1 /* 2 * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. 3 * Copyright 2012, 2014 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/stubCodeGenerator.hpp" 33 #include "utilities/defaultStream.hpp" 34 #include "vm_version_ppc.hpp" 35 #ifdef TARGET_OS_FAMILY_aix 36 # include "os_aix.inline.hpp" 37 #endif 38 #ifdef TARGET_OS_FAMILY_linux 39 # include "os_linux.inline.hpp" 40 #endif 41 42 # include <sys/sysinfo.h> 43 44 int VM_Version::_features = VM_Version::unknown_m; 45 int VM_Version::_measured_cache_line_size = 128; // default value 46 const char* VM_Version::_features_str = ""; 47 bool VM_Version::_is_determine_features_test_running = false; 48 49 50 #define MSG(flag) \ 51 if (flag && !FLAG_IS_DEFAULT(flag)) \ 52 jio_fprintf(defaultStream::error_stream(), \ 53 "warning: -XX:+" #flag " requires -XX:+UseSIGTRAP\n" \ 54 " -XX:+" #flag " will be disabled!\n"); 55 56 void VM_Version::initialize() { 57 58 // Test which instructions are supported and measure cache line size. 59 determine_features(); 60 61 // If PowerArchitecturePPC64 hasn't been specified explicitly determine from features. 62 if (FLAG_IS_DEFAULT(PowerArchitecturePPC64)) { 63 if (VM_Version::has_popcntw()) { 64 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 7); 65 } else if (VM_Version::has_cmpb()) { 66 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 6); 67 } else if (VM_Version::has_popcntb()) { 68 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 5); 69 } else { 70 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 0); 71 } 72 } 73 guarantee(PowerArchitecturePPC64 == 0 || PowerArchitecturePPC64 == 5 || 74 PowerArchitecturePPC64 == 6 || PowerArchitecturePPC64 == 7, 75 "PowerArchitecturePPC64 should be 0, 5, 6 or 7"); 76 77 if (!UseSIGTRAP) { 78 MSG(TrapBasedICMissChecks); 79 MSG(TrapBasedNotEntrantChecks); 80 MSG(TrapBasedNullChecks); 81 FLAG_SET_ERGO(bool, TrapBasedNotEntrantChecks, false); 82 FLAG_SET_ERGO(bool, TrapBasedNullChecks, false); 83 FLAG_SET_ERGO(bool, TrapBasedICMissChecks, false); 84 } 85 86 #ifdef COMPILER2 87 if (!UseSIGTRAP) { 88 MSG(TrapBasedRangeChecks); 89 FLAG_SET_ERGO(bool, TrapBasedRangeChecks, false); 90 } 91 92 // On Power6 test for section size. 93 if (PowerArchitecturePPC64 == 6) { 94 determine_section_size(); 95 // TODO: PPC port } else { 96 // TODO: PPC port PdScheduling::power6SectorSize = 0x20; 97 } 98 99 MaxVectorSize = 8; 100 #endif 101 102 // Create and print feature-string. 103 char buf[(num_features+1) * 16]; // Max 16 chars per feature. 104 jio_snprintf(buf, sizeof(buf), 105 "ppc64%s%s%s%s%s%s%s%s%s%s", 106 (has_fsqrt() ? " fsqrt" : ""), 107 (has_isel() ? " isel" : ""), 108 (has_lxarxeh() ? " lxarxeh" : ""), 109 (has_cmpb() ? " cmpb" : ""), 110 //(has_mftgpr()? " mftgpr" : ""), 111 (has_popcntb() ? " popcntb" : ""), 112 (has_popcntw() ? " popcntw" : ""), 113 (has_fcfids() ? " fcfids" : ""), 114 (has_vand() ? " vand" : ""), 115 (has_vcipher() ? " aes" : ""), 116 (has_vpmsumb() ? " vpmsumb" : "") 117 // Make sure number of %s matches num_features! 118 ); 119 _features_str = strdup(buf); 120 NOT_PRODUCT(if (Verbose) print_features();); 121 122 // PPC64 supports 8-byte compare-exchange operations (see 123 // Atomic::cmpxchg and StubGenerator::generate_atomic_cmpxchg_ptr) 124 // and 'atomic long memory ops' (see Unsafe_GetLongVolatile). 125 _supports_cx8 = true; 126 127 UseSSE = 0; // Only on x86 and x64 128 129 intx cache_line_size = _measured_cache_line_size; 130 131 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) AllocatePrefetchStyle = 1; 132 133 if (AllocatePrefetchStyle == 4) { 134 AllocatePrefetchStepSize = cache_line_size; // Need exact value. 135 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 12; // Use larger blocks by default. 136 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 2*cache_line_size; // Default is not defined? 137 } else { 138 if (cache_line_size > AllocatePrefetchStepSize) AllocatePrefetchStepSize = cache_line_size; 139 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 3; // Optimistic value. 140 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 3*cache_line_size; // Default is not defined? 141 } 142 143 assert(AllocatePrefetchLines > 0, "invalid value"); 144 if (AllocatePrefetchLines < 1) { // Set valid value in product VM. 145 AllocatePrefetchLines = 1; // Conservative value. 146 } 147 148 if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size) { 149 AllocatePrefetchStyle = 1; // Fall back if inappropriate. 150 } 151 152 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive"); 153 154 // Implementation does not use any of the vector instructions 155 // available with Power8. Their exploitation is still pending. 156 if (!UseCRC32Intrinsics) { 157 if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) { 158 FLAG_SET_DEFAULT(UseCRC32Intrinsics, true); 159 } 160 } 161 162 // The AES intrinsic stubs require AES instruction support. 163 #if defined(VM_LITTLE_ENDIAN) 164 if (has_vcipher()) { 165 if (FLAG_IS_DEFAULT(UseAES)) { 166 UseAES = true; 167 } 168 } else if (UseAES) { 169 if (!FLAG_IS_DEFAULT(UseAES)) 170 warning("AES instructions are not available on this CPU"); 171 FLAG_SET_DEFAULT(UseAES, false); 172 } 173 174 if (UseAES && has_vcipher()) { 175 if (FLAG_IS_DEFAULT(UseAESIntrinsics)) { 176 UseAESIntrinsics = true; 177 } 178 } else if (UseAESIntrinsics) { 179 if (!FLAG_IS_DEFAULT(UseAESIntrinsics)) 180 warning("AES intrinsics are not available on this CPU"); 181 FLAG_SET_DEFAULT(UseAESIntrinsics, false); 182 } 183 184 #else 185 if (UseAES) { 186 warning("AES instructions are not available on this CPU"); 187 FLAG_SET_DEFAULT(UseAES, false); 188 } 189 if (UseAESIntrinsics) { 190 if (!FLAG_IS_DEFAULT(UseAESIntrinsics)) 191 warning("AES intrinsics are not available on this CPU"); 192 FLAG_SET_DEFAULT(UseAESIntrinsics, false); 193 } 194 #endif 195 196 if (UseSHA) { 197 warning("SHA instructions are not available on this CPU"); 198 FLAG_SET_DEFAULT(UseSHA, false); 199 } 200 if (UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics) { 201 warning("SHA intrinsics are not available on this CPU"); 202 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false); 203 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false); 204 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false); 205 } 206 207 if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) { 208 UseMontgomeryMultiplyIntrinsic = true; 209 } 210 if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) { 211 UseMontgomerySquareIntrinsic = true; 212 } 213 } 214 215 void VM_Version::print_features() { 216 tty->print_cr("Version: %s cache_line_size = %d", cpu_features(), (int) get_cache_line_size()); 217 } 218 219 #ifdef COMPILER2 220 // Determine section size on power6: If section size is 8 instructions, 221 // there should be a difference between the two testloops of ~15 %. If 222 // no difference is detected the section is assumed to be 32 instructions. 223 void VM_Version::determine_section_size() { 224 225 int unroll = 80; 226 227 const int code_size = (2* unroll * 32 + 100)*BytesPerInstWord; 228 229 // Allocate space for the code. 230 ResourceMark rm; 231 CodeBuffer cb("detect_section_size", code_size, 0); 232 MacroAssembler* a = new MacroAssembler(&cb); 233 234 uint32_t *code = (uint32_t *)a->pc(); 235 // Emit code. 236 void (*test1)() = (void(*)())(void *)a->function_entry(); 237 238 Label l1; 239 240 a->li(R4, 1); 241 a->sldi(R4, R4, 28); 242 a->b(l1); 243 a->align(CodeEntryAlignment); 244 245 a->bind(l1); 246 247 for (int i = 0; i < unroll; i++) { 248 // Schleife 1 249 // ------- sector 0 ------------ 250 // ;; 0 251 a->nop(); // 1 252 a->fpnop0(); // 2 253 a->fpnop1(); // 3 254 a->addi(R4,R4, -1); // 4 255 256 // ;; 1 257 a->nop(); // 5 258 a->fmr(F6, F6); // 6 259 a->fmr(F7, F7); // 7 260 a->endgroup(); // 8 261 // ------- sector 8 ------------ 262 263 // ;; 2 264 a->nop(); // 9 265 a->nop(); // 10 266 a->fmr(F8, F8); // 11 267 a->fmr(F9, F9); // 12 268 269 // ;; 3 270 a->nop(); // 13 271 a->fmr(F10, F10); // 14 272 a->fmr(F11, F11); // 15 273 a->endgroup(); // 16 274 // -------- sector 16 ------------- 275 276 // ;; 4 277 a->nop(); // 17 278 a->nop(); // 18 279 a->fmr(F15, F15); // 19 280 a->fmr(F16, F16); // 20 281 282 // ;; 5 283 a->nop(); // 21 284 a->fmr(F17, F17); // 22 285 a->fmr(F18, F18); // 23 286 a->endgroup(); // 24 287 // ------- sector 24 ------------ 288 289 // ;; 6 290 a->nop(); // 25 291 a->nop(); // 26 292 a->fmr(F19, F19); // 27 293 a->fmr(F20, F20); // 28 294 295 // ;; 7 296 a->nop(); // 29 297 a->fmr(F21, F21); // 30 298 a->fmr(F22, F22); // 31 299 a->brnop0(); // 32 300 301 // ------- sector 32 ------------ 302 } 303 304 // ;; 8 305 a->cmpdi(CCR0, R4, unroll); // 33 306 a->bge(CCR0, l1); // 34 307 a->blr(); 308 309 // Emit code. 310 void (*test2)() = (void(*)())(void *)a->function_entry(); 311 // uint32_t *code = (uint32_t *)a->pc(); 312 313 Label l2; 314 315 a->li(R4, 1); 316 a->sldi(R4, R4, 28); 317 a->b(l2); 318 a->align(CodeEntryAlignment); 319 320 a->bind(l2); 321 322 for (int i = 0; i < unroll; i++) { 323 // Schleife 2 324 // ------- sector 0 ------------ 325 // ;; 0 326 a->brnop0(); // 1 327 a->nop(); // 2 328 //a->cmpdi(CCR0, R4, unroll); 329 a->fpnop0(); // 3 330 a->fpnop1(); // 4 331 a->addi(R4,R4, -1); // 5 332 333 // ;; 1 334 335 a->nop(); // 6 336 a->fmr(F6, F6); // 7 337 a->fmr(F7, F7); // 8 338 // ------- sector 8 --------------- 339 340 // ;; 2 341 a->endgroup(); // 9 342 343 // ;; 3 344 a->nop(); // 10 345 a->nop(); // 11 346 a->fmr(F8, F8); // 12 347 348 // ;; 4 349 a->fmr(F9, F9); // 13 350 a->nop(); // 14 351 a->fmr(F10, F10); // 15 352 353 // ;; 5 354 a->fmr(F11, F11); // 16 355 // -------- sector 16 ------------- 356 357 // ;; 6 358 a->endgroup(); // 17 359 360 // ;; 7 361 a->nop(); // 18 362 a->nop(); // 19 363 a->fmr(F15, F15); // 20 364 365 // ;; 8 366 a->fmr(F16, F16); // 21 367 a->nop(); // 22 368 a->fmr(F17, F17); // 23 369 370 // ;; 9 371 a->fmr(F18, F18); // 24 372 // -------- sector 24 ------------- 373 374 // ;; 10 375 a->endgroup(); // 25 376 377 // ;; 11 378 a->nop(); // 26 379 a->nop(); // 27 380 a->fmr(F19, F19); // 28 381 382 // ;; 12 383 a->fmr(F20, F20); // 29 384 a->nop(); // 30 385 a->fmr(F21, F21); // 31 386 387 // ;; 13 388 a->fmr(F22, F22); // 32 389 } 390 391 // -------- sector 32 ------------- 392 // ;; 14 393 a->cmpdi(CCR0, R4, unroll); // 33 394 a->bge(CCR0, l2); // 34 395 396 a->blr(); 397 uint32_t *code_end = (uint32_t *)a->pc(); 398 a->flush(); 399 400 double loop1_seconds,loop2_seconds, rel_diff; 401 uint64_t start1, stop1; 402 403 start1 = os::current_thread_cpu_time(false); 404 (*test1)(); 405 stop1 = os::current_thread_cpu_time(false); 406 loop1_seconds = (stop1- start1) / (1000 *1000 *1000.0); 407 408 409 start1 = os::current_thread_cpu_time(false); 410 (*test2)(); 411 stop1 = os::current_thread_cpu_time(false); 412 413 loop2_seconds = (stop1 - start1) / (1000 *1000 *1000.0); 414 415 rel_diff = (loop2_seconds - loop1_seconds) / loop1_seconds *100; 416 417 if (PrintAssembly) { 418 ttyLocker ttyl; 419 tty->print_cr("Decoding section size detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 420 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 421 tty->print_cr("Time loop1 :%f", loop1_seconds); 422 tty->print_cr("Time loop2 :%f", loop2_seconds); 423 tty->print_cr("(time2 - time1) / time1 = %f %%", rel_diff); 424 425 if (rel_diff > 12.0) { 426 tty->print_cr("Section Size 8 Instructions"); 427 } else{ 428 tty->print_cr("Section Size 32 Instructions or Power5"); 429 } 430 } 431 432 #if 0 // TODO: PPC port 433 // Set sector size (if not set explicitly). 434 if (FLAG_IS_DEFAULT(Power6SectorSize128PPC64)) { 435 if (rel_diff > 12.0) { 436 PdScheduling::power6SectorSize = 0x20; 437 } else { 438 PdScheduling::power6SectorSize = 0x80; 439 } 440 } else if (Power6SectorSize128PPC64) { 441 PdScheduling::power6SectorSize = 0x80; 442 } else { 443 PdScheduling::power6SectorSize = 0x20; 444 } 445 #endif 446 if (UsePower6SchedulerPPC64) Unimplemented(); 447 } 448 #endif // COMPILER2 449 450 void VM_Version::determine_features() { 451 #if defined(ABI_ELFv2) 452 const int code_size = (num_features+1+2*7)*BytesPerInstWord; // TODO(asmundak): calculation is incorrect. 453 #else 454 // 7 InstWords for each call (function descriptor + blr instruction). 455 const int code_size = (num_features+1+2*7)*BytesPerInstWord; 456 #endif 457 int features = 0; 458 459 // create test area 460 enum { BUFFER_SIZE = 2*4*K }; // Needs to be >=2* max cache line size (cache line size can't exceed min page size). 461 char test_area[BUFFER_SIZE]; 462 char *mid_of_test_area = &test_area[BUFFER_SIZE>>1]; 463 464 // Allocate space for the code. 465 ResourceMark rm; 466 CodeBuffer cb("detect_cpu_features", code_size, 0); 467 MacroAssembler* a = new MacroAssembler(&cb); 468 469 // Must be set to true so we can generate the test code. 470 _features = VM_Version::all_features_m; 471 472 // Emit code. 473 void (*test)(address addr, uint64_t offset)=(void(*)(address addr, uint64_t offset))(void *)a->function_entry(); 474 uint32_t *code = (uint32_t *)a->pc(); 475 // Don't use R0 in ldarx. 476 // Keep R3_ARG1 unmodified, it contains &field (see below). 477 // Keep R4_ARG2 unmodified, it contains offset = 0 (see below). 478 a->fsqrt(F3, F4); // code[0] -> fsqrt_m 479 a->fsqrts(F3, F4); // code[1] -> fsqrts_m 480 a->isel(R7, R5, R6, 0); // code[2] -> isel_m 481 a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[3] -> lxarx_m 482 a->cmpb(R7, R5, R6); // code[4] -> bcmp 483 //a->mftgpr(R7, F3); // code[5] -> mftgpr 484 a->popcntb(R7, R5); // code[6] -> popcntb 485 a->popcntw(R7, R5); // code[7] -> popcntw 486 a->fcfids(F3, F4); // code[8] -> fcfids 487 a->vand(VR0, VR0, VR0); // code[9] -> vand 488 a->vcipher(VR0, VR1, VR2); // code[10] -> vcipher 489 a->vpmsumb(VR0, VR1, VR2); // code[11] -> vpmsumb 490 a->blr(); 491 492 // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it. 493 void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->function_entry(); 494 a->dcbz(R3_ARG1); // R3_ARG1 = addr 495 a->blr(); 496 497 uint32_t *code_end = (uint32_t *)a->pc(); 498 a->flush(); 499 _features = VM_Version::unknown_m; 500 501 // Print the detection code. 502 if (PrintAssembly) { 503 ttyLocker ttyl; 504 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 505 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 506 } 507 508 // Measure cache line size. 509 memset(test_area, 0xFF, BUFFER_SIZE); // Fill test area with 0xFF. 510 (*zero_cacheline_func_ptr)(mid_of_test_area); // Call function which executes dcbz to the middle. 511 int count = 0; // count zeroed bytes 512 for (int i = 0; i < BUFFER_SIZE; i++) if (test_area[i] == 0) count++; 513 guarantee(is_power_of_2(count), "cache line size needs to be a power of 2"); 514 _measured_cache_line_size = count; 515 516 // Execute code. Illegal instructions will be replaced by 0 in the signal handler. 517 VM_Version::_is_determine_features_test_running = true; 518 (*test)((address)mid_of_test_area, (uint64_t)0); 519 VM_Version::_is_determine_features_test_running = false; 520 521 // determine which instructions are legal. 522 int feature_cntr = 0; 523 if (code[feature_cntr++]) features |= fsqrt_m; 524 if (code[feature_cntr++]) features |= fsqrts_m; 525 if (code[feature_cntr++]) features |= isel_m; 526 if (code[feature_cntr++]) features |= lxarxeh_m; 527 if (code[feature_cntr++]) features |= cmpb_m; 528 //if(code[feature_cntr++])features |= mftgpr_m; 529 if (code[feature_cntr++]) features |= popcntb_m; 530 if (code[feature_cntr++]) features |= popcntw_m; 531 if (code[feature_cntr++]) features |= fcfids_m; 532 if (code[feature_cntr++]) features |= vand_m; 533 if (code[feature_cntr++]) features |= vcipher_m; 534 if (code[feature_cntr++]) features |= vpmsumb_m; 535 536 // Print the detection code. 537 if (PrintAssembly) { 538 ttyLocker ttyl; 539 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " after execution:", p2i(code)); 540 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 541 } 542 543 _features = features; 544 } 545 546 547 static int saved_features = 0; 548 549 void VM_Version::allow_all() { 550 saved_features = _features; 551 _features = all_features_m; 552 } 553 554 void VM_Version::revert() { 555 _features = saved_features; 556 }