]>
Commit | Line | Data |
---|---|---|
b1322259 RS |
1 | /* |
2 | * Copyright 2005-2016 The OpenSSL Project Authors. All Rights Reserved. | |
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 | ||
c06b0f3d AP |
10 | #include <stdio.h> |
11 | #include <stdlib.h> | |
12 | #include <string.h> | |
c32fcca6 AP |
13 | #include <setjmp.h> |
14 | #include <signal.h> | |
c06b0f3d | 15 | #include <sys/time.h> |
68c06bf6 | 16 | #include <unistd.h> |
c06b0f3d AP |
17 | #include <openssl/bn.h> |
18 | ||
1fda639a | 19 | #include "sparc_arch.h" |
c32fcca6 | 20 | |
1fda639a | 21 | #if defined(__GNUC__) && defined(__linux) |
0f113f3e | 22 | __attribute__ ((visibility("hidden"))) |
1fda639a | 23 | #endif |
0f113f3e MC |
24 | unsigned int OPENSSL_sparcv9cap_P[2] = { SPARCV9_TICK_PRIVILEGED, 0 }; |
25 | ||
26 | int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, | |
27 | const BN_ULONG *np, const BN_ULONG *n0, int num) | |
28 | { | |
29 | int bn_mul_mont_vis3(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, | |
30 | const BN_ULONG *np, const BN_ULONG *n0, int num); | |
31 | int bn_mul_mont_fpu(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, | |
32 | const BN_ULONG *np, const BN_ULONG *n0, int num); | |
33 | int bn_mul_mont_int(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, | |
34 | const BN_ULONG *np, const BN_ULONG *n0, int num); | |
35 | ||
36 | if (!(num & 1) && num >= 6) { | |
37 | if ((num & 15) == 0 && num <= 64 && | |
38 | (OPENSSL_sparcv9cap_P[1] & (CFR_MONTMUL | CFR_MONTSQR)) == | |
39 | (CFR_MONTMUL | CFR_MONTSQR)) { | |
40 | typedef int (*bn_mul_mont_f) (BN_ULONG *rp, const BN_ULONG *ap, | |
41 | const BN_ULONG *bp, | |
42 | const BN_ULONG *np, | |
43 | const BN_ULONG *n0); | |
44 | int bn_mul_mont_t4_8(BN_ULONG *rp, const BN_ULONG *ap, | |
45 | const BN_ULONG *bp, const BN_ULONG *np, | |
46 | const BN_ULONG *n0); | |
47 | int bn_mul_mont_t4_16(BN_ULONG *rp, const BN_ULONG *ap, | |
48 | const BN_ULONG *bp, const BN_ULONG *np, | |
49 | const BN_ULONG *n0); | |
50 | int bn_mul_mont_t4_24(BN_ULONG *rp, const BN_ULONG *ap, | |
51 | const BN_ULONG *bp, const BN_ULONG *np, | |
52 | const BN_ULONG *n0); | |
53 | int bn_mul_mont_t4_32(BN_ULONG *rp, const BN_ULONG *ap, | |
54 | const BN_ULONG *bp, const BN_ULONG *np, | |
55 | const BN_ULONG *n0); | |
56 | static const bn_mul_mont_f funcs[4] = { | |
57 | bn_mul_mont_t4_8, bn_mul_mont_t4_16, | |
58 | bn_mul_mont_t4_24, bn_mul_mont_t4_32 | |
59 | }; | |
60 | bn_mul_mont_f worker = funcs[num / 16 - 1]; | |
61 | ||
62 | if ((*worker) (rp, ap, bp, np, n0)) | |
63 | return 1; | |
64 | /* retry once and fall back */ | |
65 | if ((*worker) (rp, ap, bp, np, n0)) | |
66 | return 1; | |
67 | return bn_mul_mont_vis3(rp, ap, bp, np, n0, num); | |
68 | } | |
69 | if ((OPENSSL_sparcv9cap_P[0] & SPARCV9_VIS3)) | |
70 | return bn_mul_mont_vis3(rp, ap, bp, np, n0, num); | |
71 | else if (num >= 8 && | |
33ea23dc AP |
72 | /* |
73 | * bn_mul_mont_fpu doesn't use FMADD, we just use the | |
74 | * flag to detect when FPU path is preferable in cases | |
75 | * when current heuristics is unreliable. [it works | |
76 | * out because FMADD-capable processors where FPU | |
77 | * code path is undesirable are also VIS3-capable and | |
78 | * VIS3 code path takes precedence.] | |
79 | */ | |
80 | ( (OPENSSL_sparcv9cap_P[0] & SPARCV9_FMADD) || | |
81 | (OPENSSL_sparcv9cap_P[0] & | |
82 | (SPARCV9_PREFER_FPU | SPARCV9_VIS1)) == | |
83 | (SPARCV9_PREFER_FPU | SPARCV9_VIS1) )) | |
0f113f3e MC |
84 | return bn_mul_mont_fpu(rp, ap, bp, np, n0, num); |
85 | } | |
86 | return bn_mul_mont_int(rp, ap, bp, np, n0, num); | |
87 | } | |
88 | ||
89 | unsigned long _sparcv9_rdtick(void); | |
90 | void _sparcv9_vis1_probe(void); | |
91 | unsigned long _sparcv9_vis1_instrument(void); | |
92 | void _sparcv9_vis2_probe(void); | |
93 | void _sparcv9_fmadd_probe(void); | |
94 | unsigned long _sparcv9_rdcfr(void); | |
95 | void _sparcv9_vis3_probe(void); | |
d40a13af | 96 | void _sparcv9_fjaesx_probe(void); |
0f113f3e MC |
97 | unsigned long _sparcv9_random(void); |
98 | size_t _sparcv9_vis1_instrument_bus(unsigned int *, size_t); | |
99 | size_t _sparcv9_vis1_instrument_bus2(unsigned int *, size_t, size_t); | |
c32fcca6 | 100 | |
c06b0f3d | 101 | unsigned long OPENSSL_rdtsc(void) |
0f113f3e MC |
102 | { |
103 | if (OPENSSL_sparcv9cap_P[0] & SPARCV9_TICK_PRIVILEGED) | |
c06b0f3d | 104 | #if defined(__sun) && defined(__SVR4) |
0f113f3e | 105 | return gethrtime(); |
c06b0f3d | 106 | #else |
0f113f3e | 107 | return 0; |
c06b0f3d | 108 | #endif |
0f113f3e MC |
109 | else |
110 | return _sparcv9_rdtick(); | |
111 | } | |
112 | ||
113 | size_t OPENSSL_instrument_bus(unsigned int *out, size_t cnt) | |
114 | { | |
115 | if ((OPENSSL_sparcv9cap_P[0] & (SPARCV9_TICK_PRIVILEGED | SPARCV9_BLK)) == | |
116 | SPARCV9_BLK) | |
117 | return _sparcv9_vis1_instrument_bus(out, cnt); | |
118 | else | |
119 | return 0; | |
120 | } | |
121 | ||
122 | size_t OPENSSL_instrument_bus2(unsigned int *out, size_t cnt, size_t max) | |
123 | { | |
124 | if ((OPENSSL_sparcv9cap_P[0] & (SPARCV9_TICK_PRIVILEGED | SPARCV9_BLK)) == | |
125 | SPARCV9_BLK) | |
126 | return _sparcv9_vis1_instrument_bus2(out, cnt, max); | |
127 | else | |
128 | return 0; | |
129 | } | |
5fabb88a | 130 | |
c32fcca6 | 131 | static sigjmp_buf common_jmp; |
0f113f3e MC |
132 | static void common_handler(int sig) |
133 | { | |
134 | siglongjmp(common_jmp, sig); | |
135 | } | |
c32fcca6 | 136 | |
2238e0e4 AP |
137 | #if defined(__sun) && defined(__SVR4) |
138 | # if defined(__GNUC__) && __GNUC__>=2 | |
139 | extern unsigned int getisax(unsigned int vec[], unsigned int sz) __attribute__ ((weak)); | |
140 | # elif defined(__SUNPRO_C) | |
141 | #pragma weak getisax | |
142 | extern unsigned int getisax(unsigned int vec[], unsigned int sz); | |
143 | # else | |
144 | static unsigned int (*getisax) (unsigned int vec[], unsigned int sz) = NULL; | |
145 | # endif | |
146 | #endif | |
147 | ||
5d7324e4 | 148 | void OPENSSL_cpuid_setup(void) |
0f113f3e MC |
149 | { |
150 | char *e; | |
151 | struct sigaction common_act, ill_oact, bus_oact; | |
152 | sigset_t all_masked, oset; | |
153 | static int trigger = 0; | |
154 | ||
155 | if (trigger) | |
156 | return; | |
157 | trigger = 1; | |
158 | ||
159 | if ((e = getenv("OPENSSL_sparcv9cap"))) { | |
160 | OPENSSL_sparcv9cap_P[0] = strtoul(e, NULL, 0); | |
161 | if ((e = strchr(e, ':'))) | |
162 | OPENSSL_sparcv9cap_P[1] = strtoul(e + 1, NULL, 0); | |
163 | return; | |
164 | } | |
165 | ||
2238e0e4 AP |
166 | #if defined(__sun) && defined(__SVR4) |
167 | if (getisax != NULL) { | |
a5a95f8d | 168 | unsigned int vec[2] = { 0, 0 }; |
2238e0e4 | 169 | |
a5a95f8d | 170 | if (getisax (vec,2)) { |
299ccadc AP |
171 | if (vec[0]&0x00020) OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS1; |
172 | if (vec[0]&0x00040) OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS2; | |
173 | if (vec[0]&0x00080) OPENSSL_sparcv9cap_P[0] |= SPARCV9_BLK; | |
174 | if (vec[0]&0x00100) OPENSSL_sparcv9cap_P[0] |= SPARCV9_FMADD; | |
175 | if (vec[0]&0x00400) OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS3; | |
176 | if (vec[0]&0x01000) OPENSSL_sparcv9cap_P[0] |= SPARCV9_FJHPCACE; | |
177 | if (vec[0]&0x02000) OPENSSL_sparcv9cap_P[0] |= SPARCV9_FJDESX; | |
178 | if (vec[0]&0x08000) OPENSSL_sparcv9cap_P[0] |= SPARCV9_IMA; | |
4400f6c6 | 179 | if (vec[0]&0x10000) OPENSSL_sparcv9cap_P[0] |= SPARCV9_FJAESX; |
299ccadc | 180 | if (vec[1]&0x00008) OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS4; |
2238e0e4 AP |
181 | |
182 | /* reconstruct %cfr copy */ | |
183 | OPENSSL_sparcv9cap_P[1] = (vec[0]>>17)&0x3ff; | |
184 | OPENSSL_sparcv9cap_P[1] |= (OPENSSL_sparcv9cap_P[1]&CFR_MONTMUL)<<1; | |
185 | if (vec[0]&0x20000000) OPENSSL_sparcv9cap_P[1] |= CFR_CRC32C; | |
299ccadc AP |
186 | if (vec[1]&0x00000020) OPENSSL_sparcv9cap_P[1] |= CFR_XMPMUL; |
187 | if (vec[1]&0x00000040) | |
188 | OPENSSL_sparcv9cap_P[1] |= CFR_XMONTMUL|CFR_XMONTSQR; | |
2238e0e4 AP |
189 | |
190 | /* Some heuristics */ | |
191 | /* all known VIS2-capable CPUs have unprivileged tick counter */ | |
192 | if (OPENSSL_sparcv9cap_P[0]&SPARCV9_VIS2) | |
193 | OPENSSL_sparcv9cap_P[0] &= ~SPARCV9_TICK_PRIVILEGED; | |
194 | ||
195 | OPENSSL_sparcv9cap_P[0] |= SPARCV9_PREFER_FPU; | |
196 | ||
197 | /* detect UltraSPARC-Tx, see sparccpud.S for details... */ | |
198 | if ((OPENSSL_sparcv9cap_P[0]&SPARCV9_VIS1) && | |
199 | _sparcv9_vis1_instrument() >= 12) | |
200 | OPENSSL_sparcv9cap_P[0] &= ~(SPARCV9_VIS1 | SPARCV9_PREFER_FPU); | |
201 | } | |
202 | ||
203 | if (sizeof(size_t) == 8) | |
204 | OPENSSL_sparcv9cap_P[0] |= SPARCV9_64BIT_STACK; | |
205 | ||
206 | return; | |
207 | } | |
208 | #endif | |
209 | ||
0f113f3e MC |
210 | /* Initial value, fits UltraSPARC-I&II... */ |
211 | OPENSSL_sparcv9cap_P[0] = SPARCV9_PREFER_FPU | SPARCV9_TICK_PRIVILEGED; | |
212 | ||
213 | sigfillset(&all_masked); | |
214 | sigdelset(&all_masked, SIGILL); | |
215 | sigdelset(&all_masked, SIGTRAP); | |
216 | # ifdef SIGEMT | |
217 | sigdelset(&all_masked, SIGEMT); | |
218 | # endif | |
219 | sigdelset(&all_masked, SIGFPE); | |
220 | sigdelset(&all_masked, SIGBUS); | |
221 | sigdelset(&all_masked, SIGSEGV); | |
222 | sigprocmask(SIG_SETMASK, &all_masked, &oset); | |
223 | ||
224 | memset(&common_act, 0, sizeof(common_act)); | |
225 | common_act.sa_handler = common_handler; | |
226 | common_act.sa_mask = all_masked; | |
227 | ||
228 | sigaction(SIGILL, &common_act, &ill_oact); | |
229 | sigaction(SIGBUS, &common_act, &bus_oact); /* T1 fails 16-bit ldda [on | |
230 | * Linux] */ | |
231 | ||
232 | if (sigsetjmp(common_jmp, 1) == 0) { | |
233 | _sparcv9_rdtick(); | |
234 | OPENSSL_sparcv9cap_P[0] &= ~SPARCV9_TICK_PRIVILEGED; | |
235 | } | |
236 | ||
237 | if (sigsetjmp(common_jmp, 1) == 0) { | |
238 | _sparcv9_vis1_probe(); | |
239 | OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS1 | SPARCV9_BLK; | |
240 | /* detect UltraSPARC-Tx, see sparccpud.S for details... */ | |
241 | if (_sparcv9_vis1_instrument() >= 12) | |
242 | OPENSSL_sparcv9cap_P[0] &= ~(SPARCV9_VIS1 | SPARCV9_PREFER_FPU); | |
243 | else { | |
244 | _sparcv9_vis2_probe(); | |
245 | OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS2; | |
246 | } | |
247 | } | |
248 | ||
249 | if (sigsetjmp(common_jmp, 1) == 0) { | |
250 | _sparcv9_fmadd_probe(); | |
251 | OPENSSL_sparcv9cap_P[0] |= SPARCV9_FMADD; | |
252 | } | |
253 | ||
254 | /* | |
255 | * VIS3 flag is tested independently from VIS1, unlike VIS2 that is, | |
256 | * because VIS3 defines even integer instructions. | |
257 | */ | |
258 | if (sigsetjmp(common_jmp, 1) == 0) { | |
259 | _sparcv9_vis3_probe(); | |
260 | OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS3; | |
261 | } | |
0f113f3e | 262 | |
4400f6c6 AP |
263 | if (sigsetjmp(common_jmp, 1) == 0) { |
264 | _sparcv9_fjaesx_probe(); | |
265 | OPENSSL_sparcv9cap_P[0] |= SPARCV9_FJAESX; | |
266 | } | |
267 | ||
0f113f3e MC |
268 | /* |
269 | * In wait for better solution _sparcv9_rdcfr is masked by | |
270 | * VIS3 flag, because it goes to uninterruptable endless | |
271 | * loop on UltraSPARC II running Solaris. Things might be | |
272 | * different on Linux... | |
273 | */ | |
274 | if ((OPENSSL_sparcv9cap_P[0] & SPARCV9_VIS3) && | |
275 | sigsetjmp(common_jmp, 1) == 0) { | |
276 | OPENSSL_sparcv9cap_P[1] = (unsigned int)_sparcv9_rdcfr(); | |
277 | } | |
278 | ||
279 | sigaction(SIGBUS, &bus_oact, NULL); | |
280 | sigaction(SIGILL, &ill_oact, NULL); | |
281 | ||
282 | sigprocmask(SIG_SETMASK, &oset, NULL); | |
283 | ||
284 | if (sizeof(size_t) == 8) | |
285 | OPENSSL_sparcv9cap_P[0] |= SPARCV9_64BIT_STACK; | |
286 | # ifdef __linux | |
287 | else { | |
288 | int ret = syscall(340); | |
289 | ||
290 | if (ret >= 0 && ret & 1) | |
291 | OPENSSL_sparcv9cap_P[0] |= SPARCV9_64BIT_STACK; | |
292 | } | |
293 | # endif | |
294 | } |