Mercurial > jdk8-mips64-public > hotspot / file revision
src/cpu/x86/vm/vm_version_x86.cpp@0118c8c7b80f
src/cpu/x86/vm/vm_version_x86.cpp
Mon, 31 Mar 2014 13:08:03 -0700
- author
- kvn
- date
- Mon, 31 Mar 2014 13:08:03 -0700
- changeset 6537
- 0118c8c7b80f
- parent 6429
- 606acabe7b5c
- child 6656
- 1eba0601f0dd
- permissions
- -rw-r--r--
8038633: crash in VM_Version::get_processor_features() on startup
Summary: Windows need an exception wrapper around getPsrInfo_stub() call in order to properly handle SEGV for YMM registers test.
Reviewed-by: iveresov, iignatyev
1 /*
2 * Copyright (c) 1997, 2013, Oracle and/or its affiliates. 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 Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "memory/resourceArea.hpp"
29 #include "runtime/java.hpp"
30 #include "runtime/stubCodeGenerator.hpp"
31 #include "vm_version_x86.hpp"
32 #ifdef TARGET_OS_FAMILY_linux
33 # include "os_linux.inline.hpp"
34 #endif
35 #ifdef TARGET_OS_FAMILY_solaris
36 # include "os_solaris.inline.hpp"
37 #endif
38 #ifdef TARGET_OS_FAMILY_windows
39 # include "os_windows.inline.hpp"
40 #endif
41 #ifdef TARGET_OS_FAMILY_bsd
42 # include "os_bsd.inline.hpp"
43 #endif
46 int VM_Version::_cpu;
47 int VM_Version::_model;
48 int VM_Version::_stepping;
49 int VM_Version::_cpuFeatures;
50 const char* VM_Version::_features_str = "";
51 VM_Version::CpuidInfo VM_Version::_cpuid_info = { 0, };
53 // Address of instruction which causes SEGV
54 address VM_Version::_cpuinfo_segv_addr = 0;
55 // Address of instruction after the one which causes SEGV
56 address VM_Version::_cpuinfo_cont_addr = 0;
58 static BufferBlob* stub_blob;
59 static const int stub_size = 600;
61 extern "C" {
62 typedef void (*get_cpu_info_stub_t)(void*);
63 }
64 static get_cpu_info_stub_t get_cpu_info_stub = NULL;
67 class VM_Version_StubGenerator: public StubCodeGenerator {
68 public:
70 VM_Version_StubGenerator(CodeBuffer *c) : StubCodeGenerator(c) {}
72 address generate_get_cpu_info() {
73 // Flags to test CPU type.
74 const uint32_t HS_EFL_AC = 0x40000;
75 const uint32_t HS_EFL_ID = 0x200000;
76 // Values for when we don't have a CPUID instruction.
77 const int CPU_FAMILY_SHIFT = 8;
78 const uint32_t CPU_FAMILY_386 = (3 << CPU_FAMILY_SHIFT);
79 const uint32_t CPU_FAMILY_486 = (4 << CPU_FAMILY_SHIFT);
81 Label detect_486, cpu486, detect_586, std_cpuid1, std_cpuid4;
82 Label sef_cpuid, ext_cpuid, ext_cpuid1, ext_cpuid5, ext_cpuid7, done;
84 StubCodeMark mark(this, "VM_Version", "get_cpu_info_stub");
85 # define __ _masm->
87 address start = __ pc();
89 //
90 // void get_cpu_info(VM_Version::CpuidInfo* cpuid_info);
91 //
92 // LP64: rcx and rdx are first and second argument registers on windows
94 __ push(rbp);
95 #ifdef _LP64
96 __ mov(rbp, c_rarg0); // cpuid_info address
97 #else
98 __ movptr(rbp, Address(rsp, 8)); // cpuid_info address
99 #endif
100 __ push(rbx);
101 __ push(rsi);
102 __ pushf(); // preserve rbx, and flags
103 __ pop(rax);
104 __ push(rax);
105 __ mov(rcx, rax);
106 //
107 // if we are unable to change the AC flag, we have a 386
108 //
109 __ xorl(rax, HS_EFL_AC);
110 __ push(rax);
111 __ popf();
112 __ pushf();
113 __ pop(rax);
114 __ cmpptr(rax, rcx);
115 __ jccb(Assembler::notEqual, detect_486);
117 __ movl(rax, CPU_FAMILY_386);
118 __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax);
119 __ jmp(done);
121 //
122 // If we are unable to change the ID flag, we have a 486 which does
123 // not support the "cpuid" instruction.
124 //
125 __ bind(detect_486);
126 __ mov(rax, rcx);
127 __ xorl(rax, HS_EFL_ID);
128 __ push(rax);
129 __ popf();
130 __ pushf();
131 __ pop(rax);
132 __ cmpptr(rcx, rax);
133 __ jccb(Assembler::notEqual, detect_586);
135 __ bind(cpu486);
136 __ movl(rax, CPU_FAMILY_486);
137 __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax);
138 __ jmp(done);
140 //
141 // At this point, we have a chip which supports the "cpuid" instruction
142 //
143 __ bind(detect_586);
144 __ xorl(rax, rax);
145 __ cpuid();
146 __ orl(rax, rax);
147 __ jcc(Assembler::equal, cpu486); // if cpuid doesn't support an input
148 // value of at least 1, we give up and
149 // assume a 486
150 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset())));
151 __ movl(Address(rsi, 0), rax);
152 __ movl(Address(rsi, 4), rbx);
153 __ movl(Address(rsi, 8), rcx);
154 __ movl(Address(rsi,12), rdx);
156 __ cmpl(rax, 0xa); // Is cpuid(0xB) supported?
157 __ jccb(Assembler::belowEqual, std_cpuid4);
159 //
160 // cpuid(0xB) Processor Topology
161 //
162 __ movl(rax, 0xb);
163 __ xorl(rcx, rcx); // Threads level
164 __ cpuid();
166 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB0_offset())));
167 __ movl(Address(rsi, 0), rax);
168 __ movl(Address(rsi, 4), rbx);
169 __ movl(Address(rsi, 8), rcx);
170 __ movl(Address(rsi,12), rdx);
172 __ movl(rax, 0xb);
173 __ movl(rcx, 1); // Cores level
174 __ cpuid();
175 __ push(rax);
176 __ andl(rax, 0x1f); // Determine if valid topology level
177 __ orl(rax, rbx); // eax[4:0] | ebx[0:15] == 0 indicates invalid level
178 __ andl(rax, 0xffff);
179 __ pop(rax);
180 __ jccb(Assembler::equal, std_cpuid4);
182 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB1_offset())));
183 __ movl(Address(rsi, 0), rax);
184 __ movl(Address(rsi, 4), rbx);
185 __ movl(Address(rsi, 8), rcx);
186 __ movl(Address(rsi,12), rdx);
188 __ movl(rax, 0xb);
189 __ movl(rcx, 2); // Packages level
190 __ cpuid();
191 __ push(rax);
192 __ andl(rax, 0x1f); // Determine if valid topology level
193 __ orl(rax, rbx); // eax[4:0] | ebx[0:15] == 0 indicates invalid level
194 __ andl(rax, 0xffff);
195 __ pop(rax);
196 __ jccb(Assembler::equal, std_cpuid4);
198 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB2_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);
204 //
205 // cpuid(0x4) Deterministic cache params
206 //
207 __ bind(std_cpuid4);
208 __ movl(rax, 4);
209 __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x4) supported?
210 __ jccb(Assembler::greater, std_cpuid1);
212 __ xorl(rcx, rcx); // L1 cache
213 __ cpuid();
214 __ push(rax);
215 __ andl(rax, 0x1f); // Determine if valid cache parameters used
216 __ orl(rax, rax); // eax[4:0] == 0 indicates invalid cache
217 __ pop(rax);
218 __ jccb(Assembler::equal, std_cpuid1);
220 __ lea(rsi, Address(rbp, in_bytes(VM_Version::dcp_cpuid4_offset())));
221 __ movl(Address(rsi, 0), rax);
222 __ movl(Address(rsi, 4), rbx);
223 __ movl(Address(rsi, 8), rcx);
224 __ movl(Address(rsi,12), rdx);
226 //
227 // Standard cpuid(0x1)
228 //
229 __ bind(std_cpuid1);
230 __ movl(rax, 1);
231 __ cpuid();
232 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
233 __ movl(Address(rsi, 0), rax);
234 __ movl(Address(rsi, 4), rbx);
235 __ movl(Address(rsi, 8), rcx);
236 __ movl(Address(rsi,12), rdx);
238 //
239 // Check if OS has enabled XGETBV instruction to access XCR0
240 // (OSXSAVE feature flag) and CPU supports AVX
241 //
242 __ andl(rcx, 0x18000000); // cpuid1 bits osxsave | avx
243 __ cmpl(rcx, 0x18000000);
244 __ jccb(Assembler::notEqual, sef_cpuid); // jump if AVX is not supported
246 //
247 // XCR0, XFEATURE_ENABLED_MASK register
248 //
249 __ xorl(rcx, rcx); // zero for XCR0 register
250 __ xgetbv();
251 __ lea(rsi, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset())));
252 __ movl(Address(rsi, 0), rax);
253 __ movl(Address(rsi, 4), rdx);
255 __ andl(rax, 0x6); // xcr0 bits sse | ymm
256 __ cmpl(rax, 0x6);
257 __ jccb(Assembler::notEqual, sef_cpuid); // jump if AVX is not supported
259 //
260 // Some OSs have a bug when upper 128bits of YMM
261 // registers are not restored after a signal processing.
262 // Generate SEGV here (reference through NULL)
263 // and check upper YMM bits after it.
264 //
265 VM_Version::set_avx_cpuFeatures(); // Enable temporary to pass asserts
267 // load value into all 32 bytes of ymm7 register
268 __ movl(rcx, VM_Version::ymm_test_value());
270 __ movdl(xmm0, rcx);
271 __ pshufd(xmm0, xmm0, 0x00);
272 __ vinsertf128h(xmm0, xmm0, xmm0);
273 __ vmovdqu(xmm7, xmm0);
274 #ifdef _LP64
275 __ vmovdqu(xmm8, xmm0);
276 __ vmovdqu(xmm15, xmm0);
277 #endif
279 __ xorl(rsi, rsi);
280 VM_Version::set_cpuinfo_segv_addr( __ pc() );
281 // Generate SEGV
282 __ movl(rax, Address(rsi, 0));
284 VM_Version::set_cpuinfo_cont_addr( __ pc() );
285 // Returns here after signal. Save xmm0 to check it later.
286 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ymm_save_offset())));
287 __ vmovdqu(Address(rsi, 0), xmm0);
288 __ vmovdqu(Address(rsi, 32), xmm7);
289 #ifdef _LP64
290 __ vmovdqu(Address(rsi, 64), xmm8);
291 __ vmovdqu(Address(rsi, 96), xmm15);
292 #endif
294 VM_Version::clean_cpuFeatures();
296 //
297 // cpuid(0x7) Structured Extended Features
298 //
299 __ bind(sef_cpuid);
300 __ movl(rax, 7);
301 __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x7) supported?
302 __ jccb(Assembler::greater, ext_cpuid);
304 __ xorl(rcx, rcx);
305 __ cpuid();
306 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset())));
307 __ movl(Address(rsi, 0), rax);
308 __ movl(Address(rsi, 4), rbx);
310 //
311 // Extended cpuid(0x80000000)
312 //
313 __ bind(ext_cpuid);
314 __ movl(rax, 0x80000000);
315 __ cpuid();
316 __ cmpl(rax, 0x80000000); // Is cpuid(0x80000001) supported?
317 __ jcc(Assembler::belowEqual, done);
318 __ cmpl(rax, 0x80000004); // Is cpuid(0x80000005) supported?
319 __ jccb(Assembler::belowEqual, ext_cpuid1);
320 __ cmpl(rax, 0x80000006); // Is cpuid(0x80000007) supported?
321 __ jccb(Assembler::belowEqual, ext_cpuid5);
322 __ cmpl(rax, 0x80000007); // Is cpuid(0x80000008) supported?
323 __ jccb(Assembler::belowEqual, ext_cpuid7);
324 //
325 // Extended cpuid(0x80000008)
326 //
327 __ movl(rax, 0x80000008);
328 __ cpuid();
329 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid8_offset())));
330 __ movl(Address(rsi, 0), rax);
331 __ movl(Address(rsi, 4), rbx);
332 __ movl(Address(rsi, 8), rcx);
333 __ movl(Address(rsi,12), rdx);
335 //
336 // Extended cpuid(0x80000007)
337 //
338 __ bind(ext_cpuid7);
339 __ movl(rax, 0x80000007);
340 __ cpuid();
341 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid7_offset())));
342 __ movl(Address(rsi, 0), rax);
343 __ movl(Address(rsi, 4), rbx);
344 __ movl(Address(rsi, 8), rcx);
345 __ movl(Address(rsi,12), rdx);
347 //
348 // Extended cpuid(0x80000005)
349 //
350 __ bind(ext_cpuid5);
351 __ movl(rax, 0x80000005);
352 __ cpuid();
353 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid5_offset())));
354 __ movl(Address(rsi, 0), rax);
355 __ movl(Address(rsi, 4), rbx);
356 __ movl(Address(rsi, 8), rcx);
357 __ movl(Address(rsi,12), rdx);
359 //
360 // Extended cpuid(0x80000001)
361 //
362 __ bind(ext_cpuid1);
363 __ movl(rax, 0x80000001);
364 __ cpuid();
365 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid1_offset())));
366 __ movl(Address(rsi, 0), rax);
367 __ movl(Address(rsi, 4), rbx);
368 __ movl(Address(rsi, 8), rcx);
369 __ movl(Address(rsi,12), rdx);
371 //
372 // return
373 //
374 __ bind(done);
375 __ popf();
376 __ pop(rsi);
377 __ pop(rbx);
378 __ pop(rbp);
379 __ ret(0);
381 # undef __
383 return start;
384 };
385 };
388 void VM_Version::get_cpu_info_wrapper() {
389 get_cpu_info_stub(&_cpuid_info);
390 }
392 #ifndef CALL_TEST_FUNC_WITH_WRAPPER_IF_NEEDED
393 #define CALL_TEST_FUNC_WITH_WRAPPER_IF_NEEDED(f) f()
394 #endif
396 void VM_Version::get_processor_features() {
398 _cpu = 4; // 486 by default
399 _model = 0;
400 _stepping = 0;
401 _cpuFeatures = 0;
402 _logical_processors_per_package = 1;
404 if (!Use486InstrsOnly) {
405 // Get raw processor info
407 // Some platforms (like Win*) need a wrapper around here
408 // in order to properly handle SEGV for YMM registers test.
409 CALL_TEST_FUNC_WITH_WRAPPER_IF_NEEDED(get_cpu_info_wrapper);
411 assert_is_initialized();
412 _cpu = extended_cpu_family();
413 _model = extended_cpu_model();
414 _stepping = cpu_stepping();
416 if (cpu_family() > 4) { // it supports CPUID
417 _cpuFeatures = feature_flags();
418 // Logical processors are only available on P4s and above,
419 // and only if hyperthreading is available.
420 _logical_processors_per_package = logical_processor_count();
421 }
422 }
424 _supports_cx8 = supports_cmpxchg8();
425 // xchg and xadd instructions
426 _supports_atomic_getset4 = true;
427 _supports_atomic_getadd4 = true;
428 LP64_ONLY(_supports_atomic_getset8 = true);
429 LP64_ONLY(_supports_atomic_getadd8 = true);
431 #ifdef _LP64
432 // OS should support SSE for x64 and hardware should support at least SSE2.
433 if (!VM_Version::supports_sse2()) {
434 vm_exit_during_initialization("Unknown x64 processor: SSE2 not supported");
435 }
436 // in 64 bit the use of SSE2 is the minimum
437 if (UseSSE < 2) UseSSE = 2;
438 #endif
440 #ifdef AMD64
441 // flush_icache_stub have to be generated first.
442 // That is why Icache line size is hard coded in ICache class,
443 // see icache_x86.hpp. It is also the reason why we can't use
444 // clflush instruction in 32-bit VM since it could be running
445 // on CPU which does not support it.
446 //
447 // The only thing we can do is to verify that flushed
448 // ICache::line_size has correct value.
449 guarantee(_cpuid_info.std_cpuid1_edx.bits.clflush != 0, "clflush is not supported");
450 // clflush_size is size in quadwords (8 bytes).
451 guarantee(_cpuid_info.std_cpuid1_ebx.bits.clflush_size == 8, "such clflush size is not supported");
452 #endif
454 // If the OS doesn't support SSE, we can't use this feature even if the HW does
455 if (!os::supports_sse())
456 _cpuFeatures &= ~(CPU_SSE|CPU_SSE2|CPU_SSE3|CPU_SSSE3|CPU_SSE4A|CPU_SSE4_1|CPU_SSE4_2);
458 if (UseSSE < 4) {
459 _cpuFeatures &= ~CPU_SSE4_1;
460 _cpuFeatures &= ~CPU_SSE4_2;
461 }
463 if (UseSSE < 3) {
464 _cpuFeatures &= ~CPU_SSE3;
465 _cpuFeatures &= ~CPU_SSSE3;
466 _cpuFeatures &= ~CPU_SSE4A;
467 }
469 if (UseSSE < 2)
470 _cpuFeatures &= ~CPU_SSE2;
472 if (UseSSE < 1)
473 _cpuFeatures &= ~CPU_SSE;
475 if (UseAVX < 2)
476 _cpuFeatures &= ~CPU_AVX2;
478 if (UseAVX < 1)
479 _cpuFeatures &= ~CPU_AVX;
481 if (!UseAES && !FLAG_IS_DEFAULT(UseAES))
482 _cpuFeatures &= ~CPU_AES;
484 if (logical_processors_per_package() == 1) {
485 // HT processor could be installed on a system which doesn't support HT.
486 _cpuFeatures &= ~CPU_HT;
487 }
489 char buf[256];
490 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%s%s%s%s%s%s%s%s%s%s%s%s",
491 cores_per_cpu(), threads_per_core(),
492 cpu_family(), _model, _stepping,
493 (supports_cmov() ? ", cmov" : ""),
494 (supports_cmpxchg8() ? ", cx8" : ""),
495 (supports_fxsr() ? ", fxsr" : ""),
496 (supports_mmx() ? ", mmx" : ""),
497 (supports_sse() ? ", sse" : ""),
498 (supports_sse2() ? ", sse2" : ""),
499 (supports_sse3() ? ", sse3" : ""),
500 (supports_ssse3()? ", ssse3": ""),
501 (supports_sse4_1() ? ", sse4.1" : ""),
502 (supports_sse4_2() ? ", sse4.2" : ""),
503 (supports_popcnt() ? ", popcnt" : ""),
504 (supports_avx() ? ", avx" : ""),
505 (supports_avx2() ? ", avx2" : ""),
506 (supports_aes() ? ", aes" : ""),
507 (supports_clmul() ? ", clmul" : ""),
508 (supports_erms() ? ", erms" : ""),
509 (supports_rtm() ? ", rtm" : ""),
510 (supports_mmx_ext() ? ", mmxext" : ""),
511 (supports_3dnow_prefetch() ? ", 3dnowpref" : ""),
512 (supports_lzcnt() ? ", lzcnt": ""),
513 (supports_sse4a() ? ", sse4a": ""),
514 (supports_ht() ? ", ht": ""),
515 (supports_tsc() ? ", tsc": ""),
516 (supports_tscinv_bit() ? ", tscinvbit": ""),
517 (supports_tscinv() ? ", tscinv": ""),
518 (supports_bmi1() ? ", bmi1" : ""),
519 (supports_bmi2() ? ", bmi2" : ""));
520 _features_str = strdup(buf);
522 // UseSSE is set to the smaller of what hardware supports and what
523 // the command line requires. I.e., you cannot set UseSSE to 2 on
524 // older Pentiums which do not support it.
525 if (UseSSE > 4) UseSSE=4;
526 if (UseSSE < 0) UseSSE=0;
527 if (!supports_sse4_1()) // Drop to 3 if no SSE4 support
528 UseSSE = MIN2((intx)3,UseSSE);
529 if (!supports_sse3()) // Drop to 2 if no SSE3 support
530 UseSSE = MIN2((intx)2,UseSSE);
531 if (!supports_sse2()) // Drop to 1 if no SSE2 support
532 UseSSE = MIN2((intx)1,UseSSE);
533 if (!supports_sse ()) // Drop to 0 if no SSE support
534 UseSSE = 0;
536 if (UseAVX > 2) UseAVX=2;
537 if (UseAVX < 0) UseAVX=0;
538 if (!supports_avx2()) // Drop to 1 if no AVX2 support
539 UseAVX = MIN2((intx)1,UseAVX);
540 if (!supports_avx ()) // Drop to 0 if no AVX support
541 UseAVX = 0;
543 // Use AES instructions if available.
544 if (supports_aes()) {
545 if (FLAG_IS_DEFAULT(UseAES)) {
546 UseAES = true;
547 }
548 } else if (UseAES) {
549 if (!FLAG_IS_DEFAULT(UseAES))
550 warning("AES instructions are not available on this CPU");
551 FLAG_SET_DEFAULT(UseAES, false);
552 }
554 // Use CLMUL instructions if available.
555 if (supports_clmul()) {
556 if (FLAG_IS_DEFAULT(UseCLMUL)) {
557 UseCLMUL = true;
558 }
559 } else if (UseCLMUL) {
560 if (!FLAG_IS_DEFAULT(UseCLMUL))
561 warning("CLMUL instructions not available on this CPU (AVX may also be required)");
562 FLAG_SET_DEFAULT(UseCLMUL, false);
563 }
565 if (UseCLMUL && (UseAVX > 0) && (UseSSE > 2)) {
566 if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) {
567 UseCRC32Intrinsics = true;
568 }
569 } else if (UseCRC32Intrinsics) {
570 if (!FLAG_IS_DEFAULT(UseCRC32Intrinsics))
571 warning("CRC32 Intrinsics requires AVX and CLMUL instructions (not available on this CPU)");
572 FLAG_SET_DEFAULT(UseCRC32Intrinsics, false);
573 }
575 // The AES intrinsic stubs require AES instruction support (of course)
576 // but also require sse3 mode for instructions it use.
577 if (UseAES && (UseSSE > 2)) {
578 if (FLAG_IS_DEFAULT(UseAESIntrinsics)) {
579 UseAESIntrinsics = true;
580 }
581 } else if (UseAESIntrinsics) {
582 if (!FLAG_IS_DEFAULT(UseAESIntrinsics))
583 warning("AES intrinsics are not available on this CPU");
584 FLAG_SET_DEFAULT(UseAESIntrinsics, false);
585 }
587 // Adjust RTM (Restricted Transactional Memory) flags
588 if (!supports_rtm() && UseRTMLocking) {
589 // Can't continue because UseRTMLocking affects UseBiasedLocking flag
590 // setting during arguments processing. See use_biased_locking().
591 // VM_Version_init() is executed after UseBiasedLocking is used
592 // in Thread::allocate().
593 vm_exit_during_initialization("RTM instructions are not available on this CPU");
594 }
596 #if INCLUDE_RTM_OPT
597 if (UseRTMLocking) {
598 if (!FLAG_IS_CMDLINE(UseRTMLocking)) {
599 // RTM locking should be used only for applications with
600 // high lock contention. For now we do not use it by default.
601 vm_exit_during_initialization("UseRTMLocking flag should be only set on command line");
602 }
603 if (!is_power_of_2(RTMTotalCountIncrRate)) {
604 warning("RTMTotalCountIncrRate must be a power of 2, resetting it to 64");
605 FLAG_SET_DEFAULT(RTMTotalCountIncrRate, 64);
606 }
607 if (RTMAbortRatio < 0 || RTMAbortRatio > 100) {
608 warning("RTMAbortRatio must be in the range 0 to 100, resetting it to 50");
609 FLAG_SET_DEFAULT(RTMAbortRatio, 50);
610 }
611 } else { // !UseRTMLocking
612 if (UseRTMForStackLocks) {
613 if (!FLAG_IS_DEFAULT(UseRTMForStackLocks)) {
614 warning("UseRTMForStackLocks flag should be off when UseRTMLocking flag is off");
615 }
616 FLAG_SET_DEFAULT(UseRTMForStackLocks, false);
617 }
618 if (UseRTMDeopt) {
619 FLAG_SET_DEFAULT(UseRTMDeopt, false);
620 }
621 if (PrintPreciseRTMLockingStatistics) {
622 FLAG_SET_DEFAULT(PrintPreciseRTMLockingStatistics, false);
623 }
624 }
625 #else
626 if (UseRTMLocking) {
627 // Only C2 does RTM locking optimization.
628 // Can't continue because UseRTMLocking affects UseBiasedLocking flag
629 // setting during arguments processing. See use_biased_locking().
630 vm_exit_during_initialization("RTM locking optimization is not supported in this VM");
631 }
632 #endif
634 #ifdef COMPILER2
635 if (UseFPUForSpilling) {
636 if (UseSSE < 2) {
637 // Only supported with SSE2+
638 FLAG_SET_DEFAULT(UseFPUForSpilling, false);
639 }
640 }
641 if (MaxVectorSize > 0) {
642 if (!is_power_of_2(MaxVectorSize)) {
643 warning("MaxVectorSize must be a power of 2");
644 FLAG_SET_DEFAULT(MaxVectorSize, 32);
645 }
646 if (MaxVectorSize > 32) {
647 FLAG_SET_DEFAULT(MaxVectorSize, 32);
648 }
649 if (MaxVectorSize > 16 && (UseAVX == 0 || !os_supports_avx_vectors())) {
650 // 32 bytes vectors (in YMM) are only supported with AVX+
651 FLAG_SET_DEFAULT(MaxVectorSize, 16);
652 }
653 if (UseSSE < 2) {
654 // Vectors (in XMM) are only supported with SSE2+
655 FLAG_SET_DEFAULT(MaxVectorSize, 0);
656 }
657 #ifdef ASSERT
658 if (supports_avx() && PrintMiscellaneous && Verbose && TraceNewVectors) {
659 tty->print_cr("State of YMM registers after signal handle:");
660 int nreg = 2 LP64_ONLY(+2);
661 const char* ymm_name[4] = {"0", "7", "8", "15"};
662 for (int i = 0; i < nreg; i++) {
663 tty->print("YMM%s:", ymm_name[i]);
664 for (int j = 7; j >=0; j--) {
665 tty->print(" %x", _cpuid_info.ymm_save[i*8 + j]);
666 }
667 tty->cr();
668 }
669 }
670 #endif
671 }
672 #endif
674 // On new cpus instructions which update whole XMM register should be used
675 // to prevent partial register stall due to dependencies on high half.
676 //
677 // UseXmmLoadAndClearUpper == true --> movsd(xmm, mem)
678 // UseXmmLoadAndClearUpper == false --> movlpd(xmm, mem)
679 // UseXmmRegToRegMoveAll == true --> movaps(xmm, xmm), movapd(xmm, xmm).
680 // UseXmmRegToRegMoveAll == false --> movss(xmm, xmm), movsd(xmm, xmm).
682 if( is_amd() ) { // AMD cpus specific settings
683 if( supports_sse2() && FLAG_IS_DEFAULT(UseAddressNop) ) {
684 // Use it on new AMD cpus starting from Opteron.
685 UseAddressNop = true;
686 }
687 if( supports_sse2() && FLAG_IS_DEFAULT(UseNewLongLShift) ) {
688 // Use it on new AMD cpus starting from Opteron.
689 UseNewLongLShift = true;
690 }
691 if( FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper) ) {
692 if( supports_sse4a() ) {
693 UseXmmLoadAndClearUpper = true; // use movsd only on '10h' Opteron
694 } else {
695 UseXmmLoadAndClearUpper = false;
696 }
697 }
698 if( FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll) ) {
699 if( supports_sse4a() ) {
700 UseXmmRegToRegMoveAll = true; // use movaps, movapd only on '10h'
701 } else {
702 UseXmmRegToRegMoveAll = false;
703 }
704 }
705 if( FLAG_IS_DEFAULT(UseXmmI2F) ) {
706 if( supports_sse4a() ) {
707 UseXmmI2F = true;
708 } else {
709 UseXmmI2F = false;
710 }
711 }
712 if( FLAG_IS_DEFAULT(UseXmmI2D) ) {
713 if( supports_sse4a() ) {
714 UseXmmI2D = true;
715 } else {
716 UseXmmI2D = false;
717 }
718 }
719 if( FLAG_IS_DEFAULT(UseSSE42Intrinsics) ) {
720 if( supports_sse4_2() && UseSSE >= 4 ) {
721 UseSSE42Intrinsics = true;
722 }
723 }
725 // some defaults for AMD family 15h
726 if ( cpu_family() == 0x15 ) {
727 // On family 15h processors default is no sw prefetch
728 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) {
729 AllocatePrefetchStyle = 0;
730 }
731 // Also, if some other prefetch style is specified, default instruction type is PREFETCHW
732 if (FLAG_IS_DEFAULT(AllocatePrefetchInstr)) {
733 AllocatePrefetchInstr = 3;
734 }
735 // On family 15h processors use XMM and UnalignedLoadStores for Array Copy
736 if (supports_sse2() && FLAG_IS_DEFAULT(UseXMMForArrayCopy)) {
737 UseXMMForArrayCopy = true;
738 }
739 if (supports_sse2() && FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
740 UseUnalignedLoadStores = true;
741 }
742 }
744 #ifdef COMPILER2
745 if (MaxVectorSize > 16) {
746 // Limit vectors size to 16 bytes on current AMD cpus.
747 FLAG_SET_DEFAULT(MaxVectorSize, 16);
748 }
749 #endif // COMPILER2
750 }
752 if( is_intel() ) { // Intel cpus specific settings
753 if( FLAG_IS_DEFAULT(UseStoreImmI16) ) {
754 UseStoreImmI16 = false; // don't use it on Intel cpus
755 }
756 if( cpu_family() == 6 || cpu_family() == 15 ) {
757 if( FLAG_IS_DEFAULT(UseAddressNop) ) {
758 // Use it on all Intel cpus starting from PentiumPro
759 UseAddressNop = true;
760 }
761 }
762 if( FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper) ) {
763 UseXmmLoadAndClearUpper = true; // use movsd on all Intel cpus
764 }
765 if( FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll) ) {
766 if( supports_sse3() ) {
767 UseXmmRegToRegMoveAll = true; // use movaps, movapd on new Intel cpus
768 } else {
769 UseXmmRegToRegMoveAll = false;
770 }
771 }
772 if( cpu_family() == 6 && supports_sse3() ) { // New Intel cpus
773 #ifdef COMPILER2
774 if( FLAG_IS_DEFAULT(MaxLoopPad) ) {
775 // For new Intel cpus do the next optimization:
776 // don't align the beginning of a loop if there are enough instructions
777 // left (NumberOfLoopInstrToAlign defined in c2_globals.hpp)
778 // in current fetch line (OptoLoopAlignment) or the padding
779 // is big (> MaxLoopPad).
780 // Set MaxLoopPad to 11 for new Intel cpus to reduce number of
781 // generated NOP instructions. 11 is the largest size of one
782 // address NOP instruction '0F 1F' (see Assembler::nop(i)).
783 MaxLoopPad = 11;
784 }
785 #endif // COMPILER2
786 if (FLAG_IS_DEFAULT(UseXMMForArrayCopy)) {
787 UseXMMForArrayCopy = true; // use SSE2 movq on new Intel cpus
788 }
789 if (supports_sse4_2() && supports_ht()) { // Newest Intel cpus
790 if (FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
791 UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus
792 }
793 }
794 if (supports_sse4_2() && UseSSE >= 4) {
795 if (FLAG_IS_DEFAULT(UseSSE42Intrinsics)) {
796 UseSSE42Intrinsics = true;
797 }
798 }
799 }
800 }
802 // Use count leading zeros count instruction if available.
803 if (supports_lzcnt()) {
804 if (FLAG_IS_DEFAULT(UseCountLeadingZerosInstruction)) {
805 UseCountLeadingZerosInstruction = true;
806 }
807 } else if (UseCountLeadingZerosInstruction) {
808 warning("lzcnt instruction is not available on this CPU");
809 FLAG_SET_DEFAULT(UseCountLeadingZerosInstruction, false);
810 }
812 if (supports_bmi1()) {
813 if (FLAG_IS_DEFAULT(UseBMI1Instructions)) {
814 UseBMI1Instructions = true;
815 }
816 } else if (UseBMI1Instructions) {
817 warning("BMI1 instructions are not available on this CPU");
818 FLAG_SET_DEFAULT(UseBMI1Instructions, false);
819 }
821 // Use count trailing zeros instruction if available
822 if (supports_bmi1()) {
823 if (FLAG_IS_DEFAULT(UseCountTrailingZerosInstruction)) {
824 UseCountTrailingZerosInstruction = UseBMI1Instructions;
825 }
826 } else if (UseCountTrailingZerosInstruction) {
827 warning("tzcnt instruction is not available on this CPU");
828 FLAG_SET_DEFAULT(UseCountTrailingZerosInstruction, false);
829 }
831 // Use population count instruction if available.
832 if (supports_popcnt()) {
833 if (FLAG_IS_DEFAULT(UsePopCountInstruction)) {
834 UsePopCountInstruction = true;
835 }
836 } else if (UsePopCountInstruction) {
837 warning("POPCNT instruction is not available on this CPU");
838 FLAG_SET_DEFAULT(UsePopCountInstruction, false);
839 }
841 // Use fast-string operations if available.
842 if (supports_erms()) {
843 if (FLAG_IS_DEFAULT(UseFastStosb)) {
844 UseFastStosb = true;
845 }
846 } else if (UseFastStosb) {
847 warning("fast-string operations are not available on this CPU");
848 FLAG_SET_DEFAULT(UseFastStosb, false);
849 }
851 #ifdef COMPILER2
852 if (FLAG_IS_DEFAULT(AlignVector)) {
853 // Modern processors allow misaligned memory operations for vectors.
854 AlignVector = !UseUnalignedLoadStores;
855 }
856 #endif // COMPILER2
858 assert(0 <= ReadPrefetchInstr && ReadPrefetchInstr <= 3, "invalid value");
859 assert(0 <= AllocatePrefetchInstr && AllocatePrefetchInstr <= 3, "invalid value");
861 // set valid Prefetch instruction
862 if( ReadPrefetchInstr < 0 ) ReadPrefetchInstr = 0;
863 if( ReadPrefetchInstr > 3 ) ReadPrefetchInstr = 3;
864 if( ReadPrefetchInstr == 3 && !supports_3dnow_prefetch() ) ReadPrefetchInstr = 0;
865 if( !supports_sse() && supports_3dnow_prefetch() ) ReadPrefetchInstr = 3;
867 if( AllocatePrefetchInstr < 0 ) AllocatePrefetchInstr = 0;
868 if( AllocatePrefetchInstr > 3 ) AllocatePrefetchInstr = 3;
869 if( AllocatePrefetchInstr == 3 && !supports_3dnow_prefetch() ) AllocatePrefetchInstr=0;
870 if( !supports_sse() && supports_3dnow_prefetch() ) AllocatePrefetchInstr = 3;
872 // Allocation prefetch settings
873 intx cache_line_size = prefetch_data_size();
874 if( cache_line_size > AllocatePrefetchStepSize )
875 AllocatePrefetchStepSize = cache_line_size;
877 assert(AllocatePrefetchLines > 0, "invalid value");
878 if( AllocatePrefetchLines < 1 ) // set valid value in product VM
879 AllocatePrefetchLines = 3;
880 assert(AllocateInstancePrefetchLines > 0, "invalid value");
881 if( AllocateInstancePrefetchLines < 1 ) // set valid value in product VM
882 AllocateInstancePrefetchLines = 1;
884 AllocatePrefetchDistance = allocate_prefetch_distance();
885 AllocatePrefetchStyle = allocate_prefetch_style();
887 if( is_intel() && cpu_family() == 6 && supports_sse3() ) {
888 if( AllocatePrefetchStyle == 2 ) { // watermark prefetching on Core
889 #ifdef _LP64
890 AllocatePrefetchDistance = 384;
891 #else
892 AllocatePrefetchDistance = 320;
893 #endif
894 }
895 if( supports_sse4_2() && supports_ht() ) { // Nehalem based cpus
896 AllocatePrefetchDistance = 192;
897 AllocatePrefetchLines = 4;
898 #ifdef COMPILER2
899 if (AggressiveOpts && FLAG_IS_DEFAULT(UseFPUForSpilling)) {
900 FLAG_SET_DEFAULT(UseFPUForSpilling, true);
901 }
902 #endif
903 }
904 }
905 assert(AllocatePrefetchDistance % AllocatePrefetchStepSize == 0, "invalid value");
907 #ifdef _LP64
908 // Prefetch settings
909 PrefetchCopyIntervalInBytes = prefetch_copy_interval_in_bytes();
910 PrefetchScanIntervalInBytes = prefetch_scan_interval_in_bytes();
911 PrefetchFieldsAhead = prefetch_fields_ahead();
912 #endif
914 if (FLAG_IS_DEFAULT(ContendedPaddingWidth) &&
915 (cache_line_size > ContendedPaddingWidth))
916 ContendedPaddingWidth = cache_line_size;
918 #ifndef PRODUCT
919 if (PrintMiscellaneous && Verbose) {
920 tty->print_cr("Logical CPUs per core: %u",
921 logical_processors_per_package());
922 tty->print("UseSSE=%d",UseSSE);
923 if (UseAVX > 0) {
924 tty->print(" UseAVX=%d",UseAVX);
925 }
926 if (UseAES) {
927 tty->print(" UseAES=1");
928 }
929 #ifdef COMPILER2
930 if (MaxVectorSize > 0) {
931 tty->print(" MaxVectorSize=%d", MaxVectorSize);
932 }
933 #endif
934 tty->cr();
935 tty->print("Allocation");
936 if (AllocatePrefetchStyle <= 0 || UseSSE == 0 && !supports_3dnow_prefetch()) {
937 tty->print_cr(": no prefetching");
938 } else {
939 tty->print(" prefetching: ");
940 if (UseSSE == 0 && supports_3dnow_prefetch()) {
941 tty->print("PREFETCHW");
942 } else if (UseSSE >= 1) {
943 if (AllocatePrefetchInstr == 0) {
944 tty->print("PREFETCHNTA");
945 } else if (AllocatePrefetchInstr == 1) {
946 tty->print("PREFETCHT0");
947 } else if (AllocatePrefetchInstr == 2) {
948 tty->print("PREFETCHT2");
949 } else if (AllocatePrefetchInstr == 3) {
950 tty->print("PREFETCHW");
951 }
952 }
953 if (AllocatePrefetchLines > 1) {
954 tty->print_cr(" at distance %d, %d lines of %d bytes", AllocatePrefetchDistance, AllocatePrefetchLines, AllocatePrefetchStepSize);
955 } else {
956 tty->print_cr(" at distance %d, one line of %d bytes", AllocatePrefetchDistance, AllocatePrefetchStepSize);
957 }
958 }
960 if (PrefetchCopyIntervalInBytes > 0) {
961 tty->print_cr("PrefetchCopyIntervalInBytes %d", PrefetchCopyIntervalInBytes);
962 }
963 if (PrefetchScanIntervalInBytes > 0) {
964 tty->print_cr("PrefetchScanIntervalInBytes %d", PrefetchScanIntervalInBytes);
965 }
966 if (PrefetchFieldsAhead > 0) {
967 tty->print_cr("PrefetchFieldsAhead %d", PrefetchFieldsAhead);
968 }
969 if (ContendedPaddingWidth > 0) {
970 tty->print_cr("ContendedPaddingWidth %d", ContendedPaddingWidth);
971 }
972 }
973 #endif // !PRODUCT
974 }
976 bool VM_Version::use_biased_locking() {
977 #if INCLUDE_RTM_OPT
978 // RTM locking is most useful when there is high lock contention and
979 // low data contention. With high lock contention the lock is usually
980 // inflated and biased locking is not suitable for that case.
981 // RTM locking code requires that biased locking is off.
982 // Note: we can't switch off UseBiasedLocking in get_processor_features()
983 // because it is used by Thread::allocate() which is called before
984 // VM_Version::initialize().
985 if (UseRTMLocking && UseBiasedLocking) {
986 if (FLAG_IS_DEFAULT(UseBiasedLocking)) {
987 FLAG_SET_DEFAULT(UseBiasedLocking, false);
988 } else {
989 warning("Biased locking is not supported with RTM locking; ignoring UseBiasedLocking flag." );
990 UseBiasedLocking = false;
991 }
992 }
993 #endif
994 return UseBiasedLocking;
995 }
997 void VM_Version::initialize() {
998 ResourceMark rm;
999 // Making this stub must be FIRST use of assembler
1001 stub_blob = BufferBlob::create("get_cpu_info_stub", stub_size);
1002 if (stub_blob == NULL) {
1003 vm_exit_during_initialization("Unable to allocate get_cpu_info_stub");
1004 }
1005 CodeBuffer c(stub_blob);
1006 VM_Version_StubGenerator g(&c);
1007 get_cpu_info_stub = CAST_TO_FN_PTR(get_cpu_info_stub_t,
1008 g.generate_get_cpu_info());
1010 get_processor_features();
1011 }
