 397 398 a->bind(l2); 399 400 for (int i = 0; i < unroll; i++) { 401 // Schleife 2 402 // ------- sector 0 ------------ 403 // ;; 0 404 a->brnop0(); // 1 405 a->nop(); // 2 406 //a->cmpdi(CCR0, R4, unroll); 407 a->fpnop0(); // 3 408 a->fpnop1(); // 4 409 a->addi(R4,R4, -1); // 5 410 411 // ;; 1 412 413 a->nop(); // 6 414 a->fmr(F6, F6); // 7 415 a->fmr(F7, F7); // 8 416 // ------- sector 8 --------------- 417 418 // ;; 2 419 a->endgroup(); // 9 420 421 // ;; 3 422 a->nop(); // 10 423 a->nop(); // 11 424 a->fmr(F8, F8); // 12 425 426 // ;; 4 427 a->fmr(F9, F9); // 13 428 a->nop(); // 14 429 a->fmr(F10, F10); // 15 430 431 // ;; 5 432 a->fmr(F11, F11); // 16 433 // -------- sector 16 ------------- 434 435 // ;; 6 436 a->endgroup(); // 17 437 438 // ;; 7 439 a->nop(); // 18 440 a->nop(); // 19 441 a->fmr(F15, F15); // 20 442 443 // ;; 8 444 a->fmr(F16, F16); // 21 445 a->nop(); // 22 446 a->fmr(F17, F17); // 23 447 448 // ;; 9 449 a->fmr(F18, F18); // 24 450 // -------- sector 24 ------------- 451 452 // ;; 10 453 a->endgroup(); // 25 454 455 // ;; 11 456 a->nop(); // 26 457 a->nop(); // 27 458 a->fmr(F19, F19); // 28 459 460 // ;; 12 461 a->fmr(F20, F20); // 29 462 a->nop(); // 30 463 a->fmr(F21, F21); // 31 464 465 // ;; 13 466 a->fmr(F22, F22); // 32 467 } 468 469 // -------- sector 32 ------------- 470 // ;; 14 471 a->cmpdi(CCR0, R4, unroll); // 33 472 a->bge(CCR0, l2); // 34 473 474 a->blr(); 475 uint32_t *code_end = (uint32_t *)a->pc(); 476 a->flush(); 477 478 double loop1_seconds,loop2_seconds, rel_diff; 479 uint64_t start1, stop1; 480 481 start1 = os::current_thread_cpu_time(false); 482 (*test1)(); 483 stop1 = os::current_thread_cpu_time(false); 484 loop1_seconds = (stop1- start1) / (1000 *1000 *1000.0); 485 486 487 start1 = os::current_thread_cpu_time(false); 488 (*test2)(); 489 stop1 = os::current_thread_cpu_time(false); 490 491 loop2_seconds = (stop1 - start1) / (1000 *1000 *1000.0); 492 493 rel_diff = (loop2_seconds - loop1_seconds) / loop1_seconds *100; 494 495 if (PrintAssembly) { 496 ttyLocker ttyl; 497 tty->print_cr("Decoding section size detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 498 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 499 tty->print_cr("Time loop1 :%f", loop1_seconds); 500 tty->print_cr("Time loop2 :%f", loop2_seconds); 501 tty->print_cr("(time2 - time1) / time1 = %f %%", rel_diff); 502 503 if (rel_diff > 12.0) { 504 tty->print_cr("Section Size 8 Instructions"); 505 } else{ 506 tty->print_cr("Section Size 32 Instructions or Power5"); 507 } 508 } 509 510 #if 0 // TODO: PPC port 511 // Set sector size (if not set explicitly). 512 if (FLAG_IS_DEFAULT(Power6SectorSize128PPC64)) { 513 if (rel_diff > 12.0) { 514 PdScheduling::power6SectorSize = 0x20; 515 } else { 516 PdScheduling::power6SectorSize = 0x80; 517 } 518 } else if (Power6SectorSize128PPC64) { 519 PdScheduling::power6SectorSize = 0x80; 520 } else { 521 PdScheduling::power6SectorSize = 0x20; 522 } 523 #endif 524 if (UsePower6SchedulerPPC64) Unimplemented(); 525 } 526 #endif // COMPILER2 527 528 void VM_Version::determine_features() { 529 #if defined(ABI_ELFv2) 530 // 1 InstWord per call for the blr instruction. 531 const int code_size = (num_features+1+2*1)*BytesPerInstWord; 532 #else 533 // 7 InstWords for each call (function descriptor + blr instruction). 534 const int code_size = (num_features+1+2*7)*BytesPerInstWord; 535 #endif 536 int features = 0; 537 538 // create test area 539 enum { BUFFER_SIZE = 2*4*K }; // Needs to be >=2* max cache line size (cache line size can't exceed min page size). 540 char test_area[BUFFER_SIZE]; 541 char *mid_of_test_area = &test_area[BUFFER_SIZE>>1]; 542 543 // Allocate space for the code. 544 ResourceMark rm; 545 CodeBuffer cb("detect_cpu_features", code_size, 0); 546 MacroAssembler* a = new MacroAssembler(&cb); 547 548 // Must be set to true so we can generate the test code. 549 _features = VM_Version::all_features_m; 550 551 // Emit code. 552 void (*test)(address addr, uint64_t offset)=(void(*)(address addr, uint64_t offset))(void *)a->function_entry(); 553 uint32_t *code = (uint32_t *)a->pc(); 554 // Don't use R0 in ldarx. 555 // Keep R3_ARG1 unmodified, it contains &field (see below). 556 // Keep R4_ARG2 unmodified, it contains offset = 0 (see below). 557 a->fsqrt(F3, F4); // code[0] -> fsqrt_m 558 a->fsqrts(F3, F4); // code[1] -> fsqrts_m 559 a->isel(R7, R5, R6, 0); // code[2] -> isel_m 560 a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[3] -> lxarx_m 561 a->cmpb(R7, R5, R6); // code[4] -> cmpb 562 a->popcntb(R7, R5); // code[5] -> popcntb 563 a->popcntw(R7, R5); // code[6] -> popcntw 564 a->fcfids(F3, F4); // code[7] -> fcfids 565 a->vand(VR0, VR0, VR0); // code[8] -> vand 566 // arg0 of lqarx must be an even register, (arg1 + arg2) must be a multiple of 16 567 a->lqarx_unchecked(R6, R3_ARG1, R4_ARG2, 1); // code[9] -> lqarx_m 568 a->vcipher(VR0, VR1, VR2); // code[10] -> vcipher 569 a->vpmsumb(VR0, VR1, VR2); // code[11] -> vpmsumb 570 a->tcheck(0); // code[12] -> tcheck 571 a->blr(); 572 573 // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it. 574 void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->function_entry(); 575 a->dcbz(R3_ARG1); // R3_ARG1 = addr 576 a->blr(); 577 578 uint32_t *code_end = (uint32_t *)a->pc(); 579 a->flush(); 580 _features = VM_Version::unknown_m; 581 582 // Print the detection code. 583 if (PrintAssembly) { 584 ttyLocker ttyl; 585 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 586 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 587 } 588 589 // Measure cache line size. 590 memset(test_area, 0xFF, BUFFER_SIZE); // Fill test area with 0xFF. 591 (*zero_cacheline_func_ptr)(mid_of_test_area); // Call function which executes dcbz to the middle. 592 int count = 0; // count zeroed bytes 593 for (int i = 0; i < BUFFER_SIZE; i++) if (test_area[i] == 0) count++; 594 guarantee(is_power_of_2(count), "cache line size needs to be a power of 2"); 595 _measured_cache_line_size = count; 596 597 // Execute code. Illegal instructions will be replaced by 0 in the signal handler. 598 VM_Version::_is_determine_features_test_running = true; 599 // We must align the first argument to 16 bytes because of the lqarx check. 600 (*test)((address)align_size_up((intptr_t)mid_of_test_area, 16), (uint64_t)0); 601 VM_Version::_is_determine_features_test_running = false; 602 603 // determine which instructions are legal. 604 int feature_cntr = 0; 605 if (code[feature_cntr++]) features |= fsqrt_m; 606 if (code[feature_cntr++]) features |= fsqrts_m; 607 if (code[feature_cntr++]) features |= isel_m; 608 if (code[feature_cntr++]) features |= lxarxeh_m; 609 if (code[feature_cntr++]) features |= cmpb_m; 610 if (code[feature_cntr++]) features |= popcntb_m; 611 if (code[feature_cntr++]) features |= popcntw_m; 612 if (code[feature_cntr++]) features |= fcfids_m; 613 if (code[feature_cntr++]) features |= vand_m; 614 if (code[feature_cntr++]) features |= lqarx_m; 615 if (code[feature_cntr++]) features |= vcipher_m; 616 if (code[feature_cntr++]) features |= vpmsumb_m; 617 if (code[feature_cntr++]) features |= tcheck_m; 618 619 // Print the detection code. 620 if (PrintAssembly) { 621 ttyLocker ttyl; 622 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " after execution:", p2i(code)); 623 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 624 } 625 626 _features = features; 627 } 628 629 // Power 8: Configure Data Stream Control Register. 630 void VM_Version::config_dscr() { 631 assert(has_tcheck(), "Only execute on Power 8 or later!"); 632 633 // 7 InstWords for each call (function descriptor + blr instruction). 634 const int code_size = (2+2*7)*BytesPerInstWord; 635 636 // Allocate space for the code. 637 ResourceMark rm; 638 CodeBuffer cb("config_dscr", code_size, 0); 639 MacroAssembler* a = new MacroAssembler(&cb); 640 641 // Emit code. 642 uint64_t (*get_dscr)() = (uint64_t(*)())(void *)a->function_entry(); 643 uint32_t *code = (uint32_t *)a->pc(); 644 a->mfdscr(R3); 645 a->blr(); 646 647 void (*set_dscr)(long) = (void(*)(long))(void *)a->function_entry(); 648 a->mtdscr(R3); 649 a->blr(); 650 651 uint32_t *code_end = (uint32_t *)a->pc(); 652 a->flush(); 653 654 // Print the detection code. 655 if (PrintAssembly) { 656 ttyLocker ttyl; 657 tty->print_cr("Decoding dscr configuration stub at " INTPTR_FORMAT " before execution:", p2i(code)); 658 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 659 } 660 661 // Apply the configuration if needed. 662 uint64_t dscr_val = (*get_dscr)(); 663 if (Verbose) { 664 tty->print_cr("dscr value was 0x%lx" , dscr_val); 665 } 666 bool change_requested = false; 667 if (DSCR_PPC64 != (uintx)-1) { 668 dscr_val = DSCR_PPC64; 669 change_requested = true; 670 } 671 if (DSCR_DPFD_PPC64 <= 7) { 672 uint64_t mask = 0x7; 673 if ((dscr_val & mask) != DSCR_DPFD_PPC64) { 674 dscr_val = (dscr_val & ~mask) | (DSCR_DPFD_PPC64); 675 change_requested = true; 676 } 677 } 678 if (DSCR_URG_PPC64 <= 7) { 679 uint64_t mask = 0x7 << 6; 680 if ((dscr_val & mask) != DSCR_DPFD_PPC64 << 6) { 681 dscr_val = (dscr_val & ~mask) | (DSCR_URG_PPC64 << 6); 682 change_requested = true; 683 } 684 } 685 if (change_requested) { 686 (*set_dscr)(dscr_val); 687 if (Verbose) { 688 tty->print_cr("dscr was set to 0x%lx" , (*get_dscr)()); 689 } 690 } 691 } 692 693 static int saved_features = 0; 694 695 void VM_Version::allow_all() { 696 saved_features = _features; 697 _features = all_features_m; 698 } 699 700 void VM_Version::revert() { 701 _features = saved_features; 702 }