diff options
Diffstat (limited to 'src/lib/libcrypto/x86cpuid.pl')
-rw-r--r-- | src/lib/libcrypto/x86cpuid.pl | 312 |
1 files changed, 0 insertions, 312 deletions
diff --git a/src/lib/libcrypto/x86cpuid.pl b/src/lib/libcrypto/x86cpuid.pl deleted file mode 100644 index a7464af19b..0000000000 --- a/src/lib/libcrypto/x86cpuid.pl +++ /dev/null | |||
@@ -1,312 +0,0 @@ | |||
1 | #!/usr/bin/env perl | ||
2 | |||
3 | $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; | ||
4 | push(@INC, "${dir}perlasm", "perlasm"); | ||
5 | require "x86asm.pl"; | ||
6 | |||
7 | &asm_init($ARGV[0],"x86cpuid"); | ||
8 | |||
9 | for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); } | ||
10 | |||
11 | &function_begin("OPENSSL_ia32_cpuid"); | ||
12 | &xor ("edx","edx"); | ||
13 | &pushf (); | ||
14 | &pop ("eax"); | ||
15 | &mov ("ecx","eax"); | ||
16 | &xor ("eax",1<<21); | ||
17 | &push ("eax"); | ||
18 | &popf (); | ||
19 | &pushf (); | ||
20 | &pop ("eax"); | ||
21 | &xor ("ecx","eax"); | ||
22 | &bt ("ecx",21); | ||
23 | &jnc (&label("done")); | ||
24 | &xor ("eax","eax"); | ||
25 | &cpuid (); | ||
26 | &mov ("edi","eax"); # max value for standard query level | ||
27 | |||
28 | &xor ("eax","eax"); | ||
29 | &cmp ("ebx",0x756e6547); # "Genu" | ||
30 | &setne (&LB("eax")); | ||
31 | &mov ("ebp","eax"); | ||
32 | &cmp ("edx",0x49656e69); # "ineI" | ||
33 | &setne (&LB("eax")); | ||
34 | &or ("ebp","eax"); | ||
35 | &cmp ("ecx",0x6c65746e); # "ntel" | ||
36 | &setne (&LB("eax")); | ||
37 | &or ("ebp","eax"); # 0 indicates Intel CPU | ||
38 | &jz (&label("intel")); | ||
39 | |||
40 | &cmp ("ebx",0x68747541); # "Auth" | ||
41 | &setne (&LB("eax")); | ||
42 | &mov ("esi","eax"); | ||
43 | &cmp ("edx",0x69746E65); # "enti" | ||
44 | &setne (&LB("eax")); | ||
45 | &or ("esi","eax"); | ||
46 | &cmp ("ecx",0x444D4163); # "cAMD" | ||
47 | &setne (&LB("eax")); | ||
48 | &or ("esi","eax"); # 0 indicates AMD CPU | ||
49 | &jnz (&label("intel")); | ||
50 | |||
51 | # AMD specific | ||
52 | &mov ("eax",0x80000000); | ||
53 | &cpuid (); | ||
54 | &cmp ("eax",0x80000008); | ||
55 | &jb (&label("intel")); | ||
56 | |||
57 | &mov ("eax",0x80000008); | ||
58 | &cpuid (); | ||
59 | &movz ("esi",&LB("ecx")); # number of cores - 1 | ||
60 | &inc ("esi"); # number of cores | ||
61 | |||
62 | &mov ("eax",1); | ||
63 | &cpuid (); | ||
64 | &bt ("edx",28); | ||
65 | &jnc (&label("done")); | ||
66 | &shr ("ebx",16); | ||
67 | &and ("ebx",0xff); | ||
68 | &cmp ("ebx","esi"); | ||
69 | &ja (&label("done")); | ||
70 | &and ("edx",0xefffffff); # clear hyper-threading bit | ||
71 | &jmp (&label("done")); | ||
72 | |||
73 | &set_label("intel"); | ||
74 | &cmp ("edi",4); | ||
75 | &mov ("edi",-1); | ||
76 | &jb (&label("nocacheinfo")); | ||
77 | |||
78 | &mov ("eax",4); | ||
79 | &mov ("ecx",0); # query L1D | ||
80 | &cpuid (); | ||
81 | &mov ("edi","eax"); | ||
82 | &shr ("edi",14); | ||
83 | &and ("edi",0xfff); # number of cores -1 per L1D | ||
84 | |||
85 | &set_label("nocacheinfo"); | ||
86 | &mov ("eax",1); | ||
87 | &cpuid (); | ||
88 | &cmp ("ebp",0); | ||
89 | &jne (&label("notP4")); | ||
90 | &and (&HB("eax"),15); # familiy ID | ||
91 | &cmp (&HB("eax"),15); # P4? | ||
92 | &jne (&label("notP4")); | ||
93 | &or ("edx",1<<20); # use reserved bit to engage RC4_CHAR | ||
94 | &set_label("notP4"); | ||
95 | &bt ("edx",28); # test hyper-threading bit | ||
96 | &jnc (&label("done")); | ||
97 | &and ("edx",0xefffffff); | ||
98 | &cmp ("edi",0); | ||
99 | &je (&label("done")); | ||
100 | |||
101 | &or ("edx",0x10000000); | ||
102 | &shr ("ebx",16); | ||
103 | &cmp (&LB("ebx"),1); | ||
104 | &ja (&label("done")); | ||
105 | &and ("edx",0xefffffff); # clear hyper-threading bit if not | ||
106 | &set_label("done"); | ||
107 | &mov ("eax","edx"); | ||
108 | &mov ("edx","ecx"); | ||
109 | &function_end("OPENSSL_ia32_cpuid"); | ||
110 | |||
111 | &external_label("OPENSSL_ia32cap_P"); | ||
112 | |||
113 | &function_begin_B("OPENSSL_rdtsc","EXTRN\t_OPENSSL_ia32cap_P:DWORD"); | ||
114 | &xor ("eax","eax"); | ||
115 | &xor ("edx","edx"); | ||
116 | &picmeup("ecx","OPENSSL_ia32cap_P"); | ||
117 | &bt (&DWP(0,"ecx"),4); | ||
118 | &jnc (&label("notsc")); | ||
119 | &rdtsc (); | ||
120 | &set_label("notsc"); | ||
121 | &ret (); | ||
122 | &function_end_B("OPENSSL_rdtsc"); | ||
123 | |||
124 | # This works in Ring 0 only [read DJGPP+MS-DOS+privileged DPMI host], | ||
125 | # but it's safe to call it on any [supported] 32-bit platform... | ||
126 | # Just check for [non-]zero return value... | ||
127 | &function_begin_B("OPENSSL_instrument_halt","EXTRN\t_OPENSSL_ia32cap_P:DWORD"); | ||
128 | &picmeup("ecx","OPENSSL_ia32cap_P"); | ||
129 | &bt (&DWP(0,"ecx"),4); | ||
130 | &jnc (&label("nohalt")); # no TSC | ||
131 | |||
132 | &data_word(0x9058900e); # push %cs; pop %eax | ||
133 | &and ("eax",3); | ||
134 | &jnz (&label("nohalt")); # not enough privileges | ||
135 | |||
136 | &pushf (); | ||
137 | &pop ("eax") | ||
138 | &bt ("eax",9); | ||
139 | &jnc (&label("nohalt")); # interrupts are disabled | ||
140 | |||
141 | &rdtsc (); | ||
142 | &push ("edx"); | ||
143 | &push ("eax"); | ||
144 | &halt (); | ||
145 | &rdtsc (); | ||
146 | |||
147 | &sub ("eax",&DWP(0,"esp")); | ||
148 | &sbb ("edx",&DWP(4,"esp")); | ||
149 | &add ("esp",8); | ||
150 | &ret (); | ||
151 | |||
152 | &set_label("nohalt"); | ||
153 | &xor ("eax","eax"); | ||
154 | &xor ("edx","edx"); | ||
155 | &ret (); | ||
156 | &function_end_B("OPENSSL_instrument_halt"); | ||
157 | |||
158 | # Essentially there is only one use for this function. Under DJGPP: | ||
159 | # | ||
160 | # #include <go32.h> | ||
161 | # ... | ||
162 | # i=OPENSSL_far_spin(_dos_ds,0x46c); | ||
163 | # ... | ||
164 | # to obtain the number of spins till closest timer interrupt. | ||
165 | |||
166 | &function_begin_B("OPENSSL_far_spin"); | ||
167 | &pushf (); | ||
168 | &pop ("eax") | ||
169 | &bt ("eax",9); | ||
170 | &jnc (&label("nospin")); # interrupts are disabled | ||
171 | |||
172 | &mov ("eax",&DWP(4,"esp")); | ||
173 | &mov ("ecx",&DWP(8,"esp")); | ||
174 | &data_word (0x90d88e1e); # push %ds, mov %eax,%ds | ||
175 | &xor ("eax","eax"); | ||
176 | &mov ("edx",&DWP(0,"ecx")); | ||
177 | &jmp (&label("spin")); | ||
178 | |||
179 | &align (16); | ||
180 | &set_label("spin"); | ||
181 | &inc ("eax"); | ||
182 | &cmp ("edx",&DWP(0,"ecx")); | ||
183 | &je (&label("spin")); | ||
184 | |||
185 | &data_word (0x1f909090); # pop %ds | ||
186 | &ret (); | ||
187 | |||
188 | &set_label("nospin"); | ||
189 | &xor ("eax","eax"); | ||
190 | &xor ("edx","edx"); | ||
191 | &ret (); | ||
192 | &function_end_B("OPENSSL_far_spin"); | ||
193 | |||
194 | &function_begin_B("OPENSSL_wipe_cpu","EXTRN\t_OPENSSL_ia32cap_P:DWORD"); | ||
195 | &xor ("eax","eax"); | ||
196 | &xor ("edx","edx"); | ||
197 | &picmeup("ecx","OPENSSL_ia32cap_P"); | ||
198 | &mov ("ecx",&DWP(0,"ecx")); | ||
199 | &bt (&DWP(0,"ecx"),1); | ||
200 | &jnc (&label("no_x87")); | ||
201 | if ($sse2) { | ||
202 | &bt (&DWP(0,"ecx"),26); | ||
203 | &jnc (&label("no_sse2")); | ||
204 | &pxor ("xmm0","xmm0"); | ||
205 | &pxor ("xmm1","xmm1"); | ||
206 | &pxor ("xmm2","xmm2"); | ||
207 | &pxor ("xmm3","xmm3"); | ||
208 | &pxor ("xmm4","xmm4"); | ||
209 | &pxor ("xmm5","xmm5"); | ||
210 | &pxor ("xmm6","xmm6"); | ||
211 | &pxor ("xmm7","xmm7"); | ||
212 | &set_label("no_sse2"); | ||
213 | } | ||
214 | # just a bunch of fldz to zap the fp/mm bank followed by finit... | ||
215 | &data_word(0xeed9eed9,0xeed9eed9,0xeed9eed9,0xeed9eed9,0x90e3db9b); | ||
216 | &set_label("no_x87"); | ||
217 | &lea ("eax",&DWP(4,"esp")); | ||
218 | &ret (); | ||
219 | &function_end_B("OPENSSL_wipe_cpu"); | ||
220 | |||
221 | &function_begin_B("OPENSSL_atomic_add"); | ||
222 | &mov ("edx",&DWP(4,"esp")); # fetch the pointer, 1st arg | ||
223 | &mov ("ecx",&DWP(8,"esp")); # fetch the increment, 2nd arg | ||
224 | &push ("ebx"); | ||
225 | &nop (); | ||
226 | &mov ("eax",&DWP(0,"edx")); | ||
227 | &set_label("spin"); | ||
228 | &lea ("ebx",&DWP(0,"eax","ecx")); | ||
229 | &nop (); | ||
230 | &data_word(0x1ab10ff0); # lock; cmpxchg %ebx,(%edx) # %eax is envolved and is always reloaded | ||
231 | &jne (&label("spin")); | ||
232 | &mov ("eax","ebx"); # OpenSSL expects the new value | ||
233 | &pop ("ebx"); | ||
234 | &ret (); | ||
235 | &function_end_B("OPENSSL_atomic_add"); | ||
236 | |||
237 | # This function can become handy under Win32 in situations when | ||
238 | # we don't know which calling convention, __stdcall or __cdecl(*), | ||
239 | # indirect callee is using. In C it can be deployed as | ||
240 | # | ||
241 | #ifdef OPENSSL_CPUID_OBJ | ||
242 | # type OPENSSL_indirect_call(void *f,...); | ||
243 | # ... | ||
244 | # OPENSSL_indirect_call(func,[up to $max arguments]); | ||
245 | #endif | ||
246 | # | ||
247 | # (*) it's designed to work even for __fastcall if number of | ||
248 | # arguments is 1 or 2! | ||
249 | &function_begin_B("OPENSSL_indirect_call"); | ||
250 | { | ||
251 | my $i,$max=7; # $max has to be chosen as 4*n-1 | ||
252 | # in order to preserve eventual | ||
253 | # stack alignment | ||
254 | &push ("ebp"); | ||
255 | &mov ("ebp","esp"); | ||
256 | &sub ("esp",$max*4); | ||
257 | &mov ("ecx",&DWP(12,"ebp")); | ||
258 | &mov (&DWP(0,"esp"),"ecx"); | ||
259 | &mov ("edx",&DWP(16,"ebp")); | ||
260 | &mov (&DWP(4,"esp"),"edx"); | ||
261 | for($i=2;$i<$max;$i++) | ||
262 | { | ||
263 | # Some copies will be redundant/bogus... | ||
264 | &mov ("eax",&DWP(12+$i*4,"ebp")); | ||
265 | &mov (&DWP(0+$i*4,"esp"),"eax"); | ||
266 | } | ||
267 | &call_ptr (&DWP(8,"ebp"));# make the call... | ||
268 | &mov ("esp","ebp"); # ... and just restore the stack pointer | ||
269 | # without paying attention to what we called, | ||
270 | # (__cdecl *func) or (__stdcall *one). | ||
271 | &pop ("ebp"); | ||
272 | &ret (); | ||
273 | } | ||
274 | &function_end_B("OPENSSL_indirect_call"); | ||
275 | |||
276 | &function_begin_B("OPENSSL_cleanse"); | ||
277 | &mov ("edx",&wparam(0)); | ||
278 | &mov ("ecx",&wparam(1)); | ||
279 | &xor ("eax","eax"); | ||
280 | &cmp ("ecx",7); | ||
281 | &jae (&label("lot")); | ||
282 | &cmp ("ecx",0); | ||
283 | &je (&label("ret")); | ||
284 | &set_label("little"); | ||
285 | &mov (&BP(0,"edx"),"al"); | ||
286 | &sub ("ecx",1); | ||
287 | &lea ("edx",&DWP(1,"edx")); | ||
288 | &jnz (&label("little")); | ||
289 | &set_label("ret"); | ||
290 | &ret (); | ||
291 | |||
292 | &set_label("lot",16); | ||
293 | &test ("edx",3); | ||
294 | &jz (&label("aligned")); | ||
295 | &mov (&BP(0,"edx"),"al"); | ||
296 | &lea ("ecx",&DWP(-1,"ecx")); | ||
297 | &lea ("edx",&DWP(1,"edx")); | ||
298 | &jmp (&label("lot")); | ||
299 | &set_label("aligned"); | ||
300 | &mov (&DWP(0,"edx"),"eax"); | ||
301 | &lea ("ecx",&DWP(-4,"ecx")); | ||
302 | &test ("ecx",-4); | ||
303 | &lea ("edx",&DWP(4,"edx")); | ||
304 | &jnz (&label("aligned")); | ||
305 | &cmp ("ecx",0); | ||
306 | &jne (&label("little")); | ||
307 | &ret (); | ||
308 | &function_end_B("OPENSSL_cleanse"); | ||
309 | |||
310 | &initseg("OPENSSL_cpuid_setup"); | ||
311 | |||
312 | &asm_finish(); | ||