Fri, 22 Aug 2014 12:03:49 -0700
8055069: TSX and RTM should be deprecated more strongly until hardware is corrected
Summary: Require to specify UnlockExperimentalVMOptions flag together with UseRTMLocking flag on un-patched systems where CPUID allows it but is unsupported otherwise.
Reviewed-by: iveresov, fzhinkin
1 /*
2 * Copyright (c) 1997, 2014, 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
266 intx saved_useavx = UseAVX;
267 intx saved_usesse = UseSSE;
268 UseAVX = 1;
269 UseSSE = 2;
271 // load value into all 32 bytes of ymm7 register
272 __ movl(rcx, VM_Version::ymm_test_value());
274 __ movdl(xmm0, rcx);
275 __ pshufd(xmm0, xmm0, 0x00);
276 __ vinsertf128h(xmm0, xmm0, xmm0);
277 __ vmovdqu(xmm7, xmm0);
278 #ifdef _LP64
279 __ vmovdqu(xmm8, xmm0);
280 __ vmovdqu(xmm15, xmm0);
281 #endif
283 __ xorl(rsi, rsi);
284 VM_Version::set_cpuinfo_segv_addr( __ pc() );
285 // Generate SEGV
286 __ movl(rax, Address(rsi, 0));
288 VM_Version::set_cpuinfo_cont_addr( __ pc() );
289 // Returns here after signal. Save xmm0 to check it later.
290 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ymm_save_offset())));
291 __ vmovdqu(Address(rsi, 0), xmm0);
292 __ vmovdqu(Address(rsi, 32), xmm7);
293 #ifdef _LP64
294 __ vmovdqu(Address(rsi, 64), xmm8);
295 __ vmovdqu(Address(rsi, 96), xmm15);
296 #endif
298 VM_Version::clean_cpuFeatures();
299 UseAVX = saved_useavx;
300 UseSSE = saved_usesse;
302 //
303 // cpuid(0x7) Structured Extended Features
304 //
305 __ bind(sef_cpuid);
306 __ movl(rax, 7);
307 __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x7) supported?
308 __ jccb(Assembler::greater, ext_cpuid);
310 __ xorl(rcx, rcx);
311 __ cpuid();
312 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset())));
313 __ movl(Address(rsi, 0), rax);
314 __ movl(Address(rsi, 4), rbx);
316 //
317 // Extended cpuid(0x80000000)
318 //
319 __ bind(ext_cpuid);
320 __ movl(rax, 0x80000000);
321 __ cpuid();
322 __ cmpl(rax, 0x80000000); // Is cpuid(0x80000001) supported?
323 __ jcc(Assembler::belowEqual, done);
324 __ cmpl(rax, 0x80000004); // Is cpuid(0x80000005) supported?
325 __ jccb(Assembler::belowEqual, ext_cpuid1);
326 __ cmpl(rax, 0x80000006); // Is cpuid(0x80000007) supported?
327 __ jccb(Assembler::belowEqual, ext_cpuid5);
328 __ cmpl(rax, 0x80000007); // Is cpuid(0x80000008) supported?
329 __ jccb(Assembler::belowEqual, ext_cpuid7);
330 //
331 // Extended cpuid(0x80000008)
332 //
333 __ movl(rax, 0x80000008);
334 __ cpuid();
335 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid8_offset())));
336 __ movl(Address(rsi, 0), rax);
337 __ movl(Address(rsi, 4), rbx);
338 __ movl(Address(rsi, 8), rcx);
339 __ movl(Address(rsi,12), rdx);
341 //
342 // Extended cpuid(0x80000007)
343 //
344 __ bind(ext_cpuid7);
345 __ movl(rax, 0x80000007);
346 __ cpuid();
347 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid7_offset())));
348 __ movl(Address(rsi, 0), rax);
349 __ movl(Address(rsi, 4), rbx);
350 __ movl(Address(rsi, 8), rcx);
351 __ movl(Address(rsi,12), rdx);
353 //
354 // Extended cpuid(0x80000005)
355 //
356 __ bind(ext_cpuid5);
357 __ movl(rax, 0x80000005);
358 __ cpuid();
359 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid5_offset())));
360 __ movl(Address(rsi, 0), rax);
361 __ movl(Address(rsi, 4), rbx);
362 __ movl(Address(rsi, 8), rcx);
363 __ movl(Address(rsi,12), rdx);
365 //
366 // Extended cpuid(0x80000001)
367 //
368 __ bind(ext_cpuid1);
369 __ movl(rax, 0x80000001);
370 __ cpuid();
371 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid1_offset())));
372 __ movl(Address(rsi, 0), rax);
373 __ movl(Address(rsi, 4), rbx);
374 __ movl(Address(rsi, 8), rcx);
375 __ movl(Address(rsi,12), rdx);
377 //
378 // return
379 //
380 __ bind(done);
381 __ popf();
382 __ pop(rsi);
383 __ pop(rbx);
384 __ pop(rbp);
385 __ ret(0);
387 # undef __
389 return start;
390 };
391 };
394 void VM_Version::get_cpu_info_wrapper() {
395 get_cpu_info_stub(&_cpuid_info);
396 }
398 #ifndef CALL_TEST_FUNC_WITH_WRAPPER_IF_NEEDED
399 #define CALL_TEST_FUNC_WITH_WRAPPER_IF_NEEDED(f) f()
400 #endif
402 void VM_Version::get_processor_features() {
404 _cpu = 4; // 486 by default
405 _model = 0;
406 _stepping = 0;
407 _cpuFeatures = 0;
408 _logical_processors_per_package = 1;
410 if (!Use486InstrsOnly) {
411 // Get raw processor info
413 // Some platforms (like Win*) need a wrapper around here
414 // in order to properly handle SEGV for YMM registers test.
415 CALL_TEST_FUNC_WITH_WRAPPER_IF_NEEDED(get_cpu_info_wrapper);
417 assert_is_initialized();
418 _cpu = extended_cpu_family();
419 _model = extended_cpu_model();
420 _stepping = cpu_stepping();
422 if (cpu_family() > 4) { // it supports CPUID
423 _cpuFeatures = feature_flags();
424 // Logical processors are only available on P4s and above,
425 // and only if hyperthreading is available.
426 _logical_processors_per_package = logical_processor_count();
427 }
428 }
430 _supports_cx8 = supports_cmpxchg8();
431 // xchg and xadd instructions
432 _supports_atomic_getset4 = true;
433 _supports_atomic_getadd4 = true;
434 LP64_ONLY(_supports_atomic_getset8 = true);
435 LP64_ONLY(_supports_atomic_getadd8 = true);
437 #ifdef _LP64
438 // OS should support SSE for x64 and hardware should support at least SSE2.
439 if (!VM_Version::supports_sse2()) {
440 vm_exit_during_initialization("Unknown x64 processor: SSE2 not supported");
441 }
442 // in 64 bit the use of SSE2 is the minimum
443 if (UseSSE < 2) UseSSE = 2;
444 #endif
446 #ifdef AMD64
447 // flush_icache_stub have to be generated first.
448 // That is why Icache line size is hard coded in ICache class,
449 // see icache_x86.hpp. It is also the reason why we can't use
450 // clflush instruction in 32-bit VM since it could be running
451 // on CPU which does not support it.
452 //
453 // The only thing we can do is to verify that flushed
454 // ICache::line_size has correct value.
455 guarantee(_cpuid_info.std_cpuid1_edx.bits.clflush != 0, "clflush is not supported");
456 // clflush_size is size in quadwords (8 bytes).
457 guarantee(_cpuid_info.std_cpuid1_ebx.bits.clflush_size == 8, "such clflush size is not supported");
458 #endif
460 // If the OS doesn't support SSE, we can't use this feature even if the HW does
461 if (!os::supports_sse())
462 _cpuFeatures &= ~(CPU_SSE|CPU_SSE2|CPU_SSE3|CPU_SSSE3|CPU_SSE4A|CPU_SSE4_1|CPU_SSE4_2);
464 if (UseSSE < 4) {
465 _cpuFeatures &= ~CPU_SSE4_1;
466 _cpuFeatures &= ~CPU_SSE4_2;
467 }
469 if (UseSSE < 3) {
470 _cpuFeatures &= ~CPU_SSE3;
471 _cpuFeatures &= ~CPU_SSSE3;
472 _cpuFeatures &= ~CPU_SSE4A;
473 }
475 if (UseSSE < 2)
476 _cpuFeatures &= ~CPU_SSE2;
478 if (UseSSE < 1)
479 _cpuFeatures &= ~CPU_SSE;
481 if (UseAVX < 2)
482 _cpuFeatures &= ~CPU_AVX2;
484 if (UseAVX < 1)
485 _cpuFeatures &= ~CPU_AVX;
487 if (!UseAES && !FLAG_IS_DEFAULT(UseAES))
488 _cpuFeatures &= ~CPU_AES;
490 if (logical_processors_per_package() == 1) {
491 // HT processor could be installed on a system which doesn't support HT.
492 _cpuFeatures &= ~CPU_HT;
493 }
495 char buf[256];
496 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",
497 cores_per_cpu(), threads_per_core(),
498 cpu_family(), _model, _stepping,
499 (supports_cmov() ? ", cmov" : ""),
500 (supports_cmpxchg8() ? ", cx8" : ""),
501 (supports_fxsr() ? ", fxsr" : ""),
502 (supports_mmx() ? ", mmx" : ""),
503 (supports_sse() ? ", sse" : ""),
504 (supports_sse2() ? ", sse2" : ""),
505 (supports_sse3() ? ", sse3" : ""),
506 (supports_ssse3()? ", ssse3": ""),
507 (supports_sse4_1() ? ", sse4.1" : ""),
508 (supports_sse4_2() ? ", sse4.2" : ""),
509 (supports_popcnt() ? ", popcnt" : ""),
510 (supports_avx() ? ", avx" : ""),
511 (supports_avx2() ? ", avx2" : ""),
512 (supports_aes() ? ", aes" : ""),
513 (supports_clmul() ? ", clmul" : ""),
514 (supports_erms() ? ", erms" : ""),
515 (supports_rtm() ? ", rtm" : ""),
516 (supports_mmx_ext() ? ", mmxext" : ""),
517 (supports_3dnow_prefetch() ? ", 3dnowpref" : ""),
518 (supports_lzcnt() ? ", lzcnt": ""),
519 (supports_sse4a() ? ", sse4a": ""),
520 (supports_ht() ? ", ht": ""),
521 (supports_tsc() ? ", tsc": ""),
522 (supports_tscinv_bit() ? ", tscinvbit": ""),
523 (supports_tscinv() ? ", tscinv": ""),
524 (supports_bmi1() ? ", bmi1" : ""),
525 (supports_bmi2() ? ", bmi2" : ""));
526 _features_str = strdup(buf);
528 // UseSSE is set to the smaller of what hardware supports and what
529 // the command line requires. I.e., you cannot set UseSSE to 2 on
530 // older Pentiums which do not support it.
531 if (UseSSE > 4) UseSSE=4;
532 if (UseSSE < 0) UseSSE=0;
533 if (!supports_sse4_1()) // Drop to 3 if no SSE4 support
534 UseSSE = MIN2((intx)3,UseSSE);
535 if (!supports_sse3()) // Drop to 2 if no SSE3 support
536 UseSSE = MIN2((intx)2,UseSSE);
537 if (!supports_sse2()) // Drop to 1 if no SSE2 support
538 UseSSE = MIN2((intx)1,UseSSE);
539 if (!supports_sse ()) // Drop to 0 if no SSE support
540 UseSSE = 0;
542 if (UseAVX > 2) UseAVX=2;
543 if (UseAVX < 0) UseAVX=0;
544 if (!supports_avx2()) // Drop to 1 if no AVX2 support
545 UseAVX = MIN2((intx)1,UseAVX);
546 if (!supports_avx ()) // Drop to 0 if no AVX support
547 UseAVX = 0;
549 // Use AES instructions if available.
550 if (supports_aes()) {
551 if (FLAG_IS_DEFAULT(UseAES)) {
552 UseAES = true;
553 }
554 } else if (UseAES) {
555 if (!FLAG_IS_DEFAULT(UseAES))
556 warning("AES instructions are not available on this CPU");
557 FLAG_SET_DEFAULT(UseAES, false);
558 }
560 // Use CLMUL instructions if available.
561 if (supports_clmul()) {
562 if (FLAG_IS_DEFAULT(UseCLMUL)) {
563 UseCLMUL = true;
564 }
565 } else if (UseCLMUL) {
566 if (!FLAG_IS_DEFAULT(UseCLMUL))
567 warning("CLMUL instructions not available on this CPU (AVX may also be required)");
568 FLAG_SET_DEFAULT(UseCLMUL, false);
569 }
571 if (UseCLMUL && (UseSSE > 2)) {
572 if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) {
573 UseCRC32Intrinsics = true;
574 }
575 } else if (UseCRC32Intrinsics) {
576 if (!FLAG_IS_DEFAULT(UseCRC32Intrinsics))
577 warning("CRC32 Intrinsics requires AVX and CLMUL instructions (not available on this CPU)");
578 FLAG_SET_DEFAULT(UseCRC32Intrinsics, false);
579 }
581 // The AES intrinsic stubs require AES instruction support (of course)
582 // but also require sse3 mode for instructions it use.
583 if (UseAES && (UseSSE > 2)) {
584 if (FLAG_IS_DEFAULT(UseAESIntrinsics)) {
585 UseAESIntrinsics = true;
586 }
587 } else if (UseAESIntrinsics) {
588 if (!FLAG_IS_DEFAULT(UseAESIntrinsics))
589 warning("AES intrinsics are not available on this CPU");
590 FLAG_SET_DEFAULT(UseAESIntrinsics, false);
591 }
593 if (UseSHA) {
594 warning("SHA instructions are not available on this CPU");
595 FLAG_SET_DEFAULT(UseSHA, false);
596 }
597 if (UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics) {
598 warning("SHA intrinsics are not available on this CPU");
599 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false);
600 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false);
601 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false);
602 }
604 // Adjust RTM (Restricted Transactional Memory) flags
605 if (!supports_rtm() && UseRTMLocking) {
606 // Can't continue because UseRTMLocking affects UseBiasedLocking flag
607 // setting during arguments processing. See use_biased_locking().
608 // VM_Version_init() is executed after UseBiasedLocking is used
609 // in Thread::allocate().
610 vm_exit_during_initialization("RTM instructions are not available on this CPU");
611 }
613 #if INCLUDE_RTM_OPT
614 if (UseRTMLocking) {
615 if (is_intel_family_core()) {
616 if ((_model == CPU_MODEL_HASWELL_E3) ||
617 (_model == CPU_MODEL_HASWELL_E7 && _stepping < 3) ||
618 (_model == CPU_MODEL_BROADWELL && _stepping < 4)) {
619 if (!UnlockExperimentalVMOptions) {
620 vm_exit_during_initialization("UseRTMLocking is only available as experimental option on this platform. It must be enabled via -XX:+UnlockExperimentalVMOptions flag.");
621 } else {
622 warning("UseRTMLocking is only available as experimental option on this platform.");
623 }
624 }
625 }
626 if (!FLAG_IS_CMDLINE(UseRTMLocking)) {
627 // RTM locking should be used only for applications with
628 // high lock contention. For now we do not use it by default.
629 vm_exit_during_initialization("UseRTMLocking flag should be only set on command line");
630 }
631 if (!is_power_of_2(RTMTotalCountIncrRate)) {
632 warning("RTMTotalCountIncrRate must be a power of 2, resetting it to 64");
633 FLAG_SET_DEFAULT(RTMTotalCountIncrRate, 64);
634 }
635 if (RTMAbortRatio < 0 || RTMAbortRatio > 100) {
636 warning("RTMAbortRatio must be in the range 0 to 100, resetting it to 50");
637 FLAG_SET_DEFAULT(RTMAbortRatio, 50);
638 }
639 } else { // !UseRTMLocking
640 if (UseRTMForStackLocks) {
641 if (!FLAG_IS_DEFAULT(UseRTMForStackLocks)) {
642 warning("UseRTMForStackLocks flag should be off when UseRTMLocking flag is off");
643 }
644 FLAG_SET_DEFAULT(UseRTMForStackLocks, false);
645 }
646 if (UseRTMDeopt) {
647 FLAG_SET_DEFAULT(UseRTMDeopt, false);
648 }
649 if (PrintPreciseRTMLockingStatistics) {
650 FLAG_SET_DEFAULT(PrintPreciseRTMLockingStatistics, false);
651 }
652 }
653 #else
654 if (UseRTMLocking) {
655 // Only C2 does RTM locking optimization.
656 // Can't continue because UseRTMLocking affects UseBiasedLocking flag
657 // setting during arguments processing. See use_biased_locking().
658 vm_exit_during_initialization("RTM locking optimization is not supported in this VM");
659 }
660 #endif
662 #ifdef COMPILER2
663 if (UseFPUForSpilling) {
664 if (UseSSE < 2) {
665 // Only supported with SSE2+
666 FLAG_SET_DEFAULT(UseFPUForSpilling, false);
667 }
668 }
669 if (MaxVectorSize > 0) {
670 if (!is_power_of_2(MaxVectorSize)) {
671 warning("MaxVectorSize must be a power of 2");
672 FLAG_SET_DEFAULT(MaxVectorSize, 32);
673 }
674 if (MaxVectorSize > 32) {
675 FLAG_SET_DEFAULT(MaxVectorSize, 32);
676 }
677 if (MaxVectorSize > 16 && (UseAVX == 0 || !os_supports_avx_vectors())) {
678 // 32 bytes vectors (in YMM) are only supported with AVX+
679 FLAG_SET_DEFAULT(MaxVectorSize, 16);
680 }
681 if (UseSSE < 2) {
682 // Vectors (in XMM) are only supported with SSE2+
683 FLAG_SET_DEFAULT(MaxVectorSize, 0);
684 }
685 #ifdef ASSERT
686 if (supports_avx() && PrintMiscellaneous && Verbose && TraceNewVectors) {
687 tty->print_cr("State of YMM registers after signal handle:");
688 int nreg = 2 LP64_ONLY(+2);
689 const char* ymm_name[4] = {"0", "7", "8", "15"};
690 for (int i = 0; i < nreg; i++) {
691 tty->print("YMM%s:", ymm_name[i]);
692 for (int j = 7; j >=0; j--) {
693 tty->print(" %x", _cpuid_info.ymm_save[i*8 + j]);
694 }
695 tty->cr();
696 }
697 }
698 #endif
699 }
700 #endif
702 // On new cpus instructions which update whole XMM register should be used
703 // to prevent partial register stall due to dependencies on high half.
704 //
705 // UseXmmLoadAndClearUpper == true --> movsd(xmm, mem)
706 // UseXmmLoadAndClearUpper == false --> movlpd(xmm, mem)
707 // UseXmmRegToRegMoveAll == true --> movaps(xmm, xmm), movapd(xmm, xmm).
708 // UseXmmRegToRegMoveAll == false --> movss(xmm, xmm), movsd(xmm, xmm).
710 if( is_amd() ) { // AMD cpus specific settings
711 if( supports_sse2() && FLAG_IS_DEFAULT(UseAddressNop) ) {
712 // Use it on new AMD cpus starting from Opteron.
713 UseAddressNop = true;
714 }
715 if( supports_sse2() && FLAG_IS_DEFAULT(UseNewLongLShift) ) {
716 // Use it on new AMD cpus starting from Opteron.
717 UseNewLongLShift = true;
718 }
719 if( FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper) ) {
720 if( supports_sse4a() ) {
721 UseXmmLoadAndClearUpper = true; // use movsd only on '10h' Opteron
722 } else {
723 UseXmmLoadAndClearUpper = false;
724 }
725 }
726 if( FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll) ) {
727 if( supports_sse4a() ) {
728 UseXmmRegToRegMoveAll = true; // use movaps, movapd only on '10h'
729 } else {
730 UseXmmRegToRegMoveAll = false;
731 }
732 }
733 if( FLAG_IS_DEFAULT(UseXmmI2F) ) {
734 if( supports_sse4a() ) {
735 UseXmmI2F = true;
736 } else {
737 UseXmmI2F = false;
738 }
739 }
740 if( FLAG_IS_DEFAULT(UseXmmI2D) ) {
741 if( supports_sse4a() ) {
742 UseXmmI2D = true;
743 } else {
744 UseXmmI2D = false;
745 }
746 }
747 if( FLAG_IS_DEFAULT(UseSSE42Intrinsics) ) {
748 if( supports_sse4_2() && UseSSE >= 4 ) {
749 UseSSE42Intrinsics = true;
750 }
751 }
753 // some defaults for AMD family 15h
754 if ( cpu_family() == 0x15 ) {
755 // On family 15h processors default is no sw prefetch
756 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) {
757 AllocatePrefetchStyle = 0;
758 }
759 // Also, if some other prefetch style is specified, default instruction type is PREFETCHW
760 if (FLAG_IS_DEFAULT(AllocatePrefetchInstr)) {
761 AllocatePrefetchInstr = 3;
762 }
763 // On family 15h processors use XMM and UnalignedLoadStores for Array Copy
764 if (supports_sse2() && FLAG_IS_DEFAULT(UseXMMForArrayCopy)) {
765 UseXMMForArrayCopy = true;
766 }
767 if (supports_sse2() && FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
768 UseUnalignedLoadStores = true;
769 }
770 }
772 #ifdef COMPILER2
773 if (MaxVectorSize > 16) {
774 // Limit vectors size to 16 bytes on current AMD cpus.
775 FLAG_SET_DEFAULT(MaxVectorSize, 16);
776 }
777 #endif // COMPILER2
778 }
780 if( is_intel() ) { // Intel cpus specific settings
781 if( FLAG_IS_DEFAULT(UseStoreImmI16) ) {
782 UseStoreImmI16 = false; // don't use it on Intel cpus
783 }
784 if( cpu_family() == 6 || cpu_family() == 15 ) {
785 if( FLAG_IS_DEFAULT(UseAddressNop) ) {
786 // Use it on all Intel cpus starting from PentiumPro
787 UseAddressNop = true;
788 }
789 }
790 if( FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper) ) {
791 UseXmmLoadAndClearUpper = true; // use movsd on all Intel cpus
792 }
793 if( FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll) ) {
794 if( supports_sse3() ) {
795 UseXmmRegToRegMoveAll = true; // use movaps, movapd on new Intel cpus
796 } else {
797 UseXmmRegToRegMoveAll = false;
798 }
799 }
800 if( cpu_family() == 6 && supports_sse3() ) { // New Intel cpus
801 #ifdef COMPILER2
802 if( FLAG_IS_DEFAULT(MaxLoopPad) ) {
803 // For new Intel cpus do the next optimization:
804 // don't align the beginning of a loop if there are enough instructions
805 // left (NumberOfLoopInstrToAlign defined in c2_globals.hpp)
806 // in current fetch line (OptoLoopAlignment) or the padding
807 // is big (> MaxLoopPad).
808 // Set MaxLoopPad to 11 for new Intel cpus to reduce number of
809 // generated NOP instructions. 11 is the largest size of one
810 // address NOP instruction '0F 1F' (see Assembler::nop(i)).
811 MaxLoopPad = 11;
812 }
813 #endif // COMPILER2
814 if (FLAG_IS_DEFAULT(UseXMMForArrayCopy)) {
815 UseXMMForArrayCopy = true; // use SSE2 movq on new Intel cpus
816 }
817 if (supports_sse4_2() && supports_ht()) { // Newest Intel cpus
818 if (FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
819 UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus
820 }
821 }
822 if (supports_sse4_2() && UseSSE >= 4) {
823 if (FLAG_IS_DEFAULT(UseSSE42Intrinsics)) {
824 UseSSE42Intrinsics = true;
825 }
826 }
827 }
828 if ((cpu_family() == 0x06) &&
829 ((extended_cpu_model() == 0x36) || // Centerton
830 (extended_cpu_model() == 0x37) || // Silvermont
831 (extended_cpu_model() == 0x4D))) {
832 #ifdef COMPILER2
833 if (FLAG_IS_DEFAULT(OptoScheduling)) {
834 OptoScheduling = true;
835 }
836 #endif
837 if (supports_sse4_2()) { // Silvermont
838 if (FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
839 UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus
840 }
841 }
842 }
843 }
845 // Use count leading zeros count instruction if available.
846 if (supports_lzcnt()) {
847 if (FLAG_IS_DEFAULT(UseCountLeadingZerosInstruction)) {
848 UseCountLeadingZerosInstruction = true;
849 }
850 } else if (UseCountLeadingZerosInstruction) {
851 warning("lzcnt instruction is not available on this CPU");
852 FLAG_SET_DEFAULT(UseCountLeadingZerosInstruction, false);
853 }
855 if (supports_bmi1()) {
856 if (FLAG_IS_DEFAULT(UseBMI1Instructions)) {
857 UseBMI1Instructions = true;
858 }
859 } else if (UseBMI1Instructions) {
860 warning("BMI1 instructions are not available on this CPU");
861 FLAG_SET_DEFAULT(UseBMI1Instructions, false);
862 }
864 // Use count trailing zeros instruction if available
865 if (supports_bmi1()) {
866 if (FLAG_IS_DEFAULT(UseCountTrailingZerosInstruction)) {
867 UseCountTrailingZerosInstruction = UseBMI1Instructions;
868 }
869 } else if (UseCountTrailingZerosInstruction) {
870 warning("tzcnt instruction is not available on this CPU");
871 FLAG_SET_DEFAULT(UseCountTrailingZerosInstruction, false);
872 }
874 // Use population count instruction if available.
875 if (supports_popcnt()) {
876 if (FLAG_IS_DEFAULT(UsePopCountInstruction)) {
877 UsePopCountInstruction = true;
878 }
879 } else if (UsePopCountInstruction) {
880 warning("POPCNT instruction is not available on this CPU");
881 FLAG_SET_DEFAULT(UsePopCountInstruction, false);
882 }
884 // Use fast-string operations if available.
885 if (supports_erms()) {
886 if (FLAG_IS_DEFAULT(UseFastStosb)) {
887 UseFastStosb = true;
888 }
889 } else if (UseFastStosb) {
890 warning("fast-string operations are not available on this CPU");
891 FLAG_SET_DEFAULT(UseFastStosb, false);
892 }
894 #ifdef COMPILER2
895 if (FLAG_IS_DEFAULT(AlignVector)) {
896 // Modern processors allow misaligned memory operations for vectors.
897 AlignVector = !UseUnalignedLoadStores;
898 }
899 #endif // COMPILER2
901 assert(0 <= ReadPrefetchInstr && ReadPrefetchInstr <= 3, "invalid value");
902 assert(0 <= AllocatePrefetchInstr && AllocatePrefetchInstr <= 3, "invalid value");
904 // set valid Prefetch instruction
905 if( ReadPrefetchInstr < 0 ) ReadPrefetchInstr = 0;
906 if( ReadPrefetchInstr > 3 ) ReadPrefetchInstr = 3;
907 if( ReadPrefetchInstr == 3 && !supports_3dnow_prefetch() ) ReadPrefetchInstr = 0;
908 if( !supports_sse() && supports_3dnow_prefetch() ) ReadPrefetchInstr = 3;
910 if( AllocatePrefetchInstr < 0 ) AllocatePrefetchInstr = 0;
911 if( AllocatePrefetchInstr > 3 ) AllocatePrefetchInstr = 3;
912 if( AllocatePrefetchInstr == 3 && !supports_3dnow_prefetch() ) AllocatePrefetchInstr=0;
913 if( !supports_sse() && supports_3dnow_prefetch() ) AllocatePrefetchInstr = 3;
915 // Allocation prefetch settings
916 intx cache_line_size = prefetch_data_size();
917 if( cache_line_size > AllocatePrefetchStepSize )
918 AllocatePrefetchStepSize = cache_line_size;
920 assert(AllocatePrefetchLines > 0, "invalid value");
921 if( AllocatePrefetchLines < 1 ) // set valid value in product VM
922 AllocatePrefetchLines = 3;
923 assert(AllocateInstancePrefetchLines > 0, "invalid value");
924 if( AllocateInstancePrefetchLines < 1 ) // set valid value in product VM
925 AllocateInstancePrefetchLines = 1;
927 AllocatePrefetchDistance = allocate_prefetch_distance();
928 AllocatePrefetchStyle = allocate_prefetch_style();
930 if (is_intel() && cpu_family() == 6 && supports_sse3()) {
931 if (AllocatePrefetchStyle == 2) { // watermark prefetching on Core
932 #ifdef _LP64
933 AllocatePrefetchDistance = 384;
934 #else
935 AllocatePrefetchDistance = 320;
936 #endif
937 }
938 if (supports_sse4_2() && supports_ht()) { // Nehalem based cpus
939 AllocatePrefetchDistance = 192;
940 AllocatePrefetchLines = 4;
941 }
942 #ifdef COMPILER2
943 if (supports_sse4_2()) {
944 if (FLAG_IS_DEFAULT(UseFPUForSpilling)) {
945 FLAG_SET_DEFAULT(UseFPUForSpilling, true);
946 }
947 }
948 #endif
949 }
950 assert(AllocatePrefetchDistance % AllocatePrefetchStepSize == 0, "invalid value");
952 #ifdef _LP64
953 // Prefetch settings
954 PrefetchCopyIntervalInBytes = prefetch_copy_interval_in_bytes();
955 PrefetchScanIntervalInBytes = prefetch_scan_interval_in_bytes();
956 PrefetchFieldsAhead = prefetch_fields_ahead();
957 #endif
959 if (FLAG_IS_DEFAULT(ContendedPaddingWidth) &&
960 (cache_line_size > ContendedPaddingWidth))
961 ContendedPaddingWidth = cache_line_size;
963 #ifndef PRODUCT
964 if (PrintMiscellaneous && Verbose) {
965 tty->print_cr("Logical CPUs per core: %u",
966 logical_processors_per_package());
967 tty->print("UseSSE=%d", (int) UseSSE);
968 if (UseAVX > 0) {
969 tty->print(" UseAVX=%d", (int) UseAVX);
970 }
971 if (UseAES) {
972 tty->print(" UseAES=1");
973 }
974 #ifdef COMPILER2
975 if (MaxVectorSize > 0) {
976 tty->print(" MaxVectorSize=%d", (int) MaxVectorSize);
977 }
978 #endif
979 tty->cr();
980 tty->print("Allocation");
981 if (AllocatePrefetchStyle <= 0 || UseSSE == 0 && !supports_3dnow_prefetch()) {
982 tty->print_cr(": no prefetching");
983 } else {
984 tty->print(" prefetching: ");
985 if (UseSSE == 0 && supports_3dnow_prefetch()) {
986 tty->print("PREFETCHW");
987 } else if (UseSSE >= 1) {
988 if (AllocatePrefetchInstr == 0) {
989 tty->print("PREFETCHNTA");
990 } else if (AllocatePrefetchInstr == 1) {
991 tty->print("PREFETCHT0");
992 } else if (AllocatePrefetchInstr == 2) {
993 tty->print("PREFETCHT2");
994 } else if (AllocatePrefetchInstr == 3) {
995 tty->print("PREFETCHW");
996 }
997 }
998 if (AllocatePrefetchLines > 1) {
999 tty->print_cr(" at distance %d, %d lines of %d bytes", (int) AllocatePrefetchDistance, (int) AllocatePrefetchLines, (int) AllocatePrefetchStepSize);
1000 } else {
1001 tty->print_cr(" at distance %d, one line of %d bytes", (int) AllocatePrefetchDistance, (int) AllocatePrefetchStepSize);
1002 }
1003 }
1005 if (PrefetchCopyIntervalInBytes > 0) {
1006 tty->print_cr("PrefetchCopyIntervalInBytes %d", (int) PrefetchCopyIntervalInBytes);
1007 }
1008 if (PrefetchScanIntervalInBytes > 0) {
1009 tty->print_cr("PrefetchScanIntervalInBytes %d", (int) PrefetchScanIntervalInBytes);
1010 }
1011 if (PrefetchFieldsAhead > 0) {
1012 tty->print_cr("PrefetchFieldsAhead %d", (int) PrefetchFieldsAhead);
1013 }
1014 if (ContendedPaddingWidth > 0) {
1015 tty->print_cr("ContendedPaddingWidth %d", (int) ContendedPaddingWidth);
1016 }
1017 }
1018 #endif // !PRODUCT
1019 }
1021 bool VM_Version::use_biased_locking() {
1022 #if INCLUDE_RTM_OPT
1023 // RTM locking is most useful when there is high lock contention and
1024 // low data contention. With high lock contention the lock is usually
1025 // inflated and biased locking is not suitable for that case.
1026 // RTM locking code requires that biased locking is off.
1027 // Note: we can't switch off UseBiasedLocking in get_processor_features()
1028 // because it is used by Thread::allocate() which is called before
1029 // VM_Version::initialize().
1030 if (UseRTMLocking && UseBiasedLocking) {
1031 if (FLAG_IS_DEFAULT(UseBiasedLocking)) {
1032 FLAG_SET_DEFAULT(UseBiasedLocking, false);
1033 } else {
1034 warning("Biased locking is not supported with RTM locking; ignoring UseBiasedLocking flag." );
1035 UseBiasedLocking = false;
1036 }
1037 }
1038 #endif
1039 return UseBiasedLocking;
1040 }
1042 void VM_Version::initialize() {
1043 ResourceMark rm;
1044 // Making this stub must be FIRST use of assembler
1046 stub_blob = BufferBlob::create("get_cpu_info_stub", stub_size);
1047 if (stub_blob == NULL) {
1048 vm_exit_during_initialization("Unable to allocate get_cpu_info_stub");
1049 }
1050 CodeBuffer c(stub_blob);
1051 VM_Version_StubGenerator g(&c);
1052 get_cpu_info_stub = CAST_TO_FN_PTR(get_cpu_info_stub_t,
1053 g.generate_get_cpu_info());
1055 get_processor_features();
1056 }