1 /*
   2  * Copyright 1997-2008 Sun Microsystems, Inc.  All Rights Reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  20  * CA 95054 USA or visit www.sun.com if you need additional information or
  21  * have any questions.
  22  *
  23  */
  24 
  25 # include "incls/_precompiled.incl"
  26 # include "incls/_vm_version_x86_32.cpp.incl"
  27 
  28 
  29 int VM_Version::_cpu;
  30 int VM_Version::_model;
  31 int VM_Version::_stepping;
  32 int VM_Version::_cpuFeatures;
  33 const char*           VM_Version::_features_str = "";
  34 VM_Version::CpuidInfo VM_Version::_cpuid_info   = { 0, };
  35 
  36 static BufferBlob* stub_blob;
  37 static const int stub_size = 300;
  38 
  39 extern "C" {
  40   typedef void (*getPsrInfo_stub_t)(void*);
  41 }
  42 static getPsrInfo_stub_t getPsrInfo_stub = NULL;
  43 
  44 
  45 class VM_Version_StubGenerator: public StubCodeGenerator {
  46  public:
  47 
  48   VM_Version_StubGenerator(CodeBuffer *c) : StubCodeGenerator(c) {}
  49 
  50   address generate_getPsrInfo() {
  51     // Flags to test CPU type.
  52     const uint32_t EFL_AC           = 0x40000;
  53     const uint32_t EFL_ID           = 0x200000;
  54     // Values for when we don't have a CPUID instruction.
  55     const int      CPU_FAMILY_SHIFT = 8;
  56     const uint32_t CPU_FAMILY_386   = (3 << CPU_FAMILY_SHIFT);
  57     const uint32_t CPU_FAMILY_486   = (4 << CPU_FAMILY_SHIFT);
  58 
  59     Label detect_486, cpu486, detect_586, std_cpuid1;
  60     Label ext_cpuid1, ext_cpuid5, done;
  61 
  62     StubCodeMark mark(this, "VM_Version", "getPsrInfo_stub");
  63 #   define __ _masm->
  64 
  65     address start = __ pc();
  66 
  67     //
  68     // void getPsrInfo(VM_Version::CpuidInfo* cpuid_info);
  69     //
  70     __ push(rbp);
  71     __ movptr(rbp, Address(rsp, 8)); // cpuid_info address
  72     __ push(rbx);
  73     __ push(rsi);
  74     __ pushf();          // preserve rbx, and flags
  75     __ pop(rax);
  76     __ push(rax);
  77     __ mov(rcx, rax);
  78     //
  79     // if we are unable to change the AC flag, we have a 386
  80     //
  81     __ xorl(rax, EFL_AC);
  82     __ push(rax);
  83     __ popf();
  84     __ pushf();
  85     __ pop(rax);
  86     __ cmpptr(rax, rcx);
  87     __ jccb(Assembler::notEqual, detect_486);
  88 
  89     __ movl(rax, CPU_FAMILY_386);
  90     __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax);
  91     __ jmp(done);
  92 
  93     //
  94     // If we are unable to change the ID flag, we have a 486 which does
  95     // not support the "cpuid" instruction.
  96     //
  97     __ bind(detect_486);
  98     __ mov(rax, rcx);
  99     __ xorl(rax, EFL_ID);
 100     __ push(rax);
 101     __ popf();
 102     __ pushf();
 103     __ pop(rax);
 104     __ cmpptr(rcx, rax);
 105     __ jccb(Assembler::notEqual, detect_586);
 106 
 107     __ bind(cpu486);
 108     __ movl(rax, CPU_FAMILY_486);
 109     __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax);
 110     __ jmp(done);
 111 
 112     //
 113     // at this point, we have a chip which supports the "cpuid" instruction
 114     //
 115     __ bind(detect_586);
 116     __ xorptr(rax, rax);
 117     __ cpuid();
 118     __ orptr(rax, rax);
 119     __ jcc(Assembler::equal, cpu486);   // if cpuid doesn't support an input
 120                                         // value of at least 1, we give up and
 121                                         // assume a 486
 122     __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset())));
 123     __ movl(Address(rsi, 0), rax);
 124     __ movl(Address(rsi, 4), rbx);
 125     __ movl(Address(rsi, 8), rcx);
 126     __ movl(Address(rsi,12), rdx);
 127 
 128     __ cmpl(rax, 3);     // Is cpuid(0x4) supported?
 129     __ jccb(Assembler::belowEqual, std_cpuid1);
 130 
 131     //
 132     // cpuid(0x4) Deterministic cache params
 133     //
 134     __ movl(rax, 4);     // and rcx already set to 0x0
 135     __ xorl(rcx, rcx);
 136     __ cpuid();
 137     __ push(rax);
 138     __ andl(rax, 0x1f);  // Determine if valid cache parameters used
 139     __ orl(rax, rax);    // rax,[4:0] == 0 indicates invalid cache
 140     __ pop(rax);
 141     __ jccb(Assembler::equal, std_cpuid1);
 142 
 143     __ lea(rsi, Address(rbp, in_bytes(VM_Version::dcp_cpuid4_offset())));
 144     __ movl(Address(rsi, 0), rax);
 145     __ movl(Address(rsi, 4), rbx);
 146     __ movl(Address(rsi, 8), rcx);
 147     __ movl(Address(rsi,12), rdx);
 148 
 149     //
 150     // Standard cpuid(0x1)
 151     //
 152     __ bind(std_cpuid1);
 153     __ movl(rax, 1);
 154     __ cpuid();
 155     __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
 156     __ movl(Address(rsi, 0), rax);
 157     __ movl(Address(rsi, 4), rbx);
 158     __ movl(Address(rsi, 8), rcx);
 159     __ movl(Address(rsi,12), rdx);
 160 
 161     __ movl(rax, 0x80000000);
 162     __ cpuid();
 163     __ cmpl(rax, 0x80000000);     // Is cpuid(0x80000001) supported?
 164     __ jcc(Assembler::belowEqual, done);
 165     __ cmpl(rax, 0x80000004);     // Is cpuid(0x80000005) supported?
 166     __ jccb(Assembler::belowEqual, ext_cpuid1);
 167     __ cmpl(rax, 0x80000007);     // Is cpuid(0x80000008) supported?
 168     __ jccb(Assembler::belowEqual, ext_cpuid5);
 169     //
 170     // Extended cpuid(0x80000008)
 171     //
 172     __ movl(rax, 0x80000008);
 173     __ cpuid();
 174     __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid8_offset())));
 175     __ movl(Address(rsi, 0), rax);
 176     __ movl(Address(rsi, 4), rbx);
 177     __ movl(Address(rsi, 8), rcx);
 178     __ movl(Address(rsi,12), rdx);
 179 
 180     //
 181     // Extended cpuid(0x80000005)
 182     //
 183     __ bind(ext_cpuid5);
 184     __ movl(rax, 0x80000005);
 185     __ cpuid();
 186     __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid5_offset())));
 187     __ movl(Address(rsi, 0), rax);
 188     __ movl(Address(rsi, 4), rbx);
 189     __ movl(Address(rsi, 8), rcx);
 190     __ movl(Address(rsi,12), rdx);
 191 
 192     //
 193     // Extended cpuid(0x80000001)
 194     //
 195     __ bind(ext_cpuid1);
 196     __ movl(rax, 0x80000001);
 197     __ cpuid();
 198     __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid1_offset())));
 199     __ movl(Address(rsi, 0), rax);
 200     __ movl(Address(rsi, 4), rbx);
 201     __ movl(Address(rsi, 8), rcx);
 202     __ movl(Address(rsi,12), rdx);
 203 
 204     //
 205     // return
 206     //
 207     __ bind(done);
 208     __ popf();
 209     __ pop(rsi);
 210     __ pop(rbx);
 211     __ pop(rbp);
 212     __ ret(0);
 213 
 214 #   undef __
 215 
 216     return start;
 217   };
 218 };
 219 
 220 
 221 void VM_Version::get_processor_features() {
 222 
 223   _cpu = 4; // 486 by default
 224   _model = 0;
 225   _stepping = 0;
 226   _cpuFeatures = 0;
 227   _logical_processors_per_package = 1;
 228   if (!Use486InstrsOnly) {
 229     // Get raw processor info
 230     getPsrInfo_stub(&_cpuid_info);
 231     assert_is_initialized();
 232     _cpu = extended_cpu_family();
 233     _model = extended_cpu_model();
 234     _stepping = cpu_stepping();
 235     if (cpu_family() > 4) { // it supports CPUID
 236       _cpuFeatures = feature_flags();
 237       // Logical processors are only available on P4s and above,
 238       // and only if hyperthreading is available.
 239       _logical_processors_per_package = logical_processor_count();
 240     }
 241   }
 242   _supports_cx8 = supports_cmpxchg8();
 243   // if the OS doesn't support SSE, we can't use this feature even if the HW does
 244   if( !os::supports_sse())
 245     _cpuFeatures &= ~(CPU_SSE|CPU_SSE2|CPU_SSE3|CPU_SSSE3|CPU_SSE4A|CPU_SSE4_1|CPU_SSE4_2);
 246   if (UseSSE < 4)
 247     _cpuFeatures &= ~CPU_SSE4_1;
 248     _cpuFeatures &= ~CPU_SSE4_2;
 249   if (UseSSE < 3) {
 250     _cpuFeatures &= ~CPU_SSE3;
 251     _cpuFeatures &= ~CPU_SSSE3;
 252     _cpuFeatures &= ~CPU_SSE4A;
 253   }
 254   if (UseSSE < 2)
 255     _cpuFeatures &= ~CPU_SSE2;
 256   if (UseSSE < 1)
 257     _cpuFeatures &= ~CPU_SSE;
 258 
 259   if (logical_processors_per_package() == 1) {
 260     // HT processor could be installed on a system which doesn't support HT.
 261     _cpuFeatures &= ~CPU_HT;
 262   }
 263 
 264   char buf[256];
 265   jio_snprintf(buf, sizeof(buf), "(%u cores per cpu, %u threads per core) family %d model %d stepping %d%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s",
 266                cores_per_cpu(), threads_per_core(),
 267                cpu_family(), _model, _stepping,
 268                (supports_cmov() ? ", cmov" : ""),
 269                (supports_cmpxchg8() ? ", cx8" : ""),
 270                (supports_fxsr() ? ", fxsr" : ""),
 271                (supports_mmx()  ? ", mmx"  : ""),
 272                (supports_sse()  ? ", sse"  : ""),
 273                (supports_sse2() ? ", sse2" : ""),
 274                (supports_sse3() ? ", sse3" : ""),
 275                (supports_ssse3()? ", ssse3": ""),
 276                (supports_sse4_1() ? ", sse4.1" : ""),
 277                (supports_sse4_2() ? ", sse4.2" : ""),
 278                (supports_mmx_ext() ? ", mmxext" : ""),
 279                (supports_3dnow()   ? ", 3dnow"  : ""),
 280                (supports_3dnow2()  ? ", 3dnowext" : ""),
 281                (supports_sse4a()   ? ", sse4a": ""),
 282                (supports_ht() ? ", ht": ""));
 283   _features_str = strdup(buf);
 284 
 285   // UseSSE is set to the smaller of what hardware supports and what
 286   // the command line requires.  I.e., you cannot set UseSSE to 2 on
 287   // older Pentiums which do not support it.
 288   if( UseSSE > 4 ) UseSSE=4;
 289   if( UseSSE < 0 ) UseSSE=0;
 290   if( !supports_sse4_1() ) // Drop to 3 if no SSE4 support
 291     UseSSE = MIN2((intx)3,UseSSE);
 292   if( !supports_sse3() ) // Drop to 2 if no SSE3 support
 293     UseSSE = MIN2((intx)2,UseSSE);
 294   if( !supports_sse2() ) // Drop to 1 if no SSE2 support
 295     UseSSE = MIN2((intx)1,UseSSE);
 296   if( !supports_sse () ) // Drop to 0 if no SSE  support
 297     UseSSE = 0;
 298 
 299   // On new cpus instructions which update whole XMM register should be used
 300   // to prevent partial register stall due to dependencies on high half.
 301   //
 302   // UseXmmLoadAndClearUpper == true  --> movsd(xmm, mem)
 303   // UseXmmLoadAndClearUpper == false --> movlpd(xmm, mem)
 304   // UseXmmRegToRegMoveAll == true  --> movaps(xmm, xmm), movapd(xmm, xmm).
 305   // UseXmmRegToRegMoveAll == false --> movss(xmm, xmm),  movsd(xmm, xmm).
 306 
 307   if( is_amd() ) { // AMD cpus specific settings
 308     if( supports_sse2() && FLAG_IS_DEFAULT(UseAddressNop) ) {
 309       // Use it on new AMD cpus starting from Opteron.
 310       UseAddressNop = true;
 311     }
 312     if( supports_sse2() && FLAG_IS_DEFAULT(UseNewLongLShift) ) {
 313       // Use it on new AMD cpus starting from Opteron.
 314       UseNewLongLShift = true;
 315     }
 316     if( FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper) ) {
 317       if( supports_sse4a() ) {
 318         UseXmmLoadAndClearUpper = true; // use movsd only on '10h' Opteron
 319       } else {
 320         UseXmmLoadAndClearUpper = false;
 321       }
 322     }
 323     if( FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll) ) {
 324       if( supports_sse4a() ) {
 325         UseXmmRegToRegMoveAll = true; // use movaps, movapd only on '10h'
 326       } else {
 327         UseXmmRegToRegMoveAll = false;
 328       }
 329     }
 330     if( FLAG_IS_DEFAULT(UseXmmI2F) ) {
 331       if( supports_sse4a() ) {
 332         UseXmmI2F = true;
 333       } else {
 334         UseXmmI2F = false;
 335       }
 336     }
 337     if( FLAG_IS_DEFAULT(UseXmmI2D) ) {
 338       if( supports_sse4a() ) {
 339         UseXmmI2D = true;
 340       } else {
 341         UseXmmI2D = false;
 342       }
 343     }
 344   }
 345 
 346   if( is_intel() ) { // Intel cpus specific settings
 347     if( FLAG_IS_DEFAULT(UseStoreImmI16) ) {
 348       UseStoreImmI16 = false; // don't use it on Intel cpus
 349     }
 350     if( cpu_family() == 6 || cpu_family() == 15 ) {
 351       if( FLAG_IS_DEFAULT(UseAddressNop) ) {
 352         // Use it on all Intel cpus starting from PentiumPro
 353         UseAddressNop = true;
 354       }
 355     }
 356     if( FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper) ) {
 357       UseXmmLoadAndClearUpper = true; // use movsd on all Intel cpus
 358     }
 359     if( FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll) ) {
 360       if( supports_sse3() ) {
 361         UseXmmRegToRegMoveAll = true; // use movaps, movapd on new Intel cpus
 362       } else {
 363         UseXmmRegToRegMoveAll = false;
 364       }
 365     }
 366     if( cpu_family() == 6 && supports_sse3() ) { // New Intel cpus
 367 #ifdef COMPILER2
 368       if( FLAG_IS_DEFAULT(MaxLoopPad) ) {
 369         // For new Intel cpus do the next optimization:
 370         // don't align the beginning of a loop if there are enough instructions
 371         // left (NumberOfLoopInstrToAlign defined in c2_globals.hpp)
 372         // in current fetch line (OptoLoopAlignment) or the padding
 373         // is big (> MaxLoopPad).
 374         // Set MaxLoopPad to 11 for new Intel cpus to reduce number of
 375         // generated NOP instructions. 11 is the largest size of one
 376         // address NOP instruction '0F 1F' (see Assembler::nop(i)).
 377         MaxLoopPad = 11;
 378       }
 379 #endif // COMPILER2
 380       if( FLAG_IS_DEFAULT(UseXMMForArrayCopy) ) {
 381         UseXMMForArrayCopy = true; // use SSE2 movq on new Intel cpus
 382       }
 383       if( supports_sse4_1() && supports_ht() ) { // Newest Intel cpus
 384         if( FLAG_IS_DEFAULT(UseUnalignedLoadStores) && UseXMMForArrayCopy ) {
 385           UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus
 386         }
 387       }
 388     }
 389   }
 390 
 391   assert(0 <= ReadPrefetchInstr && ReadPrefetchInstr <= 3, "invalid value");
 392   assert(0 <= AllocatePrefetchInstr && AllocatePrefetchInstr <= 3, "invalid value");
 393 
 394   // set valid Prefetch instruction
 395   if( ReadPrefetchInstr < 0 ) ReadPrefetchInstr = 0;
 396   if( ReadPrefetchInstr > 3 ) ReadPrefetchInstr = 3;
 397   if( ReadPrefetchInstr == 3 && !supports_3dnow() ) ReadPrefetchInstr = 0;
 398   if( !supports_sse() && supports_3dnow() ) ReadPrefetchInstr = 3;
 399 
 400   if( AllocatePrefetchInstr < 0 ) AllocatePrefetchInstr = 0;
 401   if( AllocatePrefetchInstr > 3 ) AllocatePrefetchInstr = 3;
 402   if( AllocatePrefetchInstr == 3 && !supports_3dnow() ) AllocatePrefetchInstr=0;
 403   if( !supports_sse() && supports_3dnow() ) AllocatePrefetchInstr = 3;
 404 
 405   // Allocation prefetch settings
 406   intx cache_line_size = L1_data_cache_line_size();
 407   if( cache_line_size > AllocatePrefetchStepSize )
 408     AllocatePrefetchStepSize = cache_line_size;
 409   if( FLAG_IS_DEFAULT(AllocatePrefetchLines) )
 410     AllocatePrefetchLines = 3; // Optimistic value
 411   assert(AllocatePrefetchLines > 0, "invalid value");
 412   if( AllocatePrefetchLines < 1 ) // set valid value in product VM
 413     AllocatePrefetchLines = 1; // Conservative value
 414 
 415   AllocatePrefetchDistance = allocate_prefetch_distance();
 416   AllocatePrefetchStyle    = allocate_prefetch_style();
 417 
 418   if( AllocatePrefetchStyle == 2 && is_intel() &&
 419       cpu_family() == 6 && supports_sse3() ) { // watermark prefetching on Core
 420     AllocatePrefetchDistance = 320;
 421   }
 422   assert(AllocatePrefetchDistance % AllocatePrefetchStepSize == 0, "invalid value");
 423 
 424 #ifndef PRODUCT
 425   if (PrintMiscellaneous && Verbose) {
 426     tty->print_cr("Logical CPUs per core: %u",
 427                   logical_processors_per_package());
 428     tty->print_cr("UseSSE=%d",UseSSE);
 429     tty->print("Allocation: ");
 430     if (AllocatePrefetchStyle <= 0 || UseSSE == 0 && !supports_3dnow()) {
 431       tty->print_cr("no prefetching");
 432     } else {
 433       if (UseSSE == 0 && supports_3dnow()) {
 434         tty->print("PREFETCHW");
 435       } else if (UseSSE >= 1) {
 436         if (AllocatePrefetchInstr == 0) {
 437           tty->print("PREFETCHNTA");
 438         } else if (AllocatePrefetchInstr == 1) {
 439           tty->print("PREFETCHT0");
 440         } else if (AllocatePrefetchInstr == 2) {
 441           tty->print("PREFETCHT2");
 442         } else if (AllocatePrefetchInstr == 3) {
 443           tty->print("PREFETCHW");
 444         }
 445       }
 446       if (AllocatePrefetchLines > 1) {
 447         tty->print_cr(" %d, %d lines with step %d bytes", AllocatePrefetchDistance, AllocatePrefetchLines, AllocatePrefetchStepSize);
 448       } else {
 449         tty->print_cr(" %d, one line", AllocatePrefetchDistance);
 450       }
 451     }
 452   }
 453 #endif // !PRODUCT
 454 }
 455 
 456 void VM_Version::initialize() {
 457   ResourceMark rm;
 458   // Making this stub must be FIRST use of assembler
 459 
 460   stub_blob = BufferBlob::create("getPsrInfo_stub", stub_size);
 461   if (stub_blob == NULL) {
 462     vm_exit_during_initialization("Unable to allocate getPsrInfo_stub");
 463   }
 464   CodeBuffer c(stub_blob->instructions_begin(),
 465                stub_blob->instructions_size());
 466   VM_Version_StubGenerator g(&c);
 467   getPsrInfo_stub = CAST_TO_FN_PTR(getPsrInfo_stub_t,
 468                                    g.generate_getPsrInfo());
 469 
 470   get_processor_features();
 471 }