1 /*
   2  * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright 2012, 2018 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 uint64_t VM_Version::_dscr_val = 0;
  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_lqarx()) {
  64       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 8);
  65     } else if (VM_Version::has_popcntw()) {
  66       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 7);
  67     } else if (VM_Version::has_cmpb()) {
  68       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 6);
  69     } else if (VM_Version::has_popcntb()) {
  70       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 5);
  71     } else {
  72       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 0);
  73     }
  74   }
  75   guarantee(PowerArchitecturePPC64 == 0 || PowerArchitecturePPC64 == 5 ||
  76             PowerArchitecturePPC64 == 6 || PowerArchitecturePPC64 == 7 ||
  77             PowerArchitecturePPC64 == 8,
  78             "PowerArchitecturePPC64 should be 0, 5, 6, 7, or 8");
  79 
  80   // Power 8: Configure Data Stream Control Register.
  81   if (PowerArchitecturePPC64 >= 8) {
  82     config_dscr();
  83   }
  84 
  85   if (!UseSIGTRAP) {
  86     MSG(TrapBasedICMissChecks);
  87     MSG(TrapBasedNotEntrantChecks);
  88     MSG(TrapBasedNullChecks);
  89     FLAG_SET_ERGO(bool, TrapBasedNotEntrantChecks, false);
  90     FLAG_SET_ERGO(bool, TrapBasedNullChecks,       false);
  91     FLAG_SET_ERGO(bool, TrapBasedICMissChecks,     false);
  92   }
  93 
  94 #ifdef COMPILER2
  95   if (!UseSIGTRAP) {
  96     MSG(TrapBasedRangeChecks);
  97     FLAG_SET_ERGO(bool, TrapBasedRangeChecks, false);
  98   }
  99 
 100   // On Power6 test for section size.
 101   if (PowerArchitecturePPC64 == 6) {
 102     determine_section_size();
 103   // TODO: PPC port } else {
 104   // TODO: PPC port PdScheduling::power6SectorSize = 0x20;
 105   }
 106 
 107   MaxVectorSize = 8;
 108 #endif
 109 
 110   // Create and print feature-string.
 111   char buf[(num_features+1) * 16]; // Max 16 chars per feature.
 112   jio_snprintf(buf, sizeof(buf),
 113                "ppc64%s%s%s%s%s%s%s%s%s%s%s%s%s%s",
 114                (has_fsqrt()   ? " fsqrt"   : ""),
 115                (has_isel()    ? " isel"    : ""),
 116                (has_lxarxeh() ? " lxarxeh" : ""),
 117                (has_cmpb()    ? " cmpb"    : ""),
 118                //(has_mftgpr()? " mftgpr"  : ""),
 119                (has_popcntb() ? " popcntb" : ""),
 120                (has_popcntw() ? " popcntw" : ""),
 121                (has_fcfids()  ? " fcfids"  : ""),
 122                (has_vand()    ? " vand"    : ""),
 123                (has_lqarx()   ? " lqarx"   : ""),
 124                (has_vcipher() ? " aes"     : ""),
 125                (has_vpmsumb() ? " vpmsumb" : ""),
 126                (has_mfdscr()  ? " mfdscr"  : ""),
 127                (has_vsx()     ? " vsx"     : ""),
 128                (has_vshasig() ? " sha"     : "")
 129                // Make sure number of %s matches num_features!
 130               );
 131   _features_str = strdup(buf);
 132   if (Verbose) {
 133     print_features();
 134   }
 135 
 136   // PPC64 supports 8-byte compare-exchange operations (see
 137   // Atomic::cmpxchg and StubGenerator::generate_atomic_cmpxchg_ptr)
 138   // and 'atomic long memory ops' (see Unsafe_GetLongVolatile).
 139   _supports_cx8 = true;
 140 
 141   UseSSE = 0; // Only on x86 and x64
 142 
 143   intx cache_line_size = _measured_cache_line_size;
 144 
 145   if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) AllocatePrefetchStyle = 1;
 146 
 147   if (AllocatePrefetchStyle == 4) {
 148     AllocatePrefetchStepSize = cache_line_size; // Need exact value.
 149     if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 12; // Use larger blocks by default.
 150     if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 2*cache_line_size; // Default is not defined?
 151   } else {
 152     if (cache_line_size > AllocatePrefetchStepSize) AllocatePrefetchStepSize = cache_line_size;
 153     if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 3; // Optimistic value.
 154     if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 3*cache_line_size; // Default is not defined?
 155   }
 156 
 157   assert(AllocatePrefetchLines > 0, "invalid value");
 158   if (AllocatePrefetchLines < 1) { // Set valid value in product VM.
 159     AllocatePrefetchLines = 1; // Conservative value.
 160   }
 161 
 162   if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size) {
 163     AllocatePrefetchStyle = 1; // Fall back if inappropriate.
 164   }
 165 
 166   assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive");
 167 
 168   // Implementation does not use any of the vector instructions
 169   // available with Power8. Their exploitation is still pending.
 170   if (!UseCRC32Intrinsics) {
 171     if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) {
 172       FLAG_SET_DEFAULT(UseCRC32Intrinsics, true);
 173     }
 174   }
 175 
 176   // The AES intrinsic stubs require AES instruction support.
 177   if (has_vcipher()) {
 178     if (FLAG_IS_DEFAULT(UseAES)) {
 179       UseAES = true;
 180     }
 181   } else if (UseAES) {
 182     if (!FLAG_IS_DEFAULT(UseAES))
 183       warning("AES instructions are not available on this CPU");
 184     FLAG_SET_DEFAULT(UseAES, false);
 185   }
 186 
 187   if (UseAES && has_vcipher()) {
 188     if (FLAG_IS_DEFAULT(UseAESIntrinsics)) {
 189       UseAESIntrinsics = true;
 190     }
 191   } else if (UseAESIntrinsics) {
 192     if (!FLAG_IS_DEFAULT(UseAESIntrinsics))
 193       warning("AES intrinsics are not available on this CPU");
 194     FLAG_SET_DEFAULT(UseAESIntrinsics, false);
 195   }
 196 
 197   if (has_vshasig()) {
 198     if (FLAG_IS_DEFAULT(UseSHA)) {
 199       UseSHA = true;
 200     }
 201   } else if (UseSHA) {
 202     if (!FLAG_IS_DEFAULT(UseSHA))
 203       warning("SHA instructions are not available on this CPU");
 204     FLAG_SET_DEFAULT(UseSHA, false);
 205   }
 206 
 207   if (UseSHA1Intrinsics) {
 208     warning("Intrinsics for SHA-1 crypto hash functions not available on this CPU.");
 209     FLAG_SET_DEFAULT(UseSHA1Intrinsics, false);
 210   }
 211 
 212   if (UseSHA && has_vshasig()) {
 213     if (FLAG_IS_DEFAULT(UseSHA256Intrinsics)) {
 214       FLAG_SET_DEFAULT(UseSHA256Intrinsics, true);
 215     }
 216   } else if (UseSHA256Intrinsics) {
 217     warning("Intrinsics for SHA-224 and SHA-256 crypto hash functions not available on this CPU.");
 218     FLAG_SET_DEFAULT(UseSHA256Intrinsics, false);
 219   }
 220 
 221   if (UseSHA && has_vshasig()) {
 222     if (FLAG_IS_DEFAULT(UseSHA512Intrinsics)) {
 223       FLAG_SET_DEFAULT(UseSHA512Intrinsics, true);
 224     }
 225   } else if (UseSHA512Intrinsics) {
 226     warning("Intrinsics for SHA-384 and SHA-512 crypto hash functions not available on this CPU.");
 227     FLAG_SET_DEFAULT(UseSHA512Intrinsics, false);
 228   }
 229 
 230   if (!(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics)) {
 231     FLAG_SET_DEFAULT(UseSHA, false);
 232   }
 233 
 234   if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) {
 235     UseMontgomeryMultiplyIntrinsic = true;
 236   }
 237   if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) {
 238     UseMontgomerySquareIntrinsic = true;
 239   }
 240 }
 241 
 242 void VM_Version::print_features() {
 243   tty->print_cr("Version: %s cache_line_size = %d", cpu_features(), (int) get_cache_line_size());
 244 }
 245 
 246 #ifdef COMPILER2
 247 // Determine section size on power6: If section size is 8 instructions,
 248 // there should be a difference between the two testloops of ~15 %. If
 249 // no difference is detected the section is assumed to be 32 instructions.
 250 void VM_Version::determine_section_size() {
 251 
 252   int unroll = 80;
 253 
 254   const int code_size = (2* unroll * 32 + 100)*BytesPerInstWord;
 255 
 256   // Allocate space for the code.
 257   ResourceMark rm;
 258   CodeBuffer cb("detect_section_size", code_size, 0);
 259   MacroAssembler* a = new MacroAssembler(&cb);
 260 
 261   uint32_t *code = (uint32_t *)a->pc();
 262   // Emit code.
 263   void (*test1)() = (void(*)())(void *)a->function_entry();
 264 
 265   Label l1;
 266 
 267   a->li(R4, 1);
 268   a->sldi(R4, R4, 28);
 269   a->b(l1);
 270   a->align(CodeEntryAlignment);
 271 
 272   a->bind(l1);
 273 
 274   for (int i = 0; i < unroll; i++) {
 275     // Schleife 1
 276     // ------- sector 0 ------------
 277     // ;; 0
 278     a->nop();                   // 1
 279     a->fpnop0();                // 2
 280     a->fpnop1();                // 3
 281     a->addi(R4,R4, -1); // 4
 282 
 283     // ;;  1
 284     a->nop();                   // 5
 285     a->fmr(F6, F6);             // 6
 286     a->fmr(F7, F7);             // 7
 287     a->endgroup();              // 8
 288     // ------- sector 8 ------------
 289 
 290     // ;;  2
 291     a->nop();                   // 9
 292     a->nop();                   // 10
 293     a->fmr(F8, F8);             // 11
 294     a->fmr(F9, F9);             // 12
 295 
 296     // ;;  3
 297     a->nop();                   // 13
 298     a->fmr(F10, F10);           // 14
 299     a->fmr(F11, F11);           // 15
 300     a->endgroup();              // 16
 301     // -------- sector 16 -------------
 302 
 303     // ;;  4
 304     a->nop();                   // 17
 305     a->nop();                   // 18
 306     a->fmr(F15, F15);           // 19
 307     a->fmr(F16, F16);           // 20
 308 
 309     // ;;  5
 310     a->nop();                   // 21
 311     a->fmr(F17, F17);           // 22
 312     a->fmr(F18, F18);           // 23
 313     a->endgroup();              // 24
 314     // ------- sector 24  ------------
 315 
 316     // ;;  6
 317     a->nop();                   // 25
 318     a->nop();                   // 26
 319     a->fmr(F19, F19);           // 27
 320     a->fmr(F20, F20);           // 28
 321 
 322     // ;;  7
 323     a->nop();                   // 29
 324     a->fmr(F21, F21);           // 30
 325     a->fmr(F22, F22);           // 31
 326     a->brnop0();                // 32
 327 
 328     // ------- sector 32 ------------
 329   }
 330 
 331   // ;; 8
 332   a->cmpdi(CCR0, R4, unroll);   // 33
 333   a->bge(CCR0, l1);             // 34
 334   a->blr();
 335 
 336   // Emit code.
 337   void (*test2)() = (void(*)())(void *)a->function_entry();
 338   // uint32_t *code = (uint32_t *)a->pc();
 339 
 340   Label l2;
 341 
 342   a->li(R4, 1);
 343   a->sldi(R4, R4, 28);
 344   a->b(l2);
 345   a->align(CodeEntryAlignment);
 346 
 347   a->bind(l2);
 348 
 349   for (int i = 0; i < unroll; i++) {
 350     // Schleife 2
 351     // ------- sector 0 ------------
 352     // ;; 0
 353     a->brnop0();                  // 1
 354     a->nop();                     // 2
 355     //a->cmpdi(CCR0, R4, unroll);
 356     a->fpnop0();                  // 3
 357     a->fpnop1();                  // 4
 358     a->addi(R4,R4, -1);           // 5
 359 
 360     // ;; 1
 361 
 362     a->nop();                     // 6
 363     a->fmr(F6, F6);               // 7
 364     a->fmr(F7, F7);               // 8
 365     // ------- sector 8 ---------------
 366 
 367     // ;; 2
 368     a->endgroup();                // 9
 369 
 370     // ;; 3
 371     a->nop();                     // 10
 372     a->nop();                     // 11
 373     a->fmr(F8, F8);               // 12
 374 
 375     // ;; 4
 376     a->fmr(F9, F9);               // 13
 377     a->nop();                     // 14
 378     a->fmr(F10, F10);             // 15
 379 
 380     // ;; 5
 381     a->fmr(F11, F11);             // 16
 382     // -------- sector 16 -------------
 383 
 384     // ;; 6
 385     a->endgroup();                // 17
 386 
 387     // ;; 7
 388     a->nop();                     // 18
 389     a->nop();                     // 19
 390     a->fmr(F15, F15);             // 20
 391 
 392     // ;; 8
 393     a->fmr(F16, F16);             // 21
 394     a->nop();                     // 22
 395     a->fmr(F17, F17);             // 23
 396 
 397     // ;; 9
 398     a->fmr(F18, F18);             // 24
 399     // -------- sector 24 -------------
 400 
 401     // ;; 10
 402     a->endgroup();                // 25
 403 
 404     // ;; 11
 405     a->nop();                     // 26
 406     a->nop();                     // 27
 407     a->fmr(F19, F19);             // 28
 408 
 409     // ;; 12
 410     a->fmr(F20, F20);             // 29
 411     a->nop();                     // 30
 412     a->fmr(F21, F21);             // 31
 413 
 414     // ;; 13
 415     a->fmr(F22, F22);             // 32
 416   }
 417 
 418   // -------- sector 32 -------------
 419   // ;; 14
 420   a->cmpdi(CCR0, R4, unroll); // 33
 421   a->bge(CCR0, l2);           // 34
 422 
 423   a->blr();
 424   uint32_t *code_end = (uint32_t *)a->pc();
 425   a->flush();
 426 
 427   double loop1_seconds,loop2_seconds, rel_diff;
 428   uint64_t start1, stop1;
 429 
 430   start1 = os::current_thread_cpu_time(false);
 431   (*test1)();
 432   stop1 = os::current_thread_cpu_time(false);
 433   loop1_seconds = (stop1- start1) / (1000 *1000 *1000.0);
 434 
 435 
 436   start1 = os::current_thread_cpu_time(false);
 437   (*test2)();
 438   stop1 = os::current_thread_cpu_time(false);
 439 
 440   loop2_seconds = (stop1 - start1) / (1000 *1000 *1000.0);
 441 
 442   rel_diff = (loop2_seconds - loop1_seconds) / loop1_seconds *100;
 443 
 444   if (PrintAssembly) {
 445     ttyLocker ttyl;
 446     tty->print_cr("Decoding section size detection stub at " INTPTR_FORMAT " before execution:", p2i(code));
 447     Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
 448     tty->print_cr("Time loop1 :%f", loop1_seconds);
 449     tty->print_cr("Time loop2 :%f", loop2_seconds);
 450     tty->print_cr("(time2 - time1) / time1 = %f %%", rel_diff);
 451 
 452     if (rel_diff > 12.0) {
 453       tty->print_cr("Section Size 8 Instructions");
 454     } else{
 455       tty->print_cr("Section Size 32 Instructions or Power5");
 456     }
 457   }
 458 
 459 #if 0 // TODO: PPC port
 460   // Set sector size (if not set explicitly).
 461   if (FLAG_IS_DEFAULT(Power6SectorSize128PPC64)) {
 462     if (rel_diff > 12.0) {
 463       PdScheduling::power6SectorSize = 0x20;
 464     } else {
 465       PdScheduling::power6SectorSize = 0x80;
 466     }
 467   } else if (Power6SectorSize128PPC64) {
 468     PdScheduling::power6SectorSize = 0x80;
 469   } else {
 470     PdScheduling::power6SectorSize = 0x20;
 471   }
 472 #endif
 473   if (UsePower6SchedulerPPC64) Unimplemented();
 474 }
 475 #endif // COMPILER2
 476 
 477 void VM_Version::determine_features() {
 478 #if defined(ABI_ELFv2)
 479   const int code_size = (num_features+1+2*7)*BytesPerInstWord; // TODO(asmundak): calculation is incorrect.
 480 #else
 481   // 7 InstWords for each call (function descriptor + blr instruction).
 482   const int code_size = (num_features+1+2*7)*BytesPerInstWord;
 483 #endif
 484   int features = 0;
 485 
 486   // create test area
 487   enum { BUFFER_SIZE = 2*4*K }; // Needs to be >=2* max cache line size (cache line size can't exceed min page size).
 488   char test_area[BUFFER_SIZE];
 489   char *mid_of_test_area = &test_area[BUFFER_SIZE>>1];
 490 
 491   // Allocate space for the code.
 492   ResourceMark rm;
 493   CodeBuffer cb("detect_cpu_features", code_size, 0);
 494   MacroAssembler* a = new MacroAssembler(&cb);
 495 
 496   // Must be set to true so we can generate the test code.
 497   _features = VM_Version::all_features_m;
 498 
 499   // Emit code.
 500   void (*test)(address addr, uint64_t offset)=(void(*)(address addr, uint64_t offset))(void *)a->function_entry();
 501   uint32_t *code = (uint32_t *)a->pc();
 502   // Don't use R0 in ldarx.
 503   // Keep R3_ARG1 unmodified, it contains &field (see below).
 504   // Keep R4_ARG2 unmodified, it contains offset = 0 (see below).
 505   a->fsqrt(F3, F4);                            // code[0] -> fsqrt_m
 506   a->fsqrts(F3, F4);                           // code[1] -> fsqrts_m
 507   a->isel(R7, R5, R6, 0);                      // code[2] -> isel_m
 508   a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[3] -> lxarx_m
 509   a->cmpb(R7, R5, R6);                         // code[4] -> bcmp
 510   //a->mftgpr(R7, F3);                         // code[5] -> mftgpr
 511   a->popcntb(R7, R5);                          // code[6] -> popcntb
 512   a->popcntw(R7, R5);                          // code[7] -> popcntw
 513   a->fcfids(F3, F4);                           // code[8] -> fcfids
 514   a->vand(VR0, VR0, VR0);                      // code[9] -> vand
 515   a->lqarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[10] -> lqarx_m
 516   a->vcipher(VR0, VR1, VR2);                   // code[11] -> vcipher
 517   a->vpmsumb(VR0, VR1, VR2);                   // code[12] -> vpmsumb
 518   a->mfdscr(R0);                               // code[13] -> mfdscr
 519   a->lxvd2x(VSR0, R3_ARG1);                    // code[14] -> vsx
 520   a->vshasigmaw(VR0, VR1, 1, 0xF);             // code[15] -> vshasig
 521   a->blr();
 522 
 523   // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it.
 524   void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->function_entry();
 525   a->dcbz(R3_ARG1); // R3_ARG1 = addr
 526   a->blr();
 527 
 528   uint32_t *code_end = (uint32_t *)a->pc();
 529   a->flush();
 530   _features = VM_Version::unknown_m;
 531 
 532   // Print the detection code.
 533   if (PrintAssembly) {
 534     ttyLocker ttyl;
 535     tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " before execution:", p2i(code));
 536     Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
 537   }
 538 
 539   // Measure cache line size.
 540   memset(test_area, 0xFF, BUFFER_SIZE); // Fill test area with 0xFF.
 541   (*zero_cacheline_func_ptr)(mid_of_test_area); // Call function which executes dcbz to the middle.
 542   int count = 0; // count zeroed bytes
 543   for (int i = 0; i < BUFFER_SIZE; i++) if (test_area[i] == 0) count++;
 544   guarantee(is_power_of_2(count), "cache line size needs to be a power of 2");
 545   _measured_cache_line_size = count;
 546 
 547   // Execute code. Illegal instructions will be replaced by 0 in the signal handler.
 548   VM_Version::_is_determine_features_test_running = true;
 549   (*test)((address)mid_of_test_area, (uint64_t)0);
 550   VM_Version::_is_determine_features_test_running = false;
 551 
 552   // determine which instructions are legal.
 553   int feature_cntr = 0;
 554   if (code[feature_cntr++]) features |= fsqrt_m;
 555   if (code[feature_cntr++]) features |= fsqrts_m;
 556   if (code[feature_cntr++]) features |= isel_m;
 557   if (code[feature_cntr++]) features |= lxarxeh_m;
 558   if (code[feature_cntr++]) features |= cmpb_m;
 559   //if(code[feature_cntr++])features |= mftgpr_m;
 560   if (code[feature_cntr++]) features |= popcntb_m;
 561   if (code[feature_cntr++]) features |= popcntw_m;
 562   if (code[feature_cntr++]) features |= fcfids_m;
 563   if (code[feature_cntr++]) features |= vand_m;
 564   if (code[feature_cntr++]) features |= lqarx_m;
 565   if (code[feature_cntr++]) features |= vcipher_m;
 566   if (code[feature_cntr++]) features |= vpmsumb_m;
 567   if (code[feature_cntr++]) features |= mfdscr_m;
 568   if (code[feature_cntr++]) features |= vsx_m;
 569   if (code[feature_cntr++]) features |= vshasig_m;
 570 
 571   // Print the detection code.
 572   if (PrintAssembly) {
 573     ttyLocker ttyl;
 574     tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " after execution:", p2i(code));
 575     Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
 576   }
 577 
 578   _features = features;
 579 }
 580 
 581 // Power 8: Configure Data Stream Control Register.
 582 void VM_Version::config_dscr() {
 583   assert(has_lqarx(), "Only execute on Power 8 or later!");
 584 
 585   // 7 InstWords for each call (function descriptor + blr instruction).
 586   const int code_size = (2+2*7)*BytesPerInstWord;
 587 
 588   // Allocate space for the code.
 589   ResourceMark rm;
 590   CodeBuffer cb("config_dscr", code_size, 0);
 591   MacroAssembler* a = new MacroAssembler(&cb);
 592 
 593   // Emit code.
 594   uint64_t (*get_dscr)() = (uint64_t(*)())(void *)a->function_entry();
 595   uint32_t *code = (uint32_t *)a->pc();
 596   a->mfdscr(R3);
 597   a->blr();
 598 
 599   void (*set_dscr)(long) = (void(*)(long))(void *)a->function_entry();
 600   a->mtdscr(R3);
 601   a->blr();
 602 
 603   uint32_t *code_end = (uint32_t *)a->pc();
 604   a->flush();
 605 
 606   // Print the detection code.
 607   if (PrintAssembly) {
 608     ttyLocker ttyl;
 609     tty->print_cr("Decoding dscr configuration stub at " INTPTR_FORMAT " before execution:", p2i(code));
 610     Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
 611   }
 612 
 613   // Apply the configuration if needed.
 614   _dscr_val = (*get_dscr)();
 615   if (Verbose) {
 616     tty->print_cr("dscr value was 0x%lx" , _dscr_val);
 617   }
 618   bool change_requested = false;
 619   if (DSCR_PPC64 != (uintx)-1) {
 620     _dscr_val = DSCR_PPC64;
 621     change_requested = true;
 622   }
 623   if (DSCR_DPFD_PPC64 <= 7) {
 624     uint64_t mask = 0x7;
 625     if ((_dscr_val & mask) != DSCR_DPFD_PPC64) {
 626       _dscr_val = (_dscr_val & ~mask) | (DSCR_DPFD_PPC64);
 627       change_requested = true;
 628     }
 629   }
 630   if (DSCR_URG_PPC64 <= 7) {
 631     uint64_t mask = 0x7 << 6;
 632     if ((_dscr_val & mask) != DSCR_DPFD_PPC64 << 6) {
 633       _dscr_val = (_dscr_val & ~mask) | (DSCR_URG_PPC64 << 6);
 634       change_requested = true;
 635     }
 636   }
 637   if (change_requested) {
 638     (*set_dscr)(_dscr_val);
 639     if (Verbose) {
 640       tty->print_cr("dscr was set to 0x%lx" , (*get_dscr)());
 641     }
 642   }
 643 }
 644 
 645 static int saved_features = 0;
 646 
 647 void VM_Version::allow_all() {
 648   saved_features = _features;
 649   _features      = all_features_m;
 650 }
 651 
 652 void VM_Version::revert() {
 653   _features = saved_features;
 654 }