x86cpuid.pl 12.8 KB
Newer Older
1 2 3 4 5 6 7
#! /usr/bin/env perl
# Copyright 2004-2016 The OpenSSL Project Authors. All Rights Reserved.
#
# Licensed under the OpenSSL license (the "License").  You may not use
# this file except in compliance with the License.  You can obtain a copy
# in the file LICENSE in the source distribution or at
# https://www.openssl.org/source/license.html
8

9 10
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
push(@INC, "${dir}perlasm", "perlasm");
11 12
require "x86asm.pl";

13 14 15 16
$output = pop;
open OUT,">$output";
*STDOUT=*OUT;

17 18
&asm_init($ARGV[0],"x86cpuid");

19 20
for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }

21 22 23 24 25 26 27 28 29 30 31
&function_begin("OPENSSL_ia32_cpuid");
	&xor	("edx","edx");
	&pushf	();
	&pop	("eax");
	&mov	("ecx","eax");
	&xor	("eax",1<<21);
	&push	("eax");
	&popf	();
	&pushf	();
	&pop	("eax");
	&xor	("ecx","eax");
32
	&xor	("eax","eax");
33 34
	&bt	("ecx",21);
	&jnc	(&label("nocpuid"));
35 36
	&mov	("esi",&wparam(0));
	&mov	(&DWP(8,"esi"),"eax");	# clear 3rd word
37
	&cpuid	();
38 39
	&mov	("edi","eax");		# max value for standard query level

40 41 42 43 44 45 46 47 48
	&xor	("eax","eax");
	&cmp	("ebx",0x756e6547);	# "Genu"
	&setne	(&LB("eax"));
	&mov	("ebp","eax");
	&cmp	("edx",0x49656e69);	# "ineI"
	&setne	(&LB("eax"));
	&or	("ebp","eax");
	&cmp	("ecx",0x6c65746e);	# "ntel"
	&setne	(&LB("eax"));
49 50 51 52 53 54 55 56 57 58 59 60 61 62
	&or	("ebp","eax");		# 0 indicates Intel CPU
	&jz	(&label("intel"));

	&cmp	("ebx",0x68747541);	# "Auth"
	&setne	(&LB("eax"));
	&mov	("esi","eax");
	&cmp	("edx",0x69746E65);	# "enti"
	&setne	(&LB("eax"));
	&or	("esi","eax");
	&cmp	("ecx",0x444D4163);	# "cAMD"
	&setne	(&LB("eax"));
	&or	("esi","eax");		# 0 indicates AMD CPU
	&jnz	(&label("intel"));

63
	# AMD specific
64 65
	&mov	("eax",0x80000000);
	&cpuid	();
66 67 68 69 70 71 72 73
	&cmp	("eax",0x80000001);
	&jb	(&label("intel"));
	&mov	("esi","eax");
	&mov	("eax",0x80000001);
	&cpuid	();
	&or	("ebp","ecx");
	&and	("ebp",1<<11|1);	# isolate XOP bit
	&cmp	("esi",0x80000008);
74 75 76 77 78 79 80
	&jb	(&label("intel"));

	&mov	("eax",0x80000008);
	&cpuid	();
	&movz	("esi",&LB("ecx"));	# number of cores - 1
	&inc	("esi");		# number of cores

81
	&mov	("eax",1);
82
	&xor	("ecx","ecx");
83 84
	&cpuid	();
	&bt	("edx",28);
85
	&jnc	(&label("generic"));
86 87 88
	&shr	("ebx",16);
	&and	("ebx",0xff);
	&cmp	("ebx","esi");
89
	&ja	(&label("generic"));
90
	&and	("edx",0xefffffff);	# clear hyper-threading bit
91
	&jmp	(&label("generic"));
92
	
93
&set_label("intel");
94 95 96 97 98 99 100 101 102 103
	&cmp	("edi",7);
	&jb	(&label("cacheinfo"));

	&mov	("esi",&wparam(0));
	&mov	("eax",7);
	&xor	("ecx","ecx");
	&cpuid	();
	&mov	(&DWP(8,"esi"),"ebx");

&set_label("cacheinfo");
104 105 106 107 108 109 110 111 112 113 114 115
	&cmp	("edi",4);
	&mov	("edi",-1);
	&jb	(&label("nocacheinfo"));

	&mov	("eax",4);
	&mov	("ecx",0);		# query L1D
	&cpuid	();
	&mov	("edi","eax");
	&shr	("edi",14);
	&and	("edi",0xfff);		# number of cores -1 per L1D

&set_label("nocacheinfo");
116
	&mov	("eax",1);
117
	&xor	("ecx","ecx");
118
	&cpuid	();
119
	&and	("edx",0xbfefffff);	# force reserved bits #20, #30 to 0
120
	&cmp	("ebp",0);
121 122
	&jne	(&label("notintel"));
	&or	("edx",1<<30);		# set reserved bit#30 on Intel CPUs
A
Andy Polyakov 已提交
123 124
	&and	(&HB("eax"),15);	# familiy ID
	&cmp	(&HB("eax"),15);	# P4?
125 126 127
	&jne	(&label("notintel"));
	&or	("edx",1<<20);		# set reserved bit#20 to engage RC4_CHAR
&set_label("notintel");
A
Andy Polyakov 已提交
128
	&bt	("edx",28);		# test hyper-threading bit
129
	&jnc	(&label("generic"));
130 131
	&and	("edx",0xefffffff);
	&cmp	("edi",0);
132
	&je	(&label("generic"));
133 134

	&or	("edx",0x10000000);
135
	&shr	("ebx",16);
136
	&cmp	(&LB("ebx"),1);
137
	&ja	(&label("generic"));
A
Andy Polyakov 已提交
138
	&and	("edx",0xefffffff);	# clear hyper-threading bit if not
139 140 141

&set_label("generic");
	&and	("ebp",1<<11);		# isolate AMD XOP flag
142
	&and	("ecx",0xfffff7ff);	# force 11th bit to 0
143 144 145 146
	&mov	("esi","edx");
	&or	("ebp","ecx");		# merge AMD XOP flag

	&bt	("ecx",27);		# check OSXSAVE bit
147
	&jnc	(&label("clear_avx"));
148 149 150 151 152 153 154 155
	&xor	("ecx","ecx");
	&data_byte(0x0f,0x01,0xd0);	# xgetbv
	&and	("eax",6);
	&cmp	("eax",6);
	&je	(&label("done"));
	&cmp	("eax",2);
	&je	(&label("clear_avx"));
&set_label("clear_xmm");
156 157
	&and	("ebp",0xfdfffffd);	# clear AESNI and PCLMULQDQ bits
	&and	("esi",0xfeffffff);	# clear FXSR
158
&set_label("clear_avx");
159
	&and	("ebp",0xefffe7ff);	# clear AVX, FMA and AMD XOP bits
160 161
	&mov	("edi",&wparam(0));
	&and	(&DWP(8,"edi"),0xffffffdf);	# clear AVX2
A
Andy Polyakov 已提交
162
&set_label("done");
163 164
	&mov	("eax","esi");
	&mov	("edx","ebp");
165
&set_label("nocpuid");
166 167
&function_end("OPENSSL_ia32_cpuid");

168
&external_label("OPENSSL_ia32cap_P");
169

170
&function_begin_B("OPENSSL_rdtsc","EXTRN\t_OPENSSL_ia32cap_P:DWORD");
171 172
	&xor	("eax","eax");
	&xor	("edx","edx");
173
	&picmeup("ecx","OPENSSL_ia32cap_P");
174 175 176 177 178 179 180
	&bt	(&DWP(0,"ecx"),4);
	&jnc	(&label("notsc"));
	&rdtsc	();
&set_label("notsc");
	&ret	();
&function_end_B("OPENSSL_rdtsc");

181
# This works in Ring 0 only [read DJGPP+MS-DOS+privileged DPMI host],
182 183 184 185 186 187 188 189 190 191 192 193
# but it's safe to call it on any [supported] 32-bit platform...
# Just check for [non-]zero return value...
&function_begin_B("OPENSSL_instrument_halt","EXTRN\t_OPENSSL_ia32cap_P:DWORD");
	&picmeup("ecx","OPENSSL_ia32cap_P");
	&bt	(&DWP(0,"ecx"),4);
	&jnc	(&label("nohalt"));	# no TSC

	&data_word(0x9058900e);		# push %cs; pop %eax
	&and	("eax",3);
	&jnz	(&label("nohalt"));	# not enough privileges

	&pushf	();
194
	&pop	("eax");
195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214
	&bt	("eax",9);
	&jnc	(&label("nohalt"));	# interrupts are disabled

	&rdtsc	();
	&push	("edx");
	&push	("eax");
	&halt	();
	&rdtsc	();

	&sub	("eax",&DWP(0,"esp"));
	&sbb	("edx",&DWP(4,"esp"));
	&add	("esp",8);
	&ret	();

&set_label("nohalt");
	&xor	("eax","eax");
	&xor	("edx","edx");
	&ret	();
&function_end_B("OPENSSL_instrument_halt");

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
# Essentially there is only one use for this function. Under DJGPP:
#
#	#include <go32.h>
#	...
#	i=OPENSSL_far_spin(_dos_ds,0x46c);
#	...
# to obtain the number of spins till closest timer interrupt.

&function_begin_B("OPENSSL_far_spin");
	&pushf	();
	&pop	("eax");
	&bt	("eax",9);
	&jnc	(&label("nospin"));	# interrupts are disabled

	&mov	("eax",&DWP(4,"esp"));
	&mov	("ecx",&DWP(8,"esp"));
	&data_word (0x90d88e1e);	# push %ds, mov %eax,%ds
	&xor	("eax","eax");
	&mov	("edx",&DWP(0,"ecx"));
	&jmp	(&label("spin"));

	&align	(16);
&set_label("spin");
	&inc	("eax");
	&cmp	("edx",&DWP(0,"ecx"));
	&je	(&label("spin"));

	&data_word (0x1f909090);	# pop	%ds
	&ret	();

&set_label("nospin");
	&xor	("eax","eax");
	&xor	("edx","edx");
	&ret	();
&function_end_B("OPENSSL_far_spin");
A
Andy Polyakov 已提交
250 251 252 253 254 255 256 257

&function_begin_B("OPENSSL_wipe_cpu","EXTRN\t_OPENSSL_ia32cap_P:DWORD");
	&xor	("eax","eax");
	&xor	("edx","edx");
	&picmeup("ecx","OPENSSL_ia32cap_P");
	&mov	("ecx",&DWP(0,"ecx"));
	&bt	(&DWP(0,"ecx"),1);
	&jnc	(&label("no_x87"));
258
	if ($sse2) {
259 260 261
		&and	("ecx",1<<26|1<<24);	# check SSE2 and FXSR bits
		&cmp	("ecx",1<<26|1<<24);
		&jne	(&label("no_sse2"));
262 263 264 265 266 267 268 269 270 271
		&pxor	("xmm0","xmm0");
		&pxor	("xmm1","xmm1");
		&pxor	("xmm2","xmm2");
		&pxor	("xmm3","xmm3");
		&pxor	("xmm4","xmm4");
		&pxor	("xmm5","xmm5");
		&pxor	("xmm6","xmm6");
		&pxor	("xmm7","xmm7");
	&set_label("no_sse2");
	}
272 273
	# just a bunch of fldz to zap the fp/mm bank followed by finit...
	&data_word(0xeed9eed9,0xeed9eed9,0xeed9eed9,0xeed9eed9,0x90e3db9b);
A
Andy Polyakov 已提交
274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294
&set_label("no_x87");
	&lea	("eax",&DWP(4,"esp"));
	&ret	();
&function_end_B("OPENSSL_wipe_cpu");

&function_begin_B("OPENSSL_atomic_add");
	&mov	("edx",&DWP(4,"esp"));	# fetch the pointer, 1st arg
	&mov	("ecx",&DWP(8,"esp"));	# fetch the increment, 2nd arg
	&push	("ebx");
	&nop	();
	&mov	("eax",&DWP(0,"edx"));
&set_label("spin");
	&lea	("ebx",&DWP(0,"eax","ecx"));
	&nop	();
	&data_word(0x1ab10ff0);	# lock;	cmpxchg	%ebx,(%edx)	# %eax is envolved and is always reloaded
	&jne	(&label("spin"));
	&mov	("eax","ebx");	# OpenSSL expects the new value
	&pop	("ebx");
	&ret	();
&function_end_B("OPENSSL_atomic_add");

A
Andy Polyakov 已提交
295 296 297 298 299 300 301 302 303 304 305 306 307 308
# This function can become handy under Win32 in situations when
# we don't know which calling convention, __stdcall or __cdecl(*),
# indirect callee is using. In C it can be deployed as
#
#ifdef OPENSSL_CPUID_OBJ
#	type OPENSSL_indirect_call(void *f,...);
#	...
#	OPENSSL_indirect_call(func,[up to $max arguments]);
#endif
#
# (*)	it's designed to work even for __fastcall if number of
#	arguments is 1 or 2!
&function_begin_B("OPENSSL_indirect_call");
	{
309
	my ($max,$i)=(7,);	# $max has to be chosen as 4*n-1
A
Andy Polyakov 已提交
310 311 312 313 314 315 316 317 318 319 320 321 322 323 324
				# in order to preserve eventual
				# stack alignment
	&push	("ebp");
	&mov	("ebp","esp");
	&sub	("esp",$max*4);
	&mov	("ecx",&DWP(12,"ebp"));
	&mov	(&DWP(0,"esp"),"ecx");
	&mov	("edx",&DWP(16,"ebp"));
	&mov	(&DWP(4,"esp"),"edx");
	for($i=2;$i<$max;$i++)
		{
		# Some copies will be redundant/bogus...
		&mov	("eax",&DWP(12+$i*4,"ebp"));
		&mov	(&DWP(0+$i*4,"esp"),"eax");
		}
325
	&call_ptr	(&DWP(8,"ebp"));# make the call...
A
Andy Polyakov 已提交
326 327 328 329 330 331 332 333
	&mov	("esp","ebp");	# ... and just restore the stack pointer
				# without paying attention to what we called,
				# (__cdecl *func) or (__stdcall *one).
	&pop	("ebp");
	&ret	();
	}
&function_end_B("OPENSSL_indirect_call");

334 335 336 337 338 339
&function_begin_B("OPENSSL_cleanse");
	&mov	("edx",&wparam(0));
	&mov	("ecx",&wparam(1));
	&xor	("eax","eax");
	&cmp	("ecx",7);
	&jae	(&label("lot"));
340 341
	&cmp	("ecx",0);
	&je	(&label("ret"));
342 343 344 345 346
&set_label("little");
	&mov	(&BP(0,"edx"),"al");
	&sub	("ecx",1);
	&lea	("edx",&DWP(1,"edx"));
	&jnz	(&label("little"));
347
&set_label("ret");
348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367
	&ret	();

&set_label("lot",16);
	&test	("edx",3);
	&jz	(&label("aligned"));
	&mov	(&BP(0,"edx"),"al");
	&lea	("ecx",&DWP(-1,"ecx"));
	&lea	("edx",&DWP(1,"edx"));
	&jmp	(&label("lot"));
&set_label("aligned");
	&mov	(&DWP(0,"edx"),"eax");
	&lea	("ecx",&DWP(-4,"ecx"));
	&test	("ecx",-4);
	&lea	("edx",&DWP(4,"edx"));
	&jnz	(&label("aligned"));
	&cmp	("ecx",0);
	&jne	(&label("little"));
	&ret	();
&function_end_B("OPENSSL_cleanse");

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
&function_begin_B("CRYPTO_memcmp");
	&push	("esi");
	&push	("edi");
	&mov	("esi",&wparam(0));
	&mov	("edi",&wparam(1));
	&mov	("ecx",&wparam(2));
	&xor	("eax","eax");
	&xor	("edx","edx");
	&cmp	("ecx",0);
	&je	(&label("no_data"));
&set_label("loop");
	&mov	("dl",&BP(0,"esi"));
	&lea	("esi",&DWP(1,"esi"));
	&xor	("dl",&BP(0,"edi"));
	&lea	("edi",&DWP(1,"edi"));
	&or	("al","dl");
	&dec	("ecx");
	&jnz	(&label("loop"));
	&neg	("eax");
	&shr	("eax",31);
&set_label("no_data");
	&pop	("edi");
	&pop	("esi");
	&ret	();
&function_end_B("CRYPTO_memcmp");
393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416
{
my $lasttick = "esi";
my $lastdiff = "ebx";
my $out = "edi";
my $cnt = "ecx";
my $max = "ebp";

&function_begin("OPENSSL_instrument_bus");
    &mov	("eax",0);
    if ($sse2) {
	&picmeup("edx","OPENSSL_ia32cap_P");
	&bt	(&DWP(0,"edx"),4);
	&jnc	(&label("nogo"));	# no TSC
	&bt	(&DWP(0,"edx"),19);
	&jnc	(&label("nogo"));	# no CLFLUSH

	&mov	($out,&wparam(0));	# load arguments
	&mov	($cnt,&wparam(1));

	# collect 1st tick
	&rdtsc	();
	&mov	($lasttick,"eax");	# lasttick = tick
	&mov	($lastdiff,0);		# lastdiff = 0
	&clflush(&DWP(0,$out));
417
	&data_byte(0xf0);		# lock
418 419 420 421 422 423 424 425 426 427
	&add	(&DWP(0,$out),$lastdiff);
	&jmp	(&label("loop"));

&set_label("loop",16);
	&rdtsc	();
	&mov	("edx","eax");		# put aside tick (yes, I neglect edx)
	&sub	("eax",$lasttick);	# diff
	&mov	($lasttick,"edx");	# lasttick = tick
	&mov	($lastdiff,"eax");	# lastdiff = diff
	&clflush(&DWP(0,$out));
428
	&data_byte(0xf0);		# lock
429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456
	&add	(&DWP(0,$out),"eax");	# accumulate diff
	&lea	($out,&DWP(4,$out));	# ++$out
	&sub	($cnt,1);		# --$cnt
	&jnz	(&label("loop"));

	&mov	("eax",&wparam(1));
&set_label("nogo");
    }
&function_end("OPENSSL_instrument_bus");

&function_begin("OPENSSL_instrument_bus2");
    &mov	("eax",0);
    if ($sse2) {
	&picmeup("edx","OPENSSL_ia32cap_P");
	&bt	(&DWP(0,"edx"),4);
	&jnc	(&label("nogo"));	# no TSC
	&bt	(&DWP(0,"edx"),19);
	&jnc	(&label("nogo"));	# no CLFLUSH

	&mov	($out,&wparam(0));	# load arguments
	&mov	($cnt,&wparam(1));
	&mov	($max,&wparam(2));

	&rdtsc	();			# collect 1st tick
	&mov	($lasttick,"eax");	# lasttick = tick
	&mov	($lastdiff,0);		# lastdiff = 0

	&clflush(&DWP(0,$out));
457
	&data_byte(0xf0);		# lock
458 459 460 461 462 463 464 465 466 467 468
	&add	(&DWP(0,$out),$lastdiff);

	&rdtsc	();			# collect 1st diff
	&mov	("edx","eax");		# put aside tick (yes, I neglect edx)
	&sub	("eax",$lasttick);	# diff
	&mov	($lasttick,"edx");	# lasttick = tick
	&mov	($lastdiff,"eax");	# lastdiff = diff
	&jmp	(&label("loop2"));

&set_label("loop2",16);
	&clflush(&DWP(0,$out));
469
	&data_byte(0xf0);		# lock
470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494
	&add	(&DWP(0,$out),"eax");	# accumulate diff

	&sub	($max,1);
	&jz	(&label("done2"));

	&rdtsc	();
	&mov	("edx","eax");		# put aside tick (yes, I neglect edx)
	&sub	("eax",$lasttick);	# diff
	&mov	($lasttick,"edx");	# lasttick = tick
	&cmp	("eax",$lastdiff);
	&mov	($lastdiff,"eax");	# lastdiff = diff
	&mov	("edx",0);
	&setne	("dl");
	&sub	($cnt,"edx");		# conditional --$cnt
	&lea	($out,&DWP(0,$out,"edx",4));	# conditional ++$out
	&jnz	(&label("loop2"));

&set_label("done2");
	&mov	("eax",&wparam(1));
	&sub	("eax",$cnt);
&set_label("nogo");
    }
&function_end("OPENSSL_instrument_bus2");
}

495 496 497 498 499 500 501 502 503 504 505 506
&function_begin_B("OPENSSL_ia32_rdrand");
	&mov	("ecx",8);
&set_label("loop");
	&rdrand	("eax");
	&jc	(&label("break"));
	&loop	(&label("loop"));
&set_label("break");
	&cmp	("eax",0);
	&cmove	("eax","ecx");
	&ret	();
&function_end_B("OPENSSL_ia32_rdrand");

507 508 509 510 511 512 513 514 515 516 517 518
&function_begin_B("OPENSSL_ia32_rdseed");
	&mov	("ecx",8);
&set_label("loop");
	&rdseed	("eax");
	&jc	(&label("break"));
	&loop	(&label("loop"));
&set_label("break");
	&cmp	("eax",0);
	&cmove	("eax","ecx");
	&ret	();
&function_end_B("OPENSSL_ia32_rdseed");

519
&initseg("OPENSSL_cpuid_setup");
520

521 522 523
&hidden("OPENSSL_cpuid_setup");
&hidden("OPENSSL_ia32cap_P");

524
&asm_finish();
525 526

close STDOUT;
新手
引导
客服 返回
顶部