1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
|
/*
* Runtime CPU detection
* (C) 2009,2010,2013,2017 Jack Lloyd
*
* Botan is released under the Simplified BSD License (see license.txt)
*/
#include <botan/cpuid.h>
#include <botan/types.h>
#include <botan/loadstor.h>
#include <botan/exceptn.h>
#include <botan/mem_ops.h>
#include <ostream>
#if defined(BOTAN_TARGET_CPU_IS_PPC_FAMILY)
/*
* On Darwin and OpenBSD ppc, use sysctl to detect AltiVec
*/
#if defined(BOTAN_TARGET_OS_IS_DARWIN)
#include <sys/sysctl.h>
#elif defined(BOTAN_TARGET_OS_IS_OPENBSD)
#include <sys/param.h>
#include <sys/sysctl.h>
#include <machine/cpu.h>
#endif
#elif defined(BOTAN_TARGET_CPU_IS_ARM_FAMILY)
/*
* On ARM, use getauxval if available, otherwise fall back to
* running probe functions with a SIGILL handler.
*/
#if defined(BOTAN_TARGET_OS_HAS_GETAUXVAL)
#include <sys/auxv.h>
#endif
#include <botan/internal/os_utils.h>
#elif defined(BOTAN_TARGET_CPU_IS_X86_FAMILY)
#if defined(BOTAN_BUILD_COMPILER_IS_MSVC)
#include <intrin.h>
#define X86_CPUID(type, out) do { __cpuid((int*)out, type); } while(0)
#define X86_CPUID_SUBLEVEL(type, level, out) do { __cpuidex((int*)out, type, level); } while(0)
#elif defined(BOTAN_BUILD_COMPILER_IS_INTEL)
#include <ia32intrin.h>
#define X86_CPUID(type, out) do { __cpuid(out, type); } while(0)
#define X86_CPUID_SUBLEVEL(type, level, out) do { __cpuidex((int*)out, type, level); } while(0)
#elif defined(BOTAN_TARGET_ARCH_IS_X86_64) && defined(BOTAN_USE_GCC_INLINE_ASM)
#define X86_CPUID(type, out) \
asm("cpuid\n\t" : "=a" (out[0]), "=b" (out[1]), "=c" (out[2]), "=d" (out[3]) \
: "0" (type))
#define X86_CPUID_SUBLEVEL(type, level, out) \
asm("cpuid\n\t" : "=a" (out[0]), "=b" (out[1]), "=c" (out[2]), "=d" (out[3]) \
: "0" (type), "2" (level))
#elif defined(BOTAN_BUILD_COMPILER_IS_GCC) || defined(BOTAN_BUILD_COMPILER_IS_CLANG)
#include <cpuid.h>
#define X86_CPUID(type, out) do { __get_cpuid(type, out, out+1, out+2, out+3); } while(0)
#define X86_CPUID_SUBLEVEL(type, level, out) \
do { __cpuid_count(type, level, out[0], out[1], out[2], out[3]); } while(0)
#else
#warning "No way of calling cpuid for this compiler"
#define X86_CPUID(type, out) do { clear_mem(out, 4); } while(0)
#define X86_CPUID_SUBLEVEL(type, level, out) do { clear_mem(out, 4); } while(0)
#endif
#endif
namespace Botan {
uint64_t CPUID::g_processor_features = 0;
size_t CPUID::g_cache_line_size = BOTAN_TARGET_CPU_DEFAULT_CACHE_LINE_SIZE;
bool CPUID::g_little_endian = false;
namespace {
#if defined(BOTAN_TARGET_CPU_IS_PPC_FAMILY)
/*
* PowerPC specific block: check for AltiVec using either
* sysctl or by reading processor version number register.
*/
uint64_t powerpc_detect_cpu_featutures()
{
#if defined(BOTAN_TARGET_OS_IS_DARWIN) || defined(BOTAN_TARGET_OS_IS_OPENBSD)
// On Darwin/OS X and OpenBSD, use sysctl
#if defined(BOTAN_TARGET_OS_IS_OPENBSD)
int sels[2] = { CTL_MACHDEP, CPU_ALTIVEC };
#else
// From Apple's docs
int sels[2] = { CTL_HW, HW_VECTORUNIT };
#endif
int vector_type = 0;
size_t length = sizeof(vector_type);
int error = sysctl(sels, 2, &vector_type, &length, NULL, 0);
if(error == 0 && vector_type > 0)
return (1ULL << CPUID::CPUID_ALTIVEC_BIT);
#elif defined(BOTAN_TARGET_OS_IS_LINUX) || defined(BOTAN_TARGET_OS_IS_NETBSD)
/*
On PowerPC, MSR 287 is PVR, the Processor Version Number
Normally it is only accessible to ring 0, but Linux and NetBSD
(others, too, maybe?) will trap and emulate it for us.
PVR identifiers for various AltiVec enabled CPUs. Taken from
PearPC and Linux sources, mostly.
*/
uint32_t pvr = 0;
// TODO: we could run inside SIGILL handler block
asm volatile("mfspr %0, 287" : "=r" (pvr));
// Top 16 bit suffice to identify model
pvr >>= 16;
const uint16_t PVR_G4_7400 = 0x000C;
const uint16_t PVR_G5_970 = 0x0039;
const uint16_t PVR_G5_970FX = 0x003C;
const uint16_t PVR_G5_970MP = 0x0044;
const uint16_t PVR_G5_970GX = 0x0045;
const uint16_t PVR_POWER6 = 0x003E;
const uint16_t PVR_POWER7 = 0x003F;
const uint16_t PVR_POWER8 = 0x004B;
const uint16_t PVR_CELL_PPU = 0x0070;
if(pvr == PVR_G4_7400 ||
pvr == PVR_G5_970 || pvr == PVR_G5_970FX ||
pvr == PVR_G5_970MP || pvr == PVR_G5_970GX ||
pvr == PVR_POWER6 || pvr == PVR_POWER7 || pvr == PVR_POWER8 ||
pvr == PVR_CELL_PPU)
{
return (1ULL << CPUID::CPUID_ALTIVEC_BIT);
}
#else
#warning "No PowerPC feature detection available for this platform"
#endif
return 0;
}
#elif defined(BOTAN_TARGET_CPU_IS_ARM_FAMILY)
uint64_t arm_detect_cpu_features(size_t* cache_line_size)
{
uint64_t detected_features = 0;
*cache_line_size = BOTAN_TARGET_CPU_DEFAULT_CACHE_LINE_SIZE;
#if defined(BOTAN_TARGET_OS_HAS_GETAUXVAL)
errno = 0;
/*
* On systems with getauxval these bits should normally be defined
* in bits/auxv.h but some buggy? glibc installs seem to miss them.
* These following values are all fixed, for the Linux ELF format,
* so we just hardcode them in ARM_hwcap_bit enum.
*/
enum ARM_hwcap_bit {
#if defined(BOTAN_TARGET_ARCH_IS_ARM32)
NEON_bit = (1 << 12),
AES_bit = (1 << 0),
PMULL_bit = (1 << 1),
SHA1_bit = (1 << 2),
SHA2_bit = (1 << 3),
ARCH_hwcap_neon = 16, // AT_HWCAP
ARCH_hwcap_crypto = 26, // AT_HWCAP2
#elif defined(BOTAN_TARGET_ARCH_IS_ARM64)
NEON_bit = (1 << 1),
AES_bit = (1 << 3),
PMULL_bit = (1 << 4),
SHA1_bit = (1 << 5),
SHA2_bit = (1 << 6),
ARCH_hwcap_neon = 16, // AT_HWCAP
ARCH_hwcap_crypto = 16, // AT_HWCAP
#endif
};
const unsigned long hwcap_neon = ::getauxval(ARM_hwcap_bit::ARCH_hwcap_neon);
if(hwcap_neon & ARM_hwcap_bit::NEON_bit)
detected_features |= CPUID::CPUID_ARM_NEON_BIT;
/*
On aarch64 this ends up calling getauxval twice with AT_HWCAP
This doesn't seem worth optimizing this out, since getauxval is
just reading a field in the ELF header.
*/
const unsigned long hwcap_crypto = ::getauxval(ARM_hwcap_bit::ARCH_hwcap_crypto);
if(hwcap_crypto & ARM_hwcap_bit::AES_bit)
detected_features |= CPUID::CPUID_ARM_AES_BIT;
if(hwcap_crypto & ARM_hwcap_bit::PMULL_bit)
detected_features |= CPUID::CPUID_ARM_PMULL_BIT;
if(hwcap_crypto & ARM_hwcap_bit::SHA1_bit)
detected_features |= CPUID::CPUID_ARM_SHA1_BIT;
if(hwcap_crypto & ARM_hwcap_bit::SHA2_bit)
detected_features |= CPUID::CPUID_ARM_SHA2_BIT;
const unsigned long dcache_line = ::getauxval(AT_DCACHEBSIZE);
// plausibility check
if(dcache_line == 32 || dcache_line == 64 || dcache_line == 128)
*cache_line_size = static_cast<size_t>(dcache_line);
#endif
// TODO: probe functions
return detected_features;
}
#elif defined(BOTAN_TARGET_CPU_IS_X86_FAMILY)
uint64_t x86_detect_cpu_features(size_t* cache_line_size)
{
uint64_t features_detected = 0;
uint32_t cpuid[4] = { 0 };
// CPUID 0: vendor identification, max sublevel
X86_CPUID(0, cpuid);
const uint32_t max_supported_sublevel = cpuid[0];
const uint32_t INTEL_CPUID[3] = { 0x756E6547, 0x6C65746E, 0x49656E69 };
const uint32_t AMD_CPUID[3] = { 0x68747541, 0x444D4163, 0x69746E65 };
const bool is_intel = same_mem(cpuid + 1, INTEL_CPUID, 3);
const bool is_amd = same_mem(cpuid + 1, AMD_CPUID, 3);
if(max_supported_sublevel >= 1)
{
// CPUID 1: feature bits
X86_CPUID(1, cpuid);
const uint64_t flags0 = (static_cast<uint64_t>(cpuid[2]) << 32) | cpuid[3];
enum x86_CPUID_1_bits : uint64_t {
RDTSC = (1ULL << 4),
SSE2 = (1ULL << 26),
CLMUL = (1ULL << 33),
SSSE3 = (1ULL << 41),
SSE41 = (1ULL << 51),
SSE42 = (1ULL << 52),
AESNI = (1ULL << 57),
RDRAND = (1ULL << 62)
};
if(flags0 & x86_CPUID_1_bits::RDTSC)
features_detected |= CPUID::CPUID_RDTSC_BIT;
if(flags0 & x86_CPUID_1_bits::SSE2)
features_detected |= CPUID::CPUID_SSE2_BIT;
if(flags0 & x86_CPUID_1_bits::CLMUL)
features_detected |= CPUID::CPUID_CLMUL_BIT;
if(flags0 & x86_CPUID_1_bits::SSSE3)
features_detected |= CPUID::CPUID_SSSE3_BIT;
if(flags0 & x86_CPUID_1_bits::SSE41)
features_detected |= CPUID::CPUID_SSE41_BIT;
if(flags0 & x86_CPUID_1_bits::SSE42)
features_detected |= CPUID::CPUID_SSE42_BIT;
if(flags0 & x86_CPUID_1_bits::AESNI)
features_detected |= CPUID::CPUID_AESNI_BIT;
if(flags0 & x86_CPUID_1_bits::RDRAND)
features_detected |= CPUID::CPUID_RDRAND_BIT;
}
if(is_intel)
{
// Intel cache line size is in cpuid(1) output
*cache_line_size = 8 * get_byte(2, cpuid[1]);
}
else if(is_amd)
{
// AMD puts it in vendor zone
X86_CPUID(0x80000005, cpuid);
*cache_line_size = get_byte(3, cpuid[2]);
}
if(max_supported_sublevel >= 7)
{
clear_mem(cpuid, 4);
X86_CPUID_SUBLEVEL(7, 0, cpuid);
enum x86_CPUID_7_bits : uint64_t {
AVX2 = (1ULL << 5),
BMI2 = (1ULL << 8),
AVX512F = (1ULL << 16),
RDSEED = (1ULL << 18),
ADX = (1ULL << 19),
SHA = (1ULL << 29),
};
uint64_t flags7 = (static_cast<uint64_t>(cpuid[2]) << 32) | cpuid[1];
if(flags7 & x86_CPUID_7_bits::AVX2)
features_detected |= CPUID::CPUID_AVX2_BIT;
if(flags7 & x86_CPUID_7_bits::BMI2)
features_detected |= CPUID::CPUID_BMI2_BIT;
if(flags7 & x86_CPUID_7_bits::AVX512F)
features_detected |= CPUID::CPUID_AVX512F_BIT;
if(flags7 & x86_CPUID_7_bits::RDSEED)
features_detected |= CPUID::CPUID_RDSEED_BIT;
if(flags7 & x86_CPUID_7_bits::ADX)
features_detected |= CPUID::CPUID_ADX_BIT;
if(flags7 & x86_CPUID_7_bits::SHA)
features_detected |= CPUID::CPUID_SHA_BIT;
}
/*
* If we don't have access to CPUID, we can still safely assume that
* any x86-64 processor has SSE2 and RDTSC
*/
#if defined(BOTAN_TARGET_ARCH_IS_X86_64)
if(features_detected == 0)
{
features_detected |= CPUID::CPUID_SSE2_BIT;
features_detected |= CPUID::CPUID_RDTSC_BIT;
}
#endif
return features_detected;
}
#endif
}
bool CPUID::has_simd_32()
{
#if defined(BOTAN_TARGET_SUPPORTS_SSE2)
return CPUID::has_sse2();
#elif defined(BOTAN_TARGET_SUPPORTS_ALTIVEC)
return CPUID::has_altivec();
#else
return true;
#endif
}
void CPUID::print(std::ostream& o)
{
o << "CPUID flags: ";
#define CPUID_PRINT(flag) do { if(has_##flag()) o << #flag << " "; } while(0)
#if defined(BOTAN_TARGET_CPU_IS_X86_FAMILY)
CPUID_PRINT(sse2);
CPUID_PRINT(ssse3);
CPUID_PRINT(sse41);
CPUID_PRINT(sse42);
CPUID_PRINT(avx2);
CPUID_PRINT(avx512f);
CPUID_PRINT(rdtsc);
CPUID_PRINT(bmi2);
CPUID_PRINT(adx);
CPUID_PRINT(aes_ni);
CPUID_PRINT(clmul);
CPUID_PRINT(rdrand);
CPUID_PRINT(rdseed);
CPUID_PRINT(intel_sha);
#endif
#if defined(BOTAN_TARGET_CPU_IS_PPC_FAMILY)
CPUID_PRINT(altivec);
#endif
#if defined(BOTAN_TARGET_CPU_IS_ARM_FAMILY)
CPUID_PRINT(neon);
CPUID_PRINT(arm_sha1);
CPUID_PRINT(arm_sha2);
CPUID_PRINT(arm_aes);
CPUID_PRINT(arm_pmull);
#endif
#undef CPUID_PRINT
o << "\n";
}
void CPUID::initialize()
{
g_processor_features = 0;
#if defined(BOTAN_TARGET_CPU_IS_PPC_FAMILY)
g_processor_features = powerpc_detect_cpu_featutures();
#elif defined(BOTAN_TARGET_CPU_IS_ARM_FAMILY)
g_processor_features = arm_detect_cpu_features(&g_cache_line_size);
#elif defined(BOTAN_TARGET_CPU_IS_X86_FAMILY)
g_processor_features = x86_detect_cpu_features(&g_cache_line_size);
#endif
g_processor_features |= CPUID::CPUID_INITIALIZED_BIT;
// Check runtime endian
const uint32_t endian32 = 0x01234567;
const uint8_t* e8 = reinterpret_cast<const uint8_t*>(&endian32);
if(e8[0] == 0x01 && e8[1] == 0x23 && e8[2] == 0x45 && e8[3] == 0x67)
{
g_little_endian = false;
}
else if(e8[0] == 0x67 && e8[1] == 0x45 && e8[2] == 0x23 && e8[3] == 0x01)
{
g_little_endian = true;
}
else
{
throw Internal_Error("Unexpected endian at runtime, neither big nor little");
}
// If we were compiled with a known endian, verify it matches at runtime
#if defined(BOTAN_TARGET_CPU_IS_LITTLE_ENDIAN)
BOTAN_ASSERT(g_little_endian == true, "Build and runtime endian match");
#elif defined(BOTAN_TARGET_CPU_IS_BIG_ENDIAN)
BOTAN_ASSERT(g_little_endian == false, "Build and runtime endian match");
#endif
}
}
|