1 /*
2 * Copyright (c) 1997, 2025, 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 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "asm/macroAssembler.inline.hpp"
27 #include "classfile/vmIntrinsics.hpp"
28 #include "code/codeBlob.hpp"
29 #include "compiler/compilerDefinitions.inline.hpp"
30 #include "jvm.h"
31 #include "logging/log.hpp"
32 #include "logging/logStream.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "memory/universe.hpp"
35 #include "runtime/globals_extension.hpp"
36 #include "runtime/java.hpp"
37 #include "runtime/os.inline.hpp"
38 #include "runtime/stubCodeGenerator.hpp"
39 #include "runtime/vm_version.hpp"
40 #include "utilities/checkedCast.hpp"
41 #include "utilities/powerOfTwo.hpp"
42 #include "utilities/virtualizationSupport.hpp"
43
44 int VM_Version::_cpu;
45 int VM_Version::_model;
46 int VM_Version::_stepping;
47 bool VM_Version::_has_intel_jcc_erratum;
48 VM_Version::CpuidInfo VM_Version::_cpuid_info = { 0, };
49
50 #define DECLARE_CPU_FEATURE_NAME(id, name, bit) name,
51 const char* VM_Version::_features_names[] = { CPU_FEATURE_FLAGS(DECLARE_CPU_FEATURE_NAME)};
52 #undef DECLARE_CPU_FEATURE_FLAG
53
54 // Address of instruction which causes SEGV
55 address VM_Version::_cpuinfo_segv_addr = nullptr;
56 // Address of instruction after the one which causes SEGV
57 address VM_Version::_cpuinfo_cont_addr = nullptr;
58 // Address of instruction which causes APX specific SEGV
59 address VM_Version::_cpuinfo_segv_addr_apx = nullptr;
60 // Address of instruction after the one which causes APX specific SEGV
61 address VM_Version::_cpuinfo_cont_addr_apx = nullptr;
62
63 static BufferBlob* stub_blob;
64 static const int stub_size = 2000;
65
66 extern "C" {
67 typedef void (*get_cpu_info_stub_t)(void*);
68 typedef void (*detect_virt_stub_t)(uint32_t, uint32_t*);
69 typedef void (*clear_apx_test_state_t)(void);
70 }
71 static get_cpu_info_stub_t get_cpu_info_stub = nullptr;
72 static detect_virt_stub_t detect_virt_stub = nullptr;
73 static clear_apx_test_state_t clear_apx_test_state_stub = nullptr;
74
75 #ifdef _LP64
76
77 bool VM_Version::supports_clflush() {
78 // clflush should always be available on x86_64
79 // if not we are in real trouble because we rely on it
80 // to flush the code cache.
81 // Unfortunately, Assembler::clflush is currently called as part
82 // of generation of the code cache flush routine. This happens
83 // under Universe::init before the processor features are set
84 // up. Assembler::flush calls this routine to check that clflush
85 // is allowed. So, we give the caller a free pass if Universe init
86 // is still in progress.
87 assert ((!Universe::is_fully_initialized() || (_features & CPU_FLUSH) != 0), "clflush should be available");
88 return true;
89 }
90 #endif
91
92 #define CPUID_STANDARD_FN 0x0
93 #define CPUID_STANDARD_FN_1 0x1
94 #define CPUID_STANDARD_FN_4 0x4
95 #define CPUID_STANDARD_FN_B 0xb
96
97 #define CPUID_EXTENDED_FN 0x80000000
98 #define CPUID_EXTENDED_FN_1 0x80000001
99 #define CPUID_EXTENDED_FN_2 0x80000002
100 #define CPUID_EXTENDED_FN_3 0x80000003
101 #define CPUID_EXTENDED_FN_4 0x80000004
102 #define CPUID_EXTENDED_FN_7 0x80000007
103 #define CPUID_EXTENDED_FN_8 0x80000008
104
105 class VM_Version_StubGenerator: public StubCodeGenerator {
106 public:
107
108 VM_Version_StubGenerator(CodeBuffer *c) : StubCodeGenerator(c) {}
109
110 #if defined(_LP64)
111 address clear_apx_test_state() {
112 # define __ _masm->
113 address start = __ pc();
114 // EGPRs are call clobbered registers, Explicit clearing of r16 and r31 during signal
115 // handling guarantees that preserved register values post signal handling were
116 // re-instantiated by operating system and not because they were not modified externally.
117
118 bool save_apx = UseAPX;
119 VM_Version::set_apx_cpuFeatures();
120 UseAPX = true;
121 // EGPR state save/restoration.
122 __ mov64(r16, 0L);
123 __ mov64(r31, 0L);
124 UseAPX = save_apx;
125 VM_Version::clean_cpuFeatures();
126 __ ret(0);
127 return start;
128 }
129 #endif
130
131 address generate_get_cpu_info() {
132 // Flags to test CPU type.
133 const uint32_t HS_EFL_AC = 0x40000;
134 const uint32_t HS_EFL_ID = 0x200000;
135 // Values for when we don't have a CPUID instruction.
136 const int CPU_FAMILY_SHIFT = 8;
137 const uint32_t CPU_FAMILY_386 = (3 << CPU_FAMILY_SHIFT);
138 const uint32_t CPU_FAMILY_486 = (4 << CPU_FAMILY_SHIFT);
139 bool use_evex = FLAG_IS_DEFAULT(UseAVX) || (UseAVX > 2);
140
141 Label detect_486, cpu486, detect_586, std_cpuid1, std_cpuid4;
142 Label sef_cpuid, sefsl1_cpuid, ext_cpuid, ext_cpuid1, ext_cpuid5, ext_cpuid7;
143 Label ext_cpuid8, done, wrapup, vector_save_restore, apx_save_restore_warning;
144 Label legacy_setup, save_restore_except, legacy_save_restore, start_simd_check;
145
146 StubCodeMark mark(this, "VM_Version", "get_cpu_info_stub");
147 # define __ _masm->
148
149 address start = __ pc();
150
151 //
152 // void get_cpu_info(VM_Version::CpuidInfo* cpuid_info);
153 //
154 // LP64: rcx and rdx are first and second argument registers on windows
155
156 __ push(rbp);
157 #ifdef _LP64
158 __ mov(rbp, c_rarg0); // cpuid_info address
159 #else
160 __ movptr(rbp, Address(rsp, 8)); // cpuid_info address
161 #endif
162 __ push(rbx);
163 __ push(rsi);
164 __ pushf(); // preserve rbx, and flags
165 __ pop(rax);
166 __ push(rax);
167 __ mov(rcx, rax);
168 //
169 // if we are unable to change the AC flag, we have a 386
170 //
171 __ xorl(rax, HS_EFL_AC);
172 __ push(rax);
173 __ popf();
174 __ pushf();
175 __ pop(rax);
176 __ cmpptr(rax, rcx);
177 __ jccb(Assembler::notEqual, detect_486);
178
179 __ movl(rax, CPU_FAMILY_386);
180 __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax);
181 __ jmp(done);
182
183 //
184 // If we are unable to change the ID flag, we have a 486 which does
185 // not support the "cpuid" instruction.
186 //
187 __ bind(detect_486);
188 __ mov(rax, rcx);
189 __ xorl(rax, HS_EFL_ID);
190 __ push(rax);
191 __ popf();
192 __ pushf();
193 __ pop(rax);
194 __ cmpptr(rcx, rax);
195 __ jccb(Assembler::notEqual, detect_586);
196
197 __ bind(cpu486);
198 __ movl(rax, CPU_FAMILY_486);
199 __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax);
200 __ jmp(done);
201
202 //
203 // At this point, we have a chip which supports the "cpuid" instruction
204 //
205 __ bind(detect_586);
206 __ xorl(rax, rax);
207 __ cpuid();
208 __ orl(rax, rax);
209 __ jcc(Assembler::equal, cpu486); // if cpuid doesn't support an input
210 // value of at least 1, we give up and
211 // assume a 486
212 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset())));
213 __ movl(Address(rsi, 0), rax);
214 __ movl(Address(rsi, 4), rbx);
215 __ movl(Address(rsi, 8), rcx);
216 __ movl(Address(rsi,12), rdx);
217
218 __ cmpl(rax, 0xa); // Is cpuid(0xB) supported?
219 __ jccb(Assembler::belowEqual, std_cpuid4);
220
221 //
222 // cpuid(0xB) Processor Topology
223 //
224 __ movl(rax, 0xb);
225 __ xorl(rcx, rcx); // Threads level
226 __ cpuid();
227
228 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB0_offset())));
229 __ movl(Address(rsi, 0), rax);
230 __ movl(Address(rsi, 4), rbx);
231 __ movl(Address(rsi, 8), rcx);
232 __ movl(Address(rsi,12), rdx);
233
234 __ movl(rax, 0xb);
235 __ movl(rcx, 1); // Cores level
236 __ cpuid();
237 __ push(rax);
238 __ andl(rax, 0x1f); // Determine if valid topology level
239 __ orl(rax, rbx); // eax[4:0] | ebx[0:15] == 0 indicates invalid level
240 __ andl(rax, 0xffff);
241 __ pop(rax);
242 __ jccb(Assembler::equal, std_cpuid4);
243
244 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB1_offset())));
245 __ movl(Address(rsi, 0), rax);
246 __ movl(Address(rsi, 4), rbx);
247 __ movl(Address(rsi, 8), rcx);
248 __ movl(Address(rsi,12), rdx);
249
250 __ movl(rax, 0xb);
251 __ movl(rcx, 2); // Packages level
252 __ cpuid();
253 __ push(rax);
254 __ andl(rax, 0x1f); // Determine if valid topology level
255 __ orl(rax, rbx); // eax[4:0] | ebx[0:15] == 0 indicates invalid level
256 __ andl(rax, 0xffff);
257 __ pop(rax);
258 __ jccb(Assembler::equal, std_cpuid4);
259
260 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB2_offset())));
261 __ movl(Address(rsi, 0), rax);
262 __ movl(Address(rsi, 4), rbx);
263 __ movl(Address(rsi, 8), rcx);
264 __ movl(Address(rsi,12), rdx);
265
266 //
267 // cpuid(0x4) Deterministic cache params
268 //
269 __ bind(std_cpuid4);
270 __ movl(rax, 4);
271 __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x4) supported?
272 __ jccb(Assembler::greater, std_cpuid1);
273
274 __ xorl(rcx, rcx); // L1 cache
275 __ cpuid();
276 __ push(rax);
277 __ andl(rax, 0x1f); // Determine if valid cache parameters used
278 __ orl(rax, rax); // eax[4:0] == 0 indicates invalid cache
279 __ pop(rax);
280 __ jccb(Assembler::equal, std_cpuid1);
281
282 __ lea(rsi, Address(rbp, in_bytes(VM_Version::dcp_cpuid4_offset())));
283 __ movl(Address(rsi, 0), rax);
284 __ movl(Address(rsi, 4), rbx);
285 __ movl(Address(rsi, 8), rcx);
286 __ movl(Address(rsi,12), rdx);
287
288 //
289 // Standard cpuid(0x1)
290 //
291 __ bind(std_cpuid1);
292 __ movl(rax, 1);
293 __ cpuid();
294 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
295 __ movl(Address(rsi, 0), rax);
296 __ movl(Address(rsi, 4), rbx);
297 __ movl(Address(rsi, 8), rcx);
298 __ movl(Address(rsi,12), rdx);
299
300 //
301 // Check if OS has enabled XGETBV instruction to access XCR0
302 // (OSXSAVE feature flag) and CPU supports AVX
303 //
304 __ andl(rcx, 0x18000000); // cpuid1 bits osxsave | avx
305 __ cmpl(rcx, 0x18000000);
306 __ jccb(Assembler::notEqual, sef_cpuid); // jump if AVX is not supported
307
308 //
309 // XCR0, XFEATURE_ENABLED_MASK register
310 //
311 __ xorl(rcx, rcx); // zero for XCR0 register
312 __ xgetbv();
313 __ lea(rsi, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset())));
314 __ movl(Address(rsi, 0), rax);
315 __ movl(Address(rsi, 4), rdx);
316
317 //
318 // cpuid(0x7) Structured Extended Features Enumeration Leaf.
319 //
320 __ bind(sef_cpuid);
321 __ movl(rax, 7);
322 __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x7) supported?
323 __ jccb(Assembler::greater, ext_cpuid);
324 // ECX = 0
325 __ xorl(rcx, rcx);
326 __ cpuid();
327 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset())));
328 __ movl(Address(rsi, 0), rax);
329 __ movl(Address(rsi, 4), rbx);
330 __ movl(Address(rsi, 8), rcx);
331 __ movl(Address(rsi, 12), rdx);
332
333 //
334 // cpuid(0x7) Structured Extended Features Enumeration Sub-Leaf 1.
335 //
336 __ bind(sefsl1_cpuid);
337 __ movl(rax, 7);
338 __ movl(rcx, 1);
339 __ cpuid();
340 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sefsl1_cpuid7_offset())));
341 __ movl(Address(rsi, 0), rax);
342 __ movl(Address(rsi, 4), rdx);
343
344 //
345 // Extended cpuid(0x80000000)
346 //
347 __ bind(ext_cpuid);
348 __ movl(rax, 0x80000000);
349 __ cpuid();
350 __ cmpl(rax, 0x80000000); // Is cpuid(0x80000001) supported?
351 __ jcc(Assembler::belowEqual, done);
352 __ cmpl(rax, 0x80000004); // Is cpuid(0x80000005) supported?
353 __ jcc(Assembler::belowEqual, ext_cpuid1);
354 __ cmpl(rax, 0x80000006); // Is cpuid(0x80000007) supported?
355 __ jccb(Assembler::belowEqual, ext_cpuid5);
356 __ cmpl(rax, 0x80000007); // Is cpuid(0x80000008) supported?
357 __ jccb(Assembler::belowEqual, ext_cpuid7);
358 __ cmpl(rax, 0x80000008); // Is cpuid(0x80000009 and above) supported?
359 __ jccb(Assembler::belowEqual, ext_cpuid8);
360 __ cmpl(rax, 0x8000001E); // Is cpuid(0x8000001E) supported?
361 __ jccb(Assembler::below, ext_cpuid8);
362 //
363 // Extended cpuid(0x8000001E)
364 //
365 __ movl(rax, 0x8000001E);
366 __ cpuid();
367 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid1E_offset())));
368 __ movl(Address(rsi, 0), rax);
369 __ movl(Address(rsi, 4), rbx);
370 __ movl(Address(rsi, 8), rcx);
371 __ movl(Address(rsi,12), rdx);
372
373 //
374 // Extended cpuid(0x80000008)
375 //
376 __ bind(ext_cpuid8);
377 __ movl(rax, 0x80000008);
378 __ cpuid();
379 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid8_offset())));
380 __ movl(Address(rsi, 0), rax);
381 __ movl(Address(rsi, 4), rbx);
382 __ movl(Address(rsi, 8), rcx);
383 __ movl(Address(rsi,12), rdx);
384
385 //
386 // Extended cpuid(0x80000007)
387 //
388 __ bind(ext_cpuid7);
389 __ movl(rax, 0x80000007);
390 __ cpuid();
391 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid7_offset())));
392 __ movl(Address(rsi, 0), rax);
393 __ movl(Address(rsi, 4), rbx);
394 __ movl(Address(rsi, 8), rcx);
395 __ movl(Address(rsi,12), rdx);
396
397 //
398 // Extended cpuid(0x80000005)
399 //
400 __ bind(ext_cpuid5);
401 __ movl(rax, 0x80000005);
402 __ cpuid();
403 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid5_offset())));
404 __ movl(Address(rsi, 0), rax);
405 __ movl(Address(rsi, 4), rbx);
406 __ movl(Address(rsi, 8), rcx);
407 __ movl(Address(rsi,12), rdx);
408
409 //
410 // Extended cpuid(0x80000001)
411 //
412 __ bind(ext_cpuid1);
413 __ movl(rax, 0x80000001);
414 __ cpuid();
415 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid1_offset())));
416 __ movl(Address(rsi, 0), rax);
417 __ movl(Address(rsi, 4), rbx);
418 __ movl(Address(rsi, 8), rcx);
419 __ movl(Address(rsi,12), rdx);
420
421 #if defined(_LP64)
422 //
423 // Check if OS has enabled XGETBV instruction to access XCR0
424 // (OSXSAVE feature flag) and CPU supports APX
425 //
426 // To enable APX, check CPUID.EAX=7.ECX=1.EDX[21] bit for HW support
427 // and XCRO[19] bit for OS support to save/restore extended GPR state.
428 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sefsl1_cpuid7_offset())));
429 __ movl(rax, 0x200000);
430 __ andl(rax, Address(rsi, 4));
431 __ cmpl(rax, 0x200000);
432 __ jcc(Assembler::notEqual, vector_save_restore);
433 // check _cpuid_info.xem_xcr0_eax.bits.apx_f
434 __ movl(rax, 0x80000);
435 __ andl(rax, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); // xcr0 bits apx_f
436 __ cmpl(rax, 0x80000);
437 __ jcc(Assembler::notEqual, vector_save_restore);
438
439 #ifndef PRODUCT
440 bool save_apx = UseAPX;
441 VM_Version::set_apx_cpuFeatures();
442 UseAPX = true;
443 __ mov64(r16, VM_Version::egpr_test_value());
444 __ mov64(r31, VM_Version::egpr_test_value());
445 __ xorl(rsi, rsi);
446 VM_Version::set_cpuinfo_segv_addr_apx(__ pc());
447 // Generate SEGV
448 __ movl(rax, Address(rsi, 0));
449
450 VM_Version::set_cpuinfo_cont_addr_apx(__ pc());
451 __ lea(rsi, Address(rbp, in_bytes(VM_Version::apx_save_offset())));
452 __ movq(Address(rsi, 0), r16);
453 __ movq(Address(rsi, 8), r31);
454
455 UseAPX = save_apx;
456 #endif
457 #endif
458 __ bind(vector_save_restore);
459 //
460 // Check if OS has enabled XGETBV instruction to access XCR0
461 // (OSXSAVE feature flag) and CPU supports AVX
462 //
463 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
464 __ movl(rcx, 0x18000000); // cpuid1 bits osxsave | avx
465 __ andl(rcx, Address(rsi, 8)); // cpuid1 bits osxsave | avx
466 __ cmpl(rcx, 0x18000000);
467 __ jccb(Assembler::notEqual, done); // jump if AVX is not supported
468
469 __ movl(rax, 0x6);
470 __ andl(rax, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); // xcr0 bits sse | ymm
471 __ cmpl(rax, 0x6);
472 __ jccb(Assembler::equal, start_simd_check); // return if AVX is not supported
473
474 // we need to bridge farther than imm8, so we use this island as a thunk
475 __ bind(done);
476 __ jmp(wrapup);
477
478 __ bind(start_simd_check);
479 //
480 // Some OSs have a bug when upper 128/256bits of YMM/ZMM
481 // registers are not restored after a signal processing.
482 // Generate SEGV here (reference through null)
483 // and check upper YMM/ZMM bits after it.
484 //
485 int saved_useavx = UseAVX;
486 int saved_usesse = UseSSE;
487
488 // If UseAVX is uninitialized or is set by the user to include EVEX
489 if (use_evex) {
490 // check _cpuid_info.sef_cpuid7_ebx.bits.avx512f
491 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset())));
492 __ movl(rax, 0x10000);
493 __ andl(rax, Address(rsi, 4)); // xcr0 bits sse | ymm
494 __ cmpl(rax, 0x10000);
495 __ jccb(Assembler::notEqual, legacy_setup); // jump if EVEX is not supported
496 // check _cpuid_info.xem_xcr0_eax.bits.opmask
497 // check _cpuid_info.xem_xcr0_eax.bits.zmm512
498 // check _cpuid_info.xem_xcr0_eax.bits.zmm32
499 __ movl(rax, 0xE0);
500 __ andl(rax, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); // xcr0 bits sse | ymm
501 __ cmpl(rax, 0xE0);
502 __ jccb(Assembler::notEqual, legacy_setup); // jump if EVEX is not supported
503
504 if (FLAG_IS_DEFAULT(UseAVX)) {
505 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
506 __ movl(rax, Address(rsi, 0));
507 __ cmpl(rax, 0x50654); // If it is Skylake
508 __ jcc(Assembler::equal, legacy_setup);
509 }
510 // EVEX setup: run in lowest evex mode
511 VM_Version::set_evex_cpuFeatures(); // Enable temporary to pass asserts
512 UseAVX = 3;
513 UseSSE = 2;
514 #ifdef _WINDOWS
515 // xmm5-xmm15 are not preserved by caller on windows
516 // https://msdn.microsoft.com/en-us/library/9z1stfyw.aspx
517 __ subptr(rsp, 64);
518 __ evmovdqul(Address(rsp, 0), xmm7, Assembler::AVX_512bit);
519 __ subptr(rsp, 64);
520 __ evmovdqul(Address(rsp, 0), xmm8, Assembler::AVX_512bit);
521 __ subptr(rsp, 64);
522 __ evmovdqul(Address(rsp, 0), xmm31, Assembler::AVX_512bit);
523 #endif // _WINDOWS
524
525 // load value into all 64 bytes of zmm7 register
526 __ movl(rcx, VM_Version::ymm_test_value());
527 __ movdl(xmm0, rcx);
528 __ vpbroadcastd(xmm0, xmm0, Assembler::AVX_512bit);
529 __ evmovdqul(xmm7, xmm0, Assembler::AVX_512bit);
530 #ifdef _LP64
531 __ evmovdqul(xmm8, xmm0, Assembler::AVX_512bit);
532 __ evmovdqul(xmm31, xmm0, Assembler::AVX_512bit);
533 #endif
534 VM_Version::clean_cpuFeatures();
535 __ jmp(save_restore_except);
536 }
537
538 __ bind(legacy_setup);
539 // AVX setup
540 VM_Version::set_avx_cpuFeatures(); // Enable temporary to pass asserts
541 UseAVX = 1;
542 UseSSE = 2;
543 #ifdef _WINDOWS
544 __ subptr(rsp, 32);
545 __ vmovdqu(Address(rsp, 0), xmm7);
546 __ subptr(rsp, 32);
547 __ vmovdqu(Address(rsp, 0), xmm8);
548 __ subptr(rsp, 32);
549 __ vmovdqu(Address(rsp, 0), xmm15);
550 #endif // _WINDOWS
551
552 // load value into all 32 bytes of ymm7 register
553 __ movl(rcx, VM_Version::ymm_test_value());
554
555 __ movdl(xmm0, rcx);
556 __ pshufd(xmm0, xmm0, 0x00);
557 __ vinsertf128_high(xmm0, xmm0);
558 __ vmovdqu(xmm7, xmm0);
559 #ifdef _LP64
560 __ vmovdqu(xmm8, xmm0);
561 __ vmovdqu(xmm15, xmm0);
562 #endif
563 VM_Version::clean_cpuFeatures();
564
565 __ bind(save_restore_except);
566 __ xorl(rsi, rsi);
567 VM_Version::set_cpuinfo_segv_addr(__ pc());
568 // Generate SEGV
569 __ movl(rax, Address(rsi, 0));
570
571 VM_Version::set_cpuinfo_cont_addr(__ pc());
572 // Returns here after signal. Save xmm0 to check it later.
573
574 // If UseAVX is uninitialized or is set by the user to include EVEX
575 if (use_evex) {
576 // check _cpuid_info.sef_cpuid7_ebx.bits.avx512f
577 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset())));
578 __ movl(rax, 0x10000);
579 __ andl(rax, Address(rsi, 4));
580 __ cmpl(rax, 0x10000);
581 __ jcc(Assembler::notEqual, legacy_save_restore);
582 // check _cpuid_info.xem_xcr0_eax.bits.opmask
583 // check _cpuid_info.xem_xcr0_eax.bits.zmm512
584 // check _cpuid_info.xem_xcr0_eax.bits.zmm32
585 __ movl(rax, 0xE0);
586 __ andl(rax, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); // xcr0 bits sse | ymm
587 __ cmpl(rax, 0xE0);
588 __ jcc(Assembler::notEqual, legacy_save_restore);
589
590 if (FLAG_IS_DEFAULT(UseAVX)) {
591 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
592 __ movl(rax, Address(rsi, 0));
593 __ cmpl(rax, 0x50654); // If it is Skylake
594 __ jcc(Assembler::equal, legacy_save_restore);
595 }
596 // EVEX check: run in lowest evex mode
597 VM_Version::set_evex_cpuFeatures(); // Enable temporary to pass asserts
598 UseAVX = 3;
599 UseSSE = 2;
600 __ lea(rsi, Address(rbp, in_bytes(VM_Version::zmm_save_offset())));
601 __ evmovdqul(Address(rsi, 0), xmm0, Assembler::AVX_512bit);
602 __ evmovdqul(Address(rsi, 64), xmm7, Assembler::AVX_512bit);
603 #ifdef _LP64
604 __ evmovdqul(Address(rsi, 128), xmm8, Assembler::AVX_512bit);
605 __ evmovdqul(Address(rsi, 192), xmm31, Assembler::AVX_512bit);
606 #endif
607
608 #ifdef _WINDOWS
609 __ evmovdqul(xmm31, Address(rsp, 0), Assembler::AVX_512bit);
610 __ addptr(rsp, 64);
611 __ evmovdqul(xmm8, Address(rsp, 0), Assembler::AVX_512bit);
612 __ addptr(rsp, 64);
613 __ evmovdqul(xmm7, Address(rsp, 0), Assembler::AVX_512bit);
614 __ addptr(rsp, 64);
615 #endif // _WINDOWS
616 generate_vzeroupper(wrapup);
617 VM_Version::clean_cpuFeatures();
618 UseAVX = saved_useavx;
619 UseSSE = saved_usesse;
620 __ jmp(wrapup);
621 }
622
623 __ bind(legacy_save_restore);
624 // AVX check
625 VM_Version::set_avx_cpuFeatures(); // Enable temporary to pass asserts
626 UseAVX = 1;
627 UseSSE = 2;
628 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ymm_save_offset())));
629 __ vmovdqu(Address(rsi, 0), xmm0);
630 __ vmovdqu(Address(rsi, 32), xmm7);
631 #ifdef _LP64
632 __ vmovdqu(Address(rsi, 64), xmm8);
633 __ vmovdqu(Address(rsi, 96), xmm15);
634 #endif
635
636 #ifdef _WINDOWS
637 __ vmovdqu(xmm15, Address(rsp, 0));
638 __ addptr(rsp, 32);
639 __ vmovdqu(xmm8, Address(rsp, 0));
640 __ addptr(rsp, 32);
641 __ vmovdqu(xmm7, Address(rsp, 0));
642 __ addptr(rsp, 32);
643 #endif // _WINDOWS
644
645 generate_vzeroupper(wrapup);
646 VM_Version::clean_cpuFeatures();
647 UseAVX = saved_useavx;
648 UseSSE = saved_usesse;
649
650 __ bind(wrapup);
651 __ popf();
652 __ pop(rsi);
653 __ pop(rbx);
654 __ pop(rbp);
655 __ ret(0);
656
657 # undef __
658
659 return start;
660 };
661 void generate_vzeroupper(Label& L_wrapup) {
662 # define __ _masm->
663 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset())));
664 __ cmpl(Address(rsi, 4), 0x756e6547); // 'uneG'
665 __ jcc(Assembler::notEqual, L_wrapup);
666 __ movl(rcx, 0x0FFF0FF0);
667 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
668 __ andl(rcx, Address(rsi, 0));
669 __ cmpl(rcx, 0x00050670); // If it is Xeon Phi 3200/5200/7200
670 __ jcc(Assembler::equal, L_wrapup);
671 __ cmpl(rcx, 0x00080650); // If it is Future Xeon Phi
672 __ jcc(Assembler::equal, L_wrapup);
673 // vzeroupper() will use a pre-computed instruction sequence that we
674 // can't compute until after we've determined CPU capabilities. Use
675 // uncached variant here directly to be able to bootstrap correctly
676 __ vzeroupper_uncached();
677 # undef __
678 }
679 address generate_detect_virt() {
680 StubCodeMark mark(this, "VM_Version", "detect_virt_stub");
681 # define __ _masm->
682
683 address start = __ pc();
684
685 // Evacuate callee-saved registers
686 __ push(rbp);
687 __ push(rbx);
688 __ push(rsi); // for Windows
689
690 #ifdef _LP64
691 __ mov(rax, c_rarg0); // CPUID leaf
692 __ mov(rsi, c_rarg1); // register array address (eax, ebx, ecx, edx)
693 #else
694 __ movptr(rax, Address(rsp, 16)); // CPUID leaf
695 __ movptr(rsi, Address(rsp, 20)); // register array address
696 #endif
697
698 __ cpuid();
699
700 // Store result to register array
701 __ movl(Address(rsi, 0), rax);
702 __ movl(Address(rsi, 4), rbx);
703 __ movl(Address(rsi, 8), rcx);
704 __ movl(Address(rsi, 12), rdx);
705
706 // Epilogue
707 __ pop(rsi);
708 __ pop(rbx);
709 __ pop(rbp);
710 __ ret(0);
711
712 # undef __
713
714 return start;
715 };
716
717
718 address generate_getCPUIDBrandString(void) {
719 // Flags to test CPU type.
720 const uint32_t HS_EFL_AC = 0x40000;
721 const uint32_t HS_EFL_ID = 0x200000;
722 // Values for when we don't have a CPUID instruction.
723 const int CPU_FAMILY_SHIFT = 8;
724 const uint32_t CPU_FAMILY_386 = (3 << CPU_FAMILY_SHIFT);
725 const uint32_t CPU_FAMILY_486 = (4 << CPU_FAMILY_SHIFT);
726
727 Label detect_486, cpu486, detect_586, done, ext_cpuid;
728
729 StubCodeMark mark(this, "VM_Version", "getCPUIDNameInfo_stub");
730 # define __ _masm->
731
732 address start = __ pc();
733
734 //
735 // void getCPUIDBrandString(VM_Version::CpuidInfo* cpuid_info);
736 //
737 // LP64: rcx and rdx are first and second argument registers on windows
738
739 __ push(rbp);
740 #ifdef _LP64
741 __ mov(rbp, c_rarg0); // cpuid_info address
742 #else
743 __ movptr(rbp, Address(rsp, 8)); // cpuid_info address
744 #endif
745 __ push(rbx);
746 __ push(rsi);
747 __ pushf(); // preserve rbx, and flags
748 __ pop(rax);
749 __ push(rax);
750 __ mov(rcx, rax);
751 //
752 // if we are unable to change the AC flag, we have a 386
753 //
754 __ xorl(rax, HS_EFL_AC);
755 __ push(rax);
756 __ popf();
757 __ pushf();
758 __ pop(rax);
759 __ cmpptr(rax, rcx);
760 __ jccb(Assembler::notEqual, detect_486);
761
762 __ movl(rax, CPU_FAMILY_386);
763 __ jmp(done);
764
765 //
766 // If we are unable to change the ID flag, we have a 486 which does
767 // not support the "cpuid" instruction.
768 //
769 __ bind(detect_486);
770 __ mov(rax, rcx);
771 __ xorl(rax, HS_EFL_ID);
772 __ push(rax);
773 __ popf();
774 __ pushf();
775 __ pop(rax);
776 __ cmpptr(rcx, rax);
777 __ jccb(Assembler::notEqual, detect_586);
778
779 __ bind(cpu486);
780 __ movl(rax, CPU_FAMILY_486);
781 __ jmp(done);
782
783 //
784 // At this point, we have a chip which supports the "cpuid" instruction
785 //
786 __ bind(detect_586);
787 __ xorl(rax, rax);
788 __ cpuid();
789 __ orl(rax, rax);
790 __ jcc(Assembler::equal, cpu486); // if cpuid doesn't support an input
791 // value of at least 1, we give up and
792 // assume a 486
793
794 //
795 // Extended cpuid(0x80000000) for processor brand string detection
796 //
797 __ bind(ext_cpuid);
798 __ movl(rax, CPUID_EXTENDED_FN);
799 __ cpuid();
800 __ cmpl(rax, CPUID_EXTENDED_FN_4);
801 __ jcc(Assembler::below, done);
802
803 //
804 // Extended cpuid(0x80000002) // first 16 bytes in brand string
805 //
806 __ movl(rax, CPUID_EXTENDED_FN_2);
807 __ cpuid();
808 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_0_offset())));
809 __ movl(Address(rsi, 0), rax);
810 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_1_offset())));
811 __ movl(Address(rsi, 0), rbx);
812 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_2_offset())));
813 __ movl(Address(rsi, 0), rcx);
814 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_3_offset())));
815 __ movl(Address(rsi,0), rdx);
816
817 //
818 // Extended cpuid(0x80000003) // next 16 bytes in brand string
819 //
820 __ movl(rax, CPUID_EXTENDED_FN_3);
821 __ cpuid();
822 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_4_offset())));
823 __ movl(Address(rsi, 0), rax);
824 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_5_offset())));
825 __ movl(Address(rsi, 0), rbx);
826 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_6_offset())));
827 __ movl(Address(rsi, 0), rcx);
828 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_7_offset())));
829 __ movl(Address(rsi,0), rdx);
830
831 //
832 // Extended cpuid(0x80000004) // last 16 bytes in brand string
833 //
834 __ movl(rax, CPUID_EXTENDED_FN_4);
835 __ cpuid();
836 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_8_offset())));
837 __ movl(Address(rsi, 0), rax);
838 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_9_offset())));
839 __ movl(Address(rsi, 0), rbx);
840 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_10_offset())));
841 __ movl(Address(rsi, 0), rcx);
842 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_11_offset())));
843 __ movl(Address(rsi,0), rdx);
844
845 //
846 // return
847 //
848 __ bind(done);
849 __ popf();
850 __ pop(rsi);
851 __ pop(rbx);
852 __ pop(rbp);
853 __ ret(0);
854
855 # undef __
856
857 return start;
858 };
859 };
860
861 void VM_Version::get_processor_features() {
862
863 _cpu = 4; // 486 by default
864 _model = 0;
865 _stepping = 0;
866 _features = 0;
867 _logical_processors_per_package = 1;
868 // i486 internal cache is both I&D and has a 16-byte line size
869 _L1_data_cache_line_size = 16;
870
871 // Get raw processor info
872
873 get_cpu_info_stub(&_cpuid_info);
874
875 assert_is_initialized();
876 _cpu = extended_cpu_family();
877 _model = extended_cpu_model();
878 _stepping = cpu_stepping();
879
880 if (cpu_family() > 4) { // it supports CPUID
881 _features = _cpuid_info.feature_flags(); // These can be changed by VM settings
882 _cpu_features = _features; // Preserve features
883 // Logical processors are only available on P4s and above,
884 // and only if hyperthreading is available.
885 _logical_processors_per_package = logical_processor_count();
886 _L1_data_cache_line_size = L1_line_size();
887 }
888
889 // xchg and xadd instructions
890 _supports_atomic_getset4 = true;
891 _supports_atomic_getadd4 = true;
892 LP64_ONLY(_supports_atomic_getset8 = true);
893 LP64_ONLY(_supports_atomic_getadd8 = true);
894
895 #ifdef _LP64
896 // OS should support SSE for x64 and hardware should support at least SSE2.
897 if (!VM_Version::supports_sse2()) {
898 vm_exit_during_initialization("Unknown x64 processor: SSE2 not supported");
899 }
900 // in 64 bit the use of SSE2 is the minimum
901 if (UseSSE < 2) UseSSE = 2;
902 #endif
903
904 #ifdef AMD64
905 // flush_icache_stub have to be generated first.
906 // That is why Icache line size is hard coded in ICache class,
907 // see icache_x86.hpp. It is also the reason why we can't use
908 // clflush instruction in 32-bit VM since it could be running
909 // on CPU which does not support it.
910 //
911 // The only thing we can do is to verify that flushed
912 // ICache::line_size has correct value.
913 guarantee(_cpuid_info.std_cpuid1_edx.bits.clflush != 0, "clflush is not supported");
914 // clflush_size is size in quadwords (8 bytes).
915 guarantee(_cpuid_info.std_cpuid1_ebx.bits.clflush_size == 8, "such clflush size is not supported");
916 #endif
917
918 #ifdef _LP64
919 // assigning this field effectively enables Unsafe.writebackMemory()
920 // by initing UnsafeConstant.DATA_CACHE_LINE_FLUSH_SIZE to non-zero
921 // that is only implemented on x86_64 and only if the OS plays ball
922 if (os::supports_map_sync()) {
923 // publish data cache line flush size to generic field, otherwise
924 // let if default to zero thereby disabling writeback
925 _data_cache_line_flush_size = _cpuid_info.std_cpuid1_ebx.bits.clflush_size * 8;
926 }
927 #endif
928
929 // Check if processor has Intel Ecore
930 if (FLAG_IS_DEFAULT(EnableX86ECoreOpts) && is_intel() && cpu_family() == 6 &&
931 (_model == 0x97 || _model == 0xAA || _model == 0xAC || _model == 0xAF ||
932 _model == 0xCC || _model == 0xDD)) {
933 FLAG_SET_DEFAULT(EnableX86ECoreOpts, true);
934 }
935
936 if (UseSSE < 4) {
937 _features &= ~CPU_SSE4_1;
938 _features &= ~CPU_SSE4_2;
939 }
940
941 if (UseSSE < 3) {
942 _features &= ~CPU_SSE3;
943 _features &= ~CPU_SSSE3;
944 _features &= ~CPU_SSE4A;
945 }
946
947 if (UseSSE < 2)
948 _features &= ~CPU_SSE2;
949
950 if (UseSSE < 1)
951 _features &= ~CPU_SSE;
952
953 //since AVX instructions is slower than SSE in some ZX cpus, force USEAVX=0.
954 if (is_zx() && ((cpu_family() == 6) || (cpu_family() == 7))) {
955 UseAVX = 0;
956 }
957
958 // UseSSE is set to the smaller of what hardware supports and what
959 // the command line requires. I.e., you cannot set UseSSE to 2 on
960 // older Pentiums which do not support it.
961 int use_sse_limit = 0;
962 if (UseSSE > 0) {
963 if (UseSSE > 3 && supports_sse4_1()) {
964 use_sse_limit = 4;
965 } else if (UseSSE > 2 && supports_sse3()) {
966 use_sse_limit = 3;
967 } else if (UseSSE > 1 && supports_sse2()) {
968 use_sse_limit = 2;
969 } else if (UseSSE > 0 && supports_sse()) {
970 use_sse_limit = 1;
971 } else {
972 use_sse_limit = 0;
973 }
974 }
975 if (FLAG_IS_DEFAULT(UseSSE)) {
976 FLAG_SET_DEFAULT(UseSSE, use_sse_limit);
977 } else if (UseSSE > use_sse_limit) {
978 warning("UseSSE=%d is not supported on this CPU, setting it to UseSSE=%d", UseSSE, use_sse_limit);
979 FLAG_SET_DEFAULT(UseSSE, use_sse_limit);
980 }
981
982 // first try initial setting and detect what we can support
983 int use_avx_limit = 0;
984 if (UseAVX > 0) {
985 if (UseSSE < 4) {
986 // Don't use AVX if SSE is unavailable or has been disabled.
987 use_avx_limit = 0;
988 } else if (UseAVX > 2 && supports_evex()) {
989 use_avx_limit = 3;
990 } else if (UseAVX > 1 && supports_avx2()) {
991 use_avx_limit = 2;
992 } else if (UseAVX > 0 && supports_avx()) {
993 use_avx_limit = 1;
994 } else {
995 use_avx_limit = 0;
996 }
997 }
998 if (FLAG_IS_DEFAULT(UseAVX)) {
999 // Don't use AVX-512 on older Skylakes unless explicitly requested.
1000 if (use_avx_limit > 2 && is_intel_skylake() && _stepping < 5) {
1001 FLAG_SET_DEFAULT(UseAVX, 2);
1002 } else {
1003 FLAG_SET_DEFAULT(UseAVX, use_avx_limit);
1004 }
1005 }
1006
1007 if (UseAVX > use_avx_limit) {
1008 if (UseSSE < 4) {
1009 warning("UseAVX=%d requires UseSSE=4, setting it to UseAVX=0", UseAVX);
1010 } else {
1011 warning("UseAVX=%d is not supported on this CPU, setting it to UseAVX=%d", UseAVX, use_avx_limit);
1012 }
1013 FLAG_SET_DEFAULT(UseAVX, use_avx_limit);
1014 }
1015
1016 if (UseAVX < 3) {
1017 _features &= ~CPU_AVX512F;
1018 _features &= ~CPU_AVX512DQ;
1019 _features &= ~CPU_AVX512CD;
1020 _features &= ~CPU_AVX512BW;
1021 _features &= ~CPU_AVX512VL;
1022 _features &= ~CPU_AVX512_VPOPCNTDQ;
1023 _features &= ~CPU_AVX512_VPCLMULQDQ;
1024 _features &= ~CPU_AVX512_VAES;
1025 _features &= ~CPU_AVX512_VNNI;
1026 _features &= ~CPU_AVX512_VBMI;
1027 _features &= ~CPU_AVX512_VBMI2;
1028 _features &= ~CPU_AVX512_BITALG;
1029 _features &= ~CPU_AVX512_IFMA;
1030 _features &= ~CPU_APX_F;
1031 _features &= ~CPU_AVX512_FP16;
1032 }
1033
1034 // Currently APX support is only enabled for targets supporting AVX512VL feature.
1035 bool apx_supported = os_supports_apx_egprs() && supports_apx_f() && supports_avx512vl();
1036 if (UseAPX && !apx_supported) {
1037 warning("UseAPX is not supported on this CPU, setting it to false");
1038 FLAG_SET_DEFAULT(UseAPX, false);
1039 } else if (FLAG_IS_DEFAULT(UseAPX)) {
1040 FLAG_SET_DEFAULT(UseAPX, apx_supported ? true : false);
1041 }
1042
1043 if (!UseAPX) {
1044 _features &= ~CPU_APX_F;
1045 }
1046
1047 if (UseAVX < 2) {
1048 _features &= ~CPU_AVX2;
1049 _features &= ~CPU_AVX_IFMA;
1050 }
1051
1052 if (UseAVX < 1) {
1053 _features &= ~CPU_AVX;
1054 _features &= ~CPU_VZEROUPPER;
1055 _features &= ~CPU_F16C;
1056 _features &= ~CPU_SHA512;
1057 }
1058
1059 if (logical_processors_per_package() == 1) {
1060 // HT processor could be installed on a system which doesn't support HT.
1061 _features &= ~CPU_HT;
1062 }
1063
1064 if (is_intel()) { // Intel cpus specific settings
1065 if (is_knights_family()) {
1066 _features &= ~CPU_VZEROUPPER;
1067 _features &= ~CPU_AVX512BW;
1068 _features &= ~CPU_AVX512VL;
1069 _features &= ~CPU_AVX512DQ;
1070 _features &= ~CPU_AVX512_VNNI;
1071 _features &= ~CPU_AVX512_VAES;
1072 _features &= ~CPU_AVX512_VPOPCNTDQ;
1073 _features &= ~CPU_AVX512_VPCLMULQDQ;
1074 _features &= ~CPU_AVX512_VBMI;
1075 _features &= ~CPU_AVX512_VBMI2;
1076 _features &= ~CPU_CLWB;
1077 _features &= ~CPU_FLUSHOPT;
1078 _features &= ~CPU_GFNI;
1079 _features &= ~CPU_AVX512_BITALG;
1080 _features &= ~CPU_AVX512_IFMA;
1081 _features &= ~CPU_AVX_IFMA;
1082 _features &= ~CPU_AVX512_FP16;
1083 }
1084 }
1085
1086 if (FLAG_IS_DEFAULT(IntelJccErratumMitigation)) {
1087 _has_intel_jcc_erratum = compute_has_intel_jcc_erratum();
1088 } else {
1089 _has_intel_jcc_erratum = IntelJccErratumMitigation;
1090 }
1091
1092 char buf[1024];
1093 int res = jio_snprintf(
1094 buf, sizeof(buf),
1095 "(%u cores per cpu, %u threads per core) family %d model %d stepping %d microcode 0x%x",
1096 cores_per_cpu(), threads_per_core(),
1097 cpu_family(), _model, _stepping, os::cpu_microcode_revision());
1098 assert(res > 0, "not enough temporary space allocated");
1099 insert_features_names(buf + res, sizeof(buf) - res, _features_names);
1100
1101 _features_string = os::strdup(buf);
1102
1103 // Use AES instructions if available.
1104 if (supports_aes()) {
1105 if (FLAG_IS_DEFAULT(UseAES)) {
1106 FLAG_SET_DEFAULT(UseAES, true);
1107 }
1108 if (!UseAES) {
1109 if (UseAESIntrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) {
1110 warning("AES intrinsics require UseAES flag to be enabled. Intrinsics will be disabled.");
1111 }
1112 FLAG_SET_DEFAULT(UseAESIntrinsics, false);
1113 } else {
1114 if (UseSSE > 2) {
1115 if (FLAG_IS_DEFAULT(UseAESIntrinsics)) {
1116 FLAG_SET_DEFAULT(UseAESIntrinsics, true);
1117 }
1118 } else {
1119 // The AES intrinsic stubs require AES instruction support (of course)
1120 // but also require sse3 mode or higher for instructions it use.
1121 if (UseAESIntrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) {
1122 warning("X86 AES intrinsics require SSE3 instructions or higher. Intrinsics will be disabled.");
1123 }
1124 FLAG_SET_DEFAULT(UseAESIntrinsics, false);
1125 }
1126
1127 // --AES-CTR begins--
1128 if (!UseAESIntrinsics) {
1129 if (UseAESCTRIntrinsics && !FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) {
1130 warning("AES-CTR intrinsics require UseAESIntrinsics flag to be enabled. Intrinsics will be disabled.");
1131 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false);
1132 }
1133 } else {
1134 if (supports_sse4_1()) {
1135 if (FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) {
1136 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, true);
1137 }
1138 } else {
1139 // The AES-CTR intrinsic stubs require AES instruction support (of course)
1140 // but also require sse4.1 mode or higher for instructions it use.
1141 if (UseAESCTRIntrinsics && !FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) {
1142 warning("X86 AES-CTR intrinsics require SSE4.1 instructions or higher. Intrinsics will be disabled.");
1143 }
1144 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false);
1145 }
1146 }
1147 // --AES-CTR ends--
1148 }
1149 } else if (UseAES || UseAESIntrinsics || UseAESCTRIntrinsics) {
1150 if (UseAES && !FLAG_IS_DEFAULT(UseAES)) {
1151 warning("AES instructions are not available on this CPU");
1152 FLAG_SET_DEFAULT(UseAES, false);
1153 }
1154 if (UseAESIntrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) {
1155 warning("AES intrinsics are not available on this CPU");
1156 FLAG_SET_DEFAULT(UseAESIntrinsics, false);
1157 }
1158 if (UseAESCTRIntrinsics && !FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) {
1159 warning("AES-CTR intrinsics are not available on this CPU");
1160 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false);
1161 }
1162 }
1163
1164 // Use CLMUL instructions if available.
1165 if (supports_clmul()) {
1166 if (FLAG_IS_DEFAULT(UseCLMUL)) {
1167 UseCLMUL = true;
1168 }
1169 } else if (UseCLMUL) {
1170 if (!FLAG_IS_DEFAULT(UseCLMUL))
1171 warning("CLMUL instructions not available on this CPU (AVX may also be required)");
1172 FLAG_SET_DEFAULT(UseCLMUL, false);
1173 }
1174
1175 if (UseCLMUL && (UseSSE > 2)) {
1176 if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) {
1177 UseCRC32Intrinsics = true;
1178 }
1179 } else if (UseCRC32Intrinsics) {
1180 if (!FLAG_IS_DEFAULT(UseCRC32Intrinsics))
1181 warning("CRC32 Intrinsics requires CLMUL instructions (not available on this CPU)");
1182 FLAG_SET_DEFAULT(UseCRC32Intrinsics, false);
1183 }
1184
1185 #ifdef _LP64
1186 if (supports_avx2()) {
1187 if (FLAG_IS_DEFAULT(UseAdler32Intrinsics)) {
1188 UseAdler32Intrinsics = true;
1189 }
1190 } else if (UseAdler32Intrinsics) {
1191 if (!FLAG_IS_DEFAULT(UseAdler32Intrinsics)) {
1192 warning("Adler32 Intrinsics requires avx2 instructions (not available on this CPU)");
1193 }
1194 FLAG_SET_DEFAULT(UseAdler32Intrinsics, false);
1195 }
1196 #else
1197 if (UseAdler32Intrinsics) {
1198 warning("Adler32Intrinsics not available on this CPU.");
1199 FLAG_SET_DEFAULT(UseAdler32Intrinsics, false);
1200 }
1201 #endif
1202
1203 if (supports_sse4_2() && supports_clmul()) {
1204 if (FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) {
1205 UseCRC32CIntrinsics = true;
1206 }
1207 } else if (UseCRC32CIntrinsics) {
1208 if (!FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) {
1209 warning("CRC32C intrinsics are not available on this CPU");
1210 }
1211 FLAG_SET_DEFAULT(UseCRC32CIntrinsics, false);
1212 }
1213
1214 // GHASH/GCM intrinsics
1215 if (UseCLMUL && (UseSSE > 2)) {
1216 if (FLAG_IS_DEFAULT(UseGHASHIntrinsics)) {
1217 UseGHASHIntrinsics = true;
1218 }
1219 } else if (UseGHASHIntrinsics) {
1220 if (!FLAG_IS_DEFAULT(UseGHASHIntrinsics))
1221 warning("GHASH intrinsic requires CLMUL and SSE2 instructions on this CPU");
1222 FLAG_SET_DEFAULT(UseGHASHIntrinsics, false);
1223 }
1224
1225 #ifdef _LP64
1226 // ChaCha20 Intrinsics
1227 // As long as the system supports AVX as a baseline we can do a
1228 // SIMD-enabled block function. StubGenerator makes the determination
1229 // based on the VM capabilities whether to use an AVX2 or AVX512-enabled
1230 // version.
1231 if (UseAVX >= 1) {
1232 if (FLAG_IS_DEFAULT(UseChaCha20Intrinsics)) {
1233 UseChaCha20Intrinsics = true;
1234 }
1235 } else if (UseChaCha20Intrinsics) {
1236 if (!FLAG_IS_DEFAULT(UseChaCha20Intrinsics)) {
1237 warning("ChaCha20 intrinsic requires AVX instructions");
1238 }
1239 FLAG_SET_DEFAULT(UseChaCha20Intrinsics, false);
1240 }
1241 #else
1242 // No support currently for ChaCha20 intrinsics on 32-bit platforms
1243 if (UseChaCha20Intrinsics) {
1244 warning("ChaCha20 intrinsics are not available on this CPU.");
1245 FLAG_SET_DEFAULT(UseChaCha20Intrinsics, false);
1246 }
1247 #endif // _LP64
1248
1249 // Base64 Intrinsics (Check the condition for which the intrinsic will be active)
1250 if (UseAVX >= 2) {
1251 if (FLAG_IS_DEFAULT(UseBASE64Intrinsics)) {
1252 UseBASE64Intrinsics = true;
1253 }
1254 } else if (UseBASE64Intrinsics) {
1255 if (!FLAG_IS_DEFAULT(UseBASE64Intrinsics))
1256 warning("Base64 intrinsic requires EVEX instructions on this CPU");
1257 FLAG_SET_DEFAULT(UseBASE64Intrinsics, false);
1258 }
1259
1260 if (supports_fma() && UseSSE >= 2) { // Check UseSSE since FMA code uses SSE instructions
1261 if (FLAG_IS_DEFAULT(UseFMA)) {
1262 UseFMA = true;
1263 }
1264 } else if (UseFMA) {
1265 warning("FMA instructions are not available on this CPU");
1266 FLAG_SET_DEFAULT(UseFMA, false);
1267 }
1268
1269 if (FLAG_IS_DEFAULT(UseMD5Intrinsics)) {
1270 UseMD5Intrinsics = true;
1271 }
1272
1273 if (supports_sha() LP64_ONLY(|| (supports_avx2() && supports_bmi2()))) {
1274 if (FLAG_IS_DEFAULT(UseSHA)) {
1275 UseSHA = true;
1276 }
1277 } else if (UseSHA) {
1278 warning("SHA instructions are not available on this CPU");
1279 FLAG_SET_DEFAULT(UseSHA, false);
1280 }
1281
1282 if (supports_sha() && supports_sse4_1() && UseSHA) {
1283 if (FLAG_IS_DEFAULT(UseSHA1Intrinsics)) {
1284 FLAG_SET_DEFAULT(UseSHA1Intrinsics, true);
1285 }
1286 } else if (UseSHA1Intrinsics) {
1287 warning("Intrinsics for SHA-1 crypto hash functions not available on this CPU.");
1288 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false);
1289 }
1290
1291 if (supports_sse4_1() && UseSHA) {
1292 if (FLAG_IS_DEFAULT(UseSHA256Intrinsics)) {
1293 FLAG_SET_DEFAULT(UseSHA256Intrinsics, true);
1294 }
1295 } else if (UseSHA256Intrinsics) {
1296 warning("Intrinsics for SHA-224 and SHA-256 crypto hash functions not available on this CPU.");
1297 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false);
1298 }
1299
1300 #ifdef _LP64
1301 // These are only supported on 64-bit
1302 if (UseSHA && supports_avx2() && (supports_bmi2() || supports_sha512())) {
1303 if (FLAG_IS_DEFAULT(UseSHA512Intrinsics)) {
1304 FLAG_SET_DEFAULT(UseSHA512Intrinsics, true);
1305 }
1306 } else
1307 #endif
1308 if (UseSHA512Intrinsics) {
1309 warning("Intrinsics for SHA-384 and SHA-512 crypto hash functions not available on this CPU.");
1310 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false);
1311 }
1312
1313 #ifdef _LP64
1314 if (supports_evex() && supports_avx512bw()) {
1315 if (FLAG_IS_DEFAULT(UseSHA3Intrinsics)) {
1316 UseSHA3Intrinsics = true;
1317 }
1318 } else
1319 #endif
1320 if (UseSHA3Intrinsics) {
1321 warning("Intrinsics for SHA3-224, SHA3-256, SHA3-384 and SHA3-512 crypto hash functions not available on this CPU.");
1322 FLAG_SET_DEFAULT(UseSHA3Intrinsics, false);
1323 }
1324
1325 if (!(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics)) {
1326 FLAG_SET_DEFAULT(UseSHA, false);
1327 }
1328
1329 #ifdef COMPILER2
1330 if (UseFPUForSpilling) {
1331 if (UseSSE < 2) {
1332 // Only supported with SSE2+
1333 FLAG_SET_DEFAULT(UseFPUForSpilling, false);
1334 }
1335 }
1336 #endif
1337
1338 #if COMPILER2_OR_JVMCI
1339 int max_vector_size = 0;
1340 if (UseSSE < 2) {
1341 // Vectors (in XMM) are only supported with SSE2+
1342 // SSE is always 2 on x64.
1343 max_vector_size = 0;
1344 } else if (UseAVX == 0 || !os_supports_avx_vectors()) {
1345 // 16 byte vectors (in XMM) are supported with SSE2+
1346 max_vector_size = 16;
1347 } else if (UseAVX == 1 || UseAVX == 2) {
1348 // 32 bytes vectors (in YMM) are only supported with AVX+
1349 max_vector_size = 32;
1350 } else if (UseAVX > 2) {
1351 // 64 bytes vectors (in ZMM) are only supported with AVX 3
1352 max_vector_size = 64;
1353 }
1354
1355 #ifdef _LP64
1356 int min_vector_size = 4; // We require MaxVectorSize to be at least 4 on 64bit
1357 #else
1358 int min_vector_size = 0;
1359 #endif
1360
1361 if (!FLAG_IS_DEFAULT(MaxVectorSize)) {
1362 if (MaxVectorSize < min_vector_size) {
1363 warning("MaxVectorSize must be at least %i on this platform", min_vector_size);
1364 FLAG_SET_DEFAULT(MaxVectorSize, min_vector_size);
1365 }
1366 if (MaxVectorSize > max_vector_size) {
1367 warning("MaxVectorSize must be at most %i on this platform", max_vector_size);
1368 FLAG_SET_DEFAULT(MaxVectorSize, max_vector_size);
1369 }
1370 if (!is_power_of_2(MaxVectorSize)) {
1371 warning("MaxVectorSize must be a power of 2, setting to default: %i", max_vector_size);
1372 FLAG_SET_DEFAULT(MaxVectorSize, max_vector_size);
1373 }
1374 } else {
1375 // If default, use highest supported configuration
1376 FLAG_SET_DEFAULT(MaxVectorSize, max_vector_size);
1377 }
1378
1379 #if defined(COMPILER2) && defined(ASSERT)
1380 if (MaxVectorSize > 0) {
1381 if (supports_avx() && PrintMiscellaneous && Verbose && TraceNewVectors) {
1382 tty->print_cr("State of YMM registers after signal handle:");
1383 int nreg = 2 LP64_ONLY(+2);
1384 const char* ymm_name[4] = {"0", "7", "8", "15"};
1385 for (int i = 0; i < nreg; i++) {
1386 tty->print("YMM%s:", ymm_name[i]);
1387 for (int j = 7; j >=0; j--) {
1388 tty->print(" %x", _cpuid_info.ymm_save[i*8 + j]);
1389 }
1390 tty->cr();
1391 }
1392 }
1393 }
1394 #endif // COMPILER2 && ASSERT
1395
1396 #ifdef _LP64
1397 if ((supports_avx512ifma() && supports_avx512vlbw()) || supports_avxifma()) {
1398 if (FLAG_IS_DEFAULT(UsePoly1305Intrinsics)) {
1399 FLAG_SET_DEFAULT(UsePoly1305Intrinsics, true);
1400 }
1401 } else
1402 #endif
1403 if (UsePoly1305Intrinsics) {
1404 warning("Intrinsics for Poly1305 crypto hash functions not available on this CPU.");
1405 FLAG_SET_DEFAULT(UsePoly1305Intrinsics, false);
1406 }
1407
1408 #ifdef _LP64
1409 if ((supports_avx512ifma() && supports_avx512vlbw()) || supports_avxifma()) {
1410 if (FLAG_IS_DEFAULT(UseIntPolyIntrinsics)) {
1411 FLAG_SET_DEFAULT(UseIntPolyIntrinsics, true);
1412 }
1413 } else
1414 #endif
1415 if (UseIntPolyIntrinsics) {
1416 warning("Intrinsics for Polynomial crypto functions not available on this CPU.");
1417 FLAG_SET_DEFAULT(UseIntPolyIntrinsics, false);
1418 }
1419
1420 #ifdef _LP64
1421 if (FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) {
1422 UseMultiplyToLenIntrinsic = true;
1423 }
1424 if (FLAG_IS_DEFAULT(UseSquareToLenIntrinsic)) {
1425 UseSquareToLenIntrinsic = true;
1426 }
1427 if (FLAG_IS_DEFAULT(UseMulAddIntrinsic)) {
1428 UseMulAddIntrinsic = true;
1429 }
1430 if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) {
1431 UseMontgomeryMultiplyIntrinsic = true;
1432 }
1433 if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) {
1434 UseMontgomerySquareIntrinsic = true;
1435 }
1436 #else
1437 if (UseMultiplyToLenIntrinsic) {
1438 if (!FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) {
1439 warning("multiplyToLen intrinsic is not available in 32-bit VM");
1440 }
1441 FLAG_SET_DEFAULT(UseMultiplyToLenIntrinsic, false);
1442 }
1443 if (UseMontgomeryMultiplyIntrinsic) {
1444 if (!FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) {
1445 warning("montgomeryMultiply intrinsic is not available in 32-bit VM");
1446 }
1447 FLAG_SET_DEFAULT(UseMontgomeryMultiplyIntrinsic, false);
1448 }
1449 if (UseMontgomerySquareIntrinsic) {
1450 if (!FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) {
1451 warning("montgomerySquare intrinsic is not available in 32-bit VM");
1452 }
1453 FLAG_SET_DEFAULT(UseMontgomerySquareIntrinsic, false);
1454 }
1455 if (UseSquareToLenIntrinsic) {
1456 if (!FLAG_IS_DEFAULT(UseSquareToLenIntrinsic)) {
1457 warning("squareToLen intrinsic is not available in 32-bit VM");
1458 }
1459 FLAG_SET_DEFAULT(UseSquareToLenIntrinsic, false);
1460 }
1461 if (UseMulAddIntrinsic) {
1462 if (!FLAG_IS_DEFAULT(UseMulAddIntrinsic)) {
1463 warning("mulAdd intrinsic is not available in 32-bit VM");
1464 }
1465 FLAG_SET_DEFAULT(UseMulAddIntrinsic, false);
1466 }
1467 #endif // _LP64
1468 #endif // COMPILER2_OR_JVMCI
1469
1470 // On new cpus instructions which update whole XMM register should be used
1471 // to prevent partial register stall due to dependencies on high half.
1472 //
1473 // UseXmmLoadAndClearUpper == true --> movsd(xmm, mem)
1474 // UseXmmLoadAndClearUpper == false --> movlpd(xmm, mem)
1475 // UseXmmRegToRegMoveAll == true --> movaps(xmm, xmm), movapd(xmm, xmm).
1476 // UseXmmRegToRegMoveAll == false --> movss(xmm, xmm), movsd(xmm, xmm).
1477
1478
1479 if (is_zx()) { // ZX cpus specific settings
1480 if (FLAG_IS_DEFAULT(UseStoreImmI16)) {
1481 UseStoreImmI16 = false; // don't use it on ZX cpus
1482 }
1483 if ((cpu_family() == 6) || (cpu_family() == 7)) {
1484 if (FLAG_IS_DEFAULT(UseAddressNop)) {
1485 // Use it on all ZX cpus
1486 UseAddressNop = true;
1487 }
1488 }
1489 if (FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper)) {
1490 UseXmmLoadAndClearUpper = true; // use movsd on all ZX cpus
1491 }
1492 if (FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll)) {
1493 if (supports_sse3()) {
1494 UseXmmRegToRegMoveAll = true; // use movaps, movapd on new ZX cpus
1495 } else {
1496 UseXmmRegToRegMoveAll = false;
1497 }
1498 }
1499 if (((cpu_family() == 6) || (cpu_family() == 7)) && supports_sse3()) { // new ZX cpus
1500 #ifdef COMPILER2
1501 if (FLAG_IS_DEFAULT(MaxLoopPad)) {
1502 // For new ZX cpus do the next optimization:
1503 // don't align the beginning of a loop if there are enough instructions
1504 // left (NumberOfLoopInstrToAlign defined in c2_globals.hpp)
1505 // in current fetch line (OptoLoopAlignment) or the padding
1506 // is big (> MaxLoopPad).
1507 // Set MaxLoopPad to 11 for new ZX cpus to reduce number of
1508 // generated NOP instructions. 11 is the largest size of one
1509 // address NOP instruction '0F 1F' (see Assembler::nop(i)).
1510 MaxLoopPad = 11;
1511 }
1512 #endif // COMPILER2
1513 if (FLAG_IS_DEFAULT(UseXMMForArrayCopy)) {
1514 UseXMMForArrayCopy = true; // use SSE2 movq on new ZX cpus
1515 }
1516 if (supports_sse4_2()) { // new ZX cpus
1517 if (FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
1518 UseUnalignedLoadStores = true; // use movdqu on newest ZX cpus
1519 }
1520 }
1521 if (supports_sse4_2()) {
1522 if (FLAG_IS_DEFAULT(UseSSE42Intrinsics)) {
1523 FLAG_SET_DEFAULT(UseSSE42Intrinsics, true);
1524 }
1525 } else {
1526 if (UseSSE42Intrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) {
1527 warning("SSE4.2 intrinsics require SSE4.2 instructions or higher. Intrinsics will be disabled.");
1528 }
1529 FLAG_SET_DEFAULT(UseSSE42Intrinsics, false);
1530 }
1531 }
1532
1533 if (FLAG_IS_DEFAULT(AllocatePrefetchInstr) && supports_3dnow_prefetch()) {
1534 FLAG_SET_DEFAULT(AllocatePrefetchInstr, 3);
1535 }
1536 }
1537
1538 if (is_amd_family()) { // AMD cpus specific settings
1539 if (supports_sse2() && FLAG_IS_DEFAULT(UseAddressNop)) {
1540 // Use it on new AMD cpus starting from Opteron.
1541 UseAddressNop = true;
1542 }
1543 if (supports_sse2() && FLAG_IS_DEFAULT(UseNewLongLShift)) {
1544 // Use it on new AMD cpus starting from Opteron.
1545 UseNewLongLShift = true;
1546 }
1547 if (FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper)) {
1548 if (supports_sse4a()) {
1549 UseXmmLoadAndClearUpper = true; // use movsd only on '10h' Opteron
1550 } else {
1551 UseXmmLoadAndClearUpper = false;
1552 }
1553 }
1554 if (FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll)) {
1555 if (supports_sse4a()) {
1556 UseXmmRegToRegMoveAll = true; // use movaps, movapd only on '10h'
1557 } else {
1558 UseXmmRegToRegMoveAll = false;
1559 }
1560 }
1561 if (FLAG_IS_DEFAULT(UseXmmI2F)) {
1562 if (supports_sse4a()) {
1563 UseXmmI2F = true;
1564 } else {
1565 UseXmmI2F = false;
1566 }
1567 }
1568 if (FLAG_IS_DEFAULT(UseXmmI2D)) {
1569 if (supports_sse4a()) {
1570 UseXmmI2D = true;
1571 } else {
1572 UseXmmI2D = false;
1573 }
1574 }
1575 if (supports_sse4_2()) {
1576 if (FLAG_IS_DEFAULT(UseSSE42Intrinsics)) {
1577 FLAG_SET_DEFAULT(UseSSE42Intrinsics, true);
1578 }
1579 } else {
1580 if (UseSSE42Intrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) {
1581 warning("SSE4.2 intrinsics require SSE4.2 instructions or higher. Intrinsics will be disabled.");
1582 }
1583 FLAG_SET_DEFAULT(UseSSE42Intrinsics, false);
1584 }
1585
1586 // some defaults for AMD family 15h
1587 if (cpu_family() == 0x15) {
1588 // On family 15h processors default is no sw prefetch
1589 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) {
1590 FLAG_SET_DEFAULT(AllocatePrefetchStyle, 0);
1591 }
1592 // Also, if some other prefetch style is specified, default instruction type is PREFETCHW
1593 if (FLAG_IS_DEFAULT(AllocatePrefetchInstr)) {
1594 FLAG_SET_DEFAULT(AllocatePrefetchInstr, 3);
1595 }
1596 // On family 15h processors use XMM and UnalignedLoadStores for Array Copy
1597 if (supports_sse2() && FLAG_IS_DEFAULT(UseXMMForArrayCopy)) {
1598 FLAG_SET_DEFAULT(UseXMMForArrayCopy, true);
1599 }
1600 if (supports_sse2() && FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
1601 FLAG_SET_DEFAULT(UseUnalignedLoadStores, true);
1602 }
1603 }
1604
1605 #ifdef COMPILER2
1606 if (cpu_family() < 0x17 && MaxVectorSize > 16) {
1607 // Limit vectors size to 16 bytes on AMD cpus < 17h.
1608 FLAG_SET_DEFAULT(MaxVectorSize, 16);
1609 }
1610 #endif // COMPILER2
1611
1612 // Some defaults for AMD family >= 17h && Hygon family 18h
1613 if (cpu_family() >= 0x17) {
1614 // On family >=17h processors use XMM and UnalignedLoadStores
1615 // for Array Copy
1616 if (supports_sse2() && FLAG_IS_DEFAULT(UseXMMForArrayCopy)) {
1617 FLAG_SET_DEFAULT(UseXMMForArrayCopy, true);
1618 }
1619 if (supports_sse2() && FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
1620 FLAG_SET_DEFAULT(UseUnalignedLoadStores, true);
1621 }
1622 #ifdef COMPILER2
1623 if (supports_sse4_2() && FLAG_IS_DEFAULT(UseFPUForSpilling)) {
1624 FLAG_SET_DEFAULT(UseFPUForSpilling, true);
1625 }
1626 #endif
1627 }
1628 }
1629
1630 if (is_intel()) { // Intel cpus specific settings
1631 if (FLAG_IS_DEFAULT(UseStoreImmI16)) {
1632 UseStoreImmI16 = false; // don't use it on Intel cpus
1633 }
1634 if (cpu_family() == 6 || cpu_family() == 15) {
1635 if (FLAG_IS_DEFAULT(UseAddressNop)) {
1636 // Use it on all Intel cpus starting from PentiumPro
1637 UseAddressNop = true;
1638 }
1639 }
1640 if (FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper)) {
1641 UseXmmLoadAndClearUpper = true; // use movsd on all Intel cpus
1642 }
1643 if (FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll)) {
1644 if (supports_sse3()) {
1645 UseXmmRegToRegMoveAll = true; // use movaps, movapd on new Intel cpus
1646 } else {
1647 UseXmmRegToRegMoveAll = false;
1648 }
1649 }
1650 if (cpu_family() == 6 && supports_sse3()) { // New Intel cpus
1651 #ifdef COMPILER2
1652 if (FLAG_IS_DEFAULT(MaxLoopPad)) {
1653 // For new Intel cpus do the next optimization:
1654 // don't align the beginning of a loop if there are enough instructions
1655 // left (NumberOfLoopInstrToAlign defined in c2_globals.hpp)
1656 // in current fetch line (OptoLoopAlignment) or the padding
1657 // is big (> MaxLoopPad).
1658 // Set MaxLoopPad to 11 for new Intel cpus to reduce number of
1659 // generated NOP instructions. 11 is the largest size of one
1660 // address NOP instruction '0F 1F' (see Assembler::nop(i)).
1661 MaxLoopPad = 11;
1662 }
1663 #endif // COMPILER2
1664
1665 if (FLAG_IS_DEFAULT(UseXMMForArrayCopy)) {
1666 UseXMMForArrayCopy = true; // use SSE2 movq on new Intel cpus
1667 }
1668 if ((supports_sse4_2() && supports_ht()) || supports_avx()) { // Newest Intel cpus
1669 if (FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
1670 UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus
1671 }
1672 }
1673 if (supports_sse4_2()) {
1674 if (FLAG_IS_DEFAULT(UseSSE42Intrinsics)) {
1675 FLAG_SET_DEFAULT(UseSSE42Intrinsics, true);
1676 }
1677 } else {
1678 if (UseSSE42Intrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) {
1679 warning("SSE4.2 intrinsics require SSE4.2 instructions or higher. Intrinsics will be disabled.");
1680 }
1681 FLAG_SET_DEFAULT(UseSSE42Intrinsics, false);
1682 }
1683 }
1684 if (is_atom_family() || is_knights_family()) {
1685 #ifdef COMPILER2
1686 if (FLAG_IS_DEFAULT(OptoScheduling)) {
1687 OptoScheduling = true;
1688 }
1689 #endif
1690 if (supports_sse4_2()) { // Silvermont
1691 if (FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
1692 UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus
1693 }
1694 }
1695 if (FLAG_IS_DEFAULT(UseIncDec)) {
1696 FLAG_SET_DEFAULT(UseIncDec, false);
1697 }
1698 }
1699 if (FLAG_IS_DEFAULT(AllocatePrefetchInstr) && supports_3dnow_prefetch()) {
1700 FLAG_SET_DEFAULT(AllocatePrefetchInstr, 3);
1701 }
1702 #ifdef COMPILER2
1703 if (UseAVX > 2) {
1704 if (FLAG_IS_DEFAULT(ArrayOperationPartialInlineSize) ||
1705 (!FLAG_IS_DEFAULT(ArrayOperationPartialInlineSize) &&
1706 ArrayOperationPartialInlineSize != 0 &&
1707 ArrayOperationPartialInlineSize != 16 &&
1708 ArrayOperationPartialInlineSize != 32 &&
1709 ArrayOperationPartialInlineSize != 64)) {
1710 int inline_size = 0;
1711 if (MaxVectorSize >= 64 && AVX3Threshold == 0) {
1712 inline_size = 64;
1713 } else if (MaxVectorSize >= 32) {
1714 inline_size = 32;
1715 } else if (MaxVectorSize >= 16) {
1716 inline_size = 16;
1717 }
1718 if(!FLAG_IS_DEFAULT(ArrayOperationPartialInlineSize)) {
1719 warning("Setting ArrayOperationPartialInlineSize as %d", inline_size);
1720 }
1721 ArrayOperationPartialInlineSize = inline_size;
1722 }
1723
1724 if (ArrayOperationPartialInlineSize > MaxVectorSize) {
1725 ArrayOperationPartialInlineSize = MaxVectorSize >= 16 ? MaxVectorSize : 0;
1726 if (ArrayOperationPartialInlineSize) {
1727 warning("Setting ArrayOperationPartialInlineSize as MaxVectorSize=%zd", MaxVectorSize);
1728 } else {
1729 warning("Setting ArrayOperationPartialInlineSize as %zd", ArrayOperationPartialInlineSize);
1730 }
1731 }
1732 }
1733 #endif
1734 }
1735
1736 #ifdef COMPILER2
1737 if (FLAG_IS_DEFAULT(OptimizeFill)) {
1738 if (MaxVectorSize < 32 || !VM_Version::supports_avx512vlbw()) {
1739 OptimizeFill = false;
1740 }
1741 }
1742 #endif
1743
1744 #ifdef _LP64
1745 if (UseSSE42Intrinsics) {
1746 if (FLAG_IS_DEFAULT(UseVectorizedMismatchIntrinsic)) {
1747 UseVectorizedMismatchIntrinsic = true;
1748 }
1749 } else if (UseVectorizedMismatchIntrinsic) {
1750 if (!FLAG_IS_DEFAULT(UseVectorizedMismatchIntrinsic))
1751 warning("vectorizedMismatch intrinsics are not available on this CPU");
1752 FLAG_SET_DEFAULT(UseVectorizedMismatchIntrinsic, false);
1753 }
1754 if (UseAVX >= 2) {
1755 FLAG_SET_DEFAULT(UseVectorizedHashCodeIntrinsic, true);
1756 } else if (UseVectorizedHashCodeIntrinsic) {
1757 if (!FLAG_IS_DEFAULT(UseVectorizedHashCodeIntrinsic))
1758 warning("vectorizedHashCode intrinsics are not available on this CPU");
1759 FLAG_SET_DEFAULT(UseVectorizedHashCodeIntrinsic, false);
1760 }
1761 #else
1762 if (UseVectorizedMismatchIntrinsic) {
1763 if (!FLAG_IS_DEFAULT(UseVectorizedMismatchIntrinsic)) {
1764 warning("vectorizedMismatch intrinsic is not available in 32-bit VM");
1765 }
1766 FLAG_SET_DEFAULT(UseVectorizedMismatchIntrinsic, false);
1767 }
1768 if (UseVectorizedHashCodeIntrinsic) {
1769 if (!FLAG_IS_DEFAULT(UseVectorizedHashCodeIntrinsic)) {
1770 warning("vectorizedHashCode intrinsic is not available in 32-bit VM");
1771 }
1772 FLAG_SET_DEFAULT(UseVectorizedHashCodeIntrinsic, false);
1773 }
1774 #endif // _LP64
1775
1776 // Use count leading zeros count instruction if available.
1777 if (supports_lzcnt()) {
1778 if (FLAG_IS_DEFAULT(UseCountLeadingZerosInstruction)) {
1779 UseCountLeadingZerosInstruction = true;
1780 }
1781 } else if (UseCountLeadingZerosInstruction) {
1782 warning("lzcnt instruction is not available on this CPU");
1783 FLAG_SET_DEFAULT(UseCountLeadingZerosInstruction, false);
1784 }
1785
1786 // Use count trailing zeros instruction if available
1787 if (supports_bmi1()) {
1788 // tzcnt does not require VEX prefix
1789 if (FLAG_IS_DEFAULT(UseCountTrailingZerosInstruction)) {
1790 if (!UseBMI1Instructions && !FLAG_IS_DEFAULT(UseBMI1Instructions)) {
1791 // Don't use tzcnt if BMI1 is switched off on command line.
1792 UseCountTrailingZerosInstruction = false;
1793 } else {
1794 UseCountTrailingZerosInstruction = true;
1795 }
1796 }
1797 } else if (UseCountTrailingZerosInstruction) {
1798 warning("tzcnt instruction is not available on this CPU");
1799 FLAG_SET_DEFAULT(UseCountTrailingZerosInstruction, false);
1800 }
1801
1802 // BMI instructions (except tzcnt) use an encoding with VEX prefix.
1803 // VEX prefix is generated only when AVX > 0.
1804 if (supports_bmi1() && supports_avx()) {
1805 if (FLAG_IS_DEFAULT(UseBMI1Instructions)) {
1806 UseBMI1Instructions = true;
1807 }
1808 } else if (UseBMI1Instructions) {
1809 warning("BMI1 instructions are not available on this CPU (AVX is also required)");
1810 FLAG_SET_DEFAULT(UseBMI1Instructions, false);
1811 }
1812
1813 if (supports_bmi2() && supports_avx()) {
1814 if (FLAG_IS_DEFAULT(UseBMI2Instructions)) {
1815 UseBMI2Instructions = true;
1816 }
1817 } else if (UseBMI2Instructions) {
1818 warning("BMI2 instructions are not available on this CPU (AVX is also required)");
1819 FLAG_SET_DEFAULT(UseBMI2Instructions, false);
1820 }
1821
1822 // Use population count instruction if available.
1823 if (supports_popcnt()) {
1824 if (FLAG_IS_DEFAULT(UsePopCountInstruction)) {
1825 UsePopCountInstruction = true;
1826 }
1827 } else if (UsePopCountInstruction) {
1828 warning("POPCNT instruction is not available on this CPU");
1829 FLAG_SET_DEFAULT(UsePopCountInstruction, false);
1830 }
1831
1832 // Use fast-string operations if available.
1833 if (supports_erms()) {
1834 if (FLAG_IS_DEFAULT(UseFastStosb)) {
1835 UseFastStosb = true;
1836 }
1837 } else if (UseFastStosb) {
1838 warning("fast-string operations are not available on this CPU");
1839 FLAG_SET_DEFAULT(UseFastStosb, false);
1840 }
1841
1842 // For AMD Processors use XMM/YMM MOVDQU instructions
1843 // for Object Initialization as default
1844 if (is_amd() && cpu_family() >= 0x19) {
1845 if (FLAG_IS_DEFAULT(UseFastStosb)) {
1846 UseFastStosb = false;
1847 }
1848 }
1849
1850 #ifdef COMPILER2
1851 if (is_intel() && MaxVectorSize > 16) {
1852 if (FLAG_IS_DEFAULT(UseFastStosb)) {
1853 UseFastStosb = false;
1854 }
1855 }
1856 #endif
1857
1858 // Use XMM/YMM MOVDQU instruction for Object Initialization
1859 if (!UseFastStosb && UseSSE >= 2 && UseUnalignedLoadStores) {
1860 if (FLAG_IS_DEFAULT(UseXMMForObjInit)) {
1861 UseXMMForObjInit = true;
1862 }
1863 } else if (UseXMMForObjInit) {
1864 warning("UseXMMForObjInit requires SSE2 and unaligned load/stores. Feature is switched off.");
1865 FLAG_SET_DEFAULT(UseXMMForObjInit, false);
1866 }
1867
1868 #ifdef COMPILER2
1869 if (FLAG_IS_DEFAULT(AlignVector)) {
1870 // Modern processors allow misaligned memory operations for vectors.
1871 AlignVector = !UseUnalignedLoadStores;
1872 }
1873 #endif // COMPILER2
1874
1875 if (FLAG_IS_DEFAULT(AllocatePrefetchInstr)) {
1876 if (AllocatePrefetchInstr == 3 && !supports_3dnow_prefetch()) {
1877 FLAG_SET_DEFAULT(AllocatePrefetchInstr, 0);
1878 } else if (!supports_sse() && supports_3dnow_prefetch()) {
1879 FLAG_SET_DEFAULT(AllocatePrefetchInstr, 3);
1880 }
1881 }
1882
1883 // Allocation prefetch settings
1884 int cache_line_size = checked_cast<int>(prefetch_data_size());
1885 if (FLAG_IS_DEFAULT(AllocatePrefetchStepSize) &&
1886 (cache_line_size > AllocatePrefetchStepSize)) {
1887 FLAG_SET_DEFAULT(AllocatePrefetchStepSize, cache_line_size);
1888 }
1889
1890 if ((AllocatePrefetchDistance == 0) && (AllocatePrefetchStyle != 0)) {
1891 assert(!FLAG_IS_DEFAULT(AllocatePrefetchDistance), "default value should not be 0");
1892 if (!FLAG_IS_DEFAULT(AllocatePrefetchStyle)) {
1893 warning("AllocatePrefetchDistance is set to 0 which disable prefetching. Ignoring AllocatePrefetchStyle flag.");
1894 }
1895 FLAG_SET_DEFAULT(AllocatePrefetchStyle, 0);
1896 }
1897
1898 if (FLAG_IS_DEFAULT(AllocatePrefetchDistance)) {
1899 bool use_watermark_prefetch = (AllocatePrefetchStyle == 2);
1900 FLAG_SET_DEFAULT(AllocatePrefetchDistance, allocate_prefetch_distance(use_watermark_prefetch));
1901 }
1902
1903 if (is_intel() && cpu_family() == 6 && supports_sse3()) {
1904 if (FLAG_IS_DEFAULT(AllocatePrefetchLines) &&
1905 supports_sse4_2() && supports_ht()) { // Nehalem based cpus
1906 FLAG_SET_DEFAULT(AllocatePrefetchLines, 4);
1907 }
1908 #ifdef COMPILER2
1909 if (FLAG_IS_DEFAULT(UseFPUForSpilling) && supports_sse4_2()) {
1910 FLAG_SET_DEFAULT(UseFPUForSpilling, true);
1911 }
1912 #endif
1913 }
1914
1915 if (is_zx() && ((cpu_family() == 6) || (cpu_family() == 7)) && supports_sse4_2()) {
1916 #ifdef COMPILER2
1917 if (FLAG_IS_DEFAULT(UseFPUForSpilling)) {
1918 FLAG_SET_DEFAULT(UseFPUForSpilling, true);
1919 }
1920 #endif
1921 }
1922
1923 #ifdef _LP64
1924 // Prefetch settings
1925
1926 // Prefetch interval for gc copy/scan == 9 dcache lines. Derived from
1927 // 50-warehouse specjbb runs on a 2-way 1.8ghz opteron using a 4gb heap.
1928 // Tested intervals from 128 to 2048 in increments of 64 == one cache line.
1929 // 256 bytes (4 dcache lines) was the nearest runner-up to 576.
1930
1931 // gc copy/scan is disabled if prefetchw isn't supported, because
1932 // Prefetch::write emits an inlined prefetchw on Linux.
1933 // Do not use the 3dnow prefetchw instruction. It isn't supported on em64t.
1934 // The used prefetcht0 instruction works for both amd64 and em64t.
1935
1936 if (FLAG_IS_DEFAULT(PrefetchCopyIntervalInBytes)) {
1937 FLAG_SET_DEFAULT(PrefetchCopyIntervalInBytes, 576);
1938 }
1939 if (FLAG_IS_DEFAULT(PrefetchScanIntervalInBytes)) {
1940 FLAG_SET_DEFAULT(PrefetchScanIntervalInBytes, 576);
1941 }
1942 #endif
1943
1944 if (FLAG_IS_DEFAULT(ContendedPaddingWidth) &&
1945 (cache_line_size > ContendedPaddingWidth))
1946 ContendedPaddingWidth = cache_line_size;
1947
1948 // This machine allows unaligned memory accesses
1949 if (FLAG_IS_DEFAULT(UseUnalignedAccesses)) {
1950 FLAG_SET_DEFAULT(UseUnalignedAccesses, true);
1951 }
1952
1953 #ifndef PRODUCT
1954 if (log_is_enabled(Info, os, cpu)) {
1955 LogStream ls(Log(os, cpu)::info());
1956 outputStream* log = &ls;
1957 log->print_cr("Logical CPUs per core: %u",
1958 logical_processors_per_package());
1959 log->print_cr("L1 data cache line size: %u", L1_data_cache_line_size());
1960 log->print("UseSSE=%d", UseSSE);
1961 if (UseAVX > 0) {
1962 log->print(" UseAVX=%d", UseAVX);
1963 }
1964 if (UseAES) {
1965 log->print(" UseAES=1");
1966 }
1967 #ifdef COMPILER2
1968 if (MaxVectorSize > 0) {
1969 log->print(" MaxVectorSize=%d", (int) MaxVectorSize);
1970 }
1971 #endif
1972 log->cr();
1973 log->print("Allocation");
1974 if (AllocatePrefetchStyle <= 0 || (UseSSE == 0 && !supports_3dnow_prefetch())) {
1975 log->print_cr(": no prefetching");
1976 } else {
1977 log->print(" prefetching: ");
1978 if (UseSSE == 0 && supports_3dnow_prefetch()) {
1979 log->print("PREFETCHW");
1980 } else if (UseSSE >= 1) {
1981 if (AllocatePrefetchInstr == 0) {
1982 log->print("PREFETCHNTA");
1983 } else if (AllocatePrefetchInstr == 1) {
1984 log->print("PREFETCHT0");
1985 } else if (AllocatePrefetchInstr == 2) {
1986 log->print("PREFETCHT2");
1987 } else if (AllocatePrefetchInstr == 3) {
1988 log->print("PREFETCHW");
1989 }
1990 }
1991 if (AllocatePrefetchLines > 1) {
1992 log->print_cr(" at distance %d, %d lines of %d bytes", AllocatePrefetchDistance, AllocatePrefetchLines, AllocatePrefetchStepSize);
1993 } else {
1994 log->print_cr(" at distance %d, one line of %d bytes", AllocatePrefetchDistance, AllocatePrefetchStepSize);
1995 }
1996 }
1997
1998 if (PrefetchCopyIntervalInBytes > 0) {
1999 log->print_cr("PrefetchCopyIntervalInBytes %d", (int) PrefetchCopyIntervalInBytes);
2000 }
2001 if (PrefetchScanIntervalInBytes > 0) {
2002 log->print_cr("PrefetchScanIntervalInBytes %d", (int) PrefetchScanIntervalInBytes);
2003 }
2004 if (ContendedPaddingWidth > 0) {
2005 log->print_cr("ContendedPaddingWidth %d", (int) ContendedPaddingWidth);
2006 }
2007 }
2008 #endif // !PRODUCT
2009 if (FLAG_IS_DEFAULT(UseSignumIntrinsic)) {
2010 FLAG_SET_DEFAULT(UseSignumIntrinsic, true);
2011 }
2012 if (FLAG_IS_DEFAULT(UseCopySignIntrinsic)) {
2013 FLAG_SET_DEFAULT(UseCopySignIntrinsic, true);
2014 }
2015 }
2016
2017 void VM_Version::print_platform_virtualization_info(outputStream* st) {
2018 VirtualizationType vrt = VM_Version::get_detected_virtualization();
2019 if (vrt == XenHVM) {
2020 st->print_cr("Xen hardware-assisted virtualization detected");
2021 } else if (vrt == KVM) {
2022 st->print_cr("KVM virtualization detected");
2023 } else if (vrt == VMWare) {
2024 st->print_cr("VMWare virtualization detected");
2025 VirtualizationSupport::print_virtualization_info(st);
2026 } else if (vrt == HyperV) {
2027 st->print_cr("Hyper-V virtualization detected");
2028 } else if (vrt == HyperVRole) {
2029 st->print_cr("Hyper-V role detected");
2030 }
2031 }
2032
2033 bool VM_Version::compute_has_intel_jcc_erratum() {
2034 if (!is_intel_family_core()) {
2035 // Only Intel CPUs are affected.
2036 return false;
2037 }
2038 // The following table of affected CPUs is based on the following document released by Intel:
2039 // https://www.intel.com/content/dam/support/us/en/documents/processors/mitigations-jump-conditional-code-erratum.pdf
2040 switch (_model) {
2041 case 0x8E:
2042 // 06_8EH | 9 | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Amber Lake Y
2043 // 06_8EH | 9 | 7th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Kaby Lake U
2044 // 06_8EH | 9 | 7th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Kaby Lake U 23e
2045 // 06_8EH | 9 | 7th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Kaby Lake Y
2046 // 06_8EH | A | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Coffee Lake U43e
2047 // 06_8EH | B | 8th Generation Intel(R) Core(TM) Processors based on microarchitecture code name Whiskey Lake U
2048 // 06_8EH | C | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Amber Lake Y
2049 // 06_8EH | C | 10th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Comet Lake U42
2050 // 06_8EH | C | 8th Generation Intel(R) Core(TM) Processors based on microarchitecture code name Whiskey Lake U
2051 return _stepping == 0x9 || _stepping == 0xA || _stepping == 0xB || _stepping == 0xC;
2052 case 0x4E:
2053 // 06_4E | 3 | 6th Generation Intel(R) Core(TM) Processors based on microarchitecture code name Skylake U
2054 // 06_4E | 3 | 6th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Skylake U23e
2055 // 06_4E | 3 | 6th Generation Intel(R) Core(TM) Processors based on microarchitecture code name Skylake Y
2056 return _stepping == 0x3;
2057 case 0x55:
2058 // 06_55H | 4 | Intel(R) Xeon(R) Processor D Family based on microarchitecture code name Skylake D, Bakerville
2059 // 06_55H | 4 | Intel(R) Xeon(R) Scalable Processors based on microarchitecture code name Skylake Server
2060 // 06_55H | 4 | Intel(R) Xeon(R) Processor W Family based on microarchitecture code name Skylake W
2061 // 06_55H | 4 | Intel(R) Core(TM) X-series Processors based on microarchitecture code name Skylake X
2062 // 06_55H | 4 | Intel(R) Xeon(R) Processor E3 v5 Family based on microarchitecture code name Skylake Xeon E3
2063 // 06_55 | 7 | 2nd Generation Intel(R) Xeon(R) Scalable Processors based on microarchitecture code name Cascade Lake (server)
2064 return _stepping == 0x4 || _stepping == 0x7;
2065 case 0x5E:
2066 // 06_5E | 3 | 6th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Skylake H
2067 // 06_5E | 3 | 6th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Skylake S
2068 return _stepping == 0x3;
2069 case 0x9E:
2070 // 06_9EH | 9 | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Kaby Lake G
2071 // 06_9EH | 9 | 7th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Kaby Lake H
2072 // 06_9EH | 9 | 7th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Kaby Lake S
2073 // 06_9EH | 9 | Intel(R) Core(TM) X-series Processors based on microarchitecture code name Kaby Lake X
2074 // 06_9EH | 9 | Intel(R) Xeon(R) Processor E3 v6 Family Kaby Lake Xeon E3
2075 // 06_9EH | A | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Coffee Lake H
2076 // 06_9EH | A | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Coffee Lake S
2077 // 06_9EH | A | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Coffee Lake S (6+2) x/KBP
2078 // 06_9EH | A | Intel(R) Xeon(R) Processor E Family based on microarchitecture code name Coffee Lake S (6+2)
2079 // 06_9EH | A | Intel(R) Xeon(R) Processor E Family based on microarchitecture code name Coffee Lake S (4+2)
2080 // 06_9EH | B | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Coffee Lake S (4+2)
2081 // 06_9EH | B | Intel(R) Celeron(R) Processor G Series based on microarchitecture code name Coffee Lake S (4+2)
2082 // 06_9EH | D | 9th Generation Intel(R) Core(TM) Processor Family based on microarchitecturecode name Coffee Lake H (8+2)
2083 // 06_9EH | D | 9th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Coffee Lake S (8+2)
2084 return _stepping == 0x9 || _stepping == 0xA || _stepping == 0xB || _stepping == 0xD;
2085 case 0xA5:
2086 // Not in Intel documentation.
2087 // 06_A5H | | 10th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Comet Lake S/H
2088 return true;
2089 case 0xA6:
2090 // 06_A6H | 0 | 10th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Comet Lake U62
2091 return _stepping == 0x0;
2092 case 0xAE:
2093 // 06_AEH | A | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Kaby Lake Refresh U (4+2)
2094 return _stepping == 0xA;
2095 default:
2096 // If we are running on another intel machine not recognized in the table, we are okay.
2097 return false;
2098 }
2099 }
2100
2101 // On Xen, the cpuid instruction returns
2102 // eax / registers[0]: Version of Xen
2103 // ebx / registers[1]: chars 'XenV'
2104 // ecx / registers[2]: chars 'MMXe'
2105 // edx / registers[3]: chars 'nVMM'
2106 //
2107 // On KVM / VMWare / MS Hyper-V, the cpuid instruction returns
2108 // ebx / registers[1]: chars 'KVMK' / 'VMwa' / 'Micr'
2109 // ecx / registers[2]: chars 'VMKV' / 'reVM' / 'osof'
2110 // edx / registers[3]: chars 'M' / 'ware' / 't Hv'
2111 //
2112 // more information :
2113 // https://kb.vmware.com/s/article/1009458
2114 //
2115 void VM_Version::check_virtualizations() {
2116 uint32_t registers[4] = {0};
2117 char signature[13] = {0};
2118
2119 // Xen cpuid leaves can be found 0x100 aligned boundary starting
2120 // from 0x40000000 until 0x40010000.
2121 // https://lists.linuxfoundation.org/pipermail/virtualization/2012-May/019974.html
2122 for (int leaf = 0x40000000; leaf < 0x40010000; leaf += 0x100) {
2123 detect_virt_stub(leaf, registers);
2124 memcpy(signature, ®isters[1], 12);
2125
2126 if (strncmp("VMwareVMware", signature, 12) == 0) {
2127 Abstract_VM_Version::_detected_virtualization = VMWare;
2128 // check for extended metrics from guestlib
2129 VirtualizationSupport::initialize();
2130 } else if (strncmp("Microsoft Hv", signature, 12) == 0) {
2131 Abstract_VM_Version::_detected_virtualization = HyperV;
2132 #ifdef _WINDOWS
2133 // CPUID leaf 0x40000007 is available to the root partition only.
2134 // See Hypervisor Top Level Functional Specification section 2.4.8 for more details.
2135 // https://github.com/MicrosoftDocs/Virtualization-Documentation/raw/master/tlfs/Hypervisor%20Top%20Level%20Functional%20Specification%20v6.0b.pdf
2136 detect_virt_stub(0x40000007, registers);
2137 if ((registers[0] != 0x0) ||
2138 (registers[1] != 0x0) ||
2139 (registers[2] != 0x0) ||
2140 (registers[3] != 0x0)) {
2141 Abstract_VM_Version::_detected_virtualization = HyperVRole;
2142 }
2143 #endif
2144 } else if (strncmp("KVMKVMKVM", signature, 9) == 0) {
2145 Abstract_VM_Version::_detected_virtualization = KVM;
2146 } else if (strncmp("XenVMMXenVMM", signature, 12) == 0) {
2147 Abstract_VM_Version::_detected_virtualization = XenHVM;
2148 }
2149 }
2150 }
2151
2152 #ifdef COMPILER2
2153 // Determine if it's running on Cascade Lake using default options.
2154 bool VM_Version::is_default_intel_cascade_lake() {
2155 return FLAG_IS_DEFAULT(UseAVX) &&
2156 FLAG_IS_DEFAULT(MaxVectorSize) &&
2157 UseAVX > 2 &&
2158 is_intel_cascade_lake();
2159 }
2160 #endif
2161
2162 bool VM_Version::is_intel_cascade_lake() {
2163 return is_intel_skylake() && _stepping >= 5;
2164 }
2165
2166 // avx3_threshold() sets the threshold at which 64-byte instructions are used
2167 // for implementing the array copy and clear operations.
2168 // The Intel platforms that supports the serialize instruction
2169 // has improved implementation of 64-byte load/stores and so the default
2170 // threshold is set to 0 for these platforms.
2171 int VM_Version::avx3_threshold() {
2172 return (is_intel_family_core() &&
2173 supports_serialize() &&
2174 FLAG_IS_DEFAULT(AVX3Threshold)) ? 0 : AVX3Threshold;
2175 }
2176
2177 #if defined(_LP64)
2178 void VM_Version::clear_apx_test_state() {
2179 clear_apx_test_state_stub();
2180 }
2181 #endif
2182
2183 static bool _vm_version_initialized = false;
2184
2185 void VM_Version::initialize() {
2186 ResourceMark rm;
2187 // Making this stub must be FIRST use of assembler
2188 stub_blob = BufferBlob::create("VM_Version stub", stub_size);
2189 if (stub_blob == nullptr) {
2190 vm_exit_during_initialization("Unable to allocate stub for VM_Version");
2191 }
2192 CodeBuffer c(stub_blob);
2193 VM_Version_StubGenerator g(&c);
2194
2195 get_cpu_info_stub = CAST_TO_FN_PTR(get_cpu_info_stub_t,
2196 g.generate_get_cpu_info());
2197 detect_virt_stub = CAST_TO_FN_PTR(detect_virt_stub_t,
2198 g.generate_detect_virt());
2199
2200 #if defined(_LP64)
2201 clear_apx_test_state_stub = CAST_TO_FN_PTR(clear_apx_test_state_t,
2202 g.clear_apx_test_state());
2203 #endif
2204 get_processor_features();
2205
2206 LP64_ONLY(Assembler::precompute_instructions();)
2207
2208 if (VM_Version::supports_hv()) { // Supports hypervisor
2209 check_virtualizations();
2210 }
2211 _vm_version_initialized = true;
2212 }
2213
2214 typedef enum {
2215 CPU_FAMILY_8086_8088 = 0,
2216 CPU_FAMILY_INTEL_286 = 2,
2217 CPU_FAMILY_INTEL_386 = 3,
2218 CPU_FAMILY_INTEL_486 = 4,
2219 CPU_FAMILY_PENTIUM = 5,
2220 CPU_FAMILY_PENTIUMPRO = 6, // Same family several models
2221 CPU_FAMILY_PENTIUM_4 = 0xF
2222 } FamilyFlag;
2223
2224 typedef enum {
2225 RDTSCP_FLAG = 0x08000000, // bit 27
2226 INTEL64_FLAG = 0x20000000 // bit 29
2227 } _featureExtendedEdxFlag;
2228
2229 typedef enum {
2230 FPU_FLAG = 0x00000001,
2231 VME_FLAG = 0x00000002,
2232 DE_FLAG = 0x00000004,
2233 PSE_FLAG = 0x00000008,
2234 TSC_FLAG = 0x00000010,
2235 MSR_FLAG = 0x00000020,
2236 PAE_FLAG = 0x00000040,
2237 MCE_FLAG = 0x00000080,
2238 CX8_FLAG = 0x00000100,
2239 APIC_FLAG = 0x00000200,
2240 SEP_FLAG = 0x00000800,
2241 MTRR_FLAG = 0x00001000,
2242 PGE_FLAG = 0x00002000,
2243 MCA_FLAG = 0x00004000,
2244 CMOV_FLAG = 0x00008000,
2245 PAT_FLAG = 0x00010000,
2246 PSE36_FLAG = 0x00020000,
2247 PSNUM_FLAG = 0x00040000,
2248 CLFLUSH_FLAG = 0x00080000,
2249 DTS_FLAG = 0x00200000,
2250 ACPI_FLAG = 0x00400000,
2251 MMX_FLAG = 0x00800000,
2252 FXSR_FLAG = 0x01000000,
2253 SSE_FLAG = 0x02000000,
2254 SSE2_FLAG = 0x04000000,
2255 SS_FLAG = 0x08000000,
2256 HTT_FLAG = 0x10000000,
2257 TM_FLAG = 0x20000000
2258 } FeatureEdxFlag;
2259
2260 static BufferBlob* cpuid_brand_string_stub_blob;
2261 static const int cpuid_brand_string_stub_size = 550;
2262
2263 extern "C" {
2264 typedef void (*getCPUIDBrandString_stub_t)(void*);
2265 }
2266
2267 static getCPUIDBrandString_stub_t getCPUIDBrandString_stub = nullptr;
2268
2269 // VM_Version statics
2270 enum {
2271 ExtendedFamilyIdLength_INTEL = 16,
2272 ExtendedFamilyIdLength_AMD = 24
2273 };
2274
2275 const size_t VENDOR_LENGTH = 13;
2276 const size_t CPU_EBS_MAX_LENGTH = (3 * 4 * 4 + 1);
2277 static char* _cpu_brand_string = nullptr;
2278 static int64_t _max_qualified_cpu_frequency = 0;
2279
2280 static int _no_of_threads = 0;
2281 static int _no_of_cores = 0;
2282
2283 const char* const _family_id_intel[ExtendedFamilyIdLength_INTEL] = {
2284 "8086/8088",
2285 "",
2286 "286",
2287 "386",
2288 "486",
2289 "Pentium",
2290 "Pentium Pro", //or Pentium-M/Woodcrest depending on model
2291 "",
2292 "",
2293 "",
2294 "",
2295 "",
2296 "",
2297 "",
2298 "",
2299 "Pentium 4"
2300 };
2301
2302 const char* const _family_id_amd[ExtendedFamilyIdLength_AMD] = {
2303 "",
2304 "",
2305 "",
2306 "",
2307 "5x86",
2308 "K5/K6",
2309 "Athlon/AthlonXP",
2310 "",
2311 "",
2312 "",
2313 "",
2314 "",
2315 "",
2316 "",
2317 "",
2318 "Opteron/Athlon64",
2319 "Opteron QC/Phenom", // Barcelona et.al.
2320 "",
2321 "",
2322 "",
2323 "",
2324 "",
2325 "",
2326 "Zen"
2327 };
2328 // Partially from Intel 64 and IA-32 Architecture Software Developer's Manual,
2329 // September 2013, Vol 3C Table 35-1
2330 const char* const _model_id_pentium_pro[] = {
2331 "",
2332 "Pentium Pro",
2333 "",
2334 "Pentium II model 3",
2335 "",
2336 "Pentium II model 5/Xeon/Celeron",
2337 "Celeron",
2338 "Pentium III/Pentium III Xeon",
2339 "Pentium III/Pentium III Xeon",
2340 "Pentium M model 9", // Yonah
2341 "Pentium III, model A",
2342 "Pentium III, model B",
2343 "",
2344 "Pentium M model D", // Dothan
2345 "",
2346 "Core 2", // 0xf Woodcrest/Conroe/Merom/Kentsfield/Clovertown
2347 "",
2348 "",
2349 "",
2350 "",
2351 "",
2352 "",
2353 "Celeron", // 0x16 Celeron 65nm
2354 "Core 2", // 0x17 Penryn / Harpertown
2355 "",
2356 "",
2357 "Core i7", // 0x1A CPU_MODEL_NEHALEM_EP
2358 "Atom", // 0x1B Z5xx series Silverthorn
2359 "",
2360 "Core 2", // 0x1D Dunnington (6-core)
2361 "Nehalem", // 0x1E CPU_MODEL_NEHALEM
2362 "",
2363 "",
2364 "",
2365 "",
2366 "",
2367 "",
2368 "Westmere", // 0x25 CPU_MODEL_WESTMERE
2369 "",
2370 "",
2371 "", // 0x28
2372 "",
2373 "Sandy Bridge", // 0x2a "2nd Generation Intel Core i7, i5, i3"
2374 "",
2375 "Westmere-EP", // 0x2c CPU_MODEL_WESTMERE_EP
2376 "Sandy Bridge-EP", // 0x2d CPU_MODEL_SANDYBRIDGE_EP
2377 "Nehalem-EX", // 0x2e CPU_MODEL_NEHALEM_EX
2378 "Westmere-EX", // 0x2f CPU_MODEL_WESTMERE_EX
2379 "",
2380 "",
2381 "",
2382 "",
2383 "",
2384 "",
2385 "",
2386 "",
2387 "",
2388 "",
2389 "Ivy Bridge", // 0x3a
2390 "",
2391 "Haswell", // 0x3c "4th Generation Intel Core Processor"
2392 "", // 0x3d "Next Generation Intel Core Processor"
2393 "Ivy Bridge-EP", // 0x3e "Next Generation Intel Xeon Processor E7 Family"
2394 "", // 0x3f "Future Generation Intel Xeon Processor"
2395 "",
2396 "",
2397 "",
2398 "",
2399 "",
2400 "Haswell", // 0x45 "4th Generation Intel Core Processor"
2401 "Haswell", // 0x46 "4th Generation Intel Core Processor"
2402 nullptr
2403 };
2404
2405 /* Brand ID is for back compatibility
2406 * Newer CPUs uses the extended brand string */
2407 const char* const _brand_id[] = {
2408 "",
2409 "Celeron processor",
2410 "Pentium III processor",
2411 "Intel Pentium III Xeon processor",
2412 "",
2413 "",
2414 "",
2415 "",
2416 "Intel Pentium 4 processor",
2417 nullptr
2418 };
2419
2420
2421 const char* const _feature_edx_id[] = {
2422 "On-Chip FPU",
2423 "Virtual Mode Extensions",
2424 "Debugging Extensions",
2425 "Page Size Extensions",
2426 "Time Stamp Counter",
2427 "Model Specific Registers",
2428 "Physical Address Extension",
2429 "Machine Check Exceptions",
2430 "CMPXCHG8B Instruction",
2431 "On-Chip APIC",
2432 "",
2433 "Fast System Call",
2434 "Memory Type Range Registers",
2435 "Page Global Enable",
2436 "Machine Check Architecture",
2437 "Conditional Mov Instruction",
2438 "Page Attribute Table",
2439 "36-bit Page Size Extension",
2440 "Processor Serial Number",
2441 "CLFLUSH Instruction",
2442 "",
2443 "Debug Trace Store feature",
2444 "ACPI registers in MSR space",
2445 "Intel Architecture MMX Technology",
2446 "Fast Float Point Save and Restore",
2447 "Streaming SIMD extensions",
2448 "Streaming SIMD extensions 2",
2449 "Self-Snoop",
2450 "Hyper Threading",
2451 "Thermal Monitor",
2452 "",
2453 "Pending Break Enable"
2454 };
2455
2456 const char* const _feature_extended_edx_id[] = {
2457 "",
2458 "",
2459 "",
2460 "",
2461 "",
2462 "",
2463 "",
2464 "",
2465 "",
2466 "",
2467 "",
2468 "SYSCALL/SYSRET",
2469 "",
2470 "",
2471 "",
2472 "",
2473 "",
2474 "",
2475 "",
2476 "",
2477 "Execute Disable Bit",
2478 "",
2479 "",
2480 "",
2481 "",
2482 "",
2483 "",
2484 "RDTSCP",
2485 "",
2486 "Intel 64 Architecture",
2487 "",
2488 ""
2489 };
2490
2491 const char* const _feature_ecx_id[] = {
2492 "Streaming SIMD Extensions 3",
2493 "PCLMULQDQ",
2494 "64-bit DS Area",
2495 "MONITOR/MWAIT instructions",
2496 "CPL Qualified Debug Store",
2497 "Virtual Machine Extensions",
2498 "Safer Mode Extensions",
2499 "Enhanced Intel SpeedStep technology",
2500 "Thermal Monitor 2",
2501 "Supplemental Streaming SIMD Extensions 3",
2502 "L1 Context ID",
2503 "",
2504 "Fused Multiply-Add",
2505 "CMPXCHG16B",
2506 "xTPR Update Control",
2507 "Perfmon and Debug Capability",
2508 "",
2509 "Process-context identifiers",
2510 "Direct Cache Access",
2511 "Streaming SIMD extensions 4.1",
2512 "Streaming SIMD extensions 4.2",
2513 "x2APIC",
2514 "MOVBE",
2515 "Popcount instruction",
2516 "TSC-Deadline",
2517 "AESNI",
2518 "XSAVE",
2519 "OSXSAVE",
2520 "AVX",
2521 "F16C",
2522 "RDRAND",
2523 ""
2524 };
2525
2526 const char* const _feature_extended_ecx_id[] = {
2527 "LAHF/SAHF instruction support",
2528 "Core multi-processor legacy mode",
2529 "",
2530 "",
2531 "",
2532 "Advanced Bit Manipulations: LZCNT",
2533 "SSE4A: MOVNTSS, MOVNTSD, EXTRQ, INSERTQ",
2534 "Misaligned SSE mode",
2535 "",
2536 "",
2537 "",
2538 "",
2539 "",
2540 "",
2541 "",
2542 "",
2543 "",
2544 "",
2545 "",
2546 "",
2547 "",
2548 "",
2549 "",
2550 "",
2551 "",
2552 "",
2553 "",
2554 "",
2555 "",
2556 "",
2557 "",
2558 ""
2559 };
2560
2561 void VM_Version::initialize_tsc(void) {
2562 ResourceMark rm;
2563
2564 cpuid_brand_string_stub_blob = BufferBlob::create("getCPUIDBrandString_stub", cpuid_brand_string_stub_size);
2565 if (cpuid_brand_string_stub_blob == nullptr) {
2566 vm_exit_during_initialization("Unable to allocate getCPUIDBrandString_stub");
2567 }
2568 CodeBuffer c(cpuid_brand_string_stub_blob);
2569 VM_Version_StubGenerator g(&c);
2570 getCPUIDBrandString_stub = CAST_TO_FN_PTR(getCPUIDBrandString_stub_t,
2571 g.generate_getCPUIDBrandString());
2572 }
2573
2574 const char* VM_Version::cpu_model_description(void) {
2575 uint32_t cpu_family = extended_cpu_family();
2576 uint32_t cpu_model = extended_cpu_model();
2577 const char* model = nullptr;
2578
2579 if (cpu_family == CPU_FAMILY_PENTIUMPRO) {
2580 for (uint32_t i = 0; i <= cpu_model; i++) {
2581 model = _model_id_pentium_pro[i];
2582 if (model == nullptr) {
2583 break;
2584 }
2585 }
2586 }
2587 return model;
2588 }
2589
2590 const char* VM_Version::cpu_brand_string(void) {
2591 if (_cpu_brand_string == nullptr) {
2592 _cpu_brand_string = NEW_C_HEAP_ARRAY_RETURN_NULL(char, CPU_EBS_MAX_LENGTH, mtInternal);
2593 if (nullptr == _cpu_brand_string) {
2594 return nullptr;
2595 }
2596 int ret_val = cpu_extended_brand_string(_cpu_brand_string, CPU_EBS_MAX_LENGTH);
2597 if (ret_val != OS_OK) {
2598 FREE_C_HEAP_ARRAY(char, _cpu_brand_string);
2599 _cpu_brand_string = nullptr;
2600 }
2601 }
2602 return _cpu_brand_string;
2603 }
2604
2605 const char* VM_Version::cpu_brand(void) {
2606 const char* brand = nullptr;
2607
2608 if ((_cpuid_info.std_cpuid1_ebx.value & 0xFF) > 0) {
2609 int brand_num = _cpuid_info.std_cpuid1_ebx.value & 0xFF;
2610 brand = _brand_id[0];
2611 for (int i = 0; brand != nullptr && i <= brand_num; i += 1) {
2612 brand = _brand_id[i];
2613 }
2614 }
2615 return brand;
2616 }
2617
2618 bool VM_Version::cpu_is_em64t(void) {
2619 return ((_cpuid_info.ext_cpuid1_edx.value & INTEL64_FLAG) == INTEL64_FLAG);
2620 }
2621
2622 bool VM_Version::is_netburst(void) {
2623 return (is_intel() && (extended_cpu_family() == CPU_FAMILY_PENTIUM_4));
2624 }
2625
2626 bool VM_Version::supports_tscinv_ext(void) {
2627 if (!supports_tscinv_bit()) {
2628 return false;
2629 }
2630
2631 if (is_intel()) {
2632 return true;
2633 }
2634
2635 if (is_amd()) {
2636 return !is_amd_Barcelona();
2637 }
2638
2639 if (is_hygon()) {
2640 return true;
2641 }
2642
2643 return false;
2644 }
2645
2646 void VM_Version::resolve_cpu_information_details(void) {
2647
2648 // in future we want to base this information on proper cpu
2649 // and cache topology enumeration such as:
2650 // Intel 64 Architecture Processor Topology Enumeration
2651 // which supports system cpu and cache topology enumeration
2652 // either using 2xAPICIDs or initial APICIDs
2653
2654 // currently only rough cpu information estimates
2655 // which will not necessarily reflect the exact configuration of the system
2656
2657 // this is the number of logical hardware threads
2658 // visible to the operating system
2659 _no_of_threads = os::processor_count();
2660
2661 // find out number of threads per cpu package
2662 int threads_per_package = threads_per_core() * cores_per_cpu();
2663
2664 // use amount of threads visible to the process in order to guess number of sockets
2665 _no_of_sockets = _no_of_threads / threads_per_package;
2666
2667 // process might only see a subset of the total number of threads
2668 // from a single processor package. Virtualization/resource management for example.
2669 // If so then just write a hard 1 as num of pkgs.
2670 if (0 == _no_of_sockets) {
2671 _no_of_sockets = 1;
2672 }
2673
2674 // estimate the number of cores
2675 _no_of_cores = cores_per_cpu() * _no_of_sockets;
2676 }
2677
2678
2679 const char* VM_Version::cpu_family_description(void) {
2680 int cpu_family_id = extended_cpu_family();
2681 if (is_amd()) {
2682 if (cpu_family_id < ExtendedFamilyIdLength_AMD) {
2683 return _family_id_amd[cpu_family_id];
2684 }
2685 }
2686 if (is_intel()) {
2687 if (cpu_family_id == CPU_FAMILY_PENTIUMPRO) {
2688 return cpu_model_description();
2689 }
2690 if (cpu_family_id < ExtendedFamilyIdLength_INTEL) {
2691 return _family_id_intel[cpu_family_id];
2692 }
2693 }
2694 if (is_hygon()) {
2695 return "Dhyana";
2696 }
2697 return "Unknown x86";
2698 }
2699
2700 int VM_Version::cpu_type_description(char* const buf, size_t buf_len) {
2701 assert(buf != nullptr, "buffer is null!");
2702 assert(buf_len >= CPU_TYPE_DESC_BUF_SIZE, "buffer len should at least be == CPU_TYPE_DESC_BUF_SIZE!");
2703
2704 const char* cpu_type = nullptr;
2705 const char* x64 = nullptr;
2706
2707 if (is_intel()) {
2708 cpu_type = "Intel";
2709 x64 = cpu_is_em64t() ? " Intel64" : "";
2710 } else if (is_amd()) {
2711 cpu_type = "AMD";
2712 x64 = cpu_is_em64t() ? " AMD64" : "";
2713 } else if (is_hygon()) {
2714 cpu_type = "Hygon";
2715 x64 = cpu_is_em64t() ? " AMD64" : "";
2716 } else {
2717 cpu_type = "Unknown x86";
2718 x64 = cpu_is_em64t() ? " x86_64" : "";
2719 }
2720
2721 jio_snprintf(buf, buf_len, "%s %s%s SSE SSE2%s%s%s%s%s%s%s%s",
2722 cpu_type,
2723 cpu_family_description(),
2724 supports_ht() ? " (HT)" : "",
2725 supports_sse3() ? " SSE3" : "",
2726 supports_ssse3() ? " SSSE3" : "",
2727 supports_sse4_1() ? " SSE4.1" : "",
2728 supports_sse4_2() ? " SSE4.2" : "",
2729 supports_sse4a() ? " SSE4A" : "",
2730 is_netburst() ? " Netburst" : "",
2731 is_intel_family_core() ? " Core" : "",
2732 x64);
2733
2734 return OS_OK;
2735 }
2736
2737 int VM_Version::cpu_extended_brand_string(char* const buf, size_t buf_len) {
2738 assert(buf != nullptr, "buffer is null!");
2739 assert(buf_len >= CPU_EBS_MAX_LENGTH, "buffer len should at least be == CPU_EBS_MAX_LENGTH!");
2740 assert(getCPUIDBrandString_stub != nullptr, "not initialized");
2741
2742 // invoke newly generated asm code to fetch CPU Brand String
2743 getCPUIDBrandString_stub(&_cpuid_info);
2744
2745 // fetch results into buffer
2746 *((uint32_t*) &buf[0]) = _cpuid_info.proc_name_0;
2747 *((uint32_t*) &buf[4]) = _cpuid_info.proc_name_1;
2748 *((uint32_t*) &buf[8]) = _cpuid_info.proc_name_2;
2749 *((uint32_t*) &buf[12]) = _cpuid_info.proc_name_3;
2750 *((uint32_t*) &buf[16]) = _cpuid_info.proc_name_4;
2751 *((uint32_t*) &buf[20]) = _cpuid_info.proc_name_5;
2752 *((uint32_t*) &buf[24]) = _cpuid_info.proc_name_6;
2753 *((uint32_t*) &buf[28]) = _cpuid_info.proc_name_7;
2754 *((uint32_t*) &buf[32]) = _cpuid_info.proc_name_8;
2755 *((uint32_t*) &buf[36]) = _cpuid_info.proc_name_9;
2756 *((uint32_t*) &buf[40]) = _cpuid_info.proc_name_10;
2757 *((uint32_t*) &buf[44]) = _cpuid_info.proc_name_11;
2758
2759 return OS_OK;
2760 }
2761
2762 size_t VM_Version::cpu_write_support_string(char* const buf, size_t buf_len) {
2763 guarantee(buf != nullptr, "buffer is null!");
2764 guarantee(buf_len > 0, "buffer len not enough!");
2765
2766 unsigned int flag = 0;
2767 unsigned int fi = 0;
2768 size_t written = 0;
2769 const char* prefix = "";
2770
2771 #define WRITE_TO_BUF(string) \
2772 { \
2773 int res = jio_snprintf(&buf[written], buf_len - written, "%s%s", prefix, string); \
2774 if (res < 0) { \
2775 return buf_len - 1; \
2776 } \
2777 written += res; \
2778 if (prefix[0] == '\0') { \
2779 prefix = ", "; \
2780 } \
2781 }
2782
2783 for (flag = 1, fi = 0; flag <= 0x20000000 ; flag <<= 1, fi++) {
2784 if (flag == HTT_FLAG && (((_cpuid_info.std_cpuid1_ebx.value >> 16) & 0xff) <= 1)) {
2785 continue; /* no hyperthreading */
2786 } else if (flag == SEP_FLAG && (cpu_family() == CPU_FAMILY_PENTIUMPRO && ((_cpuid_info.std_cpuid1_eax.value & 0xff) < 0x33))) {
2787 continue; /* no fast system call */
2788 }
2789 if ((_cpuid_info.std_cpuid1_edx.value & flag) && strlen(_feature_edx_id[fi]) > 0) {
2790 WRITE_TO_BUF(_feature_edx_id[fi]);
2791 }
2792 }
2793
2794 for (flag = 1, fi = 0; flag <= 0x20000000; flag <<= 1, fi++) {
2795 if ((_cpuid_info.std_cpuid1_ecx.value & flag) && strlen(_feature_ecx_id[fi]) > 0) {
2796 WRITE_TO_BUF(_feature_ecx_id[fi]);
2797 }
2798 }
2799
2800 for (flag = 1, fi = 0; flag <= 0x20000000 ; flag <<= 1, fi++) {
2801 if ((_cpuid_info.ext_cpuid1_ecx.value & flag) && strlen(_feature_extended_ecx_id[fi]) > 0) {
2802 WRITE_TO_BUF(_feature_extended_ecx_id[fi]);
2803 }
2804 }
2805
2806 for (flag = 1, fi = 0; flag <= 0x20000000; flag <<= 1, fi++) {
2807 if ((_cpuid_info.ext_cpuid1_edx.value & flag) && strlen(_feature_extended_edx_id[fi]) > 0) {
2808 WRITE_TO_BUF(_feature_extended_edx_id[fi]);
2809 }
2810 }
2811
2812 if (supports_tscinv_bit()) {
2813 WRITE_TO_BUF("Invariant TSC");
2814 }
2815
2816 return written;
2817 }
2818
2819 /**
2820 * Write a detailed description of the cpu to a given buffer, including
2821 * feature set.
2822 */
2823 int VM_Version::cpu_detailed_description(char* const buf, size_t buf_len) {
2824 assert(buf != nullptr, "buffer is null!");
2825 assert(buf_len >= CPU_DETAILED_DESC_BUF_SIZE, "buffer len should at least be == CPU_DETAILED_DESC_BUF_SIZE!");
2826
2827 static const char* unknown = "<unknown>";
2828 char vendor_id[VENDOR_LENGTH];
2829 const char* family = nullptr;
2830 const char* model = nullptr;
2831 const char* brand = nullptr;
2832 int outputLen = 0;
2833
2834 family = cpu_family_description();
2835 if (family == nullptr) {
2836 family = unknown;
2837 }
2838
2839 model = cpu_model_description();
2840 if (model == nullptr) {
2841 model = unknown;
2842 }
2843
2844 brand = cpu_brand_string();
2845
2846 if (brand == nullptr) {
2847 brand = cpu_brand();
2848 if (brand == nullptr) {
2849 brand = unknown;
2850 }
2851 }
2852
2853 *((uint32_t*) &vendor_id[0]) = _cpuid_info.std_vendor_name_0;
2854 *((uint32_t*) &vendor_id[4]) = _cpuid_info.std_vendor_name_2;
2855 *((uint32_t*) &vendor_id[8]) = _cpuid_info.std_vendor_name_1;
2856 vendor_id[VENDOR_LENGTH-1] = '\0';
2857
2858 outputLen = jio_snprintf(buf, buf_len, "Brand: %s, Vendor: %s\n"
2859 "Family: %s (0x%x), Model: %s (0x%x), Stepping: 0x%x\n"
2860 "Ext. family: 0x%x, Ext. model: 0x%x, Type: 0x%x, Signature: 0x%8.8x\n"
2861 "Features: ebx: 0x%8.8x, ecx: 0x%8.8x, edx: 0x%8.8x\n"
2862 "Ext. features: eax: 0x%8.8x, ebx: 0x%8.8x, ecx: 0x%8.8x, edx: 0x%8.8x\n"
2863 "Supports: ",
2864 brand,
2865 vendor_id,
2866 family,
2867 extended_cpu_family(),
2868 model,
2869 extended_cpu_model(),
2870 cpu_stepping(),
2871 _cpuid_info.std_cpuid1_eax.bits.ext_family,
2872 _cpuid_info.std_cpuid1_eax.bits.ext_model,
2873 _cpuid_info.std_cpuid1_eax.bits.proc_type,
2874 _cpuid_info.std_cpuid1_eax.value,
2875 _cpuid_info.std_cpuid1_ebx.value,
2876 _cpuid_info.std_cpuid1_ecx.value,
2877 _cpuid_info.std_cpuid1_edx.value,
2878 _cpuid_info.ext_cpuid1_eax,
2879 _cpuid_info.ext_cpuid1_ebx,
2880 _cpuid_info.ext_cpuid1_ecx,
2881 _cpuid_info.ext_cpuid1_edx);
2882
2883 if (outputLen < 0 || (size_t) outputLen >= buf_len - 1) {
2884 if (buf_len > 0) { buf[buf_len-1] = '\0'; }
2885 return OS_ERR;
2886 }
2887
2888 cpu_write_support_string(&buf[outputLen], buf_len - outputLen);
2889
2890 return OS_OK;
2891 }
2892
2893
2894 // Fill in Abstract_VM_Version statics
2895 void VM_Version::initialize_cpu_information() {
2896 assert(_vm_version_initialized, "should have initialized VM_Version long ago");
2897 assert(!_initialized, "shouldn't be initialized yet");
2898 resolve_cpu_information_details();
2899
2900 // initialize cpu_name and cpu_desc
2901 cpu_type_description(_cpu_name, CPU_TYPE_DESC_BUF_SIZE);
2902 cpu_detailed_description(_cpu_desc, CPU_DETAILED_DESC_BUF_SIZE);
2903 _initialized = true;
2904 }
2905
2906 /**
2907 * For information about extracting the frequency from the cpu brand string, please see:
2908 *
2909 * Intel Processor Identification and the CPUID Instruction
2910 * Application Note 485
2911 * May 2012
2912 *
2913 * The return value is the frequency in Hz.
2914 */
2915 int64_t VM_Version::max_qualified_cpu_freq_from_brand_string(void) {
2916 const char* const brand_string = cpu_brand_string();
2917 if (brand_string == nullptr) {
2918 return 0;
2919 }
2920 const int64_t MEGA = 1000000;
2921 int64_t multiplier = 0;
2922 int64_t frequency = 0;
2923 uint8_t idx = 0;
2924 // The brand string buffer is at most 48 bytes.
2925 // -2 is to prevent buffer overrun when looking for y in yHz, as z is +2 from y.
2926 for (; idx < 48-2; ++idx) {
2927 // Format is either "x.xxyHz" or "xxxxyHz", where y=M, G, T and x are digits.
2928 // Search brand string for "yHz" where y is M, G, or T.
2929 if (brand_string[idx+1] == 'H' && brand_string[idx+2] == 'z') {
2930 if (brand_string[idx] == 'M') {
2931 multiplier = MEGA;
2932 } else if (brand_string[idx] == 'G') {
2933 multiplier = MEGA * 1000;
2934 } else if (brand_string[idx] == 'T') {
2935 multiplier = MEGA * MEGA;
2936 }
2937 break;
2938 }
2939 }
2940 if (multiplier > 0) {
2941 // Compute frequency (in Hz) from brand string.
2942 if (brand_string[idx-3] == '.') { // if format is "x.xx"
2943 frequency = (brand_string[idx-4] - '0') * multiplier;
2944 frequency += (brand_string[idx-2] - '0') * multiplier / 10;
2945 frequency += (brand_string[idx-1] - '0') * multiplier / 100;
2946 } else { // format is "xxxx"
2947 frequency = (brand_string[idx-4] - '0') * 1000;
2948 frequency += (brand_string[idx-3] - '0') * 100;
2949 frequency += (brand_string[idx-2] - '0') * 10;
2950 frequency += (brand_string[idx-1] - '0');
2951 frequency *= multiplier;
2952 }
2953 }
2954 return frequency;
2955 }
2956
2957
2958 int64_t VM_Version::maximum_qualified_cpu_frequency(void) {
2959 if (_max_qualified_cpu_frequency == 0) {
2960 _max_qualified_cpu_frequency = max_qualified_cpu_freq_from_brand_string();
2961 }
2962 return _max_qualified_cpu_frequency;
2963 }
2964
2965 uint64_t VM_Version::CpuidInfo::feature_flags() const {
2966 uint64_t result = 0;
2967 if (std_cpuid1_edx.bits.cmpxchg8 != 0)
2968 result |= CPU_CX8;
2969 if (std_cpuid1_edx.bits.cmov != 0)
2970 result |= CPU_CMOV;
2971 if (std_cpuid1_edx.bits.clflush != 0)
2972 result |= CPU_FLUSH;
2973 #ifdef _LP64
2974 // clflush should always be available on x86_64
2975 // if not we are in real trouble because we rely on it
2976 // to flush the code cache.
2977 assert ((result & CPU_FLUSH) != 0, "clflush should be available");
2978 #endif
2979 if (std_cpuid1_edx.bits.fxsr != 0 || (is_amd_family() &&
2980 ext_cpuid1_edx.bits.fxsr != 0))
2981 result |= CPU_FXSR;
2982 // HT flag is set for multi-core processors also.
2983 if (threads_per_core() > 1)
2984 result |= CPU_HT;
2985 if (std_cpuid1_edx.bits.mmx != 0 || (is_amd_family() &&
2986 ext_cpuid1_edx.bits.mmx != 0))
2987 result |= CPU_MMX;
2988 if (std_cpuid1_edx.bits.sse != 0)
2989 result |= CPU_SSE;
2990 if (std_cpuid1_edx.bits.sse2 != 0)
2991 result |= CPU_SSE2;
2992 if (std_cpuid1_ecx.bits.sse3 != 0)
2993 result |= CPU_SSE3;
2994 if (std_cpuid1_ecx.bits.ssse3 != 0)
2995 result |= CPU_SSSE3;
2996 if (std_cpuid1_ecx.bits.sse4_1 != 0)
2997 result |= CPU_SSE4_1;
2998 if (std_cpuid1_ecx.bits.sse4_2 != 0)
2999 result |= CPU_SSE4_2;
3000 if (std_cpuid1_ecx.bits.popcnt != 0)
3001 result |= CPU_POPCNT;
3002 if (sefsl1_cpuid7_edx.bits.apx_f != 0 &&
3003 xem_xcr0_eax.bits.apx_f != 0) {
3004 result |= CPU_APX_F;
3005 }
3006 if (std_cpuid1_ecx.bits.avx != 0 &&
3007 std_cpuid1_ecx.bits.osxsave != 0 &&
3008 xem_xcr0_eax.bits.sse != 0 &&
3009 xem_xcr0_eax.bits.ymm != 0) {
3010 result |= CPU_AVX;
3011 result |= CPU_VZEROUPPER;
3012 if (sefsl1_cpuid7_eax.bits.sha512 != 0)
3013 result |= CPU_SHA512;
3014 if (std_cpuid1_ecx.bits.f16c != 0)
3015 result |= CPU_F16C;
3016 if (sef_cpuid7_ebx.bits.avx2 != 0) {
3017 result |= CPU_AVX2;
3018 if (sefsl1_cpuid7_eax.bits.avx_ifma != 0)
3019 result |= CPU_AVX_IFMA;
3020 }
3021 if (sef_cpuid7_ecx.bits.gfni != 0)
3022 result |= CPU_GFNI;
3023 if (sef_cpuid7_ebx.bits.avx512f != 0 &&
3024 xem_xcr0_eax.bits.opmask != 0 &&
3025 xem_xcr0_eax.bits.zmm512 != 0 &&
3026 xem_xcr0_eax.bits.zmm32 != 0) {
3027 result |= CPU_AVX512F;
3028 if (sef_cpuid7_ebx.bits.avx512cd != 0)
3029 result |= CPU_AVX512CD;
3030 if (sef_cpuid7_ebx.bits.avx512dq != 0)
3031 result |= CPU_AVX512DQ;
3032 if (sef_cpuid7_ebx.bits.avx512ifma != 0)
3033 result |= CPU_AVX512_IFMA;
3034 if (sef_cpuid7_ebx.bits.avx512pf != 0)
3035 result |= CPU_AVX512PF;
3036 if (sef_cpuid7_ebx.bits.avx512er != 0)
3037 result |= CPU_AVX512ER;
3038 if (sef_cpuid7_ebx.bits.avx512bw != 0)
3039 result |= CPU_AVX512BW;
3040 if (sef_cpuid7_ebx.bits.avx512vl != 0)
3041 result |= CPU_AVX512VL;
3042 if (sef_cpuid7_ecx.bits.avx512_vpopcntdq != 0)
3043 result |= CPU_AVX512_VPOPCNTDQ;
3044 if (sef_cpuid7_ecx.bits.avx512_vpclmulqdq != 0)
3045 result |= CPU_AVX512_VPCLMULQDQ;
3046 if (sef_cpuid7_ecx.bits.vaes != 0)
3047 result |= CPU_AVX512_VAES;
3048 if (sef_cpuid7_ecx.bits.avx512_vnni != 0)
3049 result |= CPU_AVX512_VNNI;
3050 if (sef_cpuid7_ecx.bits.avx512_bitalg != 0)
3051 result |= CPU_AVX512_BITALG;
3052 if (sef_cpuid7_ecx.bits.avx512_vbmi != 0)
3053 result |= CPU_AVX512_VBMI;
3054 if (sef_cpuid7_ecx.bits.avx512_vbmi2 != 0)
3055 result |= CPU_AVX512_VBMI2;
3056 }
3057 }
3058 if (std_cpuid1_ecx.bits.hv != 0)
3059 result |= CPU_HV;
3060 if (sef_cpuid7_ebx.bits.bmi1 != 0)
3061 result |= CPU_BMI1;
3062 if (std_cpuid1_edx.bits.tsc != 0)
3063 result |= CPU_TSC;
3064 if (ext_cpuid7_edx.bits.tsc_invariance != 0)
3065 result |= CPU_TSCINV_BIT;
3066 if (std_cpuid1_ecx.bits.aes != 0)
3067 result |= CPU_AES;
3068 if (sef_cpuid7_ebx.bits.erms != 0)
3069 result |= CPU_ERMS;
3070 if (sef_cpuid7_edx.bits.fast_short_rep_mov != 0)
3071 result |= CPU_FSRM;
3072 if (std_cpuid1_ecx.bits.clmul != 0)
3073 result |= CPU_CLMUL;
3074 if (sef_cpuid7_ebx.bits.rtm != 0)
3075 result |= CPU_RTM;
3076 if (sef_cpuid7_ebx.bits.adx != 0)
3077 result |= CPU_ADX;
3078 if (sef_cpuid7_ebx.bits.bmi2 != 0)
3079 result |= CPU_BMI2;
3080 if (sef_cpuid7_ebx.bits.sha != 0)
3081 result |= CPU_SHA;
3082 if (std_cpuid1_ecx.bits.fma != 0)
3083 result |= CPU_FMA;
3084 if (sef_cpuid7_ebx.bits.clflushopt != 0)
3085 result |= CPU_FLUSHOPT;
3086 if (ext_cpuid1_edx.bits.rdtscp != 0)
3087 result |= CPU_RDTSCP;
3088 if (sef_cpuid7_ecx.bits.rdpid != 0)
3089 result |= CPU_RDPID;
3090
3091 // AMD|Hygon features.
3092 if (is_amd_family()) {
3093 if ((ext_cpuid1_edx.bits.tdnow != 0) ||
3094 (ext_cpuid1_ecx.bits.prefetchw != 0))
3095 result |= CPU_3DNOW_PREFETCH;
3096 if (ext_cpuid1_ecx.bits.lzcnt != 0)
3097 result |= CPU_LZCNT;
3098 if (ext_cpuid1_ecx.bits.sse4a != 0)
3099 result |= CPU_SSE4A;
3100 }
3101
3102 // Intel features.
3103 if (is_intel()) {
3104 if (ext_cpuid1_ecx.bits.lzcnt != 0) {
3105 result |= CPU_LZCNT;
3106 }
3107 if (ext_cpuid1_ecx.bits.prefetchw != 0) {
3108 result |= CPU_3DNOW_PREFETCH;
3109 }
3110 if (sef_cpuid7_ebx.bits.clwb != 0) {
3111 result |= CPU_CLWB;
3112 }
3113 if (sef_cpuid7_edx.bits.serialize != 0)
3114 result |= CPU_SERIALIZE;
3115
3116 if (_cpuid_info.sef_cpuid7_edx.bits.avx512_fp16 != 0)
3117 result |= CPU_AVX512_FP16;
3118 }
3119
3120 // ZX features.
3121 if (is_zx()) {
3122 if (ext_cpuid1_ecx.bits.lzcnt != 0) {
3123 result |= CPU_LZCNT;
3124 }
3125 if (ext_cpuid1_ecx.bits.prefetchw != 0) {
3126 result |= CPU_3DNOW_PREFETCH;
3127 }
3128 }
3129
3130 // Protection key features.
3131 if (sef_cpuid7_ecx.bits.pku != 0) {
3132 result |= CPU_PKU;
3133 }
3134 if (sef_cpuid7_ecx.bits.ospke != 0) {
3135 result |= CPU_OSPKE;
3136 }
3137
3138 // Control flow enforcement (CET) features.
3139 if (sef_cpuid7_ecx.bits.cet_ss != 0) {
3140 result |= CPU_CET_SS;
3141 }
3142 if (sef_cpuid7_edx.bits.cet_ibt != 0) {
3143 result |= CPU_CET_IBT;
3144 }
3145
3146 // Composite features.
3147 if (supports_tscinv_bit() &&
3148 ((is_amd_family() && !is_amd_Barcelona()) ||
3149 is_intel_tsc_synched_at_init())) {
3150 result |= CPU_TSCINV;
3151 }
3152
3153 return result;
3154 }
3155
3156 bool VM_Version::os_supports_avx_vectors() {
3157 bool retVal = false;
3158 int nreg = 2 LP64_ONLY(+2);
3159 if (supports_evex()) {
3160 // Verify that OS save/restore all bits of EVEX registers
3161 // during signal processing.
3162 retVal = true;
3163 for (int i = 0; i < 16 * nreg; i++) { // 64 bytes per zmm register
3164 if (_cpuid_info.zmm_save[i] != ymm_test_value()) {
3165 retVal = false;
3166 break;
3167 }
3168 }
3169 } else if (supports_avx()) {
3170 // Verify that OS save/restore all bits of AVX registers
3171 // during signal processing.
3172 retVal = true;
3173 for (int i = 0; i < 8 * nreg; i++) { // 32 bytes per ymm register
3174 if (_cpuid_info.ymm_save[i] != ymm_test_value()) {
3175 retVal = false;
3176 break;
3177 }
3178 }
3179 // zmm_save will be set on a EVEX enabled machine even if we choose AVX code gen
3180 if (retVal == false) {
3181 // Verify that OS save/restore all bits of EVEX registers
3182 // during signal processing.
3183 retVal = true;
3184 for (int i = 0; i < 16 * nreg; i++) { // 64 bytes per zmm register
3185 if (_cpuid_info.zmm_save[i] != ymm_test_value()) {
3186 retVal = false;
3187 break;
3188 }
3189 }
3190 }
3191 }
3192 return retVal;
3193 }
3194
3195 bool VM_Version::os_supports_apx_egprs() {
3196 if (!supports_apx_f()) {
3197 return false;
3198 }
3199 // Enable APX support for product builds after
3200 // completion of planned features listed in JDK-8329030.
3201 #if !defined(PRODUCT)
3202 if (_cpuid_info.apx_save[0] != egpr_test_value() ||
3203 _cpuid_info.apx_save[1] != egpr_test_value()) {
3204 return false;
3205 }
3206 return true;
3207 #else
3208 return false;
3209 #endif
3210 }
3211
3212 uint VM_Version::cores_per_cpu() {
3213 uint result = 1;
3214 if (is_intel()) {
3215 bool supports_topology = supports_processor_topology();
3216 if (supports_topology) {
3217 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus /
3218 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus;
3219 }
3220 if (!supports_topology || result == 0) {
3221 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1);
3222 }
3223 } else if (is_amd_family()) {
3224 result = (_cpuid_info.ext_cpuid8_ecx.bits.cores_per_cpu + 1);
3225 } else if (is_zx()) {
3226 bool supports_topology = supports_processor_topology();
3227 if (supports_topology) {
3228 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus /
3229 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus;
3230 }
3231 if (!supports_topology || result == 0) {
3232 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1);
3233 }
3234 }
3235 return result;
3236 }
3237
3238 uint VM_Version::threads_per_core() {
3239 uint result = 1;
3240 if (is_intel() && supports_processor_topology()) {
3241 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus;
3242 } else if (is_zx() && supports_processor_topology()) {
3243 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus;
3244 } else if (_cpuid_info.std_cpuid1_edx.bits.ht != 0) {
3245 if (cpu_family() >= 0x17) {
3246 result = _cpuid_info.ext_cpuid1E_ebx.bits.threads_per_core + 1;
3247 } else {
3248 result = _cpuid_info.std_cpuid1_ebx.bits.threads_per_cpu /
3249 cores_per_cpu();
3250 }
3251 }
3252 return (result == 0 ? 1 : result);
3253 }
3254
3255 uint VM_Version::L1_line_size() {
3256 uint result = 0;
3257 if (is_intel()) {
3258 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1);
3259 } else if (is_amd_family()) {
3260 result = _cpuid_info.ext_cpuid5_ecx.bits.L1_line_size;
3261 } else if (is_zx()) {
3262 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1);
3263 }
3264 if (result < 32) // not defined ?
3265 result = 32; // 32 bytes by default on x86 and other x64
3266 return result;
3267 }
3268
3269 bool VM_Version::is_intel_tsc_synched_at_init() {
3270 if (is_intel_family_core()) {
3271 uint32_t ext_model = extended_cpu_model();
3272 if (ext_model == CPU_MODEL_NEHALEM_EP ||
3273 ext_model == CPU_MODEL_WESTMERE_EP ||
3274 ext_model == CPU_MODEL_SANDYBRIDGE_EP ||
3275 ext_model == CPU_MODEL_IVYBRIDGE_EP) {
3276 // <= 2-socket invariant tsc support. EX versions are usually used
3277 // in > 2-socket systems and likely don't synchronize tscs at
3278 // initialization.
3279 // Code that uses tsc values must be prepared for them to arbitrarily
3280 // jump forward or backward.
3281 return true;
3282 }
3283 }
3284 return false;
3285 }
3286
3287 int VM_Version::allocate_prefetch_distance(bool use_watermark_prefetch) {
3288 // Hardware prefetching (distance/size in bytes):
3289 // Pentium 3 - 64 / 32
3290 // Pentium 4 - 256 / 128
3291 // Athlon - 64 / 32 ????
3292 // Opteron - 128 / 64 only when 2 sequential cache lines accessed
3293 // Core - 128 / 64
3294 //
3295 // Software prefetching (distance in bytes / instruction with best score):
3296 // Pentium 3 - 128 / prefetchnta
3297 // Pentium 4 - 512 / prefetchnta
3298 // Athlon - 128 / prefetchnta
3299 // Opteron - 256 / prefetchnta
3300 // Core - 256 / prefetchnta
3301 // It will be used only when AllocatePrefetchStyle > 0
3302
3303 if (is_amd_family()) { // AMD | Hygon
3304 if (supports_sse2()) {
3305 return 256; // Opteron
3306 } else {
3307 return 128; // Athlon
3308 }
3309 } else { // Intel
3310 if (supports_sse3() && cpu_family() == 6) {
3311 if (supports_sse4_2() && supports_ht()) { // Nehalem based cpus
3312 return 192;
3313 } else if (use_watermark_prefetch) { // watermark prefetching on Core
3314 #ifdef _LP64
3315 return 384;
3316 #else
3317 return 320;
3318 #endif
3319 }
3320 }
3321 if (supports_sse2()) {
3322 if (cpu_family() == 6) {
3323 return 256; // Pentium M, Core, Core2
3324 } else {
3325 return 512; // Pentium 4
3326 }
3327 } else {
3328 return 128; // Pentium 3 (and all other old CPUs)
3329 }
3330 }
3331 }
3332
3333 bool VM_Version::is_intrinsic_supported(vmIntrinsicID id) {
3334 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
3335 switch (id) {
3336 case vmIntrinsics::_floatToFloat16:
3337 case vmIntrinsics::_float16ToFloat:
3338 if (!supports_float16()) {
3339 return false;
3340 }
3341 break;
3342 default:
3343 break;
3344 }
3345 return true;
3346 }