]>
Commit | Line | Data |
---|---|---|
b1322259 | 1 | /* |
83cf7abf | 2 | * Copyright 2009-2018 The OpenSSL Project Authors. All Rights Reserved. |
b1322259 RS |
3 | * |
4 | * Licensed under the OpenSSL license (the "License"). You may not use | |
5 | * this file except in compliance with the License. You can obtain a copy | |
6 | * in the file LICENSE in the source distribution or at | |
7 | * https://www.openssl.org/source/license.html | |
8 | */ | |
9 | ||
b4b48a10 AP |
10 | #include <stdio.h> |
11 | #include <stdlib.h> | |
12 | #include <string.h> | |
13 | #include <setjmp.h> | |
14 | #include <signal.h> | |
fd054957 | 15 | #include <unistd.h> |
d5630dd6 | 16 | #if defined(__linux) || defined(_AIX) |
0f113f3e | 17 | # include <sys/utsname.h> |
78c3e205 | 18 | #endif |
2688d999 AP |
19 | #if defined(_AIX53) /* defined even on post-5.3 */ |
20 | # include <sys/systemcfg.h> | |
21 | # if !defined(__power_set) | |
22 | # define __power_set(a) (_system_configuration.implementation & (a)) | |
23 | # endif | |
24 | #endif | |
0bd93bbe AP |
25 | #if defined(__APPLE__) && defined(__MACH__) |
26 | # include <sys/types.h> | |
27 | # include <sys/sysctl.h> | |
28 | #endif | |
11252459 | 29 | #include <openssl/crypto.h> |
10b0b5ec | 30 | #include <openssl/bn.h> |
8c8fbca9 BE |
31 | #include <internal/cryptlib.h> |
32 | #include <internal/chacha.h> | |
33 | #include "bn/bn_lcl.h" | |
b4b48a10 | 34 | |
0e716d92 | 35 | #include "ppc_arch.h" |
b4b48a10 | 36 | |
07f3e4f3 | 37 | unsigned int OPENSSL_ppccap_P = 0; |
b4b48a10 AP |
38 | |
39 | static sigset_t all_masked; | |
40 | ||
10b0b5ec RL |
41 | #ifdef OPENSSL_BN_ASM_MONT |
42 | int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, | |
43 | const BN_ULONG *np, const BN_ULONG *n0, int num) | |
44 | { | |
45 | int bn_mul_mont_int(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, | |
46 | const BN_ULONG *np, const BN_ULONG *n0, int num); | |
47 | int bn_mul4x_mont_int(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, | |
48 | const BN_ULONG *np, const BN_ULONG *n0, int num); | |
49 | ||
50 | if (num < 4) | |
51 | return 0; | |
52 | ||
53 | if ((num & 3) == 0) | |
54 | return bn_mul4x_mont_int(rp, ap, bp, np, n0, num); | |
55 | ||
56 | /* | |
57 | * There used to be [optional] call to bn_mul_mont_fpu64 here, | |
58 | * but above subroutine is faster on contemporary processors. | |
59 | * Formulation means that there might be old processors where | |
60 | * FPU code path would be faster, POWER6 perhaps, but there was | |
61 | * no opportunity to figure it out... | |
62 | */ | |
63 | ||
64 | return bn_mul_mont_int(rp, ap, bp, np, n0, num); | |
65 | } | |
66 | #endif | |
67 | ||
68 | void sha256_block_p8(void *ctx, const void *inp, size_t len); | |
69 | void sha256_block_ppc(void *ctx, const void *inp, size_t len); | |
8c8fbca9 | 70 | void sha256_block_data_order(void *ctx, const void *inp, size_t len); |
10b0b5ec RL |
71 | void sha256_block_data_order(void *ctx, const void *inp, size_t len) |
72 | { | |
73 | OPENSSL_ppccap_P & PPC_CRYPTO207 ? sha256_block_p8(ctx, inp, len) : | |
74 | sha256_block_ppc(ctx, inp, len); | |
75 | } | |
76 | ||
77 | void sha512_block_p8(void *ctx, const void *inp, size_t len); | |
78 | void sha512_block_ppc(void *ctx, const void *inp, size_t len); | |
8c8fbca9 | 79 | void sha512_block_data_order(void *ctx, const void *inp, size_t len); |
10b0b5ec RL |
80 | void sha512_block_data_order(void *ctx, const void *inp, size_t len) |
81 | { | |
82 | OPENSSL_ppccap_P & PPC_CRYPTO207 ? sha512_block_p8(ctx, inp, len) : | |
83 | sha512_block_ppc(ctx, inp, len); | |
84 | } | |
85 | ||
86 | #ifndef OPENSSL_NO_CHACHA | |
87 | void ChaCha20_ctr32_int(unsigned char *out, const unsigned char *inp, | |
88 | size_t len, const unsigned int key[8], | |
89 | const unsigned int counter[4]); | |
90 | void ChaCha20_ctr32_vmx(unsigned char *out, const unsigned char *inp, | |
91 | size_t len, const unsigned int key[8], | |
92 | const unsigned int counter[4]); | |
316d527f AP |
93 | void ChaCha20_ctr32_vsx(unsigned char *out, const unsigned char *inp, |
94 | size_t len, const unsigned int key[8], | |
95 | const unsigned int counter[4]); | |
10b0b5ec RL |
96 | void ChaCha20_ctr32(unsigned char *out, const unsigned char *inp, |
97 | size_t len, const unsigned int key[8], | |
98 | const unsigned int counter[4]) | |
99 | { | |
316d527f AP |
100 | OPENSSL_ppccap_P & PPC_CRYPTO207 |
101 | ? ChaCha20_ctr32_vsx(out, inp, len, key, counter) | |
102 | : OPENSSL_ppccap_P & PPC_ALTIVEC | |
103 | ? ChaCha20_ctr32_vmx(out, inp, len, key, counter) | |
104 | : ChaCha20_ctr32_int(out, inp, len, key, counter); | |
10b0b5ec RL |
105 | } |
106 | #endif | |
107 | ||
108 | #ifndef OPENSSL_NO_POLY1305 | |
109 | void poly1305_init_int(void *ctx, const unsigned char key[16]); | |
110 | void poly1305_blocks(void *ctx, const unsigned char *inp, size_t len, | |
111 | unsigned int padbit); | |
112 | void poly1305_emit(void *ctx, unsigned char mac[16], | |
113 | const unsigned int nonce[4]); | |
114 | void poly1305_init_fpu(void *ctx, const unsigned char key[16]); | |
115 | void poly1305_blocks_fpu(void *ctx, const unsigned char *inp, size_t len, | |
116 | unsigned int padbit); | |
117 | void poly1305_emit_fpu(void *ctx, unsigned char mac[16], | |
118 | const unsigned int nonce[4]); | |
8c8fbca9 | 119 | int poly1305_init(void *ctx, const unsigned char key[16], void *func[2]); |
10b0b5ec RL |
120 | int poly1305_init(void *ctx, const unsigned char key[16], void *func[2]) |
121 | { | |
122 | if (sizeof(size_t) == 4 && (OPENSSL_ppccap_P & PPC_FPU)) { | |
123 | poly1305_init_fpu(ctx, key); | |
8c8fbca9 BE |
124 | func[0] = (void*)(uintptr_t)poly1305_blocks_fpu; |
125 | func[1] = (void*)(uintptr_t)poly1305_emit_fpu; | |
10b0b5ec RL |
126 | } else { |
127 | poly1305_init_int(ctx, key); | |
8c8fbca9 BE |
128 | func[0] = (void*)(uintptr_t)poly1305_blocks; |
129 | func[1] = (void*)(uintptr_t)poly1305_emit; | |
10b0b5ec RL |
130 | } |
131 | return 1; | |
132 | } | |
133 | #endif | |
134 | ||
135 | #ifdef ECP_NISTZ256_ASM | |
136 | void ecp_nistz256_mul_mont(unsigned long res[4], const unsigned long a[4], | |
137 | const unsigned long b[4]); | |
138 | ||
139 | void ecp_nistz256_to_mont(unsigned long res[4], const unsigned long in[4]); | |
140 | void ecp_nistz256_to_mont(unsigned long res[4], const unsigned long in[4]) | |
141 | { | |
142 | static const unsigned long RR[] = { 0x0000000000000003U, | |
143 | 0xfffffffbffffffffU, | |
144 | 0xfffffffffffffffeU, | |
145 | 0x00000004fffffffdU }; | |
146 | ||
147 | ecp_nistz256_mul_mont(res, in, RR); | |
148 | } | |
149 | ||
150 | void ecp_nistz256_from_mont(unsigned long res[4], const unsigned long in[4]); | |
151 | void ecp_nistz256_from_mont(unsigned long res[4], const unsigned long in[4]) | |
152 | { | |
153 | static const unsigned long one[] = { 1, 0, 0, 0 }; | |
154 | ||
155 | ecp_nistz256_mul_mont(res, in, one); | |
156 | } | |
157 | #endif | |
158 | ||
b4b48a10 | 159 | static sigjmp_buf ill_jmp; |
0f113f3e MC |
160 | static void ill_handler(int sig) |
161 | { | |
162 | siglongjmp(ill_jmp, sig); | |
163 | } | |
b4b48a10 | 164 | |
2688d999 | 165 | void OPENSSL_fpu_probe(void); |
70b76d39 | 166 | void OPENSSL_ppc64_probe(void); |
fd054957 | 167 | void OPENSSL_altivec_probe(void); |
de51e830 | 168 | void OPENSSL_crypto207_probe(void); |
53385e1f | 169 | void OPENSSL_madd300_probe(void); |
70b76d39 | 170 | |
2688d999 AP |
171 | /* |
172 | * Use a weak reference to getauxval() so we can use it if it is available | |
173 | * but don't break the build if it is not. Note that this is *link-time* | |
174 | * feature detection, not *run-time*. In other words if we link with | |
175 | * symbol present, it's expected to be present even at run-time. | |
176 | */ | |
177 | #if defined(__GNUC__) && __GNUC__>=2 && defined(__ELF__) | |
178 | extern unsigned long getauxval(unsigned long type) __attribute__ ((weak)); | |
179 | #else | |
180 | static unsigned long (*getauxval) (unsigned long) = NULL; | |
181 | #endif | |
182 | ||
183 | /* I wish <sys/auxv.h> was universally available */ | |
184 | #define HWCAP 16 /* AT_HWCAP */ | |
185 | #define HWCAP_PPC64 (1U << 30) | |
186 | #define HWCAP_ALTIVEC (1U << 28) | |
187 | #define HWCAP_FPU (1U << 27) | |
188 | #define HWCAP_POWER6_EXT (1U << 9) | |
189 | #define HWCAP_VSX (1U << 7) | |
190 | ||
191 | #define HWCAP2 26 /* AT_HWCAP2 */ | |
192 | #define HWCAP_VEC_CRYPTO (1U << 25) | |
e0e53282 | 193 | #define HWCAP_ARCH_3_00 (1U << 23) |
2688d999 AP |
194 | |
195 | # if defined(__GNUC__) && __GNUC__>=2 | |
196 | __attribute__ ((constructor)) | |
197 | # endif | |
b4b48a10 | 198 | void OPENSSL_cpuid_setup(void) |
0f113f3e MC |
199 | { |
200 | char *e; | |
201 | struct sigaction ill_oact, ill_act; | |
202 | sigset_t oset; | |
203 | static int trigger = 0; | |
204 | ||
205 | if (trigger) | |
206 | return; | |
207 | trigger = 1; | |
208 | ||
0f113f3e MC |
209 | if ((e = getenv("OPENSSL_ppccap"))) { |
210 | OPENSSL_ppccap_P = strtoul(e, NULL, 0); | |
211 | return; | |
212 | } | |
b4b48a10 | 213 | |
0f113f3e | 214 | OPENSSL_ppccap_P = 0; |
6415dd7b | 215 | |
fd054957 | 216 | #if defined(_AIX) |
2688d999 AP |
217 | OPENSSL_ppccap_P |= PPC_FPU; |
218 | ||
0f113f3e MC |
219 | if (sizeof(size_t) == 4) { |
220 | struct utsname uts; | |
fd054957 | 221 | # if defined(_SC_AIX_KERNEL_BITMODE) |
0f113f3e MC |
222 | if (sysconf(_SC_AIX_KERNEL_BITMODE) != 64) |
223 | return; | |
fd054957 | 224 | # endif |
0f113f3e MC |
225 | if (uname(&uts) != 0 || atoi(uts.version) < 6) |
226 | return; | |
227 | } | |
2688d999 AP |
228 | |
229 | # if defined(__power_set) | |
230 | /* | |
231 | * Value used in __power_set is a single-bit 1<<n one denoting | |
232 | * specific processor class. Incidentally 0xffffffff<<n can be | |
233 | * used to denote specific processor and its successors. | |
234 | */ | |
235 | if (sizeof(size_t) == 4) { | |
236 | /* In 32-bit case PPC_FPU64 is always fastest [if option] */ | |
237 | if (__power_set(0xffffffffU<<13)) /* POWER5 and later */ | |
238 | OPENSSL_ppccap_P |= PPC_FPU64; | |
239 | } else { | |
240 | /* In 64-bit case PPC_FPU64 is fastest only on POWER6 */ | |
241 | if (__power_set(0x1U<<14)) /* POWER6 */ | |
242 | OPENSSL_ppccap_P |= PPC_FPU64; | |
243 | } | |
244 | ||
245 | if (__power_set(0xffffffffU<<14)) /* POWER6 and later */ | |
246 | OPENSSL_ppccap_P |= PPC_ALTIVEC; | |
247 | ||
248 | if (__power_set(0xffffffffU<<16)) /* POWER8 and later */ | |
249 | OPENSSL_ppccap_P |= PPC_CRYPTO207; | |
250 | ||
e0e53282 AP |
251 | if (__power_set(0xffffffffU<<17)) /* POWER9 and later */ |
252 | OPENSSL_ppccap_P |= PPC_MADD300; | |
253 | ||
2688d999 AP |
254 | return; |
255 | # endif | |
256 | #endif | |
257 | ||
0bd93bbe AP |
258 | #if defined(__APPLE__) && defined(__MACH__) |
259 | OPENSSL_ppccap_P |= PPC_FPU; | |
260 | ||
261 | { | |
262 | int val; | |
263 | size_t len = sizeof(val); | |
264 | ||
265 | if (sysctlbyname("hw.optional.64bitops", &val, &len, NULL, 0) == 0) { | |
266 | if (val) | |
267 | OPENSSL_ppccap_P |= PPC_FPU64; | |
268 | } | |
269 | ||
270 | len = sizeof(val); | |
271 | if (sysctlbyname("hw.optional.altivec", &val, &len, NULL, 0) == 0) { | |
272 | if (val) | |
273 | OPENSSL_ppccap_P |= PPC_ALTIVEC; | |
274 | } | |
275 | ||
276 | return; | |
277 | } | |
278 | #endif | |
279 | ||
2688d999 AP |
280 | if (getauxval != NULL) { |
281 | unsigned long hwcap = getauxval(HWCAP); | |
282 | ||
283 | if (hwcap & HWCAP_FPU) { | |
dccd20d1 | 284 | OPENSSL_ppccap_P |= PPC_FPU; |
2688d999 AP |
285 | |
286 | if (sizeof(size_t) == 4) { | |
287 | /* In 32-bit case PPC_FPU64 is always fastest [if option] */ | |
288 | if (hwcap & HWCAP_PPC64) | |
289 | OPENSSL_ppccap_P |= PPC_FPU64; | |
290 | } else { | |
291 | /* In 64-bit case PPC_FPU64 is fastest only on POWER6 */ | |
292 | if (hwcap & HWCAP_POWER6_EXT) | |
293 | OPENSSL_ppccap_P |= PPC_FPU64; | |
294 | } | |
295 | } | |
296 | ||
297 | if (hwcap & HWCAP_ALTIVEC) { | |
298 | OPENSSL_ppccap_P |= PPC_ALTIVEC; | |
299 | ||
300 | if ((hwcap & HWCAP_VSX) && (getauxval(HWCAP2) & HWCAP_VEC_CRYPTO)) | |
301 | OPENSSL_ppccap_P |= PPC_CRYPTO207; | |
302 | } | |
303 | ||
e0e53282 AP |
304 | if (hwcap & HWCAP_ARCH_3_00) { |
305 | OPENSSL_ppccap_P |= PPC_MADD300; | |
306 | } | |
307 | ||
2688d999 AP |
308 | return; |
309 | } | |
310 | ||
311 | sigfillset(&all_masked); | |
312 | sigdelset(&all_masked, SIGILL); | |
313 | sigdelset(&all_masked, SIGTRAP); | |
314 | #ifdef SIGEMT | |
315 | sigdelset(&all_masked, SIGEMT); | |
fd054957 | 316 | #endif |
2688d999 AP |
317 | sigdelset(&all_masked, SIGFPE); |
318 | sigdelset(&all_masked, SIGBUS); | |
319 | sigdelset(&all_masked, SIGSEGV); | |
fd054957 | 320 | |
0f113f3e MC |
321 | memset(&ill_act, 0, sizeof(ill_act)); |
322 | ill_act.sa_handler = ill_handler; | |
323 | ill_act.sa_mask = all_masked; | |
6415dd7b | 324 | |
0f113f3e MC |
325 | sigprocmask(SIG_SETMASK, &ill_act.sa_mask, &oset); |
326 | sigaction(SIGILL, &ill_act, &ill_oact); | |
6415dd7b | 327 | |
2688d999 AP |
328 | if (sigsetjmp(ill_jmp,1) == 0) { |
329 | OPENSSL_fpu_probe(); | |
330 | OPENSSL_ppccap_P |= PPC_FPU; | |
331 | ||
332 | if (sizeof(size_t) == 4) { | |
78c3e205 | 333 | #ifdef __linux |
2688d999 AP |
334 | struct utsname uts; |
335 | if (uname(&uts) == 0 && strcmp(uts.machine, "ppc64") == 0) | |
78c3e205 | 336 | #endif |
2688d999 AP |
337 | if (sigsetjmp(ill_jmp, 1) == 0) { |
338 | OPENSSL_ppc64_probe(); | |
339 | OPENSSL_ppccap_P |= PPC_FPU64; | |
340 | } | |
341 | } else { | |
342 | /* | |
343 | * Wanted code detecting POWER6 CPU and setting PPC_FPU64 | |
344 | */ | |
345 | } | |
0f113f3e MC |
346 | } |
347 | ||
348 | if (sigsetjmp(ill_jmp, 1) == 0) { | |
349 | OPENSSL_altivec_probe(); | |
350 | OPENSSL_ppccap_P |= PPC_ALTIVEC; | |
351 | if (sigsetjmp(ill_jmp, 1) == 0) { | |
352 | OPENSSL_crypto207_probe(); | |
353 | OPENSSL_ppccap_P |= PPC_CRYPTO207; | |
354 | } | |
355 | } | |
356 | ||
e0e53282 AP |
357 | if (sigsetjmp(ill_jmp, 1) == 0) { |
358 | OPENSSL_madd300_probe(); | |
359 | OPENSSL_ppccap_P |= PPC_MADD300; | |
360 | } | |
361 | ||
0f113f3e MC |
362 | sigaction(SIGILL, &ill_oact, NULL); |
363 | sigprocmask(SIG_SETMASK, &oset, NULL); | |
364 | } |