1 /*
2 * Copyright (c) 1997, 2020, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2015, 2020, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.hpp"
28 #include "asm/macroAssembler.inline.hpp"
29 #include "memory/resourceArea.hpp"
30 #include "runtime/java.hpp"
31 #include "runtime/os.hpp"
32 #include "runtime/stubCodeGenerator.hpp"
33 #include "runtime/vm_version.hpp"
34 #include "utilities/macros.hpp"
35
36 #include OS_HEADER_INLINE(os)
37
38 #include <sys/auxv.h>
39 #include <asm/hwcap.h>
40
41 #ifndef HWCAP_AES
42 #define HWCAP_AES (1<<3)
43 #endif
44
45 #ifndef HWCAP_PMULL
46 #define HWCAP_PMULL (1<<4)
47 #endif
48
49 #ifndef HWCAP_SHA1
50 #define HWCAP_SHA1 (1<<5)
51 #endif
52
53 #ifndef HWCAP_SHA2
54 #define HWCAP_SHA2 (1<<6)
55 #endif
56
57 #ifndef HWCAP_CRC32
58 #define HWCAP_CRC32 (1<<7)
59 #endif
60
61 #ifndef HWCAP_ATOMICS
62 #define HWCAP_ATOMICS (1<<8)
63 #endif
64
65 #ifndef HWCAP_SHA512
66 #define HWCAP_SHA512 (1 << 21)
67 #endif
68
69 int VM_Version::_cpu;
70 int VM_Version::_model;
71 int VM_Version::_model2;
72 int VM_Version::_variant;
73 int VM_Version::_revision;
74 int VM_Version::_stepping;
75 bool VM_Version::_dcpop;
76 VM_Version::PsrInfo VM_Version::_psr_info = { 0, };
77
78 static BufferBlob* stub_blob;
79 static const int stub_size = 550;
80
81 extern "C" {
82 typedef void (*getPsrInfo_stub_t)(void*);
83 }
84 static getPsrInfo_stub_t getPsrInfo_stub = NULL;
85
86
87 class VM_Version_StubGenerator: public StubCodeGenerator {
88 public:
89
90 VM_Version_StubGenerator(CodeBuffer *c) : StubCodeGenerator(c) {}
91
92 address generate_getPsrInfo() {
93 StubCodeMark mark(this, "VM_Version", "getPsrInfo_stub");
94 # define __ _masm->
95 address start = __ pc();
96
97 // void getPsrInfo(VM_Version::PsrInfo* psr_info);
98
99 address entry = __ pc();
100
101 __ enter();
102
103 __ get_dczid_el0(rscratch1);
104 __ strw(rscratch1, Address(c_rarg0, in_bytes(VM_Version::dczid_el0_offset())));
105
106 __ get_ctr_el0(rscratch1);
107 __ strw(rscratch1, Address(c_rarg0, in_bytes(VM_Version::ctr_el0_offset())));
108
109 __ leave();
110 __ ret(lr);
111
112 # undef __
113
114 return start;
115 }
116 };
117
118
119 void VM_Version::get_processor_features() {
120 _supports_cx8 = true;
121 _supports_atomic_getset4 = true;
122 _supports_atomic_getadd4 = true;
123 _supports_atomic_getset8 = true;
124 _supports_atomic_getadd8 = true;
125
126 getPsrInfo_stub(&_psr_info);
127
128 int dcache_line = VM_Version::dcache_line_size();
129
130 // Limit AllocatePrefetchDistance so that it does not exceed the
131 // constraint in AllocatePrefetchDistanceConstraintFunc.
132 if (FLAG_IS_DEFAULT(AllocatePrefetchDistance))
133 FLAG_SET_DEFAULT(AllocatePrefetchDistance, MIN2(512, 3*dcache_line));
134
135 if (FLAG_IS_DEFAULT(AllocatePrefetchStepSize))
136 FLAG_SET_DEFAULT(AllocatePrefetchStepSize, dcache_line);
137 if (FLAG_IS_DEFAULT(PrefetchScanIntervalInBytes))
138 FLAG_SET_DEFAULT(PrefetchScanIntervalInBytes, 3*dcache_line);
139 if (FLAG_IS_DEFAULT(PrefetchCopyIntervalInBytes))
140 FLAG_SET_DEFAULT(PrefetchCopyIntervalInBytes, 3*dcache_line);
141 if (FLAG_IS_DEFAULT(SoftwarePrefetchHintDistance))
142 FLAG_SET_DEFAULT(SoftwarePrefetchHintDistance, 3*dcache_line);
143
144 if (PrefetchCopyIntervalInBytes != -1 &&
145 ((PrefetchCopyIntervalInBytes & 7) || (PrefetchCopyIntervalInBytes >= 32768))) {
146 warning("PrefetchCopyIntervalInBytes must be -1, or a multiple of 8 and < 32768");
147 PrefetchCopyIntervalInBytes &= ~7;
148 if (PrefetchCopyIntervalInBytes >= 32768)
149 PrefetchCopyIntervalInBytes = 32760;
150 }
151
152 if (AllocatePrefetchDistance !=-1 && (AllocatePrefetchDistance & 7)) {
153 warning("AllocatePrefetchDistance must be multiple of 8");
154 AllocatePrefetchDistance &= ~7;
155 }
156
157 if (AllocatePrefetchStepSize & 7) {
158 warning("AllocatePrefetchStepSize must be multiple of 8");
159 AllocatePrefetchStepSize &= ~7;
160 }
161
162 if (SoftwarePrefetchHintDistance != -1 &&
163 (SoftwarePrefetchHintDistance & 7)) {
164 warning("SoftwarePrefetchHintDistance must be -1, or a multiple of 8");
165 SoftwarePrefetchHintDistance &= ~7;
166 }
167
168 uint64_t auxv = getauxval(AT_HWCAP);
169
170 char buf[512];
171
172 _features = auxv;
173
174 int cpu_lines = 0;
175 if (FILE *f = fopen("/proc/cpuinfo", "r")) {
176 // need a large buffer as the flags line may include lots of text
177 char buf[1024], *p;
178 while (fgets(buf, sizeof (buf), f) != NULL) {
179 if ((p = strchr(buf, ':')) != NULL) {
180 long v = strtol(p+1, NULL, 0);
181 if (strncmp(buf, "CPU implementer", sizeof "CPU implementer" - 1) == 0) {
182 _cpu = v;
183 cpu_lines++;
184 } else if (strncmp(buf, "CPU variant", sizeof "CPU variant" - 1) == 0) {
185 _variant = v;
186 } else if (strncmp(buf, "CPU part", sizeof "CPU part" - 1) == 0) {
187 if (_model != v) _model2 = _model;
188 _model = v;
189 } else if (strncmp(buf, "CPU revision", sizeof "CPU revision" - 1) == 0) {
190 _revision = v;
191 } else if (strncmp(buf, "flags", sizeof("flags") - 1) == 0) {
192 if (strstr(p+1, "dcpop")) {
193 _dcpop = true;
194 }
195 }
196 }
197 }
198 fclose(f);
199 }
200
201 if (os::supports_map_sync()) {
202 // if dcpop is available publish data cache line flush size via
203 // generic field, otherwise let if default to zero thereby
204 // disabling writeback
205 if (_dcpop) {
206 _data_cache_line_flush_size = dcache_line;
207 }
208 }
209
210 // Enable vendor specific features
211
212 // Ampere eMAG
213 if (_cpu == CPU_AMCC && (_model == 0) && (_variant == 0x3)) {
214 if (FLAG_IS_DEFAULT(AvoidUnalignedAccesses)) {
215 FLAG_SET_DEFAULT(AvoidUnalignedAccesses, true);
216 }
217 if (FLAG_IS_DEFAULT(UseSIMDForMemoryOps)) {
218 FLAG_SET_DEFAULT(UseSIMDForMemoryOps, true);
219 }
220 if (FLAG_IS_DEFAULT(UseSIMDForArrayEquals)) {
221 FLAG_SET_DEFAULT(UseSIMDForArrayEquals, !(_revision == 1 || _revision == 2));
222 }
223 }
224
225 // ThunderX
226 if (_cpu == CPU_CAVIUM && (_model == 0xA1)) {
227 guarantee(_variant != 0, "Pre-release hardware no longer supported.");
228 if (FLAG_IS_DEFAULT(AvoidUnalignedAccesses)) {
229 FLAG_SET_DEFAULT(AvoidUnalignedAccesses, true);
230 }
231 if (FLAG_IS_DEFAULT(UseSIMDForMemoryOps)) {
232 FLAG_SET_DEFAULT(UseSIMDForMemoryOps, (_variant > 0));
233 }
234 if (FLAG_IS_DEFAULT(UseSIMDForArrayEquals)) {
235 FLAG_SET_DEFAULT(UseSIMDForArrayEquals, false);
236 }
237 }
238
239 // ThunderX2
240 if ((_cpu == CPU_CAVIUM && (_model == 0xAF)) ||
241 (_cpu == CPU_BROADCOM && (_model == 0x516))) {
242 if (FLAG_IS_DEFAULT(AvoidUnalignedAccesses)) {
243 FLAG_SET_DEFAULT(AvoidUnalignedAccesses, true);
244 }
245 if (FLAG_IS_DEFAULT(UseSIMDForMemoryOps)) {
246 FLAG_SET_DEFAULT(UseSIMDForMemoryOps, true);
247 }
248 }
249
250 // HiSilicon TSV110
251 if (_cpu == CPU_HISILICON && _model == 0xd01) {
252 if (FLAG_IS_DEFAULT(AvoidUnalignedAccesses)) {
253 FLAG_SET_DEFAULT(AvoidUnalignedAccesses, true);
254 }
255 if (FLAG_IS_DEFAULT(UseSIMDForMemoryOps)) {
256 FLAG_SET_DEFAULT(UseSIMDForMemoryOps, true);
257 }
258 }
259
260 // Cortex A53
261 if (_cpu == CPU_ARM && (_model == 0xd03 || _model2 == 0xd03)) {
262 _features |= CPU_A53MAC;
263 if (FLAG_IS_DEFAULT(UseSIMDForArrayEquals)) {
264 FLAG_SET_DEFAULT(UseSIMDForArrayEquals, false);
265 }
266 }
267
268 // Cortex A73
269 if (_cpu == CPU_ARM && (_model == 0xd09 || _model2 == 0xd09)) {
270 if (FLAG_IS_DEFAULT(SoftwarePrefetchHintDistance)) {
271 FLAG_SET_DEFAULT(SoftwarePrefetchHintDistance, -1);
272 }
273 // A73 is faster with short-and-easy-for-speculative-execution-loop
274 if (FLAG_IS_DEFAULT(UseSimpleArrayEquals)) {
275 FLAG_SET_DEFAULT(UseSimpleArrayEquals, true);
276 }
277 }
278
279 if (_cpu == CPU_ARM && (_model == 0xd07 || _model2 == 0xd07)) _features |= CPU_STXR_PREFETCH;
280 // If an olde style /proc/cpuinfo (cpu_lines == 1) then if _model is an A57 (0xd07)
281 // we assume the worst and assume we could be on a big little system and have
282 // undisclosed A53 cores which we could be swapped to at any stage
283 if (_cpu == CPU_ARM && cpu_lines == 1 && _model == 0xd07) _features |= CPU_A53MAC;
284
285 sprintf(buf, "0x%02x:0x%x:0x%03x:%d", _cpu, _variant, _model, _revision);
286 if (_model2) sprintf(buf+strlen(buf), "(0x%03x)", _model2);
287 if (auxv & HWCAP_ASIMD) strcat(buf, ", simd");
288 if (auxv & HWCAP_CRC32) strcat(buf, ", crc");
289 if (auxv & HWCAP_AES) strcat(buf, ", aes");
290 if (auxv & HWCAP_SHA1) strcat(buf, ", sha1");
291 if (auxv & HWCAP_SHA2) strcat(buf, ", sha256");
292 if (auxv & HWCAP_SHA512) strcat(buf, ", sha512");
293 if (auxv & HWCAP_ATOMICS) strcat(buf, ", lse");
294
295 _features_string = os::strdup(buf);
296
297 if (FLAG_IS_DEFAULT(UseCRC32)) {
298 UseCRC32 = (auxv & HWCAP_CRC32) != 0;
299 }
300
301 if (UseCRC32 && (auxv & HWCAP_CRC32) == 0) {
302 warning("UseCRC32 specified, but not supported on this CPU");
303 FLAG_SET_DEFAULT(UseCRC32, false);
304 }
305
306 if (FLAG_IS_DEFAULT(UseAdler32Intrinsics)) {
307 FLAG_SET_DEFAULT(UseAdler32Intrinsics, true);
308 }
309
310 if (UseVectorizedMismatchIntrinsic) {
311 warning("UseVectorizedMismatchIntrinsic specified, but not available on this CPU.");
312 FLAG_SET_DEFAULT(UseVectorizedMismatchIntrinsic, false);
313 }
314
315 if (auxv & HWCAP_ATOMICS) {
316 if (FLAG_IS_DEFAULT(UseLSE))
317 FLAG_SET_DEFAULT(UseLSE, true);
318 } else {
319 if (UseLSE) {
320 warning("UseLSE specified, but not supported on this CPU");
321 FLAG_SET_DEFAULT(UseLSE, false);
322 }
323 }
324
325 if (auxv & HWCAP_AES) {
326 UseAES = UseAES || FLAG_IS_DEFAULT(UseAES);
327 UseAESIntrinsics =
328 UseAESIntrinsics || (UseAES && FLAG_IS_DEFAULT(UseAESIntrinsics));
329 if (UseAESIntrinsics && !UseAES) {
330 warning("UseAESIntrinsics enabled, but UseAES not, enabling");
331 UseAES = true;
332 }
333 } else {
334 if (UseAES) {
335 warning("AES instructions are not available on this CPU");
336 FLAG_SET_DEFAULT(UseAES, false);
337 }
338 if (UseAESIntrinsics) {
339 warning("AES intrinsics are not available on this CPU");
340 FLAG_SET_DEFAULT(UseAESIntrinsics, false);
341 }
342 }
343
344 if (UseAESCTRIntrinsics) {
345 warning("AES/CTR intrinsics are not available on this CPU");
346 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false);
347 }
348
349 if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) {
350 UseCRC32Intrinsics = true;
351 }
352
353 if (auxv & HWCAP_CRC32) {
354 if (FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) {
355 FLAG_SET_DEFAULT(UseCRC32CIntrinsics, true);
356 }
357 } else if (UseCRC32CIntrinsics) {
358 warning("CRC32C is not available on the CPU");
359 FLAG_SET_DEFAULT(UseCRC32CIntrinsics, false);
360 }
361
362 if (FLAG_IS_DEFAULT(UseFMA)) {
363 FLAG_SET_DEFAULT(UseFMA, true);
364 }
365
366 if (UseMD5Intrinsics) {
367 warning("MD5 intrinsics are not available on this CPU");
368 FLAG_SET_DEFAULT(UseMD5Intrinsics, false);
369 }
370
371 if (auxv & (HWCAP_SHA1 | HWCAP_SHA2)) {
372 if (FLAG_IS_DEFAULT(UseSHA)) {
373 FLAG_SET_DEFAULT(UseSHA, true);
374 }
375 } else if (UseSHA) {
376 warning("SHA instructions are not available on this CPU");
377 FLAG_SET_DEFAULT(UseSHA, false);
378 }
379
380 if (UseSHA && (auxv & HWCAP_SHA1)) {
381 if (FLAG_IS_DEFAULT(UseSHA1Intrinsics)) {
382 FLAG_SET_DEFAULT(UseSHA1Intrinsics, true);
383 }
384 } else if (UseSHA1Intrinsics) {
385 warning("Intrinsics for SHA-1 crypto hash functions not available on this CPU.");
386 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false);
387 }
388
389 if (UseSHA && (auxv & HWCAP_SHA2)) {
390 if (FLAG_IS_DEFAULT(UseSHA256Intrinsics)) {
391 FLAG_SET_DEFAULT(UseSHA256Intrinsics, true);
392 }
393 } else if (UseSHA256Intrinsics) {
394 warning("Intrinsics for SHA-224 and SHA-256 crypto hash functions not available on this CPU.");
395 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false);
396 }
397
398 if (UseSHA && (auxv & HWCAP_SHA512)) {
399 // Do not auto-enable UseSHA512Intrinsics until it has been fully tested on hardware
400 // if (FLAG_IS_DEFAULT(UseSHA512Intrinsics)) {
401 // FLAG_SET_DEFAULT(UseSHA512Intrinsics, true);
402 // }
403 } else if (UseSHA512Intrinsics) {
404 warning("Intrinsics for SHA-384 and SHA-512 crypto hash functions not available on this CPU.");
405 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false);
406 }
407
408 if (!(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics)) {
409 FLAG_SET_DEFAULT(UseSHA, false);
410 }
411
412 if (auxv & HWCAP_PMULL) {
413 if (FLAG_IS_DEFAULT(UseGHASHIntrinsics)) {
414 FLAG_SET_DEFAULT(UseGHASHIntrinsics, true);
415 }
416 } else if (UseGHASHIntrinsics) {
417 warning("GHASH intrinsics are not available on this CPU");
418 FLAG_SET_DEFAULT(UseGHASHIntrinsics, false);
419 }
420
421 if (is_zva_enabled()) {
422 if (FLAG_IS_DEFAULT(UseBlockZeroing)) {
423 FLAG_SET_DEFAULT(UseBlockZeroing, true);
424 }
425 if (FLAG_IS_DEFAULT(BlockZeroingLowLimit)) {
426 FLAG_SET_DEFAULT(BlockZeroingLowLimit, 4 * VM_Version::zva_length());
427 }
428 } else if (UseBlockZeroing) {
429 warning("DC ZVA is not available on this CPU");
430 FLAG_SET_DEFAULT(UseBlockZeroing, false);
431 }
432
433 // This machine allows unaligned memory accesses
434 if (FLAG_IS_DEFAULT(UseUnalignedAccesses)) {
435 FLAG_SET_DEFAULT(UseUnalignedAccesses, true);
436 }
437
438 if (FLAG_IS_DEFAULT(UsePopCountInstruction)) {
439 FLAG_SET_DEFAULT(UsePopCountInstruction, true);
440 }
441
442 if (!UsePopCountInstruction) {
443 warning("UsePopCountInstruction is always enabled on this CPU");
444 UsePopCountInstruction = true;
445 }
446
447 #ifdef COMPILER2
448 if (FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) {
449 UseMultiplyToLenIntrinsic = true;
450 }
451
452 if (FLAG_IS_DEFAULT(UseSquareToLenIntrinsic)) {
453 UseSquareToLenIntrinsic = true;
454 }
455
456 if (FLAG_IS_DEFAULT(UseMulAddIntrinsic)) {
457 UseMulAddIntrinsic = true;
458 }
459
460 if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) {
461 UseMontgomeryMultiplyIntrinsic = true;
462 }
463 if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) {
464 UseMontgomerySquareIntrinsic = true;
465 }
466
467 if (FLAG_IS_DEFAULT(OptoScheduling)) {
468 OptoScheduling = true;
469 }
470
471 if (FLAG_IS_DEFAULT(AlignVector)) {
472 AlignVector = AvoidUnalignedAccesses;
473 }
474 #endif
475 }
476
477 void VM_Version::initialize() {
478 ResourceMark rm;
479
480 stub_blob = BufferBlob::create("getPsrInfo_stub", stub_size);
481 if (stub_blob == NULL) {
482 vm_exit_during_initialization("Unable to allocate getPsrInfo_stub");
483 }
484
485 CodeBuffer c(stub_blob);
486 VM_Version_StubGenerator g(&c);
487 getPsrInfo_stub = CAST_TO_FN_PTR(getPsrInfo_stub_t,
488 g.generate_getPsrInfo());
489
490 get_processor_features();
491
492 UNSUPPORTED_OPTION(CriticalJNINatives);
493 }