1 /*
2 * Copyright (c) 1997, 2026, 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/ostream.hpp"
42 #include "utilities/powerOfTwo.hpp"
43 #include "utilities/virtualizationSupport.hpp"
44
45 int VM_Version::_cpu;
46 int VM_Version::_model;
47 int VM_Version::_stepping;
48 bool VM_Version::_has_intel_jcc_erratum;
49 VM_Version::CpuidInfo VM_Version::_cpuid_info = { 0, };
50
51 #define DECLARE_CPU_FEATURE_NAME(id, name, bit) XSTR(name),
52 const char* VM_Version::_features_names[] = { CPU_FEATURE_FLAGS(DECLARE_CPU_FEATURE_NAME)};
53 #undef DECLARE_CPU_FEATURE_NAME
54
55 // Address of instruction which causes SEGV
56 address VM_Version::_cpuinfo_segv_addr = nullptr;
57 // Address of instruction after the one which causes SEGV
58 address VM_Version::_cpuinfo_cont_addr = nullptr;
59 // Address of instruction which causes APX specific SEGV
60 address VM_Version::_cpuinfo_segv_addr_apx = nullptr;
61 // Address of instruction after the one which causes APX specific SEGV
62 address VM_Version::_cpuinfo_cont_addr_apx = nullptr;
63
64 static BufferBlob* stub_blob;
65 static const int stub_size = 2550;
66
67 int VM_Version::VM_Features::_features_bitmap_size = sizeof(VM_Version::VM_Features::_features_bitmap) / BytesPerLong;
68
69 VM_Version::VM_Features VM_Version::_features;
70 VM_Version::VM_Features VM_Version::_cpu_features;
71
72 extern "C" {
73 typedef void (*get_cpu_info_stub_t)(void*);
74 typedef void (*detect_virt_stub_t)(uint32_t, uint32_t*);
75 typedef void (*clear_apx_test_state_t)(void);
76 typedef void (*getCPUIDBrandString_stub_t)(void*);
77 }
78 static get_cpu_info_stub_t get_cpu_info_stub = nullptr;
79 static detect_virt_stub_t detect_virt_stub = nullptr;
80 static clear_apx_test_state_t clear_apx_test_state_stub = nullptr;
81 static getCPUIDBrandString_stub_t getCPUIDBrandString_stub = nullptr;
82
83 bool VM_Version::supports_clflush() {
84 // clflush should always be available on x86_64
85 // if not we are in real trouble because we rely on it
86 // to flush the code cache.
87 // Unfortunately, Assembler::clflush is currently called as part
88 // of generation of the code cache flush routine. This happens
89 // under Universe::init before the processor features are set
90 // up. Assembler::flush calls this routine to check that clflush
91 // is allowed. So, we give the caller a free pass if Universe init
92 // is still in progress.
93 assert ((!Universe::is_fully_initialized() || _features.supports_feature(CPU_FLUSH)), "clflush should be available");
94 return true;
95 }
96
97 #define CPUID_STANDARD_FN 0x0
98 #define CPUID_STANDARD_FN_1 0x1
99 #define CPUID_STANDARD_FN_4 0x4
100 #define CPUID_STANDARD_FN_B 0xb
101
102 #define CPUID_EXTENDED_FN 0x80000000
103 #define CPUID_EXTENDED_FN_1 0x80000001
104 #define CPUID_EXTENDED_FN_2 0x80000002
105 #define CPUID_EXTENDED_FN_3 0x80000003
106 #define CPUID_EXTENDED_FN_4 0x80000004
107 #define CPUID_EXTENDED_FN_7 0x80000007
108 #define CPUID_EXTENDED_FN_8 0x80000008
109
110 class VM_Version_StubGenerator: public StubCodeGenerator {
111 public:
112
113 VM_Version_StubGenerator(CodeBuffer *c) : StubCodeGenerator(c) {}
114
115 address clear_apx_test_state() {
116 # define __ _masm->
117 address start = __ pc();
118 // EGPRs are call clobbered registers, Explicit clearing of r16 and r31 during signal
119 // handling guarantees that preserved register values post signal handling were
120 // re-instantiated by operating system and not because they were not modified externally.
121
122 bool save_apx = UseAPX;
123 VM_Version::set_apx_cpuFeatures();
124 UseAPX = true;
125 // EGPR state save/restoration.
126 __ mov64(r16, 0L);
127 __ mov64(r31, 0L);
128 UseAPX = save_apx;
129 VM_Version::clean_cpuFeatures();
130 __ ret(0);
131 return start;
132 }
133
134 address generate_get_cpu_info() {
135 // Flags to test CPU type.
136 const uint32_t HS_EFL_AC = 0x40000;
137 const uint32_t HS_EFL_ID = 0x200000;
138 // Values for when we don't have a CPUID instruction.
139 const int CPU_FAMILY_SHIFT = 8;
140 const uint32_t CPU_FAMILY_386 = (3 << CPU_FAMILY_SHIFT);
141 const uint32_t CPU_FAMILY_486 = (4 << CPU_FAMILY_SHIFT);
142 bool use_evex = FLAG_IS_DEFAULT(UseAVX) || (UseAVX > 2);
143
144 Label detect_486, cpu486, detect_586, std_cpuid1, std_cpuid4, std_cpuid24, std_cpuid29;
145 Label sef_cpuid, sefsl1_cpuid, ext_cpuid, ext_cpuid1, ext_cpuid5, ext_cpuid7;
146 Label ext_cpuid8, done, wrapup, vector_save_restore, apx_save_restore_warning, apx_xstate;
147 Label legacy_setup, save_restore_except, legacy_save_restore, start_simd_check;
148
149 StubCodeMark mark(this, "VM_Version", "get_cpu_info_stub");
150 # define __ _masm->
151
152 address start = __ pc();
153
154 //
155 // void get_cpu_info(VM_Version::CpuidInfo* cpuid_info);
156 //
157 // rcx and rdx are first and second argument registers on windows
158
159 __ push(rbp);
160 __ mov(rbp, c_rarg0); // cpuid_info address
161 __ push(rbx);
162 __ push(rsi);
163 __ pushf(); // preserve rbx, and flags
164 __ pop(rax);
165 __ push(rax);
166 __ mov(rcx, rax);
167 //
168 // if we are unable to change the AC flag, we have a 386
169 //
170 __ xorl(rax, HS_EFL_AC);
171 __ push(rax);
172 __ popf();
173 __ pushf();
174 __ pop(rax);
175 __ cmpptr(rax, rcx);
176 __ jccb(Assembler::notEqual, detect_486);
177
178 __ movl(rax, CPU_FAMILY_386);
179 __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax);
180 __ jmp(done);
181
182 //
183 // If we are unable to change the ID flag, we have a 486 which does
184 // not support the "cpuid" instruction.
185 //
186 __ bind(detect_486);
187 __ mov(rax, rcx);
188 __ xorl(rax, HS_EFL_ID);
189 __ push(rax);
190 __ popf();
191 __ pushf();
192 __ pop(rax);
193 __ cmpptr(rcx, rax);
194 __ jccb(Assembler::notEqual, detect_586);
195
196 __ bind(cpu486);
197 __ movl(rax, CPU_FAMILY_486);
198 __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax);
199 __ jmp(done);
200
201 //
202 // At this point, we have a chip which supports the "cpuid" instruction
203 //
204 __ bind(detect_586);
205 __ xorl(rax, rax);
206 __ cpuid();
207 __ orl(rax, rax);
208 __ jcc(Assembler::equal, cpu486); // if cpuid doesn't support an input
209 // value of at least 1, we give up and
210 // assume a 486
211 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset())));
212 __ movl(Address(rsi, 0), rax);
213 __ movl(Address(rsi, 4), rbx);
214 __ movl(Address(rsi, 8), rcx);
215 __ movl(Address(rsi,12), rdx);
216
217 __ cmpl(rax, 0xa); // Is cpuid(0xB) supported?
218 __ jccb(Assembler::belowEqual, std_cpuid4);
219
220 //
221 // cpuid(0xB) Processor Topology
222 //
223 __ movl(rax, 0xb);
224 __ xorl(rcx, rcx); // Threads level
225 __ cpuid();
226
227 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB0_offset())));
228 __ movl(Address(rsi, 0), rax);
229 __ movl(Address(rsi, 4), rbx);
230 __ movl(Address(rsi, 8), rcx);
231 __ movl(Address(rsi,12), rdx);
232
233 __ movl(rax, 0xb);
234 __ movl(rcx, 1); // Cores level
235 __ cpuid();
236 __ push(rax);
237 __ andl(rax, 0x1f); // Determine if valid topology level
238 __ orl(rax, rbx); // eax[4:0] | ebx[0:15] == 0 indicates invalid level
239 __ andl(rax, 0xffff);
240 __ pop(rax);
241 __ jccb(Assembler::equal, std_cpuid4);
242
243 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB1_offset())));
244 __ movl(Address(rsi, 0), rax);
245 __ movl(Address(rsi, 4), rbx);
246 __ movl(Address(rsi, 8), rcx);
247 __ movl(Address(rsi,12), rdx);
248
249 __ movl(rax, 0xb);
250 __ movl(rcx, 2); // Packages level
251 __ cpuid();
252 __ push(rax);
253 __ andl(rax, 0x1f); // Determine if valid topology level
254 __ orl(rax, rbx); // eax[4:0] | ebx[0:15] == 0 indicates invalid level
255 __ andl(rax, 0xffff);
256 __ pop(rax);
257 __ jccb(Assembler::equal, std_cpuid4);
258
259 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB2_offset())));
260 __ movl(Address(rsi, 0), rax);
261 __ movl(Address(rsi, 4), rbx);
262 __ movl(Address(rsi, 8), rcx);
263 __ movl(Address(rsi,12), rdx);
264
265 //
266 // cpuid(0x4) Deterministic cache params
267 //
268 __ bind(std_cpuid4);
269 __ movl(rax, 4);
270 __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x4) supported?
271 __ jccb(Assembler::greater, std_cpuid1);
272
273 __ xorl(rcx, rcx); // L1 cache
274 __ cpuid();
275 __ push(rax);
276 __ andl(rax, 0x1f); // Determine if valid cache parameters used
277 __ orl(rax, rax); // eax[4:0] == 0 indicates invalid cache
278 __ pop(rax);
279 __ jccb(Assembler::equal, std_cpuid1);
280
281 __ lea(rsi, Address(rbp, in_bytes(VM_Version::dcp_cpuid4_offset())));
282 __ movl(Address(rsi, 0), rax);
283 __ movl(Address(rsi, 4), rbx);
284 __ movl(Address(rsi, 8), rcx);
285 __ movl(Address(rsi,12), rdx);
286
287 //
288 // Standard cpuid(0x1)
289 //
290 __ bind(std_cpuid1);
291 __ movl(rax, 1);
292 __ cpuid();
293 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
294 __ movl(Address(rsi, 0), rax);
295 __ movl(Address(rsi, 4), rbx);
296 __ movl(Address(rsi, 8), rcx);
297 __ movl(Address(rsi,12), rdx);
298
299 //
300 // Check if OS has enabled XGETBV instruction to access XCR0
301 // (OSXSAVE feature flag) and CPU supports AVX
302 //
303 __ andl(rcx, 0x18000000); // cpuid1 bits osxsave | avx
304 __ cmpl(rcx, 0x18000000);
305 __ jccb(Assembler::notEqual, sef_cpuid); // jump if AVX is not supported
306
307 //
308 // XCR0, XFEATURE_ENABLED_MASK register
309 //
310 __ xorl(rcx, rcx); // zero for XCR0 register
311 __ xgetbv();
312 __ lea(rsi, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset())));
313 __ movl(Address(rsi, 0), rax);
314 __ movl(Address(rsi, 4), rdx);
315
316 //
317 // cpuid(0x7) Structured Extended Features Enumeration Leaf.
318 //
319 __ bind(sef_cpuid);
320 __ movl(rax, 7);
321 __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x7) supported?
322 __ jccb(Assembler::greater, ext_cpuid);
323 // ECX = 0
324 __ xorl(rcx, rcx);
325 __ cpuid();
326 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset())));
327 __ movl(Address(rsi, 0), rax);
328 __ movl(Address(rsi, 4), rbx);
329 __ movl(Address(rsi, 8), rcx);
330 __ movl(Address(rsi, 12), rdx);
331
332 //
333 // cpuid(0x7) Structured Extended Features Enumeration Sub-Leaf 1.
334 //
335 __ bind(sefsl1_cpuid);
336 __ movl(rax, 7);
337 __ movl(rcx, 1);
338 __ cpuid();
339 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sefsl1_cpuid7_offset())));
340 __ movl(Address(rsi, 0), rax);
341 __ movl(Address(rsi, 4), rdx);
342
343 //
344 // cpuid(0x29) APX NCI NDD NF (EAX = 29H, ECX = 0).
345 //
346 __ bind(std_cpuid29);
347 __ movl(rax, 0x29);
348 __ movl(rcx, 0);
349 __ cpuid();
350 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid29_offset())));
351 __ movl(Address(rsi, 0), rbx);
352
353 //
354 // cpuid(0x24) Converged Vector ISA Main Leaf (EAX = 24H, ECX = 0).
355 //
356 __ bind(std_cpuid24);
357 __ movl(rax, 0x24);
358 __ movl(rcx, 0);
359 __ cpuid();
360 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid24_offset())));
361 __ movl(Address(rsi, 0), rax);
362 __ movl(Address(rsi, 4), rbx);
363
364 //
365 // Extended cpuid(0x80000000)
366 //
367 __ bind(ext_cpuid);
368 __ movl(rax, 0x80000000);
369 __ cpuid();
370 __ cmpl(rax, 0x80000000); // Is cpuid(0x80000001) supported?
371 __ jcc(Assembler::belowEqual, done);
372 __ cmpl(rax, 0x80000004); // Is cpuid(0x80000005) supported?
373 __ jcc(Assembler::belowEqual, ext_cpuid1);
374 __ cmpl(rax, 0x80000006); // Is cpuid(0x80000007) supported?
375 __ jccb(Assembler::belowEqual, ext_cpuid5);
376 __ cmpl(rax, 0x80000007); // Is cpuid(0x80000008) supported?
377 __ jccb(Assembler::belowEqual, ext_cpuid7);
378 __ cmpl(rax, 0x80000008); // Is cpuid(0x80000009 and above) supported?
379 __ jccb(Assembler::belowEqual, ext_cpuid8);
380 __ cmpl(rax, 0x8000001E); // Is cpuid(0x8000001E) supported?
381 __ jccb(Assembler::below, ext_cpuid8);
382 //
383 // Extended cpuid(0x8000001E)
384 //
385 __ movl(rax, 0x8000001E);
386 __ cpuid();
387 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid1E_offset())));
388 __ movl(Address(rsi, 0), rax);
389 __ movl(Address(rsi, 4), rbx);
390 __ movl(Address(rsi, 8), rcx);
391 __ movl(Address(rsi,12), rdx);
392
393 //
394 // Extended cpuid(0x80000008)
395 //
396 __ bind(ext_cpuid8);
397 __ movl(rax, 0x80000008);
398 __ cpuid();
399 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid8_offset())));
400 __ movl(Address(rsi, 0), rax);
401 __ movl(Address(rsi, 4), rbx);
402 __ movl(Address(rsi, 8), rcx);
403 __ movl(Address(rsi,12), rdx);
404
405 //
406 // Extended cpuid(0x80000007)
407 //
408 __ bind(ext_cpuid7);
409 __ movl(rax, 0x80000007);
410 __ cpuid();
411 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid7_offset())));
412 __ movl(Address(rsi, 0), rax);
413 __ movl(Address(rsi, 4), rbx);
414 __ movl(Address(rsi, 8), rcx);
415 __ movl(Address(rsi,12), rdx);
416
417 //
418 // Extended cpuid(0x80000005)
419 //
420 __ bind(ext_cpuid5);
421 __ movl(rax, 0x80000005);
422 __ cpuid();
423 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid5_offset())));
424 __ movl(Address(rsi, 0), rax);
425 __ movl(Address(rsi, 4), rbx);
426 __ movl(Address(rsi, 8), rcx);
427 __ movl(Address(rsi,12), rdx);
428
429 //
430 // Extended cpuid(0x80000001)
431 //
432 __ bind(ext_cpuid1);
433 __ movl(rax, 0x80000001);
434 __ cpuid();
435 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid1_offset())));
436 __ movl(Address(rsi, 0), rax);
437 __ movl(Address(rsi, 4), rbx);
438 __ movl(Address(rsi, 8), rcx);
439 __ movl(Address(rsi,12), rdx);
440
441 //
442 // Check if OS has enabled XGETBV instruction to access XCR0
443 // (OSXSAVE feature flag) and CPU supports APX
444 //
445 // To enable APX, check CPUID.EAX=7.ECX=1.EDX[21] bit for HW support
446 // and XCRO[19] bit for OS support to save/restore extended GPR state.
447 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sefsl1_cpuid7_offset())));
448 __ movl(rax, 0x200000);
449 __ andl(rax, Address(rsi, 4));
450 __ jcc(Assembler::equal, vector_save_restore);
451 // check _cpuid_info.xem_xcr0_eax.bits.apx_f
452 __ movl(rax, 0x80000);
453 __ andl(rax, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); // xcr0 bits apx_f
454 __ jcc(Assembler::equal, vector_save_restore);
455
456 bool save_apx = UseAPX;
457 VM_Version::set_apx_cpuFeatures();
458 UseAPX = true;
459 __ mov64(r16, VM_Version::egpr_test_value());
460 __ mov64(r31, VM_Version::egpr_test_value());
461 __ xorl(rsi, rsi);
462 VM_Version::set_cpuinfo_segv_addr_apx(__ pc());
463 // Generate SEGV
464 __ movl(rax, Address(rsi, 0));
465
466 VM_Version::set_cpuinfo_cont_addr_apx(__ pc());
467 __ lea(rsi, Address(rbp, in_bytes(VM_Version::apx_save_offset())));
468 __ movq(Address(rsi, 0), r16);
469 __ movq(Address(rsi, 8), r31);
470
471 //
472 // Query CPUID 0xD.19 for APX XSAVE offset
473 // Extended State Enumeration Sub-leaf 19 (APX)
474 // EAX = size of APX state (should be 128)
475 // EBX = offset in standard XSAVE format
476 //
477 __ movl(rax, 0xD);
478 __ movl(rcx, 19);
479 __ cpuid();
480 __ lea(rsi, Address(rbp, in_bytes(VM_Version::apx_xstate_size_offset())));
481 __ movl(Address(rsi, 0), rax);
482 __ lea(rsi, Address(rbp, in_bytes(VM_Version::apx_xstate_offset_offset())));
483 __ movl(Address(rsi, 0), rbx);
484
485 UseAPX = save_apx;
486 __ bind(vector_save_restore);
487 //
488 // Check if OS has enabled XGETBV instruction to access XCR0
489 // (OSXSAVE feature flag) and CPU supports AVX
490 //
491 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
492 __ movl(rcx, 0x18000000); // cpuid1 bits osxsave | avx
493 __ andl(rcx, Address(rsi, 8)); // cpuid1 bits osxsave | avx
494 __ cmpl(rcx, 0x18000000);
495 __ jccb(Assembler::notEqual, done); // jump if AVX is not supported
496
497 __ movl(rax, 0x6);
498 __ andl(rax, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); // xcr0 bits sse | ymm
499 __ cmpl(rax, 0x6);
500 __ jccb(Assembler::equal, start_simd_check); // return if AVX is not supported
501
502 // we need to bridge farther than imm8, so we use this island as a thunk
503 __ bind(done);
504 __ jmp(wrapup);
505
506 __ bind(start_simd_check);
507 //
508 // Some OSs have a bug when upper 128/256bits of YMM/ZMM
509 // registers are not restored after a signal processing.
510 // Generate SEGV here (reference through null)
511 // and check upper YMM/ZMM bits after it.
512 //
513 int saved_useavx = UseAVX;
514 int saved_usesse = UseSSE;
515
516 // If UseAVX is uninitialized or is set by the user to include EVEX
517 if (use_evex) {
518 // check _cpuid_info.sef_cpuid7_ebx.bits.avx512f
519 // OR check _cpuid_info.sefsl1_cpuid7_edx.bits.avx10
520 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset())));
521 __ movl(rax, 0x10000);
522 __ andl(rax, Address(rsi, 4));
523 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sefsl1_cpuid7_offset())));
524 __ movl(rbx, 0x80000);
525 __ andl(rbx, Address(rsi, 4));
526 __ orl(rax, rbx);
527 __ jccb(Assembler::equal, legacy_setup); // jump if EVEX is not supported
528 // check _cpuid_info.xem_xcr0_eax.bits.opmask
529 // check _cpuid_info.xem_xcr0_eax.bits.zmm512
530 // check _cpuid_info.xem_xcr0_eax.bits.zmm32
531 __ movl(rax, 0xE0);
532 __ andl(rax, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); // xcr0 bits sse | ymm
533 __ cmpl(rax, 0xE0);
534 __ jccb(Assembler::notEqual, legacy_setup); // jump if EVEX is not supported
535
536 if (FLAG_IS_DEFAULT(UseAVX)) {
537 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
538 __ movl(rax, Address(rsi, 0));
539 __ cmpl(rax, 0x50654); // If it is Skylake
540 __ jcc(Assembler::equal, legacy_setup);
541 }
542 // EVEX setup: run in lowest evex mode
543 VM_Version::set_evex_cpuFeatures(); // Enable temporary to pass asserts
544 UseAVX = 3;
545 UseSSE = 2;
546 #ifdef _WINDOWS
547 // xmm5-xmm15 are not preserved by caller on windows
548 // https://msdn.microsoft.com/en-us/library/9z1stfyw.aspx
549 __ subptr(rsp, 64);
550 __ evmovdqul(Address(rsp, 0), xmm7, Assembler::AVX_512bit);
551 __ subptr(rsp, 64);
552 __ evmovdqul(Address(rsp, 0), xmm8, Assembler::AVX_512bit);
553 __ subptr(rsp, 64);
554 __ evmovdqul(Address(rsp, 0), xmm31, Assembler::AVX_512bit);
555 #endif // _WINDOWS
556
557 // load value into all 64 bytes of zmm7 register
558 __ movl(rcx, VM_Version::ymm_test_value());
559 __ movdl(xmm0, rcx);
560 __ vpbroadcastd(xmm0, xmm0, Assembler::AVX_512bit);
561 __ evmovdqul(xmm7, xmm0, Assembler::AVX_512bit);
562 __ evmovdqul(xmm8, xmm0, Assembler::AVX_512bit);
563 __ evmovdqul(xmm31, xmm0, Assembler::AVX_512bit);
564 VM_Version::clean_cpuFeatures();
565 __ jmp(save_restore_except);
566 }
567
568 __ bind(legacy_setup);
569 // AVX setup
570 VM_Version::set_avx_cpuFeatures(); // Enable temporary to pass asserts
571 UseAVX = 1;
572 UseSSE = 2;
573 #ifdef _WINDOWS
574 __ subptr(rsp, 32);
575 __ vmovdqu(Address(rsp, 0), xmm7);
576 __ subptr(rsp, 32);
577 __ vmovdqu(Address(rsp, 0), xmm8);
578 __ subptr(rsp, 32);
579 __ vmovdqu(Address(rsp, 0), xmm15);
580 #endif // _WINDOWS
581
582 // load value into all 32 bytes of ymm7 register
583 __ movl(rcx, VM_Version::ymm_test_value());
584
585 __ movdl(xmm0, rcx);
586 __ pshufd(xmm0, xmm0, 0x00);
587 __ vinsertf128_high(xmm0, xmm0);
588 __ vmovdqu(xmm7, xmm0);
589 __ vmovdqu(xmm8, xmm0);
590 __ vmovdqu(xmm15, xmm0);
591 VM_Version::clean_cpuFeatures();
592
593 __ bind(save_restore_except);
594 __ xorl(rsi, rsi);
595 VM_Version::set_cpuinfo_segv_addr(__ pc());
596 // Generate SEGV
597 __ movl(rax, Address(rsi, 0));
598
599 VM_Version::set_cpuinfo_cont_addr(__ pc());
600 // Returns here after signal. Save xmm0 to check it later.
601
602 // If UseAVX is uninitialized or is set by the user to include EVEX
603 if (use_evex) {
604 // check _cpuid_info.sef_cpuid7_ebx.bits.avx512f
605 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset())));
606 __ movl(rax, 0x10000);
607 __ andl(rax, Address(rsi, 4));
608 __ jcc(Assembler::equal, legacy_save_restore);
609 // check _cpuid_info.xem_xcr0_eax.bits.opmask
610 // check _cpuid_info.xem_xcr0_eax.bits.zmm512
611 // check _cpuid_info.xem_xcr0_eax.bits.zmm32
612 __ movl(rax, 0xE0);
613 __ andl(rax, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); // xcr0 bits sse | ymm
614 __ cmpl(rax, 0xE0);
615 __ jcc(Assembler::notEqual, legacy_save_restore);
616
617 if (FLAG_IS_DEFAULT(UseAVX)) {
618 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
619 __ movl(rax, Address(rsi, 0));
620 __ cmpl(rax, 0x50654); // If it is Skylake
621 __ jcc(Assembler::equal, legacy_save_restore);
622 }
623 // EVEX check: run in lowest evex mode
624 VM_Version::set_evex_cpuFeatures(); // Enable temporary to pass asserts
625 UseAVX = 3;
626 UseSSE = 2;
627 __ lea(rsi, Address(rbp, in_bytes(VM_Version::zmm_save_offset())));
628 __ evmovdqul(Address(rsi, 0), xmm0, Assembler::AVX_512bit);
629 __ evmovdqul(Address(rsi, 64), xmm7, Assembler::AVX_512bit);
630 __ evmovdqul(Address(rsi, 128), xmm8, Assembler::AVX_512bit);
631 __ evmovdqul(Address(rsi, 192), xmm31, Assembler::AVX_512bit);
632
633 #ifdef _WINDOWS
634 __ evmovdqul(xmm31, Address(rsp, 0), Assembler::AVX_512bit);
635 __ addptr(rsp, 64);
636 __ evmovdqul(xmm8, Address(rsp, 0), Assembler::AVX_512bit);
637 __ addptr(rsp, 64);
638 __ evmovdqul(xmm7, Address(rsp, 0), Assembler::AVX_512bit);
639 __ addptr(rsp, 64);
640 #endif // _WINDOWS
641 generate_vzeroupper(wrapup);
642 VM_Version::clean_cpuFeatures();
643 UseAVX = saved_useavx;
644 UseSSE = saved_usesse;
645 __ jmp(wrapup);
646 }
647
648 __ bind(legacy_save_restore);
649 // AVX check
650 VM_Version::set_avx_cpuFeatures(); // Enable temporary to pass asserts
651 UseAVX = 1;
652 UseSSE = 2;
653 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ymm_save_offset())));
654 __ vmovdqu(Address(rsi, 0), xmm0);
655 __ vmovdqu(Address(rsi, 32), xmm7);
656 __ vmovdqu(Address(rsi, 64), xmm8);
657 __ vmovdqu(Address(rsi, 96), xmm15);
658
659 #ifdef _WINDOWS
660 __ vmovdqu(xmm15, Address(rsp, 0));
661 __ addptr(rsp, 32);
662 __ vmovdqu(xmm8, Address(rsp, 0));
663 __ addptr(rsp, 32);
664 __ vmovdqu(xmm7, Address(rsp, 0));
665 __ addptr(rsp, 32);
666 #endif // _WINDOWS
667
668 generate_vzeroupper(wrapup);
669 VM_Version::clean_cpuFeatures();
670 UseAVX = saved_useavx;
671 UseSSE = saved_usesse;
672
673 __ bind(wrapup);
674 __ popf();
675 __ pop(rsi);
676 __ pop(rbx);
677 __ pop(rbp);
678 __ ret(0);
679
680 # undef __
681
682 return start;
683 };
684 void generate_vzeroupper(Label& L_wrapup) {
685 # define __ _masm->
686 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset())));
687 __ cmpl(Address(rsi, 4), 0x756e6547); // 'uneG'
688 __ jcc(Assembler::notEqual, L_wrapup);
689 __ movl(rcx, 0x0FFF0FF0);
690 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
691 __ andl(rcx, Address(rsi, 0));
692 __ cmpl(rcx, 0x00050670); // If it is Xeon Phi 3200/5200/7200
693 __ jcc(Assembler::equal, L_wrapup);
694 __ cmpl(rcx, 0x00080650); // If it is Future Xeon Phi
695 __ jcc(Assembler::equal, L_wrapup);
696 // vzeroupper() will use a pre-computed instruction sequence that we
697 // can't compute until after we've determined CPU capabilities. Use
698 // uncached variant here directly to be able to bootstrap correctly
699 __ vzeroupper_uncached();
700 # undef __
701 }
702 address generate_detect_virt() {
703 StubCodeMark mark(this, "VM_Version", "detect_virt_stub");
704 # define __ _masm->
705
706 address start = __ pc();
707
708 // Evacuate callee-saved registers
709 __ push(rbp);
710 __ push(rbx);
711 __ push(rsi); // for Windows
712
713 __ mov(rax, c_rarg0); // CPUID leaf
714 __ mov(rsi, c_rarg1); // register array address (eax, ebx, ecx, edx)
715
716 __ cpuid();
717
718 // Store result to register array
719 __ movl(Address(rsi, 0), rax);
720 __ movl(Address(rsi, 4), rbx);
721 __ movl(Address(rsi, 8), rcx);
722 __ movl(Address(rsi, 12), rdx);
723
724 // Epilogue
725 __ pop(rsi);
726 __ pop(rbx);
727 __ pop(rbp);
728 __ ret(0);
729
730 # undef __
731
732 return start;
733 };
734
735
736 address generate_getCPUIDBrandString(void) {
737 // Flags to test CPU type.
738 const uint32_t HS_EFL_AC = 0x40000;
739 const uint32_t HS_EFL_ID = 0x200000;
740 // Values for when we don't have a CPUID instruction.
741 const int CPU_FAMILY_SHIFT = 8;
742 const uint32_t CPU_FAMILY_386 = (3 << CPU_FAMILY_SHIFT);
743 const uint32_t CPU_FAMILY_486 = (4 << CPU_FAMILY_SHIFT);
744
745 Label detect_486, cpu486, detect_586, done, ext_cpuid;
746
747 StubCodeMark mark(this, "VM_Version", "getCPUIDNameInfo_stub");
748 # define __ _masm->
749
750 address start = __ pc();
751
752 //
753 // void getCPUIDBrandString(VM_Version::CpuidInfo* cpuid_info);
754 //
755 // rcx and rdx are first and second argument registers on windows
756
757 __ push(rbp);
758 __ mov(rbp, c_rarg0); // cpuid_info address
759 __ push(rbx);
760 __ push(rsi);
761 __ pushf(); // preserve rbx, and flags
762 __ pop(rax);
763 __ push(rax);
764 __ mov(rcx, rax);
765 //
766 // if we are unable to change the AC flag, we have a 386
767 //
768 __ xorl(rax, HS_EFL_AC);
769 __ push(rax);
770 __ popf();
771 __ pushf();
772 __ pop(rax);
773 __ cmpptr(rax, rcx);
774 __ jccb(Assembler::notEqual, detect_486);
775
776 __ movl(rax, CPU_FAMILY_386);
777 __ jmp(done);
778
779 //
780 // If we are unable to change the ID flag, we have a 486 which does
781 // not support the "cpuid" instruction.
782 //
783 __ bind(detect_486);
784 __ mov(rax, rcx);
785 __ xorl(rax, HS_EFL_ID);
786 __ push(rax);
787 __ popf();
788 __ pushf();
789 __ pop(rax);
790 __ cmpptr(rcx, rax);
791 __ jccb(Assembler::notEqual, detect_586);
792
793 __ bind(cpu486);
794 __ movl(rax, CPU_FAMILY_486);
795 __ jmp(done);
796
797 //
798 // At this point, we have a chip which supports the "cpuid" instruction
799 //
800 __ bind(detect_586);
801 __ xorl(rax, rax);
802 __ cpuid();
803 __ orl(rax, rax);
804 __ jcc(Assembler::equal, cpu486); // if cpuid doesn't support an input
805 // value of at least 1, we give up and
806 // assume a 486
807
808 //
809 // Extended cpuid(0x80000000) for processor brand string detection
810 //
811 __ bind(ext_cpuid);
812 __ movl(rax, CPUID_EXTENDED_FN);
813 __ cpuid();
814 __ cmpl(rax, CPUID_EXTENDED_FN_4);
815 __ jcc(Assembler::below, done);
816
817 //
818 // Extended cpuid(0x80000002) // first 16 bytes in brand string
819 //
820 __ movl(rax, CPUID_EXTENDED_FN_2);
821 __ cpuid();
822 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_0_offset())));
823 __ movl(Address(rsi, 0), rax);
824 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_1_offset())));
825 __ movl(Address(rsi, 0), rbx);
826 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_2_offset())));
827 __ movl(Address(rsi, 0), rcx);
828 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_3_offset())));
829 __ movl(Address(rsi,0), rdx);
830
831 //
832 // Extended cpuid(0x80000003) // next 16 bytes in brand string
833 //
834 __ movl(rax, CPUID_EXTENDED_FN_3);
835 __ cpuid();
836 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_4_offset())));
837 __ movl(Address(rsi, 0), rax);
838 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_5_offset())));
839 __ movl(Address(rsi, 0), rbx);
840 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_6_offset())));
841 __ movl(Address(rsi, 0), rcx);
842 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_7_offset())));
843 __ movl(Address(rsi,0), rdx);
844
845 //
846 // Extended cpuid(0x80000004) // last 16 bytes in brand string
847 //
848 __ movl(rax, CPUID_EXTENDED_FN_4);
849 __ cpuid();
850 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_8_offset())));
851 __ movl(Address(rsi, 0), rax);
852 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_9_offset())));
853 __ movl(Address(rsi, 0), rbx);
854 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_10_offset())));
855 __ movl(Address(rsi, 0), rcx);
856 __ lea(rsi, Address(rbp, in_bytes(VM_Version::proc_name_11_offset())));
857 __ movl(Address(rsi,0), rdx);
858
859 //
860 // return
861 //
862 __ bind(done);
863 __ popf();
864 __ pop(rsi);
865 __ pop(rbx);
866 __ pop(rbp);
867 __ ret(0);
868
869 # undef __
870
871 return start;
872 };
873 };
874
875 void VM_Version::get_processor_features() {
876
877 _cpu = 4; // 486 by default
878 _model = 0;
879 _stepping = 0;
880 _logical_processors_per_package = 1;
881 // i486 internal cache is both I&D and has a 16-byte line size
882 _L1_data_cache_line_size = 16;
883
884 // Get raw processor info
885
886 get_cpu_info_stub(&_cpuid_info);
887
888 assert_is_initialized();
889 _cpu = extended_cpu_family();
890 _model = extended_cpu_model();
891 _stepping = cpu_stepping();
892
893 if (cpu_family() > 4) { // it supports CPUID
894 _features = _cpuid_info.feature_flags(); // These can be changed by VM settings
895 _cpu_features = _features; // Preserve features
896 // Logical processors are only available on P4s and above,
897 // and only if hyperthreading is available.
898 _logical_processors_per_package = logical_processor_count();
899 _L1_data_cache_line_size = L1_line_size();
900 }
901
902 // xchg and xadd instructions
903 _supports_atomic_getset4 = true;
904 _supports_atomic_getadd4 = true;
905 _supports_atomic_getset8 = true;
906 _supports_atomic_getadd8 = true;
907
908 // OS should support SSE for x64 and hardware should support at least SSE2.
909 if (!VM_Version::supports_sse2()) {
910 vm_exit_during_initialization("Unknown x64 processor: SSE2 not supported");
911 }
912 // in 64 bit the use of SSE2 is the minimum
913 if (UseSSE < 2) UseSSE = 2;
914
915 // flush_icache_stub have to be generated first.
916 // That is why Icache line size is hard coded in ICache class,
917 // see icache_x86.hpp. It is also the reason why we can't use
918 // clflush instruction in 32-bit VM since it could be running
919 // on CPU which does not support it.
920 //
921 // The only thing we can do is to verify that flushed
922 // ICache::line_size has correct value.
923 guarantee(_cpuid_info.std_cpuid1_edx.bits.clflush != 0, "clflush is not supported");
924 // clflush_size is size in quadwords (8 bytes).
925 guarantee(_cpuid_info.std_cpuid1_ebx.bits.clflush_size == 8, "such clflush size is not supported");
926
927 // assigning this field effectively enables Unsafe.writebackMemory()
928 // by initing UnsafeConstant.DATA_CACHE_LINE_FLUSH_SIZE to non-zero
929 // that is only implemented on x86_64 and only if the OS plays ball
930 if (os::supports_map_sync()) {
931 // publish data cache line flush size to generic field, otherwise
932 // let if default to zero thereby disabling writeback
933 _data_cache_line_flush_size = _cpuid_info.std_cpuid1_ebx.bits.clflush_size * 8;
934 }
935
936 // Check if processor has Intel Ecore
937 if (FLAG_IS_DEFAULT(EnableX86ECoreOpts) && is_intel() && is_intel_server_family() &&
938 (supports_hybrid() ||
939 _model == 0xAF /* Xeon 6 E-cores (Sierra Forest) */ ||
940 _model == 0xDD /* Xeon 6+ E-cores (Clearwater Forest) */ )) {
941 FLAG_SET_DEFAULT(EnableX86ECoreOpts, true);
942 }
943
944 if (UseSSE < 4) {
945 _features.clear_feature(CPU_SSE4_1);
946 _features.clear_feature(CPU_SSE4_2);
947 }
948
949 if (UseSSE < 3) {
950 _features.clear_feature(CPU_SSE3);
951 _features.clear_feature(CPU_SSSE3);
952 _features.clear_feature(CPU_SSE4A);
953 }
954
955 if (UseSSE < 2)
956 _features.clear_feature(CPU_SSE2);
957
958 if (UseSSE < 1)
959 _features.clear_feature(CPU_SSE);
960
961 // ZX cpus specific settings
962 if (is_zx() && FLAG_IS_DEFAULT(UseAVX)) {
963 if (cpu_family() == 7) {
964 if (extended_cpu_model() == 0x5B || extended_cpu_model() == 0x6B) {
965 UseAVX = 1;
966 } else if (extended_cpu_model() == 0x1B || extended_cpu_model() == 0x3B) {
967 UseAVX = 0;
968 }
969 } else if (cpu_family() == 6) {
970 UseAVX = 0;
971 }
972 }
973
974 // UseSSE is set to the smaller of what hardware supports and what
975 // the command line requires. I.e., you cannot set UseSSE to 2 on
976 // older Pentiums which do not support it.
977 int use_sse_limit = 0;
978 if (UseSSE > 0) {
979 if (UseSSE > 3 && supports_sse4_1()) {
980 use_sse_limit = 4;
981 } else if (UseSSE > 2 && supports_sse3()) {
982 use_sse_limit = 3;
983 } else if (UseSSE > 1 && supports_sse2()) {
984 use_sse_limit = 2;
985 } else if (UseSSE > 0 && supports_sse()) {
986 use_sse_limit = 1;
987 } else {
988 use_sse_limit = 0;
989 }
990 }
991 if (FLAG_IS_DEFAULT(UseSSE)) {
992 FLAG_SET_DEFAULT(UseSSE, use_sse_limit);
993 } else if (UseSSE > use_sse_limit) {
994 warning("UseSSE=%d is not supported on this CPU, setting it to UseSSE=%d", UseSSE, use_sse_limit);
995 FLAG_SET_DEFAULT(UseSSE, use_sse_limit);
996 }
997
998 // first try initial setting and detect what we can support
999 int use_avx_limit = 0;
1000 if (UseAVX > 0) {
1001 if (UseSSE < 4) {
1002 // Don't use AVX if SSE is unavailable or has been disabled.
1003 use_avx_limit = 0;
1004 } else if (UseAVX > 2 && supports_evex()) {
1005 use_avx_limit = 3;
1006 } else if (UseAVX > 1 && supports_avx2()) {
1007 use_avx_limit = 2;
1008 } else if (UseAVX > 0 && supports_avx()) {
1009 use_avx_limit = 1;
1010 } else {
1011 use_avx_limit = 0;
1012 }
1013 }
1014 if (FLAG_IS_DEFAULT(UseAVX)) {
1015 // Don't use AVX-512 on older Skylakes unless explicitly requested.
1016 if (use_avx_limit > 2 && is_intel_skylake() && _stepping < 5) {
1017 FLAG_SET_DEFAULT(UseAVX, 2);
1018 } else {
1019 FLAG_SET_DEFAULT(UseAVX, use_avx_limit);
1020 }
1021 }
1022
1023 if (UseAVX > use_avx_limit) {
1024 if (UseSSE < 4) {
1025 warning("UseAVX=%d requires UseSSE=4, setting it to UseAVX=0", UseAVX);
1026 } else {
1027 warning("UseAVX=%d is not supported on this CPU, setting it to UseAVX=%d", UseAVX, use_avx_limit);
1028 }
1029 FLAG_SET_DEFAULT(UseAVX, use_avx_limit);
1030 }
1031
1032 if (UseAVX < 3) {
1033 _features.clear_feature(CPU_AVX512F);
1034 _features.clear_feature(CPU_AVX512DQ);
1035 _features.clear_feature(CPU_AVX512CD);
1036 _features.clear_feature(CPU_AVX512BW);
1037 _features.clear_feature(CPU_AVX512ER);
1038 _features.clear_feature(CPU_AVX512PF);
1039 _features.clear_feature(CPU_AVX512VL);
1040 _features.clear_feature(CPU_AVX512_VPOPCNTDQ);
1041 _features.clear_feature(CPU_AVX512_VPCLMULQDQ);
1042 _features.clear_feature(CPU_AVX512_VAES);
1043 _features.clear_feature(CPU_AVX512_VNNI);
1044 _features.clear_feature(CPU_AVX512_VBMI);
1045 _features.clear_feature(CPU_AVX512_VBMI2);
1046 _features.clear_feature(CPU_AVX512_BITALG);
1047 _features.clear_feature(CPU_AVX512_IFMA);
1048 _features.clear_feature(CPU_APX_F);
1049 _features.clear_feature(CPU_AVX512_FP16);
1050 _features.clear_feature(CPU_AVX10_1);
1051 _features.clear_feature(CPU_AVX10_2);
1052 }
1053
1054
1055 if (UseAVX < 2) {
1056 _features.clear_feature(CPU_AVX2);
1057 _features.clear_feature(CPU_AVX_IFMA);
1058 }
1059
1060 if (UseAVX < 1) {
1061 _features.clear_feature(CPU_AVX);
1062 _features.clear_feature(CPU_VZEROUPPER);
1063 _features.clear_feature(CPU_F16C);
1064 _features.clear_feature(CPU_SHA512);
1065 }
1066
1067 if (logical_processors_per_package() == 1) {
1068 // HT processor could be installed on a system which doesn't support HT.
1069 _features.clear_feature(CPU_HT);
1070 }
1071
1072 if (is_intel()) { // Intel cpus specific settings
1073 if (is_knights_family()) {
1074 _features.clear_feature(CPU_VZEROUPPER);
1075 _features.clear_feature(CPU_AVX512BW);
1076 _features.clear_feature(CPU_AVX512VL);
1077 _features.clear_feature(CPU_APX_F);
1078 _features.clear_feature(CPU_AVX512DQ);
1079 _features.clear_feature(CPU_AVX512_VNNI);
1080 _features.clear_feature(CPU_AVX512_VAES);
1081 _features.clear_feature(CPU_AVX512_VPOPCNTDQ);
1082 _features.clear_feature(CPU_AVX512_VPCLMULQDQ);
1083 _features.clear_feature(CPU_AVX512_VBMI);
1084 _features.clear_feature(CPU_AVX512_VBMI2);
1085 _features.clear_feature(CPU_CLWB);
1086 _features.clear_feature(CPU_FLUSHOPT);
1087 _features.clear_feature(CPU_GFNI);
1088 _features.clear_feature(CPU_AVX512_BITALG);
1089 _features.clear_feature(CPU_AVX512_IFMA);
1090 _features.clear_feature(CPU_AVX_IFMA);
1091 _features.clear_feature(CPU_AVX512_FP16);
1092 _features.clear_feature(CPU_AVX10_1);
1093 _features.clear_feature(CPU_AVX10_2);
1094 }
1095 }
1096
1097 // Currently APX support is only enabled for targets supporting AVX512VL feature.
1098 if (supports_apx_f() && os_supports_apx_egprs() && supports_avx512vl()) {
1099 if (FLAG_IS_DEFAULT(UseAPX)) {
1100 UseAPX = false; // by default UseAPX is false
1101 _features.clear_feature(CPU_APX_F);
1102 } else if (!UseAPX) {
1103 _features.clear_feature(CPU_APX_F);
1104 }
1105 } else if (UseAPX) {
1106 if (!FLAG_IS_DEFAULT(UseAPX)) {
1107 warning("APX is not supported on this CPU, setting it to false)");
1108 }
1109 FLAG_SET_DEFAULT(UseAPX, false);
1110 }
1111
1112 CHECK_CPU_FEATURE(supports_clmul, CLMUL);
1113 CHECK_CPU_FEATURE(supports_aes, AES);
1114 CHECK_CPU_FEATURE(supports_fma, FMA);
1115
1116 if (supports_sha() || (supports_avx2() && supports_bmi2())) {
1117 if (FLAG_IS_DEFAULT(UseSHA)) {
1118 UseSHA = true;
1119 } else if (!UseSHA) {
1120 _features.clear_feature(CPU_SHA);
1121 }
1122 } else if (UseSHA) {
1123 if (!FLAG_IS_DEFAULT(UseSHA)) {
1124 warning("SHA instructions are not available on this CPU");
1125 }
1126 FLAG_SET_DEFAULT(UseSHA, false);
1127 }
1128
1129 if (FLAG_IS_DEFAULT(IntelJccErratumMitigation)) {
1130 _has_intel_jcc_erratum = compute_has_intel_jcc_erratum();
1131 FLAG_SET_ERGO(IntelJccErratumMitigation, _has_intel_jcc_erratum);
1132 } else {
1133 _has_intel_jcc_erratum = IntelJccErratumMitigation;
1134 }
1135
1136 assert(supports_clflush(), "Always present");
1137 if (X86ICacheSync == -1) {
1138 // Auto-detect, choosing the best performant one that still flushes
1139 // the cache. We could switch to CPUID/SERIALIZE ("4"/"5") going forward.
1140 if (supports_clwb()) {
1141 FLAG_SET_ERGO(X86ICacheSync, 3);
1142 } else if (supports_clflushopt()) {
1143 FLAG_SET_ERGO(X86ICacheSync, 2);
1144 } else {
1145 FLAG_SET_ERGO(X86ICacheSync, 1);
1146 }
1147 } else {
1148 if ((X86ICacheSync == 2) && !supports_clflushopt()) {
1149 vm_exit_during_initialization("CPU does not support CLFLUSHOPT, unable to use X86ICacheSync=2");
1150 }
1151 if ((X86ICacheSync == 3) && !supports_clwb()) {
1152 vm_exit_during_initialization("CPU does not support CLWB, unable to use X86ICacheSync=3");
1153 }
1154 if ((X86ICacheSync == 5) && !supports_serialize()) {
1155 vm_exit_during_initialization("CPU does not support SERIALIZE, unable to use X86ICacheSync=5");
1156 }
1157 }
1158
1159 stringStream ss(2048);
1160 if (supports_hybrid()) {
1161 ss.print("(hybrid)");
1162 } else {
1163 ss.print("(%u cores per cpu, %u threads per core)", cores_per_cpu(), threads_per_core());
1164 }
1165 ss.print(" family %d model %d stepping %d microcode 0x%x",
1166 cpu_family(), _model, _stepping, os::cpu_microcode_revision());
1167 ss.print(", ");
1168 int features_offset = (int)ss.size();
1169 insert_features_names(_features, ss);
1170
1171 _cpu_info_string = ss.as_string(true);
1172 _features_string = _cpu_info_string + features_offset;
1173
1174 // Use AES instructions if available.
1175 if (supports_aes()) {
1176 if (supports_sse3()) {
1177 if (FLAG_IS_DEFAULT(UseAESIntrinsics)) {
1178 FLAG_SET_DEFAULT(UseAESIntrinsics, true);
1179 }
1180 } else if (UseAESIntrinsics) {
1181 // The AES intrinsic stubs require AES instruction support (of course)
1182 // but also require sse3 mode or higher for instructions it use.
1183 if (!FLAG_IS_DEFAULT(UseAESIntrinsics)) {
1184 warning("X86 AES intrinsics require SSE3 instructions or higher. Intrinsics will be disabled.");
1185 }
1186 FLAG_SET_DEFAULT(UseAESIntrinsics, false);
1187 }
1188 if (!UseAESIntrinsics) {
1189 if (UseAESCTRIntrinsics) {
1190 if (!FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) {
1191 warning("AES-CTR intrinsics require UseAESIntrinsics flag to be enabled. Intrinsics will be disabled.");
1192 }
1193 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false);
1194 }
1195 } else {
1196 if (supports_sse4_1()) {
1197 if (FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) {
1198 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, true);
1199 }
1200 } else if (UseAESCTRIntrinsics) {
1201 // The AES-CTR intrinsic stubs require AES instruction support (of course)
1202 // but also require sse4.1 mode or higher for instructions it use.
1203 if (!FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) {
1204 warning("X86 AES-CTR intrinsics require SSE4.1 instructions or higher. Intrinsics will be disabled.");
1205 }
1206 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false);
1207 }
1208 }
1209 } else {
1210 if (!cpu_supports_aes()) {
1211 if (UseAESIntrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) {
1212 warning("AES intrinsics are not available on this CPU");
1213 }
1214 FLAG_SET_DEFAULT(UseAESIntrinsics, false);
1215 if (UseAESCTRIntrinsics && !FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) {
1216 warning("AES-CTR intrinsics are not available on this CPU");
1217 }
1218 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false);
1219 } else if (!UseAES) {
1220 if (UseAESIntrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) {
1221 warning("AES intrinsics require UseAES flag to be enabled. Intrinsics will be disabled.");
1222 }
1223 FLAG_SET_DEFAULT(UseAESIntrinsics, false);
1224 if (UseAESCTRIntrinsics && !FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) {
1225 warning("AES_CTR intrinsics require UseAES flag to be enabled. AES_CTR intrinsics will be disabled.");
1226 }
1227 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false);
1228 }
1229 }
1230
1231 if (UseCLMUL && (UseSSE > 2)) {
1232 if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) {
1233 UseCRC32Intrinsics = true;
1234 }
1235 } else if (UseCRC32Intrinsics) {
1236 if (!FLAG_IS_DEFAULT(UseCRC32Intrinsics))
1237 warning("CRC32 Intrinsics requires CLMUL instructions (not available on this CPU)");
1238 FLAG_SET_DEFAULT(UseCRC32Intrinsics, false);
1239 }
1240
1241 if (supports_avx2()) {
1242 if (FLAG_IS_DEFAULT(UseAdler32Intrinsics)) {
1243 UseAdler32Intrinsics = true;
1244 }
1245 } else if (UseAdler32Intrinsics) {
1246 if (!FLAG_IS_DEFAULT(UseAdler32Intrinsics)) {
1247 warning("Adler32 Intrinsics requires avx2 instructions (not available on this CPU)");
1248 }
1249 FLAG_SET_DEFAULT(UseAdler32Intrinsics, false);
1250 }
1251
1252 if (supports_sse4_2() && supports_clmul()) {
1253 if (FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) {
1254 UseCRC32CIntrinsics = true;
1255 }
1256 } else if (UseCRC32CIntrinsics) {
1257 if (!FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) {
1258 warning("CRC32C intrinsics are not available on this CPU");
1259 }
1260 FLAG_SET_DEFAULT(UseCRC32CIntrinsics, false);
1261 }
1262
1263 // GHASH/GCM intrinsics
1264 if (UseCLMUL && (UseSSE > 2)) {
1265 if (FLAG_IS_DEFAULT(UseGHASHIntrinsics)) {
1266 UseGHASHIntrinsics = true;
1267 }
1268 } else if (UseGHASHIntrinsics) {
1269 if (!FLAG_IS_DEFAULT(UseGHASHIntrinsics)) {
1270 warning("GHASH intrinsic requires CLMUL and SSE2 instructions on this CPU");
1271 }
1272 FLAG_SET_DEFAULT(UseGHASHIntrinsics, false);
1273 }
1274
1275 // ChaCha20 Intrinsics
1276 // As long as the system supports AVX as a baseline we can do a
1277 // SIMD-enabled block function. StubGenerator makes the determination
1278 // based on the VM capabilities whether to use an AVX2 or AVX512-enabled
1279 // version.
1280 if (UseAVX >= 1) {
1281 if (FLAG_IS_DEFAULT(UseChaCha20Intrinsics)) {
1282 UseChaCha20Intrinsics = true;
1283 }
1284 } else if (UseChaCha20Intrinsics) {
1285 if (!FLAG_IS_DEFAULT(UseChaCha20Intrinsics)) {
1286 warning("ChaCha20 intrinsic requires AVX instructions");
1287 }
1288 FLAG_SET_DEFAULT(UseChaCha20Intrinsics, false);
1289 }
1290
1291 // Kyber Intrinsics
1292 // Currently we only have them for AVX512
1293 if (supports_evex() && supports_avx512bw()) {
1294 if (FLAG_IS_DEFAULT(UseKyberIntrinsics)) {
1295 UseKyberIntrinsics = true;
1296 }
1297 } else if (UseKyberIntrinsics) {
1298 if (!FLAG_IS_DEFAULT(UseKyberIntrinsics)) {
1299 warning("Intrinsics for ML-KEM are not available on this CPU.");
1300 }
1301 FLAG_SET_DEFAULT(UseKyberIntrinsics, false);
1302 }
1303
1304 // Dilithium Intrinsics
1305 if (UseAVX > 1) {
1306 if (FLAG_IS_DEFAULT(UseDilithiumIntrinsics)) {
1307 UseDilithiumIntrinsics = true;
1308 }
1309 } else if (UseDilithiumIntrinsics) {
1310 if (!FLAG_IS_DEFAULT(UseDilithiumIntrinsics)) {
1311 warning("Intrinsics for ML-DSA are not available on this CPU.");
1312 }
1313 FLAG_SET_DEFAULT(UseDilithiumIntrinsics, false);
1314 }
1315
1316 // Base64 Intrinsics (Check the condition for which the intrinsic will be active)
1317 if (UseAVX >= 2) {
1318 if (FLAG_IS_DEFAULT(UseBASE64Intrinsics)) {
1319 UseBASE64Intrinsics = true;
1320 }
1321 } else if (UseBASE64Intrinsics) {
1322 if (!FLAG_IS_DEFAULT(UseBASE64Intrinsics)) {
1323 warning("Base64 intrinsic requires EVEX instructions on this CPU");
1324 }
1325 FLAG_SET_DEFAULT(UseBASE64Intrinsics, false);
1326 }
1327
1328 if (FLAG_IS_DEFAULT(UseMD5Intrinsics)) {
1329 UseMD5Intrinsics = true;
1330 }
1331
1332 if (supports_sha() && supports_sse4_1() && UseSHA) {
1333 if (FLAG_IS_DEFAULT(UseSHA1Intrinsics)) {
1334 FLAG_SET_DEFAULT(UseSHA1Intrinsics, true);
1335 }
1336 } else if (UseSHA1Intrinsics) {
1337 if (!FLAG_IS_DEFAULT(UseSHA1Intrinsics)) {
1338 warning("Intrinsics for SHA-1 crypto hash functions not available on this CPU.");
1339 }
1340 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false);
1341 }
1342
1343 if (supports_sse4_1() && UseSHA) {
1344 if (FLAG_IS_DEFAULT(UseSHA256Intrinsics)) {
1345 FLAG_SET_DEFAULT(UseSHA256Intrinsics, true);
1346 }
1347 } else if (UseSHA256Intrinsics) {
1348 if (!FLAG_IS_DEFAULT(UseSHA256Intrinsics)) {
1349 warning("Intrinsics for SHA-224 and SHA-256 crypto hash functions not available on this CPU.");
1350 }
1351 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false);
1352 }
1353
1354 if (UseSHA && supports_avx2() && (supports_bmi2() || supports_sha512())) {
1355 if (FLAG_IS_DEFAULT(UseSHA512Intrinsics)) {
1356 FLAG_SET_DEFAULT(UseSHA512Intrinsics, true);
1357 }
1358 } else if (UseSHA512Intrinsics) {
1359 if (!FLAG_IS_DEFAULT(UseSHA512Intrinsics)) {
1360 warning("Intrinsics for SHA-384 and SHA-512 crypto hash functions not available on this CPU.");
1361 }
1362 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false);
1363 }
1364
1365 if (UseSHA && supports_evex() && supports_avx512bw()) {
1366 if (FLAG_IS_DEFAULT(UseSHA3Intrinsics)) {
1367 FLAG_SET_DEFAULT(UseSHA3Intrinsics, true);
1368 }
1369 } else if (UseSHA3Intrinsics) {
1370 if (!FLAG_IS_DEFAULT(UseSHA3Intrinsics)) {
1371 warning("Intrinsics for SHA3-224, SHA3-256, SHA3-384 and SHA3-512 crypto hash functions not available on this CPU.");
1372 }
1373 FLAG_SET_DEFAULT(UseSHA3Intrinsics, false);
1374 }
1375
1376 #if COMPILER2_OR_JVMCI
1377 int max_vector_size = 0;
1378 if (UseAVX == 0 || !os_supports_avx_vectors()) {
1379 // 16 byte vectors (in XMM) are supported with SSE2+
1380 max_vector_size = 16;
1381 } else if (UseAVX == 1 || UseAVX == 2) {
1382 // 32 bytes vectors (in YMM) are only supported with AVX+
1383 max_vector_size = 32;
1384 } else if (UseAVX > 2) {
1385 // 64 bytes vectors (in ZMM) are only supported with AVX 3
1386 max_vector_size = 64;
1387 }
1388
1389 int min_vector_size = 4; // We require MaxVectorSize to be at least 4 on 64bit
1390
1391 if (!FLAG_IS_DEFAULT(MaxVectorSize)) {
1392 if (MaxVectorSize < min_vector_size) {
1393 warning("MaxVectorSize must be at least %i on this platform", min_vector_size);
1394 FLAG_SET_DEFAULT(MaxVectorSize, min_vector_size);
1395 }
1396 if (MaxVectorSize > max_vector_size) {
1397 warning("MaxVectorSize must be at most %i on this platform", max_vector_size);
1398 FLAG_SET_DEFAULT(MaxVectorSize, max_vector_size);
1399 }
1400 if (!is_power_of_2(MaxVectorSize)) {
1401 warning("MaxVectorSize must be a power of 2, setting to default: %i", max_vector_size);
1402 FLAG_SET_DEFAULT(MaxVectorSize, max_vector_size);
1403 }
1404 } else {
1405 // If default, use highest supported configuration
1406 FLAG_SET_DEFAULT(MaxVectorSize, max_vector_size);
1407 }
1408
1409 #if defined(COMPILER2) && defined(ASSERT)
1410 if (MaxVectorSize > 0) {
1411 if (supports_avx() && PrintMiscellaneous && Verbose && TraceNewVectors) {
1412 tty->print_cr("State of YMM registers after signal handle:");
1413 int nreg = 4;
1414 const char* ymm_name[4] = {"0", "7", "8", "15"};
1415 for (int i = 0; i < nreg; i++) {
1416 tty->print("YMM%s:", ymm_name[i]);
1417 for (int j = 7; j >=0; j--) {
1418 tty->print(" %x", _cpuid_info.ymm_save[i*8 + j]);
1419 }
1420 tty->cr();
1421 }
1422 }
1423 }
1424 #endif // COMPILER2 && ASSERT
1425
1426 if ((supports_avx512ifma() && supports_avx512vlbw()) || supports_avxifma()) {
1427 if (FLAG_IS_DEFAULT(UsePoly1305Intrinsics)) {
1428 FLAG_SET_DEFAULT(UsePoly1305Intrinsics, true);
1429 }
1430 } else if (UsePoly1305Intrinsics) {
1431 if (!FLAG_IS_DEFAULT(UsePoly1305Intrinsics)) {
1432 warning("Intrinsics for Poly1305 crypto hash functions not available on this CPU.");
1433 }
1434 FLAG_SET_DEFAULT(UsePoly1305Intrinsics, false);
1435 }
1436
1437 if ((supports_avx512ifma() && supports_avx512vlbw()) || supports_avxifma()) {
1438 if (FLAG_IS_DEFAULT(UseIntPolyIntrinsics)) {
1439 FLAG_SET_DEFAULT(UseIntPolyIntrinsics, true);
1440 }
1441 } else if (UseIntPolyIntrinsics) {
1442 if (!FLAG_IS_DEFAULT(UseIntPolyIntrinsics)) {
1443 warning("Intrinsics for Polynomial crypto functions not available on this CPU.");
1444 }
1445 FLAG_SET_DEFAULT(UseIntPolyIntrinsics, false);
1446 }
1447
1448 if (FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) {
1449 UseMultiplyToLenIntrinsic = true;
1450 }
1451 if (FLAG_IS_DEFAULT(UseSquareToLenIntrinsic)) {
1452 UseSquareToLenIntrinsic = true;
1453 }
1454 if (FLAG_IS_DEFAULT(UseMulAddIntrinsic)) {
1455 UseMulAddIntrinsic = true;
1456 }
1457 if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) {
1458 UseMontgomeryMultiplyIntrinsic = true;
1459 }
1460 if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) {
1461 UseMontgomerySquareIntrinsic = true;
1462 }
1463 #endif // COMPILER2_OR_JVMCI
1464
1465 // On new cpus instructions which update whole XMM register should be used
1466 // to prevent partial register stall due to dependencies on high half.
1467 //
1468 // UseXmmLoadAndClearUpper == true --> movsd(xmm, mem)
1469 // UseXmmLoadAndClearUpper == false --> movlpd(xmm, mem)
1470 // UseXmmRegToRegMoveAll == true --> movaps(xmm, xmm), movapd(xmm, xmm).
1471 // UseXmmRegToRegMoveAll == false --> movss(xmm, xmm), movsd(xmm, xmm).
1472
1473
1474 if (is_zx()) { // ZX cpus specific settings
1475 if (FLAG_IS_DEFAULT(UseStoreImmI16)) {
1476 UseStoreImmI16 = false; // don't use it on ZX cpus
1477 }
1478 if ((cpu_family() == 6) || (cpu_family() == 7)) {
1479 if (FLAG_IS_DEFAULT(UseAddressNop)) {
1480 // Use it on all ZX cpus
1481 UseAddressNop = true;
1482 }
1483 }
1484 if (FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper)) {
1485 UseXmmLoadAndClearUpper = true; // use movsd on all ZX cpus
1486 }
1487 if (FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll)) {
1488 if (supports_sse3()) {
1489 UseXmmRegToRegMoveAll = true; // use movaps, movapd on new ZX cpus
1490 } else {
1491 UseXmmRegToRegMoveAll = false;
1492 }
1493 }
1494 if (((cpu_family() == 6) || (cpu_family() == 7)) && supports_sse3()) { // new ZX cpus
1495 #ifdef COMPILER2
1496 if (FLAG_IS_DEFAULT(MaxLoopPad)) {
1497 // For new ZX cpus do the next optimization:
1498 // don't align the beginning of a loop if there are enough instructions
1499 // left (NumberOfLoopInstrToAlign defined in c2_globals.hpp)
1500 // in current fetch line (OptoLoopAlignment) or the padding
1501 // is big (> MaxLoopPad).
1502 // Set MaxLoopPad to 11 for new ZX cpus to reduce number of
1503 // generated NOP instructions. 11 is the largest size of one
1504 // address NOP instruction '0F 1F' (see Assembler::nop(i)).
1505 MaxLoopPad = 11;
1506 }
1507 #endif // COMPILER2
1508 if (supports_sse4_2()) { // new ZX cpus
1509 if (FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
1510 UseUnalignedLoadStores = true; // use movdqu on newest ZX cpus
1511 }
1512 }
1513 }
1514
1515 if (FLAG_IS_DEFAULT(AllocatePrefetchInstr) && supports_3dnow_prefetch()) {
1516 FLAG_SET_DEFAULT(AllocatePrefetchInstr, 3);
1517 }
1518 }
1519
1520 if (is_amd_family()) { // AMD cpus specific settings
1521 if (supports_sse2() && FLAG_IS_DEFAULT(UseAddressNop)) {
1522 // Use it on new AMD cpus starting from Opteron.
1523 UseAddressNop = true;
1524 }
1525 if (FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper)) {
1526 if (supports_sse4a()) {
1527 UseXmmLoadAndClearUpper = true; // use movsd only on '10h' Opteron
1528 } else {
1529 UseXmmLoadAndClearUpper = false;
1530 }
1531 }
1532 if (FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll)) {
1533 if (supports_sse4a()) {
1534 UseXmmRegToRegMoveAll = true; // use movaps, movapd only on '10h'
1535 } else {
1536 UseXmmRegToRegMoveAll = false;
1537 }
1538 }
1539 if (FLAG_IS_DEFAULT(UseXmmI2F)) {
1540 if (supports_sse4a()) {
1541 UseXmmI2F = true;
1542 } else {
1543 UseXmmI2F = false;
1544 }
1545 }
1546 if (FLAG_IS_DEFAULT(UseXmmI2D)) {
1547 if (supports_sse4a()) {
1548 UseXmmI2D = true;
1549 } else {
1550 UseXmmI2D = false;
1551 }
1552 }
1553
1554 // some defaults for AMD family 15h
1555 if (cpu_family() == 0x15) {
1556 // On family 15h processors default is no sw prefetch
1557 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) {
1558 FLAG_SET_DEFAULT(AllocatePrefetchStyle, 0);
1559 }
1560 // Also, if some other prefetch style is specified, default instruction type is PREFETCHW
1561 if (FLAG_IS_DEFAULT(AllocatePrefetchInstr)) {
1562 FLAG_SET_DEFAULT(AllocatePrefetchInstr, 3);
1563 }
1564 if (supports_sse2() && FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
1565 FLAG_SET_DEFAULT(UseUnalignedLoadStores, true);
1566 }
1567 }
1568
1569 #ifdef COMPILER2
1570 if (cpu_family() < 0x17 && MaxVectorSize > 16) {
1571 // Limit vectors size to 16 bytes on AMD cpus < 17h.
1572 FLAG_SET_DEFAULT(MaxVectorSize, 16);
1573 }
1574 #endif // COMPILER2
1575
1576 // Some defaults for AMD family >= 17h && Hygon family 18h
1577 if (cpu_family() >= 0x17) {
1578 // On family >=17h processors use XMM and UnalignedLoadStores
1579 // for Array Copy
1580 if (supports_sse2() && FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
1581 FLAG_SET_DEFAULT(UseUnalignedLoadStores, true);
1582 }
1583 #ifdef COMPILER2
1584 if (supports_sse4_2() && FLAG_IS_DEFAULT(UseFPUForSpilling)) {
1585 FLAG_SET_DEFAULT(UseFPUForSpilling, true);
1586 }
1587 #endif
1588 }
1589 }
1590
1591 if (is_intel()) { // Intel cpus specific settings
1592 if (FLAG_IS_DEFAULT(UseStoreImmI16)) {
1593 UseStoreImmI16 = false; // don't use it on Intel cpus
1594 }
1595 if (is_intel_server_family() || cpu_family() == 15) {
1596 if (FLAG_IS_DEFAULT(UseAddressNop)) {
1597 // Use it on all Intel cpus starting from PentiumPro
1598 UseAddressNop = true;
1599 }
1600 }
1601 if (FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper)) {
1602 UseXmmLoadAndClearUpper = true; // use movsd on all Intel cpus
1603 }
1604 if (FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll)) {
1605 if (supports_sse3()) {
1606 UseXmmRegToRegMoveAll = true; // use movaps, movapd on new Intel cpus
1607 } else {
1608 UseXmmRegToRegMoveAll = false;
1609 }
1610 }
1611 if (is_intel_server_family() && supports_sse3()) { // New Intel cpus
1612 #ifdef COMPILER2
1613 if (FLAG_IS_DEFAULT(MaxLoopPad)) {
1614 // For new Intel cpus do the next optimization:
1615 // don't align the beginning of a loop if there are enough instructions
1616 // left (NumberOfLoopInstrToAlign defined in c2_globals.hpp)
1617 // in current fetch line (OptoLoopAlignment) or the padding
1618 // is big (> MaxLoopPad).
1619 // Set MaxLoopPad to 11 for new Intel cpus to reduce number of
1620 // generated NOP instructions. 11 is the largest size of one
1621 // address NOP instruction '0F 1F' (see Assembler::nop(i)).
1622 MaxLoopPad = 11;
1623 }
1624 #endif // COMPILER2
1625
1626 if (is_intel_modern_cpu()) { // Newest Intel cpus
1627 if (FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
1628 UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus
1629 }
1630 }
1631 }
1632 if (is_atom_family() || is_knights_family()) {
1633 #ifdef COMPILER2
1634 if (FLAG_IS_DEFAULT(OptoScheduling)) {
1635 OptoScheduling = true;
1636 }
1637 #endif
1638 if (supports_sse4_2()) { // Silvermont
1639 if (FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
1640 UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus
1641 }
1642 }
1643 if (FLAG_IS_DEFAULT(UseIncDec)) {
1644 FLAG_SET_DEFAULT(UseIncDec, false);
1645 }
1646 }
1647 if (FLAG_IS_DEFAULT(AllocatePrefetchInstr) && supports_3dnow_prefetch()) {
1648 FLAG_SET_DEFAULT(AllocatePrefetchInstr, 3);
1649 }
1650 }
1651
1652 #ifdef COMPILER2
1653 if (UseAVX > 2) {
1654 if (FLAG_IS_DEFAULT(ArrayOperationPartialInlineSize) ||
1655 (!FLAG_IS_DEFAULT(ArrayOperationPartialInlineSize) &&
1656 ArrayOperationPartialInlineSize != 0 &&
1657 ArrayOperationPartialInlineSize != 16 &&
1658 ArrayOperationPartialInlineSize != 32 &&
1659 ArrayOperationPartialInlineSize != 64)) {
1660 int inline_size = 0;
1661 if (MaxVectorSize >= 64 && AVX3Threshold == 0) {
1662 inline_size = 64;
1663 } else if (MaxVectorSize >= 32) {
1664 inline_size = 32;
1665 } else if (MaxVectorSize >= 16) {
1666 inline_size = 16;
1667 }
1668 if(!FLAG_IS_DEFAULT(ArrayOperationPartialInlineSize)) {
1669 warning("Setting ArrayOperationPartialInlineSize as %d", inline_size);
1670 }
1671 ArrayOperationPartialInlineSize = inline_size;
1672 }
1673
1674 if (ArrayOperationPartialInlineSize > MaxVectorSize) {
1675 ArrayOperationPartialInlineSize = MaxVectorSize >= 16 ? MaxVectorSize : 0;
1676 if (ArrayOperationPartialInlineSize) {
1677 warning("Setting ArrayOperationPartialInlineSize as MaxVectorSize=%zd", MaxVectorSize);
1678 } else {
1679 warning("Setting ArrayOperationPartialInlineSize as %zd", ArrayOperationPartialInlineSize);
1680 }
1681 }
1682 }
1683
1684 if (FLAG_IS_DEFAULT(OptimizeFill)) {
1685 if (MaxVectorSize < 32 || (!EnableX86ECoreOpts && !VM_Version::supports_avx512vlbw())) {
1686 OptimizeFill = false;
1687 }
1688 }
1689 #endif
1690 if (supports_sse4_2()) {
1691 if (FLAG_IS_DEFAULT(UseSSE42Intrinsics)) {
1692 FLAG_SET_DEFAULT(UseSSE42Intrinsics, true);
1693 }
1694 } else if (UseSSE42Intrinsics) {
1695 if (!FLAG_IS_DEFAULT(UseSSE42Intrinsics)) {
1696 warning("SSE4.2 intrinsics require SSE4.2 instructions or higher. Intrinsics will be disabled.");
1697 }
1698 FLAG_SET_DEFAULT(UseSSE42Intrinsics, false);
1699 }
1700 if (UseSSE42Intrinsics) {
1701 if (FLAG_IS_DEFAULT(UseVectorizedMismatchIntrinsic)) {
1702 UseVectorizedMismatchIntrinsic = true;
1703 }
1704 } else if (UseVectorizedMismatchIntrinsic) {
1705 if (!FLAG_IS_DEFAULT(UseVectorizedMismatchIntrinsic)) {
1706 warning("vectorizedMismatch intrinsics are not available on this CPU");
1707 }
1708 FLAG_SET_DEFAULT(UseVectorizedMismatchIntrinsic, false);
1709 }
1710 if (UseAVX >= 2) {
1711 FLAG_SET_DEFAULT(UseVectorizedHashCodeIntrinsic, true);
1712 } else if (UseVectorizedHashCodeIntrinsic) {
1713 if (!FLAG_IS_DEFAULT(UseVectorizedHashCodeIntrinsic)) {
1714 warning("vectorizedHashCode intrinsics are not available on this CPU");
1715 }
1716 FLAG_SET_DEFAULT(UseVectorizedHashCodeIntrinsic, false);
1717 }
1718
1719 // Use count leading zeros count instruction if available.
1720 if (supports_lzcnt()) {
1721 if (FLAG_IS_DEFAULT(UseCountLeadingZerosInstruction)) {
1722 UseCountLeadingZerosInstruction = true;
1723 }
1724 } else if (UseCountLeadingZerosInstruction) {
1725 if (!FLAG_IS_DEFAULT(UseCountLeadingZerosInstruction)) {
1726 warning("lzcnt instruction is not available on this CPU");
1727 }
1728 FLAG_SET_DEFAULT(UseCountLeadingZerosInstruction, false);
1729 }
1730
1731 // Use count trailing zeros instruction if available
1732 if (supports_bmi1()) {
1733 // tzcnt does not require VEX prefix
1734 if (FLAG_IS_DEFAULT(UseCountTrailingZerosInstruction)) {
1735 if (!UseBMI1Instructions && !FLAG_IS_DEFAULT(UseBMI1Instructions)) {
1736 // Don't use tzcnt if BMI1 is switched off on command line.
1737 UseCountTrailingZerosInstruction = false;
1738 } else {
1739 UseCountTrailingZerosInstruction = true;
1740 }
1741 }
1742 } else if (UseCountTrailingZerosInstruction) {
1743 if (!FLAG_IS_DEFAULT(UseCountTrailingZerosInstruction)) {
1744 warning("tzcnt instruction is not available on this CPU");
1745 }
1746 FLAG_SET_DEFAULT(UseCountTrailingZerosInstruction, false);
1747 }
1748
1749 // BMI instructions (except tzcnt) use an encoding with VEX prefix.
1750 // VEX prefix is generated only when AVX > 0.
1751 if (supports_bmi1() && supports_avx()) {
1752 if (FLAG_IS_DEFAULT(UseBMI1Instructions)) {
1753 UseBMI1Instructions = true;
1754 }
1755 } else if (UseBMI1Instructions) {
1756 if (!FLAG_IS_DEFAULT(UseBMI1Instructions)) {
1757 warning("BMI1 instructions are not available on this CPU (AVX is also required)");
1758 }
1759 FLAG_SET_DEFAULT(UseBMI1Instructions, false);
1760 }
1761
1762 if (supports_bmi2() && supports_avx()) {
1763 if (FLAG_IS_DEFAULT(UseBMI2Instructions)) {
1764 UseBMI2Instructions = true;
1765 }
1766 } else if (UseBMI2Instructions) {
1767 if (!FLAG_IS_DEFAULT(UseBMI2Instructions)) {
1768 warning("BMI2 instructions are not available on this CPU (AVX is also required)");
1769 }
1770 FLAG_SET_DEFAULT(UseBMI2Instructions, false);
1771 }
1772
1773 // Use population count instruction if available.
1774 if (supports_popcnt()) {
1775 if (FLAG_IS_DEFAULT(UsePopCountInstruction)) {
1776 UsePopCountInstruction = true;
1777 }
1778 } else if (UsePopCountInstruction) {
1779 if (!FLAG_IS_DEFAULT(UsePopCountInstruction)) {
1780 warning("POPCNT instruction is not available on this CPU");
1781 }
1782 FLAG_SET_DEFAULT(UsePopCountInstruction, false);
1783 }
1784
1785 // Use fast-string operations if available.
1786 if (supports_erms()) {
1787 if (FLAG_IS_DEFAULT(UseFastStosb)) {
1788 UseFastStosb = true;
1789 }
1790 } else if (UseFastStosb) {
1791 if (!FLAG_IS_DEFAULT(UseFastStosb)) {
1792 warning("fast-string operations are not available on this CPU");
1793 }
1794 FLAG_SET_DEFAULT(UseFastStosb, false);
1795 }
1796
1797 // For AMD Processors use XMM/YMM MOVDQU instructions
1798 // for Object Initialization as default
1799 if (is_amd() && cpu_family() >= 0x19) {
1800 if (FLAG_IS_DEFAULT(UseFastStosb)) {
1801 UseFastStosb = false;
1802 }
1803 }
1804
1805 #ifdef COMPILER2
1806 if (is_intel() && MaxVectorSize > 16) {
1807 if (FLAG_IS_DEFAULT(UseFastStosb)) {
1808 UseFastStosb = false;
1809 }
1810 }
1811 #endif
1812
1813 // Use XMM/YMM MOVDQU instruction for Object Initialization
1814 if (UseUnalignedLoadStores) {
1815 if (FLAG_IS_DEFAULT(UseXMMForObjInit)) {
1816 UseXMMForObjInit = true;
1817 }
1818 } else if (UseXMMForObjInit) {
1819 if (!FLAG_IS_DEFAULT(UseXMMForObjInit)) {
1820 warning("UseXMMForObjInit requires SSE2 and unaligned load/stores. Feature is switched off.");
1821 }
1822 FLAG_SET_DEFAULT(UseXMMForObjInit, false);
1823 }
1824
1825 #ifdef COMPILER2
1826 if (FLAG_IS_DEFAULT(AlignVector)) {
1827 // Modern processors allow misaligned memory operations for vectors.
1828 AlignVector = !UseUnalignedLoadStores;
1829 }
1830 #endif // COMPILER2
1831
1832 if (FLAG_IS_DEFAULT(AllocatePrefetchInstr)) {
1833 if (AllocatePrefetchInstr == 3 && !supports_3dnow_prefetch()) {
1834 FLAG_SET_DEFAULT(AllocatePrefetchInstr, 0);
1835 } else if (!supports_sse() && supports_3dnow_prefetch()) {
1836 FLAG_SET_DEFAULT(AllocatePrefetchInstr, 3);
1837 }
1838 }
1839
1840 // Allocation prefetch settings
1841 int cache_line_size = checked_cast<int>(prefetch_data_size());
1842 if (FLAG_IS_DEFAULT(AllocatePrefetchStepSize) &&
1843 (cache_line_size > AllocatePrefetchStepSize)) {
1844 FLAG_SET_DEFAULT(AllocatePrefetchStepSize, cache_line_size);
1845 }
1846
1847 if ((AllocatePrefetchDistance == 0) && (AllocatePrefetchStyle != 0)) {
1848 assert(!FLAG_IS_DEFAULT(AllocatePrefetchDistance), "default value should not be 0");
1849 if (!FLAG_IS_DEFAULT(AllocatePrefetchStyle)) {
1850 warning("AllocatePrefetchDistance is set to 0 which disable prefetching. Ignoring AllocatePrefetchStyle flag.");
1851 }
1852 FLAG_SET_DEFAULT(AllocatePrefetchStyle, 0);
1853 }
1854
1855 if (FLAG_IS_DEFAULT(AllocatePrefetchDistance)) {
1856 bool use_watermark_prefetch = (AllocatePrefetchStyle == 2);
1857 FLAG_SET_DEFAULT(AllocatePrefetchDistance, allocate_prefetch_distance(use_watermark_prefetch));
1858 }
1859
1860 if (is_intel() && is_intel_server_family() && supports_sse3()) {
1861 if (FLAG_IS_DEFAULT(AllocatePrefetchLines) &&
1862 is_intel_modern_cpu()) { // Nehalem based cpus
1863 FLAG_SET_DEFAULT(AllocatePrefetchLines, 4);
1864 }
1865 #ifdef COMPILER2
1866 if (FLAG_IS_DEFAULT(UseFPUForSpilling) && supports_sse4_2()) {
1867 FLAG_SET_DEFAULT(UseFPUForSpilling, true);
1868 }
1869 #endif
1870 }
1871
1872 if (is_zx() && ((cpu_family() == 6) || (cpu_family() == 7)) && supports_sse4_2()) {
1873 #ifdef COMPILER2
1874 if (FLAG_IS_DEFAULT(UseFPUForSpilling)) {
1875 FLAG_SET_DEFAULT(UseFPUForSpilling, true);
1876 }
1877 #endif
1878 }
1879
1880 // Prefetch settings
1881
1882 // Prefetch interval for gc copy/scan == 9 dcache lines. Derived from
1883 // 50-warehouse specjbb runs on a 2-way 1.8ghz opteron using a 4gb heap.
1884 // Tested intervals from 128 to 2048 in increments of 64 == one cache line.
1885 // 256 bytes (4 dcache lines) was the nearest runner-up to 576.
1886
1887 // gc copy/scan is disabled if prefetchw isn't supported, because
1888 // Prefetch::write emits an inlined prefetchw on Linux.
1889 // Do not use the 3dnow prefetchw instruction. It isn't supported on em64t.
1890 // The used prefetcht0 instruction works for both amd64 and em64t.
1891
1892 if (FLAG_IS_DEFAULT(PrefetchCopyIntervalInBytes)) {
1893 FLAG_SET_DEFAULT(PrefetchCopyIntervalInBytes, 576);
1894 }
1895 if (FLAG_IS_DEFAULT(PrefetchScanIntervalInBytes)) {
1896 FLAG_SET_DEFAULT(PrefetchScanIntervalInBytes, 576);
1897 }
1898
1899 if (FLAG_IS_DEFAULT(ContendedPaddingWidth) &&
1900 (cache_line_size > ContendedPaddingWidth))
1901 ContendedPaddingWidth = cache_line_size;
1902
1903 // This machine allows unaligned memory accesses
1904 if (FLAG_IS_DEFAULT(UseUnalignedAccesses)) {
1905 FLAG_SET_DEFAULT(UseUnalignedAccesses, true);
1906 }
1907
1908 #ifndef PRODUCT
1909 if (log_is_enabled(Info, os, cpu)) {
1910 LogStream ls(Log(os, cpu)::info());
1911 outputStream* log = &ls;
1912 log->print_cr("Logical CPUs per core: %u",
1913 logical_processors_per_package());
1914 log->print_cr("L1 data cache line size: %u", L1_data_cache_line_size());
1915 log->print("UseSSE=%d", UseSSE);
1916 if (UseAVX > 0) {
1917 log->print(" UseAVX=%d", UseAVX);
1918 }
1919 if (UseAES) {
1920 log->print(" UseAES=1");
1921 }
1922 #ifdef COMPILER2
1923 if (MaxVectorSize > 0) {
1924 log->print(" MaxVectorSize=%d", (int) MaxVectorSize);
1925 }
1926 #endif
1927 log->cr();
1928 log->print("Allocation");
1929 if (AllocatePrefetchStyle <= 0) {
1930 log->print_cr(": no prefetching");
1931 } else {
1932 log->print(" prefetching: ");
1933 if (AllocatePrefetchInstr == 0) {
1934 log->print("PREFETCHNTA");
1935 } else if (AllocatePrefetchInstr == 1) {
1936 log->print("PREFETCHT0");
1937 } else if (AllocatePrefetchInstr == 2) {
1938 log->print("PREFETCHT2");
1939 } else if (AllocatePrefetchInstr == 3) {
1940 log->print("PREFETCHW");
1941 }
1942 if (AllocatePrefetchLines > 1) {
1943 log->print_cr(" at distance %d, %d lines of %d bytes", AllocatePrefetchDistance, AllocatePrefetchLines, AllocatePrefetchStepSize);
1944 } else {
1945 log->print_cr(" at distance %d, one line of %d bytes", AllocatePrefetchDistance, AllocatePrefetchStepSize);
1946 }
1947 }
1948
1949 if (PrefetchCopyIntervalInBytes > 0) {
1950 log->print_cr("PrefetchCopyIntervalInBytes %d", (int) PrefetchCopyIntervalInBytes);
1951 }
1952 if (PrefetchScanIntervalInBytes > 0) {
1953 log->print_cr("PrefetchScanIntervalInBytes %d", (int) PrefetchScanIntervalInBytes);
1954 }
1955 if (ContendedPaddingWidth > 0) {
1956 log->print_cr("ContendedPaddingWidth %d", (int) ContendedPaddingWidth);
1957 }
1958 }
1959 #endif // !PRODUCT
1960 if (FLAG_IS_DEFAULT(UseSignumIntrinsic)) {
1961 FLAG_SET_DEFAULT(UseSignumIntrinsic, true);
1962 }
1963 if (FLAG_IS_DEFAULT(UseCopySignIntrinsic)) {
1964 FLAG_SET_DEFAULT(UseCopySignIntrinsic, true);
1965 }
1966 // CopyAVX3Threshold is the threshold at which 64-byte instructions are used
1967 // for implementing the array copy and clear operations.
1968 // The Intel platforms that supports the serialize instruction
1969 // have improved implementation of 64-byte load/stores and so the default
1970 // threshold is set to 0 for these platforms.
1971 if (FLAG_IS_DEFAULT(CopyAVX3Threshold)) {
1972 if (is_intel() && is_intel_server_family() && supports_serialize()) {
1973 FLAG_SET_DEFAULT(CopyAVX3Threshold, 0);
1974 } else {
1975 FLAG_SET_DEFAULT(CopyAVX3Threshold, AVX3Threshold);
1976 }
1977 }
1978 }
1979
1980 void VM_Version::print_platform_virtualization_info(outputStream* st) {
1981 VirtualizationType vrt = VM_Version::get_detected_virtualization();
1982 if (vrt == XenHVM) {
1983 st->print_cr("Xen hardware-assisted virtualization detected");
1984 } else if (vrt == KVM) {
1985 st->print_cr("KVM virtualization detected");
1986 } else if (vrt == VMWare) {
1987 st->print_cr("VMWare virtualization detected");
1988 VirtualizationSupport::print_virtualization_info(st);
1989 } else if (vrt == HyperV) {
1990 st->print_cr("Hyper-V virtualization detected");
1991 } else if (vrt == HyperVRole) {
1992 st->print_cr("Hyper-V role detected");
1993 }
1994 }
1995
1996 bool VM_Version::compute_has_intel_jcc_erratum() {
1997 if (!is_intel_family_core()) {
1998 // Only Intel CPUs are affected.
1999 return false;
2000 }
2001 // The following table of affected CPUs is based on the following document released by Intel:
2002 // https://www.intel.com/content/dam/support/us/en/documents/processors/mitigations-jump-conditional-code-erratum.pdf
2003 switch (_model) {
2004 case 0x8E:
2005 // 06_8EH | 9 | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Amber Lake Y
2006 // 06_8EH | 9 | 7th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Kaby Lake U
2007 // 06_8EH | 9 | 7th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Kaby Lake U 23e
2008 // 06_8EH | 9 | 7th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Kaby Lake Y
2009 // 06_8EH | A | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Coffee Lake U43e
2010 // 06_8EH | B | 8th Generation Intel(R) Core(TM) Processors based on microarchitecture code name Whiskey Lake U
2011 // 06_8EH | C | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Amber Lake Y
2012 // 06_8EH | C | 10th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Comet Lake U42
2013 // 06_8EH | C | 8th Generation Intel(R) Core(TM) Processors based on microarchitecture code name Whiskey Lake U
2014 return _stepping == 0x9 || _stepping == 0xA || _stepping == 0xB || _stepping == 0xC;
2015 case 0x4E:
2016 // 06_4E | 3 | 6th Generation Intel(R) Core(TM) Processors based on microarchitecture code name Skylake U
2017 // 06_4E | 3 | 6th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Skylake U23e
2018 // 06_4E | 3 | 6th Generation Intel(R) Core(TM) Processors based on microarchitecture code name Skylake Y
2019 return _stepping == 0x3;
2020 case 0x55:
2021 // 06_55H | 4 | Intel(R) Xeon(R) Processor D Family based on microarchitecture code name Skylake D, Bakerville
2022 // 06_55H | 4 | Intel(R) Xeon(R) Scalable Processors based on microarchitecture code name Skylake Server
2023 // 06_55H | 4 | Intel(R) Xeon(R) Processor W Family based on microarchitecture code name Skylake W
2024 // 06_55H | 4 | Intel(R) Core(TM) X-series Processors based on microarchitecture code name Skylake X
2025 // 06_55H | 4 | Intel(R) Xeon(R) Processor E3 v5 Family based on microarchitecture code name Skylake Xeon E3
2026 // 06_55 | 7 | 2nd Generation Intel(R) Xeon(R) Scalable Processors based on microarchitecture code name Cascade Lake (server)
2027 return _stepping == 0x4 || _stepping == 0x7;
2028 case 0x5E:
2029 // 06_5E | 3 | 6th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Skylake H
2030 // 06_5E | 3 | 6th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Skylake S
2031 return _stepping == 0x3;
2032 case 0x9E:
2033 // 06_9EH | 9 | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Kaby Lake G
2034 // 06_9EH | 9 | 7th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Kaby Lake H
2035 // 06_9EH | 9 | 7th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Kaby Lake S
2036 // 06_9EH | 9 | Intel(R) Core(TM) X-series Processors based on microarchitecture code name Kaby Lake X
2037 // 06_9EH | 9 | Intel(R) Xeon(R) Processor E3 v6 Family Kaby Lake Xeon E3
2038 // 06_9EH | A | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Coffee Lake H
2039 // 06_9EH | A | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Coffee Lake S
2040 // 06_9EH | A | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Coffee Lake S (6+2) x/KBP
2041 // 06_9EH | A | Intel(R) Xeon(R) Processor E Family based on microarchitecture code name Coffee Lake S (6+2)
2042 // 06_9EH | A | Intel(R) Xeon(R) Processor E Family based on microarchitecture code name Coffee Lake S (4+2)
2043 // 06_9EH | B | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Coffee Lake S (4+2)
2044 // 06_9EH | B | Intel(R) Celeron(R) Processor G Series based on microarchitecture code name Coffee Lake S (4+2)
2045 // 06_9EH | D | 9th Generation Intel(R) Core(TM) Processor Family based on microarchitecturecode name Coffee Lake H (8+2)
2046 // 06_9EH | D | 9th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Coffee Lake S (8+2)
2047 return _stepping == 0x9 || _stepping == 0xA || _stepping == 0xB || _stepping == 0xD;
2048 case 0xA5:
2049 // Not in Intel documentation.
2050 // 06_A5H | | 10th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Comet Lake S/H
2051 return true;
2052 case 0xA6:
2053 // 06_A6H | 0 | 10th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Comet Lake U62
2054 return _stepping == 0x0;
2055 case 0xAE:
2056 // 06_AEH | A | 8th Generation Intel(R) Core(TM) Processor Family based on microarchitecture code name Kaby Lake Refresh U (4+2)
2057 return _stepping == 0xA;
2058 default:
2059 // If we are running on another intel machine not recognized in the table, we are okay.
2060 return false;
2061 }
2062 }
2063
2064 // On Xen, the cpuid instruction returns
2065 // eax / registers[0]: Version of Xen
2066 // ebx / registers[1]: chars 'XenV'
2067 // ecx / registers[2]: chars 'MMXe'
2068 // edx / registers[3]: chars 'nVMM'
2069 //
2070 // On KVM / VMWare / MS Hyper-V, the cpuid instruction returns
2071 // ebx / registers[1]: chars 'KVMK' / 'VMwa' / 'Micr'
2072 // ecx / registers[2]: chars 'VMKV' / 'reVM' / 'osof'
2073 // edx / registers[3]: chars 'M' / 'ware' / 't Hv'
2074 //
2075 // more information :
2076 // https://kb.vmware.com/s/article/1009458
2077 //
2078 void VM_Version::check_virtualizations() {
2079 uint32_t registers[4] = {0};
2080 char signature[13] = {0};
2081
2082 // Xen cpuid leaves can be found 0x100 aligned boundary starting
2083 // from 0x40000000 until 0x40010000.
2084 // https://lists.linuxfoundation.org/pipermail/virtualization/2012-May/019974.html
2085 for (int leaf = 0x40000000; leaf < 0x40010000; leaf += 0x100) {
2086 detect_virt_stub(leaf, registers);
2087 memcpy(signature, ®isters[1], 12);
2088
2089 if (strncmp("VMwareVMware", signature, 12) == 0) {
2090 Abstract_VM_Version::_detected_virtualization = VMWare;
2091 // check for extended metrics from guestlib
2092 VirtualizationSupport::initialize();
2093 } else if (strncmp("Microsoft Hv", signature, 12) == 0) {
2094 Abstract_VM_Version::_detected_virtualization = HyperV;
2095 #ifdef _WINDOWS
2096 // CPUID leaf 0x40000007 is available to the root partition only.
2097 // See Hypervisor Top Level Functional Specification section 2.4.8 for more details.
2098 // https://github.com/MicrosoftDocs/Virtualization-Documentation/raw/master/tlfs/Hypervisor%20Top%20Level%20Functional%20Specification%20v6.0b.pdf
2099 detect_virt_stub(0x40000007, registers);
2100 if ((registers[0] != 0x0) ||
2101 (registers[1] != 0x0) ||
2102 (registers[2] != 0x0) ||
2103 (registers[3] != 0x0)) {
2104 Abstract_VM_Version::_detected_virtualization = HyperVRole;
2105 }
2106 #endif
2107 } else if (strncmp("KVMKVMKVM", signature, 9) == 0) {
2108 Abstract_VM_Version::_detected_virtualization = KVM;
2109 } else if (strncmp("XenVMMXenVMM", signature, 12) == 0) {
2110 Abstract_VM_Version::_detected_virtualization = XenHVM;
2111 }
2112 }
2113 }
2114
2115 #ifdef COMPILER2
2116 // Determine if it's running on Cascade Lake using default options.
2117 bool VM_Version::is_default_intel_cascade_lake() {
2118 return FLAG_IS_DEFAULT(UseAVX) &&
2119 FLAG_IS_DEFAULT(MaxVectorSize) &&
2120 UseAVX > 2 &&
2121 is_intel_cascade_lake();
2122 }
2123 #endif
2124
2125 bool VM_Version::is_intel_cascade_lake() {
2126 return is_intel_skylake() && _stepping >= 5;
2127 }
2128
2129 bool VM_Version::is_intel_darkmont() {
2130 return is_intel() && is_intel_server_family() && (_model == 0xCC || _model == 0xDD);
2131 }
2132
2133 void VM_Version::clear_apx_test_state() {
2134 clear_apx_test_state_stub();
2135 }
2136
2137 static bool _vm_version_initialized = false;
2138
2139 void VM_Version::initialize() {
2140 ResourceMark rm;
2141
2142 // Making this stub must be FIRST use of assembler
2143 stub_blob = BufferBlob::create("VM_Version stub", stub_size);
2144 if (stub_blob == nullptr) {
2145 vm_exit_during_initialization("Unable to allocate stub for VM_Version");
2146 }
2147 CodeBuffer c(stub_blob);
2148 VM_Version_StubGenerator g(&c);
2149
2150 get_cpu_info_stub = CAST_TO_FN_PTR(get_cpu_info_stub_t,
2151 g.generate_get_cpu_info());
2152 detect_virt_stub = CAST_TO_FN_PTR(detect_virt_stub_t,
2153 g.generate_detect_virt());
2154 clear_apx_test_state_stub = CAST_TO_FN_PTR(clear_apx_test_state_t,
2155 g.clear_apx_test_state());
2156 getCPUIDBrandString_stub = CAST_TO_FN_PTR(getCPUIDBrandString_stub_t,
2157 g.generate_getCPUIDBrandString());
2158 get_processor_features();
2159
2160 Assembler::precompute_instructions();
2161
2162 if (VM_Version::supports_hv()) { // Supports hypervisor
2163 check_virtualizations();
2164 }
2165 _vm_version_initialized = true;
2166 }
2167
2168 typedef enum {
2169 CPU_FAMILY_8086_8088 = 0,
2170 CPU_FAMILY_INTEL_286 = 2,
2171 CPU_FAMILY_INTEL_386 = 3,
2172 CPU_FAMILY_INTEL_486 = 4,
2173 CPU_FAMILY_PENTIUM = 5,
2174 CPU_FAMILY_PENTIUMPRO = 6, // Same family several models
2175 CPU_FAMILY_PENTIUM_4 = 0xF
2176 } FamilyFlag;
2177
2178 typedef enum {
2179 RDTSCP_FLAG = 0x08000000, // bit 27
2180 INTEL64_FLAG = 0x20000000 // bit 29
2181 } _featureExtendedEdxFlag;
2182
2183 typedef enum {
2184 FPU_FLAG = 0x00000001,
2185 VME_FLAG = 0x00000002,
2186 DE_FLAG = 0x00000004,
2187 PSE_FLAG = 0x00000008,
2188 TSC_FLAG = 0x00000010,
2189 MSR_FLAG = 0x00000020,
2190 PAE_FLAG = 0x00000040,
2191 MCE_FLAG = 0x00000080,
2192 CX8_FLAG = 0x00000100,
2193 APIC_FLAG = 0x00000200,
2194 SEP_FLAG = 0x00000800,
2195 MTRR_FLAG = 0x00001000,
2196 PGE_FLAG = 0x00002000,
2197 MCA_FLAG = 0x00004000,
2198 CMOV_FLAG = 0x00008000,
2199 PAT_FLAG = 0x00010000,
2200 PSE36_FLAG = 0x00020000,
2201 PSNUM_FLAG = 0x00040000,
2202 CLFLUSH_FLAG = 0x00080000,
2203 DTS_FLAG = 0x00200000,
2204 ACPI_FLAG = 0x00400000,
2205 MMX_FLAG = 0x00800000,
2206 FXSR_FLAG = 0x01000000,
2207 SSE_FLAG = 0x02000000,
2208 SSE2_FLAG = 0x04000000,
2209 SS_FLAG = 0x08000000,
2210 HTT_FLAG = 0x10000000,
2211 TM_FLAG = 0x20000000
2212 } FeatureEdxFlag;
2213
2214 // VM_Version statics
2215 enum {
2216 ExtendedFamilyIdLength_INTEL = 16,
2217 ExtendedFamilyIdLength_AMD = 24
2218 };
2219
2220 const size_t VENDOR_LENGTH = 13;
2221 const size_t CPU_EBS_MAX_LENGTH = (3 * 4 * 4 + 1);
2222 static char* _cpu_brand_string = nullptr;
2223 static int64_t _max_qualified_cpu_frequency = 0;
2224
2225 static int _no_of_threads = 0;
2226 static int _no_of_cores = 0;
2227
2228 const char* const _family_id_intel[ExtendedFamilyIdLength_INTEL] = {
2229 "8086/8088",
2230 "",
2231 "286",
2232 "386",
2233 "486",
2234 "Pentium",
2235 "Pentium Pro", //or Pentium-M/Woodcrest depending on model
2236 "",
2237 "",
2238 "",
2239 "",
2240 "",
2241 "",
2242 "",
2243 "",
2244 "Pentium 4"
2245 };
2246
2247 const char* const _family_id_amd[ExtendedFamilyIdLength_AMD] = {
2248 "",
2249 "",
2250 "",
2251 "",
2252 "5x86",
2253 "K5/K6",
2254 "Athlon/AthlonXP",
2255 "",
2256 "",
2257 "",
2258 "",
2259 "",
2260 "",
2261 "",
2262 "",
2263 "Opteron/Athlon64",
2264 "Opteron QC/Phenom", // Barcelona et.al.
2265 "",
2266 "",
2267 "",
2268 "",
2269 "",
2270 "",
2271 "Zen"
2272 };
2273 // Partially from Intel 64 and IA-32 Architecture Software Developer's Manual,
2274 // September 2013, Vol 3C Table 35-1
2275 const char* const _model_id_pentium_pro[] = {
2276 "",
2277 "Pentium Pro",
2278 "",
2279 "Pentium II model 3",
2280 "",
2281 "Pentium II model 5/Xeon/Celeron",
2282 "Celeron",
2283 "Pentium III/Pentium III Xeon",
2284 "Pentium III/Pentium III Xeon",
2285 "Pentium M model 9", // Yonah
2286 "Pentium III, model A",
2287 "Pentium III, model B",
2288 "",
2289 "Pentium M model D", // Dothan
2290 "",
2291 "Core 2", // 0xf Woodcrest/Conroe/Merom/Kentsfield/Clovertown
2292 "",
2293 "",
2294 "",
2295 "",
2296 "",
2297 "",
2298 "Celeron", // 0x16 Celeron 65nm
2299 "Core 2", // 0x17 Penryn / Harpertown
2300 "",
2301 "",
2302 "Core i7", // 0x1A CPU_MODEL_NEHALEM_EP
2303 "Atom", // 0x1B Z5xx series Silverthorn
2304 "",
2305 "Core 2", // 0x1D Dunnington (6-core)
2306 "Nehalem", // 0x1E CPU_MODEL_NEHALEM
2307 "",
2308 "",
2309 "",
2310 "",
2311 "",
2312 "",
2313 "Westmere", // 0x25 CPU_MODEL_WESTMERE
2314 "",
2315 "",
2316 "", // 0x28
2317 "",
2318 "Sandy Bridge", // 0x2a "2nd Generation Intel Core i7, i5, i3"
2319 "",
2320 "Westmere-EP", // 0x2c CPU_MODEL_WESTMERE_EP
2321 "Sandy Bridge-EP", // 0x2d CPU_MODEL_SANDYBRIDGE_EP
2322 "Nehalem-EX", // 0x2e CPU_MODEL_NEHALEM_EX
2323 "Westmere-EX", // 0x2f CPU_MODEL_WESTMERE_EX
2324 "",
2325 "",
2326 "",
2327 "",
2328 "",
2329 "",
2330 "",
2331 "",
2332 "",
2333 "",
2334 "Ivy Bridge", // 0x3a
2335 "",
2336 "Haswell", // 0x3c "4th Generation Intel Core Processor"
2337 "", // 0x3d "Next Generation Intel Core Processor"
2338 "Ivy Bridge-EP", // 0x3e "Next Generation Intel Xeon Processor E7 Family"
2339 "", // 0x3f "Future Generation Intel Xeon Processor"
2340 "",
2341 "",
2342 "",
2343 "",
2344 "",
2345 "Haswell", // 0x45 "4th Generation Intel Core Processor"
2346 "Haswell", // 0x46 "4th Generation Intel Core Processor"
2347 nullptr
2348 };
2349
2350 /* Brand ID is for back compatibility
2351 * Newer CPUs uses the extended brand string */
2352 const char* const _brand_id[] = {
2353 "",
2354 "Celeron processor",
2355 "Pentium III processor",
2356 "Intel Pentium III Xeon processor",
2357 "",
2358 "",
2359 "",
2360 "",
2361 "Intel Pentium 4 processor",
2362 nullptr
2363 };
2364
2365
2366 const char* const _feature_edx_id[] = {
2367 "On-Chip FPU",
2368 "Virtual Mode Extensions",
2369 "Debugging Extensions",
2370 "Page Size Extensions",
2371 "Time Stamp Counter",
2372 "Model Specific Registers",
2373 "Physical Address Extension",
2374 "Machine Check Exceptions",
2375 "CMPXCHG8B Instruction",
2376 "On-Chip APIC",
2377 "",
2378 "Fast System Call",
2379 "Memory Type Range Registers",
2380 "Page Global Enable",
2381 "Machine Check Architecture",
2382 "Conditional Mov Instruction",
2383 "Page Attribute Table",
2384 "36-bit Page Size Extension",
2385 "Processor Serial Number",
2386 "CLFLUSH Instruction",
2387 "",
2388 "Debug Trace Store feature",
2389 "ACPI registers in MSR space",
2390 "Intel Architecture MMX Technology",
2391 "Fast Float Point Save and Restore",
2392 "Streaming SIMD extensions",
2393 "Streaming SIMD extensions 2",
2394 "Self-Snoop",
2395 "Hyper Threading",
2396 "Thermal Monitor",
2397 "",
2398 "Pending Break Enable"
2399 };
2400
2401 const char* const _feature_extended_edx_id[] = {
2402 "",
2403 "",
2404 "",
2405 "",
2406 "",
2407 "",
2408 "",
2409 "",
2410 "",
2411 "",
2412 "",
2413 "SYSCALL/SYSRET",
2414 "",
2415 "",
2416 "",
2417 "",
2418 "",
2419 "",
2420 "",
2421 "",
2422 "Execute Disable Bit",
2423 "",
2424 "",
2425 "",
2426 "",
2427 "",
2428 "",
2429 "RDTSCP",
2430 "",
2431 "Intel 64 Architecture",
2432 "",
2433 ""
2434 };
2435
2436 const char* const _feature_ecx_id[] = {
2437 "Streaming SIMD Extensions 3",
2438 "PCLMULQDQ",
2439 "64-bit DS Area",
2440 "MONITOR/MWAIT instructions",
2441 "CPL Qualified Debug Store",
2442 "Virtual Machine Extensions",
2443 "Safer Mode Extensions",
2444 "Enhanced Intel SpeedStep technology",
2445 "Thermal Monitor 2",
2446 "Supplemental Streaming SIMD Extensions 3",
2447 "L1 Context ID",
2448 "",
2449 "Fused Multiply-Add",
2450 "CMPXCHG16B",
2451 "xTPR Update Control",
2452 "Perfmon and Debug Capability",
2453 "",
2454 "Process-context identifiers",
2455 "Direct Cache Access",
2456 "Streaming SIMD extensions 4.1",
2457 "Streaming SIMD extensions 4.2",
2458 "x2APIC",
2459 "MOVBE",
2460 "Popcount instruction",
2461 "TSC-Deadline",
2462 "AESNI",
2463 "XSAVE",
2464 "OSXSAVE",
2465 "AVX",
2466 "F16C",
2467 "RDRAND",
2468 ""
2469 };
2470
2471 const char* const _feature_extended_ecx_id[] = {
2472 "LAHF/SAHF instruction support",
2473 "Core multi-processor legacy mode",
2474 "",
2475 "",
2476 "",
2477 "Advanced Bit Manipulations: LZCNT",
2478 "SSE4A: MOVNTSS, MOVNTSD, EXTRQ, INSERTQ",
2479 "Misaligned SSE mode",
2480 "",
2481 "",
2482 "",
2483 "",
2484 "",
2485 "",
2486 "",
2487 "",
2488 "",
2489 "",
2490 "",
2491 "",
2492 "",
2493 "",
2494 "",
2495 "",
2496 "",
2497 "",
2498 "",
2499 "",
2500 "",
2501 "",
2502 "",
2503 ""
2504 };
2505
2506 const char* VM_Version::cpu_model_description(void) {
2507 uint32_t cpu_family = extended_cpu_family();
2508 uint32_t cpu_model = extended_cpu_model();
2509 const char* model = nullptr;
2510
2511 if (cpu_family == CPU_FAMILY_PENTIUMPRO) {
2512 for (uint32_t i = 0; i <= cpu_model; i++) {
2513 model = _model_id_pentium_pro[i];
2514 if (model == nullptr) {
2515 break;
2516 }
2517 }
2518 }
2519 return model;
2520 }
2521
2522 const char* VM_Version::cpu_brand_string(void) {
2523 if (_cpu_brand_string == nullptr) {
2524 _cpu_brand_string = NEW_C_HEAP_ARRAY_RETURN_NULL(char, CPU_EBS_MAX_LENGTH, mtInternal);
2525 if (nullptr == _cpu_brand_string) {
2526 return nullptr;
2527 }
2528 int ret_val = cpu_extended_brand_string(_cpu_brand_string, CPU_EBS_MAX_LENGTH);
2529 if (ret_val != OS_OK) {
2530 FREE_C_HEAP_ARRAY(_cpu_brand_string);
2531 _cpu_brand_string = nullptr;
2532 }
2533 }
2534 return _cpu_brand_string;
2535 }
2536
2537 const char* VM_Version::cpu_brand(void) {
2538 const char* brand = nullptr;
2539
2540 if ((_cpuid_info.std_cpuid1_ebx.value & 0xFF) > 0) {
2541 int brand_num = _cpuid_info.std_cpuid1_ebx.value & 0xFF;
2542 brand = _brand_id[0];
2543 for (int i = 0; brand != nullptr && i <= brand_num; i += 1) {
2544 brand = _brand_id[i];
2545 }
2546 }
2547 return brand;
2548 }
2549
2550 bool VM_Version::cpu_is_em64t(void) {
2551 return ((_cpuid_info.ext_cpuid1_edx.value & INTEL64_FLAG) == INTEL64_FLAG);
2552 }
2553
2554 bool VM_Version::is_netburst(void) {
2555 return (is_intel() && (extended_cpu_family() == CPU_FAMILY_PENTIUM_4));
2556 }
2557
2558 bool VM_Version::supports_tscinv_ext(void) {
2559 if (!supports_tscinv_bit()) {
2560 return false;
2561 }
2562
2563 if (is_intel()) {
2564 return true;
2565 }
2566
2567 if (is_amd()) {
2568 return !is_amd_Barcelona();
2569 }
2570
2571 if (is_hygon()) {
2572 return true;
2573 }
2574
2575 return false;
2576 }
2577
2578 void VM_Version::resolve_cpu_information_details(void) {
2579
2580 // in future we want to base this information on proper cpu
2581 // and cache topology enumeration such as:
2582 // Intel 64 Architecture Processor Topology Enumeration
2583 // which supports system cpu and cache topology enumeration
2584 // either using 2xAPICIDs or initial APICIDs
2585
2586 // currently only rough cpu information estimates
2587 // which will not necessarily reflect the exact configuration of the system
2588
2589 // this is the number of logical hardware threads
2590 // visible to the operating system
2591 _no_of_threads = os::processor_count();
2592
2593 // find out number of threads per cpu package
2594 int threads_per_package = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus;
2595 if (threads_per_package == 0) {
2596 // Fallback code to avoid div by zero in subsequent code.
2597 // CPUID 0Bh (ECX = 1) might return 0 on older AMD processor (EPYC 7763 at least)
2598 threads_per_package = threads_per_core() * cores_per_cpu();
2599 }
2600
2601 // use amount of threads visible to the process in order to guess number of sockets
2602 _no_of_sockets = _no_of_threads / threads_per_package;
2603
2604 // process might only see a subset of the total number of threads
2605 // from a single processor package. Virtualization/resource management for example.
2606 // If so then just write a hard 1 as num of pkgs.
2607 if (0 == _no_of_sockets) {
2608 _no_of_sockets = 1;
2609 }
2610
2611 // estimate the number of cores
2612 _no_of_cores = cores_per_cpu() * _no_of_sockets;
2613 }
2614
2615
2616 const char* VM_Version::cpu_family_description(void) {
2617 int cpu_family_id = extended_cpu_family();
2618 if (is_amd()) {
2619 if (cpu_family_id < ExtendedFamilyIdLength_AMD) {
2620 return _family_id_amd[cpu_family_id];
2621 }
2622 }
2623 if (is_intel()) {
2624 if (cpu_family_id == CPU_FAMILY_PENTIUMPRO) {
2625 return cpu_model_description();
2626 }
2627 if (cpu_family_id < ExtendedFamilyIdLength_INTEL) {
2628 return _family_id_intel[cpu_family_id];
2629 }
2630 }
2631 if (is_zx()) {
2632 int cpu_model_id = extended_cpu_model();
2633 if (cpu_family_id == 7) {
2634 switch (cpu_model_id) {
2635 case 0x1B:
2636 return "wudaokou";
2637 case 0x3B:
2638 return "lujiazui";
2639 case 0x5B:
2640 return "yongfeng";
2641 case 0x6B:
2642 return "shijidadao";
2643 }
2644 } else if (cpu_family_id == 6) {
2645 return "zhangjiang";
2646 }
2647 }
2648 if (is_hygon()) {
2649 return "Dhyana";
2650 }
2651 return "Unknown x86";
2652 }
2653
2654 int VM_Version::cpu_type_description(char* const buf, size_t buf_len) {
2655 assert(buf != nullptr, "buffer is null!");
2656 assert(buf_len >= CPU_TYPE_DESC_BUF_SIZE, "buffer len should at least be == CPU_TYPE_DESC_BUF_SIZE!");
2657
2658 const char* cpu_type = nullptr;
2659 const char* x64 = nullptr;
2660
2661 if (is_intel()) {
2662 cpu_type = "Intel";
2663 x64 = cpu_is_em64t() ? " Intel64" : "";
2664 } else if (is_amd()) {
2665 cpu_type = "AMD";
2666 x64 = cpu_is_em64t() ? " AMD64" : "";
2667 } else if (is_zx()) {
2668 cpu_type = "Zhaoxin";
2669 x64 = cpu_is_em64t() ? " x86_64" : "";
2670 } else if (is_hygon()) {
2671 cpu_type = "Hygon";
2672 x64 = cpu_is_em64t() ? " AMD64" : "";
2673 } else {
2674 cpu_type = "Unknown x86";
2675 x64 = cpu_is_em64t() ? " x86_64" : "";
2676 }
2677
2678 jio_snprintf(buf, buf_len, "%s %s%s SSE SSE2%s%s%s%s%s%s%s%s",
2679 cpu_type,
2680 cpu_family_description(),
2681 supports_ht() ? " (HT)" : "",
2682 supports_sse3() ? " SSE3" : "",
2683 supports_ssse3() ? " SSSE3" : "",
2684 supports_sse4_1() ? " SSE4.1" : "",
2685 supports_sse4_2() ? " SSE4.2" : "",
2686 supports_sse4a() ? " SSE4A" : "",
2687 is_netburst() ? " Netburst" : "",
2688 is_intel_family_core() ? " Core" : "",
2689 x64);
2690
2691 return OS_OK;
2692 }
2693
2694 int VM_Version::cpu_extended_brand_string(char* const buf, size_t buf_len) {
2695 assert(buf != nullptr, "buffer is null!");
2696 assert(buf_len >= CPU_EBS_MAX_LENGTH, "buffer len should at least be == CPU_EBS_MAX_LENGTH!");
2697 assert(getCPUIDBrandString_stub != nullptr, "not initialized");
2698
2699 // invoke newly generated asm code to fetch CPU Brand String
2700 getCPUIDBrandString_stub(&_cpuid_info);
2701
2702 // fetch results into buffer
2703 *((uint32_t*) &buf[0]) = _cpuid_info.proc_name_0;
2704 *((uint32_t*) &buf[4]) = _cpuid_info.proc_name_1;
2705 *((uint32_t*) &buf[8]) = _cpuid_info.proc_name_2;
2706 *((uint32_t*) &buf[12]) = _cpuid_info.proc_name_3;
2707 *((uint32_t*) &buf[16]) = _cpuid_info.proc_name_4;
2708 *((uint32_t*) &buf[20]) = _cpuid_info.proc_name_5;
2709 *((uint32_t*) &buf[24]) = _cpuid_info.proc_name_6;
2710 *((uint32_t*) &buf[28]) = _cpuid_info.proc_name_7;
2711 *((uint32_t*) &buf[32]) = _cpuid_info.proc_name_8;
2712 *((uint32_t*) &buf[36]) = _cpuid_info.proc_name_9;
2713 *((uint32_t*) &buf[40]) = _cpuid_info.proc_name_10;
2714 *((uint32_t*) &buf[44]) = _cpuid_info.proc_name_11;
2715
2716 return OS_OK;
2717 }
2718
2719 size_t VM_Version::cpu_write_support_string(char* const buf, size_t buf_len) {
2720 guarantee(buf != nullptr, "buffer is null!");
2721 guarantee(buf_len > 0, "buffer len not enough!");
2722
2723 unsigned int flag = 0;
2724 unsigned int fi = 0;
2725 size_t written = 0;
2726 const char* prefix = "";
2727
2728 #define WRITE_TO_BUF(string) \
2729 { \
2730 int res = jio_snprintf(&buf[written], buf_len - written, "%s%s", prefix, string); \
2731 if (res < 0) { \
2732 return buf_len - 1; \
2733 } \
2734 written += res; \
2735 if (prefix[0] == '\0') { \
2736 prefix = ", "; \
2737 } \
2738 }
2739
2740 for (flag = 1, fi = 0; flag <= 0x20000000 ; flag <<= 1, fi++) {
2741 if (flag == HTT_FLAG && (((_cpuid_info.std_cpuid1_ebx.value >> 16) & 0xff) <= 1)) {
2742 continue; /* no hyperthreading */
2743 } else if (flag == SEP_FLAG && (cpu_family() == CPU_FAMILY_PENTIUMPRO && ((_cpuid_info.std_cpuid1_eax.value & 0xff) < 0x33))) {
2744 continue; /* no fast system call */
2745 }
2746 if ((_cpuid_info.std_cpuid1_edx.value & flag) && strlen(_feature_edx_id[fi]) > 0) {
2747 WRITE_TO_BUF(_feature_edx_id[fi]);
2748 }
2749 }
2750
2751 for (flag = 1, fi = 0; flag <= 0x20000000; flag <<= 1, fi++) {
2752 if ((_cpuid_info.std_cpuid1_ecx.value & flag) && strlen(_feature_ecx_id[fi]) > 0) {
2753 WRITE_TO_BUF(_feature_ecx_id[fi]);
2754 }
2755 }
2756
2757 for (flag = 1, fi = 0; flag <= 0x20000000 ; flag <<= 1, fi++) {
2758 if ((_cpuid_info.ext_cpuid1_ecx.value & flag) && strlen(_feature_extended_ecx_id[fi]) > 0) {
2759 WRITE_TO_BUF(_feature_extended_ecx_id[fi]);
2760 }
2761 }
2762
2763 for (flag = 1, fi = 0; flag <= 0x20000000; flag <<= 1, fi++) {
2764 if ((_cpuid_info.ext_cpuid1_edx.value & flag) && strlen(_feature_extended_edx_id[fi]) > 0) {
2765 WRITE_TO_BUF(_feature_extended_edx_id[fi]);
2766 }
2767 }
2768
2769 if (supports_tscinv_bit()) {
2770 WRITE_TO_BUF("Invariant TSC");
2771 }
2772
2773 if (supports_hybrid()) {
2774 WRITE_TO_BUF("Hybrid Architecture");
2775 }
2776
2777 return written;
2778 }
2779
2780 /**
2781 * Write a detailed description of the cpu to a given buffer, including
2782 * feature set.
2783 */
2784 int VM_Version::cpu_detailed_description(char* const buf, size_t buf_len) {
2785 assert(buf != nullptr, "buffer is null!");
2786 assert(buf_len >= CPU_DETAILED_DESC_BUF_SIZE, "buffer len should at least be == CPU_DETAILED_DESC_BUF_SIZE!");
2787
2788 static const char* unknown = "<unknown>";
2789 char vendor_id[VENDOR_LENGTH];
2790 const char* family = nullptr;
2791 const char* model = nullptr;
2792 const char* brand = nullptr;
2793 int outputLen = 0;
2794
2795 family = cpu_family_description();
2796 if (family == nullptr) {
2797 family = unknown;
2798 }
2799
2800 model = cpu_model_description();
2801 if (model == nullptr) {
2802 model = unknown;
2803 }
2804
2805 brand = cpu_brand_string();
2806
2807 if (brand == nullptr) {
2808 brand = cpu_brand();
2809 if (brand == nullptr) {
2810 brand = unknown;
2811 }
2812 }
2813
2814 *((uint32_t*) &vendor_id[0]) = _cpuid_info.std_vendor_name_0;
2815 *((uint32_t*) &vendor_id[4]) = _cpuid_info.std_vendor_name_2;
2816 *((uint32_t*) &vendor_id[8]) = _cpuid_info.std_vendor_name_1;
2817 vendor_id[VENDOR_LENGTH-1] = '\0';
2818
2819 outputLen = jio_snprintf(buf, buf_len, "Brand: %s, Vendor: %s\n"
2820 "Family: %s (0x%x), Model: %s (0x%x), Stepping: 0x%x\n"
2821 "Ext. family: 0x%x, Ext. model: 0x%x, Type: 0x%x, Signature: 0x%8.8x\n"
2822 "Features: ebx: 0x%8.8x, ecx: 0x%8.8x, edx: 0x%8.8x\n"
2823 "Ext. features: eax: 0x%8.8x, ebx: 0x%8.8x, ecx: 0x%8.8x, edx: 0x%8.8x\n"
2824 "Supports: ",
2825 brand,
2826 vendor_id,
2827 family,
2828 extended_cpu_family(),
2829 model,
2830 extended_cpu_model(),
2831 cpu_stepping(),
2832 _cpuid_info.std_cpuid1_eax.bits.ext_family,
2833 _cpuid_info.std_cpuid1_eax.bits.ext_model,
2834 _cpuid_info.std_cpuid1_eax.bits.proc_type,
2835 _cpuid_info.std_cpuid1_eax.value,
2836 _cpuid_info.std_cpuid1_ebx.value,
2837 _cpuid_info.std_cpuid1_ecx.value,
2838 _cpuid_info.std_cpuid1_edx.value,
2839 _cpuid_info.ext_cpuid1_eax,
2840 _cpuid_info.ext_cpuid1_ebx,
2841 _cpuid_info.ext_cpuid1_ecx,
2842 _cpuid_info.ext_cpuid1_edx);
2843
2844 if (outputLen < 0 || (size_t) outputLen >= buf_len - 1) {
2845 if (buf_len > 0) { buf[buf_len-1] = '\0'; }
2846 return OS_ERR;
2847 }
2848
2849 cpu_write_support_string(&buf[outputLen], buf_len - outputLen);
2850
2851 return OS_OK;
2852 }
2853
2854
2855 // Fill in Abstract_VM_Version statics
2856 void VM_Version::initialize_cpu_information() {
2857 assert(_vm_version_initialized, "should have initialized VM_Version long ago");
2858 assert(!_initialized, "shouldn't be initialized yet");
2859 resolve_cpu_information_details();
2860
2861 // initialize cpu_name and cpu_desc
2862 cpu_type_description(_cpu_name, CPU_TYPE_DESC_BUF_SIZE);
2863 cpu_detailed_description(_cpu_desc, CPU_DETAILED_DESC_BUF_SIZE);
2864 _initialized = true;
2865 }
2866
2867 /**
2868 * For information about extracting the frequency from the cpu brand string, please see:
2869 *
2870 * Intel Processor Identification and the CPUID Instruction
2871 * Application Note 485
2872 * May 2012
2873 *
2874 * The return value is the frequency in Hz.
2875 */
2876 int64_t VM_Version::max_qualified_cpu_freq_from_brand_string(void) {
2877 const char* const brand_string = cpu_brand_string();
2878 if (brand_string == nullptr) {
2879 return 0;
2880 }
2881 const int64_t MEGA = 1000000;
2882 int64_t multiplier = 0;
2883 int64_t frequency = 0;
2884 uint8_t idx = 0;
2885 // The brand string buffer is at most 48 bytes.
2886 // -2 is to prevent buffer overrun when looking for y in yHz, as z is +2 from y.
2887 for (; idx < 48-2; ++idx) {
2888 // Format is either "x.xxyHz" or "xxxxyHz", where y=M, G, T and x are digits.
2889 // Search brand string for "yHz" where y is M, G, or T.
2890 if (brand_string[idx+1] == 'H' && brand_string[idx+2] == 'z') {
2891 if (brand_string[idx] == 'M') {
2892 multiplier = MEGA;
2893 } else if (brand_string[idx] == 'G') {
2894 multiplier = MEGA * 1000;
2895 } else if (brand_string[idx] == 'T') {
2896 multiplier = MEGA * MEGA;
2897 }
2898 break;
2899 }
2900 }
2901 if (multiplier > 0) {
2902 // Compute frequency (in Hz) from brand string.
2903 if (brand_string[idx-3] == '.') { // if format is "x.xx"
2904 frequency = (brand_string[idx-4] - '0') * multiplier;
2905 frequency += (brand_string[idx-2] - '0') * multiplier / 10;
2906 frequency += (brand_string[idx-1] - '0') * multiplier / 100;
2907 } else { // format is "xxxx"
2908 frequency = (brand_string[idx-4] - '0') * 1000;
2909 frequency += (brand_string[idx-3] - '0') * 100;
2910 frequency += (brand_string[idx-2] - '0') * 10;
2911 frequency += (brand_string[idx-1] - '0');
2912 frequency *= multiplier;
2913 }
2914 }
2915 return frequency;
2916 }
2917
2918
2919 int64_t VM_Version::maximum_qualified_cpu_frequency(void) {
2920 if (_max_qualified_cpu_frequency == 0) {
2921 _max_qualified_cpu_frequency = max_qualified_cpu_freq_from_brand_string();
2922 }
2923 return _max_qualified_cpu_frequency;
2924 }
2925
2926 VM_Version::VM_Features VM_Version::CpuidInfo::feature_flags() const {
2927 VM_Features vm_features;
2928 if (std_cpuid1_edx.bits.cmpxchg8 != 0)
2929 vm_features.set_feature(CPU_CX8);
2930 if (std_cpuid1_edx.bits.cmov != 0)
2931 vm_features.set_feature(CPU_CMOV);
2932 if (std_cpuid1_edx.bits.clflush != 0)
2933 vm_features.set_feature(CPU_FLUSH);
2934 // clflush should always be available on x86_64
2935 // if not we are in real trouble because we rely on it
2936 // to flush the code cache.
2937 assert (vm_features.supports_feature(CPU_FLUSH), "clflush should be available");
2938 if (std_cpuid1_edx.bits.fxsr != 0 || (is_amd_family() &&
2939 ext_cpuid1_edx.bits.fxsr != 0))
2940 vm_features.set_feature(CPU_FXSR);
2941 // HT flag is set for multi-core processors also.
2942 if (threads_per_core() > 1)
2943 vm_features.set_feature(CPU_HT);
2944 if (std_cpuid1_edx.bits.mmx != 0 || (is_amd_family() &&
2945 ext_cpuid1_edx.bits.mmx != 0))
2946 vm_features.set_feature(CPU_MMX);
2947 if (std_cpuid1_edx.bits.sse != 0)
2948 vm_features.set_feature(CPU_SSE);
2949 if (std_cpuid1_edx.bits.sse2 != 0)
2950 vm_features.set_feature(CPU_SSE2);
2951 if (std_cpuid1_ecx.bits.sse3 != 0)
2952 vm_features.set_feature(CPU_SSE3);
2953 if (std_cpuid1_ecx.bits.ssse3 != 0)
2954 vm_features.set_feature(CPU_SSSE3);
2955 if (std_cpuid1_ecx.bits.sse4_1 != 0)
2956 vm_features.set_feature(CPU_SSE4_1);
2957 if (std_cpuid1_ecx.bits.sse4_2 != 0)
2958 vm_features.set_feature(CPU_SSE4_2);
2959 if (std_cpuid1_ecx.bits.popcnt != 0)
2960 vm_features.set_feature(CPU_POPCNT);
2961 if (sefsl1_cpuid7_edx.bits.apx_f != 0 &&
2962 xem_xcr0_eax.bits.apx_f != 0 &&
2963 std_cpuid29_ebx.bits.apx_nci_ndd_nf != 0) {
2964 vm_features.set_feature(CPU_APX_F);
2965 }
2966 if (std_cpuid1_ecx.bits.avx != 0 &&
2967 std_cpuid1_ecx.bits.osxsave != 0 &&
2968 xem_xcr0_eax.bits.sse != 0 &&
2969 xem_xcr0_eax.bits.ymm != 0) {
2970 vm_features.set_feature(CPU_AVX);
2971 vm_features.set_feature(CPU_VZEROUPPER);
2972 if (sefsl1_cpuid7_eax.bits.sha512 != 0)
2973 vm_features.set_feature(CPU_SHA512);
2974 if (std_cpuid1_ecx.bits.f16c != 0)
2975 vm_features.set_feature(CPU_F16C);
2976 if (sef_cpuid7_ebx.bits.avx2 != 0) {
2977 vm_features.set_feature(CPU_AVX2);
2978 if (sefsl1_cpuid7_eax.bits.avx_ifma != 0)
2979 vm_features.set_feature(CPU_AVX_IFMA);
2980 }
2981 if (sef_cpuid7_ecx.bits.gfni != 0)
2982 vm_features.set_feature(CPU_GFNI);
2983 if (sef_cpuid7_ebx.bits.avx512f != 0 &&
2984 xem_xcr0_eax.bits.opmask != 0 &&
2985 xem_xcr0_eax.bits.zmm512 != 0 &&
2986 xem_xcr0_eax.bits.zmm32 != 0) {
2987 vm_features.set_feature(CPU_AVX512F);
2988 if (sef_cpuid7_ebx.bits.avx512cd != 0)
2989 vm_features.set_feature(CPU_AVX512CD);
2990 if (sef_cpuid7_ebx.bits.avx512dq != 0)
2991 vm_features.set_feature(CPU_AVX512DQ);
2992 if (sef_cpuid7_ebx.bits.avx512ifma != 0)
2993 vm_features.set_feature(CPU_AVX512_IFMA);
2994 if (sef_cpuid7_ebx.bits.avx512pf != 0)
2995 vm_features.set_feature(CPU_AVX512PF);
2996 if (sef_cpuid7_ebx.bits.avx512er != 0)
2997 vm_features.set_feature(CPU_AVX512ER);
2998 if (sef_cpuid7_ebx.bits.avx512bw != 0)
2999 vm_features.set_feature(CPU_AVX512BW);
3000 if (sef_cpuid7_ebx.bits.avx512vl != 0)
3001 vm_features.set_feature(CPU_AVX512VL);
3002 if (sef_cpuid7_ecx.bits.avx512_vpopcntdq != 0)
3003 vm_features.set_feature(CPU_AVX512_VPOPCNTDQ);
3004 if (sef_cpuid7_ecx.bits.avx512_vpclmulqdq != 0)
3005 vm_features.set_feature(CPU_AVX512_VPCLMULQDQ);
3006 if (sef_cpuid7_ecx.bits.vaes != 0)
3007 vm_features.set_feature(CPU_AVX512_VAES);
3008 if (sef_cpuid7_ecx.bits.avx512_vnni != 0)
3009 vm_features.set_feature(CPU_AVX512_VNNI);
3010 if (sef_cpuid7_ecx.bits.avx512_bitalg != 0)
3011 vm_features.set_feature(CPU_AVX512_BITALG);
3012 if (sef_cpuid7_ecx.bits.avx512_vbmi != 0)
3013 vm_features.set_feature(CPU_AVX512_VBMI);
3014 if (sef_cpuid7_ecx.bits.avx512_vbmi2 != 0)
3015 vm_features.set_feature(CPU_AVX512_VBMI2);
3016 }
3017 if (is_intel()) {
3018 if (sefsl1_cpuid7_edx.bits.avx10 != 0 &&
3019 std_cpuid24_ebx.bits.avx10_vlen_512 !=0 &&
3020 std_cpuid24_ebx.bits.avx10_converged_isa_version >= 1 &&
3021 xem_xcr0_eax.bits.opmask != 0 &&
3022 xem_xcr0_eax.bits.zmm512 != 0 &&
3023 xem_xcr0_eax.bits.zmm32 != 0) {
3024 vm_features.set_feature(CPU_AVX10_1);
3025 vm_features.set_feature(CPU_AVX512F);
3026 vm_features.set_feature(CPU_AVX512CD);
3027 vm_features.set_feature(CPU_AVX512DQ);
3028 vm_features.set_feature(CPU_AVX512PF);
3029 vm_features.set_feature(CPU_AVX512ER);
3030 vm_features.set_feature(CPU_AVX512BW);
3031 vm_features.set_feature(CPU_AVX512VL);
3032 vm_features.set_feature(CPU_AVX512_VPOPCNTDQ);
3033 vm_features.set_feature(CPU_AVX512_VPCLMULQDQ);
3034 vm_features.set_feature(CPU_AVX512_VAES);
3035 vm_features.set_feature(CPU_AVX512_VNNI);
3036 vm_features.set_feature(CPU_AVX512_BITALG);
3037 vm_features.set_feature(CPU_AVX512_VBMI);
3038 vm_features.set_feature(CPU_AVX512_VBMI2);
3039 if (std_cpuid24_ebx.bits.avx10_converged_isa_version >= 2) {
3040 vm_features.set_feature(CPU_AVX10_2);
3041 }
3042 }
3043 }
3044 }
3045
3046 if (std_cpuid1_ecx.bits.hv != 0)
3047 vm_features.set_feature(CPU_HV);
3048 if (sef_cpuid7_ebx.bits.bmi1 != 0)
3049 vm_features.set_feature(CPU_BMI1);
3050 if (std_cpuid1_edx.bits.tsc != 0)
3051 vm_features.set_feature(CPU_TSC);
3052 if (ext_cpuid7_edx.bits.tsc_invariance != 0)
3053 vm_features.set_feature(CPU_TSCINV_BIT);
3054 if (std_cpuid1_ecx.bits.aes != 0)
3055 vm_features.set_feature(CPU_AES);
3056 if (ext_cpuid1_ecx.bits.lzcnt != 0)
3057 vm_features.set_feature(CPU_LZCNT);
3058 if (ext_cpuid1_ecx.bits.prefetchw != 0)
3059 vm_features.set_feature(CPU_3DNOW_PREFETCH);
3060 if (sef_cpuid7_ebx.bits.erms != 0)
3061 vm_features.set_feature(CPU_ERMS);
3062 if (sef_cpuid7_edx.bits.fast_short_rep_mov != 0)
3063 vm_features.set_feature(CPU_FSRM);
3064 if (std_cpuid1_ecx.bits.clmul != 0)
3065 vm_features.set_feature(CPU_CLMUL);
3066 if (sef_cpuid7_ebx.bits.rtm != 0)
3067 vm_features.set_feature(CPU_RTM);
3068 if (sef_cpuid7_ebx.bits.adx != 0)
3069 vm_features.set_feature(CPU_ADX);
3070 if (sef_cpuid7_ebx.bits.bmi2 != 0)
3071 vm_features.set_feature(CPU_BMI2);
3072 if (sef_cpuid7_ebx.bits.sha != 0)
3073 vm_features.set_feature(CPU_SHA);
3074 if (std_cpuid1_ecx.bits.fma != 0)
3075 vm_features.set_feature(CPU_FMA);
3076 if (sef_cpuid7_ebx.bits.clflushopt != 0)
3077 vm_features.set_feature(CPU_FLUSHOPT);
3078 if (sef_cpuid7_ebx.bits.clwb != 0)
3079 vm_features.set_feature(CPU_CLWB);
3080 if (ext_cpuid1_edx.bits.rdtscp != 0)
3081 vm_features.set_feature(CPU_RDTSCP);
3082 if (sef_cpuid7_ecx.bits.rdpid != 0)
3083 vm_features.set_feature(CPU_RDPID);
3084
3085 // AMD|Hygon additional features.
3086 if (is_amd_family()) {
3087 // PREFETCHW was checked above, check TDNOW here.
3088 if ((ext_cpuid1_edx.bits.tdnow != 0))
3089 vm_features.set_feature(CPU_3DNOW_PREFETCH);
3090 if (ext_cpuid1_ecx.bits.sse4a != 0)
3091 vm_features.set_feature(CPU_SSE4A);
3092 }
3093
3094 // Intel additional features.
3095 if (is_intel()) {
3096 if (sef_cpuid7_edx.bits.serialize != 0)
3097 vm_features.set_feature(CPU_SERIALIZE);
3098 if (sef_cpuid7_edx.bits.hybrid != 0)
3099 vm_features.set_feature(CPU_HYBRID);
3100 if (_cpuid_info.sef_cpuid7_edx.bits.avx512_fp16 != 0)
3101 vm_features.set_feature(CPU_AVX512_FP16);
3102 }
3103
3104 // ZX additional features.
3105 if (is_zx()) {
3106 // We do not know if these are supported by ZX, so we cannot trust
3107 // common CPUID bit for them.
3108 assert(vm_features.supports_feature(CPU_CLWB), "Check if it is supported?");
3109 vm_features.clear_feature(CPU_CLWB);
3110 }
3111
3112 // Protection key features.
3113 if (sef_cpuid7_ecx.bits.pku != 0) {
3114 vm_features.set_feature(CPU_PKU);
3115 }
3116 if (sef_cpuid7_ecx.bits.ospke != 0) {
3117 vm_features.set_feature(CPU_OSPKE);
3118 }
3119
3120 // Control flow enforcement (CET) features.
3121 if (sef_cpuid7_ecx.bits.cet_ss != 0) {
3122 vm_features.set_feature(CPU_CET_SS);
3123 }
3124 if (sef_cpuid7_edx.bits.cet_ibt != 0) {
3125 vm_features.set_feature(CPU_CET_IBT);
3126 }
3127
3128 // Composite features.
3129 if (supports_tscinv_bit() &&
3130 ((is_amd_family() && !is_amd_Barcelona()) ||
3131 is_intel_tsc_synched_at_init())) {
3132 vm_features.set_feature(CPU_TSCINV);
3133 }
3134 return vm_features;
3135 }
3136
3137 bool VM_Version::os_supports_avx_vectors() {
3138 bool retVal = false;
3139 int nreg = 4;
3140 if (supports_evex()) {
3141 // Verify that OS save/restore all bits of EVEX registers
3142 // during signal processing.
3143 retVal = true;
3144 for (int i = 0; i < 16 * nreg; i++) { // 64 bytes per zmm register
3145 if (_cpuid_info.zmm_save[i] != ymm_test_value()) {
3146 retVal = false;
3147 break;
3148 }
3149 }
3150 } else if (supports_avx()) {
3151 // Verify that OS save/restore all bits of AVX registers
3152 // during signal processing.
3153 retVal = true;
3154 for (int i = 0; i < 8 * nreg; i++) { // 32 bytes per ymm register
3155 if (_cpuid_info.ymm_save[i] != ymm_test_value()) {
3156 retVal = false;
3157 break;
3158 }
3159 }
3160 // zmm_save will be set on a EVEX enabled machine even if we choose AVX code gen
3161 if (retVal == false) {
3162 // Verify that OS save/restore all bits of EVEX registers
3163 // during signal processing.
3164 retVal = true;
3165 for (int i = 0; i < 16 * nreg; i++) { // 64 bytes per zmm register
3166 if (_cpuid_info.zmm_save[i] != ymm_test_value()) {
3167 retVal = false;
3168 break;
3169 }
3170 }
3171 }
3172 }
3173 return retVal;
3174 }
3175
3176 bool VM_Version::os_supports_apx_egprs() {
3177 if (!supports_apx_f()) {
3178 return false;
3179 }
3180 if (_cpuid_info.apx_save[0] != egpr_test_value() ||
3181 _cpuid_info.apx_save[1] != egpr_test_value()) {
3182 return false;
3183 }
3184 return true;
3185 }
3186
3187 uint VM_Version::cores_per_cpu() {
3188 uint result = 1;
3189 if (is_intel()) {
3190 bool supports_topology = supports_processor_topology();
3191 if (supports_topology) {
3192 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus /
3193 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus;
3194 }
3195 if (!supports_topology || result == 0) {
3196 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1);
3197 }
3198 } else if (is_amd_family()) {
3199 result = _cpuid_info.ext_cpuid8_ecx.bits.threads_per_cpu + 1;
3200 if (cpu_family() >= 0x17) { // Zen or later
3201 result /= _cpuid_info.ext_cpuid1E_ebx.bits.threads_per_core + 1;
3202 }
3203 } else if (is_zx()) {
3204 bool supports_topology = supports_processor_topology();
3205 if (supports_topology) {
3206 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus /
3207 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus;
3208 }
3209 if (!supports_topology || result == 0) {
3210 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1);
3211 }
3212 }
3213 return result;
3214 }
3215
3216 uint VM_Version::threads_per_core() {
3217 uint result = 1;
3218 if (is_intel() && supports_processor_topology()) {
3219 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus;
3220 } else if (is_zx() && supports_processor_topology()) {
3221 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus;
3222 } else if (_cpuid_info.std_cpuid1_edx.bits.ht != 0) {
3223 if (cpu_family() >= 0x17) {
3224 result = _cpuid_info.ext_cpuid1E_ebx.bits.threads_per_core + 1;
3225 } else {
3226 result = _cpuid_info.std_cpuid1_ebx.bits.threads_per_cpu /
3227 cores_per_cpu();
3228 }
3229 }
3230 return (result == 0 ? 1 : result);
3231 }
3232
3233 uint VM_Version::L1_line_size() {
3234 uint result = 0;
3235 if (is_intel()) {
3236 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1);
3237 } else if (is_amd_family()) {
3238 result = _cpuid_info.ext_cpuid5_ecx.bits.L1_line_size;
3239 } else if (is_zx()) {
3240 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1);
3241 }
3242 if (result < 32) // not defined ?
3243 result = 32; // 32 bytes by default on x86 and other x64
3244 return result;
3245 }
3246
3247 bool VM_Version::is_intel_tsc_synched_at_init() {
3248 if (is_intel_family_core()) {
3249 uint32_t ext_model = extended_cpu_model();
3250 if (ext_model == CPU_MODEL_NEHALEM_EP ||
3251 ext_model == CPU_MODEL_WESTMERE_EP ||
3252 ext_model == CPU_MODEL_SANDYBRIDGE_EP ||
3253 ext_model == CPU_MODEL_IVYBRIDGE_EP) {
3254 // <= 2-socket invariant tsc support. EX versions are usually used
3255 // in > 2-socket systems and likely don't synchronize tscs at
3256 // initialization.
3257 // Code that uses tsc values must be prepared for them to arbitrarily
3258 // jump forward or backward.
3259 return true;
3260 }
3261 }
3262 return false;
3263 }
3264
3265 int VM_Version::allocate_prefetch_distance(bool use_watermark_prefetch) {
3266 // Hardware prefetching (distance/size in bytes):
3267 // Pentium 3 - 64 / 32
3268 // Pentium 4 - 256 / 128
3269 // Athlon - 64 / 32 ????
3270 // Opteron - 128 / 64 only when 2 sequential cache lines accessed
3271 // Core - 128 / 64
3272 //
3273 // Software prefetching (distance in bytes / instruction with best score):
3274 // Pentium 3 - 128 / prefetchnta
3275 // Pentium 4 - 512 / prefetchnta
3276 // Athlon - 128 / prefetchnta
3277 // Opteron - 256 / prefetchnta
3278 // Core - 256 / prefetchnta
3279 // It will be used only when AllocatePrefetchStyle > 0
3280
3281 if (is_amd_family()) { // AMD | Hygon
3282 if (supports_sse2()) {
3283 return 256; // Opteron
3284 } else {
3285 return 128; // Athlon
3286 }
3287 } else if (is_zx()) {
3288 if (supports_sse2()) {
3289 return 256;
3290 } else {
3291 return 128;
3292 }
3293 } else { // Intel
3294 if (supports_sse3() && is_intel_server_family()) {
3295 if (is_intel_modern_cpu()) { // Nehalem based cpus
3296 return 192;
3297 } else if (use_watermark_prefetch) { // watermark prefetching on Core
3298 return 384;
3299 }
3300 }
3301 if (supports_sse2()) {
3302 if (is_intel_server_family()) {
3303 return 256; // Pentium M, Core, Core2
3304 } else {
3305 return 512; // Pentium 4
3306 }
3307 } else {
3308 return 128; // Pentium 3 (and all other old CPUs)
3309 }
3310 }
3311 }
3312
3313 bool VM_Version::is_intrinsic_supported(vmIntrinsicID id) {
3314 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
3315 switch (id) {
3316 case vmIntrinsics::_floatToFloat16:
3317 case vmIntrinsics::_float16ToFloat:
3318 if (!supports_float16()) {
3319 return false;
3320 }
3321 break;
3322 default:
3323 break;
3324 }
3325 return true;
3326 }
3327
3328 void VM_Version::insert_features_names(VM_Version::VM_Features features, stringStream& ss) {
3329 int i = 0;
3330 ss.join([&]() {
3331 const char* str = nullptr;
3332 while ((i < MAX_CPU_FEATURES) && (str == nullptr)) {
3333 if (features.supports_feature((VM_Version::Feature_Flag)i)) {
3334 str = _features_names[i];
3335 }
3336 i += 1;
3337 }
3338 return str;
3339 }, ", ");
3340 }
3341
3342 void VM_Version::get_cpu_features_name(void* features_buffer, stringStream& ss) {
3343 VM_Features* features = (VM_Features*)features_buffer;
3344 insert_features_names(*features, ss);
3345 }
3346
3347 void VM_Version::get_missing_features_name(void* features_set1, void* features_set2, stringStream& ss) {
3348 VM_Features* vm_features_set1 = (VM_Features*)features_set1;
3349 VM_Features* vm_features_set2 = (VM_Features*)features_set2;
3350 int i = 0;
3351 ss.join([&]() {
3352 const char* str = nullptr;
3353 while ((i < MAX_CPU_FEATURES) && (str == nullptr)) {
3354 Feature_Flag flag = (Feature_Flag)i;
3355 if (vm_features_set1->supports_feature(flag) && !vm_features_set2->supports_feature(flag)) {
3356 str = _features_names[i];
3357 }
3358 i += 1;
3359 }
3360 return str;
3361 }, ", ");
3362 }
3363
3364 int VM_Version::cpu_features_size() {
3365 return sizeof(VM_Features);
3366 }
3367
3368 void VM_Version::store_cpu_features(void* buf) {
3369 VM_Features copy = _features;
3370 copy.clear_feature(CPU_HT); // HT does not result in incompatibility of aot code cache
3371 memcpy(buf, ©, sizeof(VM_Features));
3372 }
3373
3374 bool VM_Version::supports_features(void* features_buffer) {
3375 VM_Features* features_to_test = (VM_Features*)features_buffer;
3376 return _features.supports_features(features_to_test);
3377 }