1 /* 2 * x86 CPU test 3 * 4 * Copyright (c) 2003 Fabrice Bellard 5 * 6 * This program is free software; you can redistribute it and/or modify 7 * it under the terms of the GNU General Public License as published by 8 * the Free Software Foundation; either version 2 of the License, or 9 * (at your option) any later version. 10 * 11 * This program is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 14 * GNU General Public License for more details. 15 * 16 * You should have received a copy of the GNU General Public License 17 * along with this program; if not, write to the Free Software 18 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 19 */ 20 #define _GNU_SOURCE 21 #include <stdlib.h> 22 #include <stdio.h> 23 #include <string.h> 24 #include <inttypes.h> 25 #include <math.h> 26 #include <signal.h> 27 #include <setjmp.h> 28 #include <errno.h> 29 #include <sys/ucontext.h> 30 #include <sys/mman.h> 31 32 #if !defined(__x86_64__) 33 //#define TEST_VM86 34 #define TEST_SEGS 35 #endif 36 //#define LINUX_VM86_IOPL_FIX 37 //#define TEST_P4_FLAGS 38 #ifdef __SSE__ 39 #define TEST_SSE 40 #define TEST_CMOV 1 41 #define TEST_FCOMI 1 42 #else 43 #undef TEST_SSE 44 #define TEST_CMOV 1 45 #define TEST_FCOMI 1 46 #endif 47 48 #if defined(__x86_64__) 49 #define FMT64X "%016lx" 50 #define FMTLX "%016lx" 51 #define X86_64_ONLY(x) x 52 #else 53 #define FMT64X "%016" PRIx64 54 #define FMTLX "%08lx" 55 #define X86_64_ONLY(x) 56 #endif 57 58 #ifdef TEST_VM86 59 #include <asm/vm86.h> 60 #endif 61 62 #define xglue(x, y) x ## y 63 #define glue(x, y) xglue(x, y) 64 #define stringify(s) tostring(s) 65 #define tostring(s) #s 66 67 #define CC_C 0x0001 68 #define CC_P 0x0004 69 #define CC_A 0x0010 70 #define CC_Z 0x0040 71 #define CC_S 0x0080 72 #define CC_O 0x0800 73 74 #define __init_call __attribute__ ((unused,__section__ ("initcall"))) 75 76 #define CC_MASK (CC_C | CC_P | CC_Z | CC_S | CC_O | CC_A) 77 78 #if defined(__x86_64__) 79 static inline long i2l(long v) 80 { 81 return v | ((v ^ 0xabcd) << 32); 82 } 83 #else 84 static inline long i2l(long v) 85 { 86 return v; 87 } 88 #endif 89 90 #define OP add 91 #include "test-i386.h" 92 93 #define OP sub 94 #include "test-i386.h" 95 96 #define OP xor 97 #include "test-i386.h" 98 99 #define OP and 100 #include "test-i386.h" 101 102 #define OP or 103 #include "test-i386.h" 104 105 #define OP cmp 106 #include "test-i386.h" 107 108 #define OP adc 109 #define OP_CC 110 #include "test-i386.h" 111 112 #define OP sbb 113 #define OP_CC 114 #include "test-i386.h" 115 116 #define OP inc 117 #define OP_CC 118 #define OP1 119 #include "test-i386.h" 120 121 #define OP dec 122 #define OP_CC 123 #define OP1 124 #include "test-i386.h" 125 126 #define OP neg 127 #define OP_CC 128 #define OP1 129 #include "test-i386.h" 130 131 #define OP not 132 #define OP_CC 133 #define OP1 134 #include "test-i386.h" 135 136 #undef CC_MASK 137 #define CC_MASK (CC_C | CC_P | CC_Z | CC_S | CC_O) 138 139 #define OP shl 140 #include "test-i386-shift.h" 141 142 #define OP shr 143 #include "test-i386-shift.h" 144 145 #define OP sar 146 #include "test-i386-shift.h" 147 148 #define OP rol 149 #include "test-i386-shift.h" 150 151 #define OP ror 152 #include "test-i386-shift.h" 153 154 #define OP rcr 155 #define OP_CC 156 #include "test-i386-shift.h" 157 158 #define OP rcl 159 #define OP_CC 160 #include "test-i386-shift.h" 161 162 #define OP shld 163 #define OP_SHIFTD 164 #define OP_NOBYTE 165 #include "test-i386-shift.h" 166 167 #define OP shrd 168 #define OP_SHIFTD 169 #define OP_NOBYTE 170 #include "test-i386-shift.h" 171 172 /* XXX: should be more precise ? */ 173 #undef CC_MASK 174 #define CC_MASK (CC_C) 175 176 #define OP bt 177 #define OP_NOBYTE 178 #include "test-i386-shift.h" 179 180 #define OP bts 181 #define OP_NOBYTE 182 #include "test-i386-shift.h" 183 184 #define OP btr 185 #define OP_NOBYTE 186 #include "test-i386-shift.h" 187 188 #define OP btc 189 #define OP_NOBYTE 190 #include "test-i386-shift.h" 191 192 /* lea test (modrm support) */ 193 #define TEST_LEAQ(STR)\ 194 {\ 195 asm("lea " STR ", %0"\ 196 : "=r" (res)\ 197 : "a" (eax), "b" (ebx), "c" (ecx), "d" (edx), "S" (esi), "D" (edi));\ 198 printf("lea %s = " FMTLX "\n", STR, res);\ 199 } 200 201 #define TEST_LEA(STR)\ 202 {\ 203 asm("lea " STR ", %0"\ 204 : "=r" (res)\ 205 : "a" (eax), "b" (ebx), "c" (ecx), "d" (edx), "S" (esi), "D" (edi));\ 206 printf("lea %s = " FMTLX "\n", STR, res);\ 207 } 208 209 #define TEST_LEA16(STR)\ 210 {\ 211 asm(".code16 ; .byte 0x67 ; leal " STR ", %0 ; .code32"\ 212 : "=wq" (res)\ 213 : "a" (eax), "b" (ebx), "c" (ecx), "d" (edx), "S" (esi), "D" (edi));\ 214 printf("lea %s = %08lx\n", STR, res);\ 215 } 216 217 218 void test_lea(void) 219 { 220 long eax, ebx, ecx, edx, esi, edi, res; 221 eax = i2l(0x0001); 222 ebx = i2l(0x0002); 223 ecx = i2l(0x0004); 224 edx = i2l(0x0008); 225 esi = i2l(0x0010); 226 edi = i2l(0x0020); 227 228 TEST_LEA("0x4000"); 229 230 TEST_LEA("(%%eax)"); 231 TEST_LEA("(%%ebx)"); 232 TEST_LEA("(%%ecx)"); 233 TEST_LEA("(%%edx)"); 234 TEST_LEA("(%%esi)"); 235 TEST_LEA("(%%edi)"); 236 237 TEST_LEA("0x40(%%eax)"); 238 TEST_LEA("0x40(%%ebx)"); 239 TEST_LEA("0x40(%%ecx)"); 240 TEST_LEA("0x40(%%edx)"); 241 TEST_LEA("0x40(%%esi)"); 242 TEST_LEA("0x40(%%edi)"); 243 244 TEST_LEA("0x4000(%%eax)"); 245 TEST_LEA("0x4000(%%ebx)"); 246 TEST_LEA("0x4000(%%ecx)"); 247 TEST_LEA("0x4000(%%edx)"); 248 TEST_LEA("0x4000(%%esi)"); 249 TEST_LEA("0x4000(%%edi)"); 250 251 TEST_LEA("(%%eax, %%ecx)"); 252 TEST_LEA("(%%ebx, %%edx)"); 253 TEST_LEA("(%%ecx, %%ecx)"); 254 TEST_LEA("(%%edx, %%ecx)"); 255 TEST_LEA("(%%esi, %%ecx)"); 256 TEST_LEA("(%%edi, %%ecx)"); 257 258 TEST_LEA("0x40(%%eax, %%ecx)"); 259 TEST_LEA("0x4000(%%ebx, %%edx)"); 260 261 TEST_LEA("(%%ecx, %%ecx, 2)"); 262 TEST_LEA("(%%edx, %%ecx, 4)"); 263 TEST_LEA("(%%esi, %%ecx, 8)"); 264 265 TEST_LEA("(,%%eax, 2)"); 266 TEST_LEA("(,%%ebx, 4)"); 267 TEST_LEA("(,%%ecx, 8)"); 268 269 TEST_LEA("0x40(,%%eax, 2)"); 270 TEST_LEA("0x40(,%%ebx, 4)"); 271 TEST_LEA("0x40(,%%ecx, 8)"); 272 273 274 TEST_LEA("-10(%%ecx, %%ecx, 2)"); 275 TEST_LEA("-10(%%edx, %%ecx, 4)"); 276 TEST_LEA("-10(%%esi, %%ecx, 8)"); 277 278 TEST_LEA("0x4000(%%ecx, %%ecx, 2)"); 279 TEST_LEA("0x4000(%%edx, %%ecx, 4)"); 280 TEST_LEA("0x4000(%%esi, %%ecx, 8)"); 281 282 #if defined(__x86_64__) 283 TEST_LEAQ("0x4000"); 284 TEST_LEAQ("0x4000(%%rip)"); 285 286 TEST_LEAQ("(%%rax)"); 287 TEST_LEAQ("(%%rbx)"); 288 TEST_LEAQ("(%%rcx)"); 289 TEST_LEAQ("(%%rdx)"); 290 TEST_LEAQ("(%%rsi)"); 291 TEST_LEAQ("(%%rdi)"); 292 293 TEST_LEAQ("0x40(%%rax)"); 294 TEST_LEAQ("0x40(%%rbx)"); 295 TEST_LEAQ("0x40(%%rcx)"); 296 TEST_LEAQ("0x40(%%rdx)"); 297 TEST_LEAQ("0x40(%%rsi)"); 298 TEST_LEAQ("0x40(%%rdi)"); 299 300 TEST_LEAQ("0x4000(%%rax)"); 301 TEST_LEAQ("0x4000(%%rbx)"); 302 TEST_LEAQ("0x4000(%%rcx)"); 303 TEST_LEAQ("0x4000(%%rdx)"); 304 TEST_LEAQ("0x4000(%%rsi)"); 305 TEST_LEAQ("0x4000(%%rdi)"); 306 307 TEST_LEAQ("(%%rax, %%rcx)"); 308 TEST_LEAQ("(%%rbx, %%rdx)"); 309 TEST_LEAQ("(%%rcx, %%rcx)"); 310 TEST_LEAQ("(%%rdx, %%rcx)"); 311 TEST_LEAQ("(%%rsi, %%rcx)"); 312 TEST_LEAQ("(%%rdi, %%rcx)"); 313 314 TEST_LEAQ("0x40(%%rax, %%rcx)"); 315 TEST_LEAQ("0x4000(%%rbx, %%rdx)"); 316 317 TEST_LEAQ("(%%rcx, %%rcx, 2)"); 318 TEST_LEAQ("(%%rdx, %%rcx, 4)"); 319 TEST_LEAQ("(%%rsi, %%rcx, 8)"); 320 321 TEST_LEAQ("(,%%rax, 2)"); 322 TEST_LEAQ("(,%%rbx, 4)"); 323 TEST_LEAQ("(,%%rcx, 8)"); 324 325 TEST_LEAQ("0x40(,%%rax, 2)"); 326 TEST_LEAQ("0x40(,%%rbx, 4)"); 327 TEST_LEAQ("0x40(,%%rcx, 8)"); 328 329 330 TEST_LEAQ("-10(%%rcx, %%rcx, 2)"); 331 TEST_LEAQ("-10(%%rdx, %%rcx, 4)"); 332 TEST_LEAQ("-10(%%rsi, %%rcx, 8)"); 333 334 TEST_LEAQ("0x4000(%%rcx, %%rcx, 2)"); 335 TEST_LEAQ("0x4000(%%rdx, %%rcx, 4)"); 336 TEST_LEAQ("0x4000(%%rsi, %%rcx, 8)"); 337 #else 338 /* limited 16 bit addressing test */ 339 TEST_LEA16("0x4000"); 340 TEST_LEA16("(%%bx)"); 341 TEST_LEA16("(%%si)"); 342 TEST_LEA16("(%%di)"); 343 TEST_LEA16("0x40(%%bx)"); 344 TEST_LEA16("0x40(%%si)"); 345 TEST_LEA16("0x40(%%di)"); 346 TEST_LEA16("0x4000(%%bx)"); 347 TEST_LEA16("0x4000(%%si)"); 348 TEST_LEA16("(%%bx,%%si)"); 349 TEST_LEA16("(%%bx,%%di)"); 350 TEST_LEA16("0x40(%%bx,%%si)"); 351 TEST_LEA16("0x40(%%bx,%%di)"); 352 TEST_LEA16("0x4000(%%bx,%%si)"); 353 TEST_LEA16("0x4000(%%bx,%%di)"); 354 #endif 355 } 356 357 #define TEST_JCC(JCC, v1, v2)\ 358 {\ 359 int res;\ 360 asm("movl $1, %0\n\t"\ 361 "cmpl %2, %1\n\t"\ 362 "j" JCC " 1f\n\t"\ 363 "movl $0, %0\n\t"\ 364 "1:\n\t"\ 365 : "=r" (res)\ 366 : "r" (v1), "r" (v2));\ 367 printf("%-10s %d\n", "j" JCC, res);\ 368 \ 369 asm("movl $0, %0\n\t"\ 370 "cmpl %2, %1\n\t"\ 371 "set" JCC " %b0\n\t"\ 372 : "=r" (res)\ 373 : "r" (v1), "r" (v2));\ 374 printf("%-10s %d\n", "set" JCC, res);\ 375 if (TEST_CMOV) {\ 376 long val = i2l(1);\ 377 long res = i2l(0x12345678);\ 378 X86_64_ONLY(\ 379 asm("cmpl %2, %1\n\t"\ 380 "cmov" JCC "q %3, %0\n\t"\ 381 : "=r" (res)\ 382 : "r" (v1), "r" (v2), "m" (val), "0" (res));\ 383 printf("%-10s R=" FMTLX "\n", "cmov" JCC "q", res);)\ 384 asm("cmpl %2, %1\n\t"\ 385 "cmov" JCC "l %k3, %k0\n\t"\ 386 : "=r" (res)\ 387 : "r" (v1), "r" (v2), "m" (val), "0" (res));\ 388 printf("%-10s R=" FMTLX "\n", "cmov" JCC "l", res);\ 389 asm("cmpl %2, %1\n\t"\ 390 "cmov" JCC "w %w3, %w0\n\t"\ 391 : "=r" (res)\ 392 : "r" (v1), "r" (v2), "r" (1), "0" (res));\ 393 printf("%-10s R=" FMTLX "\n", "cmov" JCC "w", res);\ 394 } \ 395 } 396 397 /* various jump tests */ 398 void test_jcc(void) 399 { 400 TEST_JCC("ne", 1, 1); 401 TEST_JCC("ne", 1, 0); 402 403 TEST_JCC("e", 1, 1); 404 TEST_JCC("e", 1, 0); 405 406 TEST_JCC("l", 1, 1); 407 TEST_JCC("l", 1, 0); 408 TEST_JCC("l", 1, -1); 409 410 TEST_JCC("le", 1, 1); 411 TEST_JCC("le", 1, 0); 412 TEST_JCC("le", 1, -1); 413 414 TEST_JCC("ge", 1, 1); 415 TEST_JCC("ge", 1, 0); 416 TEST_JCC("ge", -1, 1); 417 418 TEST_JCC("g", 1, 1); 419 TEST_JCC("g", 1, 0); 420 TEST_JCC("g", 1, -1); 421 422 TEST_JCC("b", 1, 1); 423 TEST_JCC("b", 1, 0); 424 TEST_JCC("b", 1, -1); 425 426 TEST_JCC("be", 1, 1); 427 TEST_JCC("be", 1, 0); 428 TEST_JCC("be", 1, -1); 429 430 TEST_JCC("ae", 1, 1); 431 TEST_JCC("ae", 1, 0); 432 TEST_JCC("ae", 1, -1); 433 434 TEST_JCC("a", 1, 1); 435 TEST_JCC("a", 1, 0); 436 TEST_JCC("a", 1, -1); 437 438 439 TEST_JCC("p", 1, 1); 440 TEST_JCC("p", 1, 0); 441 442 TEST_JCC("np", 1, 1); 443 TEST_JCC("np", 1, 0); 444 445 TEST_JCC("o", 0x7fffffff, 0); 446 TEST_JCC("o", 0x7fffffff, -1); 447 448 TEST_JCC("no", 0x7fffffff, 0); 449 TEST_JCC("no", 0x7fffffff, -1); 450 451 TEST_JCC("s", 0, 1); 452 TEST_JCC("s", 0, -1); 453 TEST_JCC("s", 0, 0); 454 455 TEST_JCC("ns", 0, 1); 456 TEST_JCC("ns", 0, -1); 457 TEST_JCC("ns", 0, 0); 458 } 459 460 #define TEST_LOOP(insn) \ 461 {\ 462 for(i = 0; i < sizeof(ecx_vals) / sizeof(long); i++) {\ 463 ecx = ecx_vals[i];\ 464 for(zf = 0; zf < 2; zf++) {\ 465 asm("test %2, %2\n\t"\ 466 "movl $1, %0\n\t"\ 467 insn " 1f\n\t" \ 468 "movl $0, %0\n\t"\ 469 "1:\n\t"\ 470 : "=a" (res)\ 471 : "c" (ecx), "b" (!zf)); \ 472 printf("%-10s ECX=" FMTLX " ZF=%ld r=%d\n", insn, ecx, zf, res); \ 473 }\ 474 }\ 475 } 476 477 void test_loop(void) 478 { 479 long ecx, zf; 480 const long ecx_vals[] = { 481 0, 482 1, 483 0x10000, 484 0x10001, 485 #if defined(__x86_64__) 486 0x100000000L, 487 0x100000001L, 488 #endif 489 }; 490 int i, res; 491 492 #if !defined(__x86_64__) 493 TEST_LOOP("jcxz"); 494 TEST_LOOP("loopw"); 495 TEST_LOOP("loopzw"); 496 TEST_LOOP("loopnzw"); 497 #endif 498 499 TEST_LOOP("jecxz"); 500 TEST_LOOP("loopl"); 501 TEST_LOOP("loopzl"); 502 TEST_LOOP("loopnzl"); 503 } 504 505 #undef CC_MASK 506 #ifdef TEST_P4_FLAGS 507 #define CC_MASK (CC_C | CC_P | CC_Z | CC_S | CC_O | CC_A) 508 #else 509 #define CC_MASK (CC_O | CC_C) 510 #endif 511 512 #define OP mul 513 #include "test-i386-muldiv.h" 514 515 #define OP imul 516 #include "test-i386-muldiv.h" 517 518 void test_imulw2(long op0, long op1) 519 { 520 long res, s1, s0, flags; 521 s0 = op0; 522 s1 = op1; 523 res = s0; 524 flags = 0; 525 asm volatile ("push %4\n\t" 526 "popf\n\t" 527 "imulw %w2, %w0\n\t" 528 "pushf\n\t" 529 "pop %1\n\t" 530 : "=q" (res), "=g" (flags) 531 : "q" (s1), "0" (res), "1" (flags)); 532 printf("%-10s A=" FMTLX " B=" FMTLX " R=" FMTLX " CC=%04lx\n", 533 "imulw", s0, s1, res, flags & CC_MASK); 534 } 535 536 void test_imull2(long op0, long op1) 537 { 538 long res, s1, s0, flags; 539 s0 = op0; 540 s1 = op1; 541 res = s0; 542 flags = 0; 543 asm volatile ("push %4\n\t" 544 "popf\n\t" 545 "imull %k2, %k0\n\t" 546 "pushf\n\t" 547 "pop %1\n\t" 548 : "=q" (res), "=g" (flags) 549 : "q" (s1), "0" (res), "1" (flags)); 550 printf("%-10s A=" FMTLX " B=" FMTLX " R=" FMTLX " CC=%04lx\n", 551 "imull", s0, s1, res, flags & CC_MASK); 552 } 553 554 #if defined(__x86_64__) 555 void test_imulq2(long op0, long op1) 556 { 557 long res, s1, s0, flags; 558 s0 = op0; 559 s1 = op1; 560 res = s0; 561 flags = 0; 562 asm volatile ("push %4\n\t" 563 "popf\n\t" 564 "imulq %2, %0\n\t" 565 "pushf\n\t" 566 "pop %1\n\t" 567 : "=q" (res), "=g" (flags) 568 : "q" (s1), "0" (res), "1" (flags)); 569 printf("%-10s A=" FMTLX " B=" FMTLX " R=" FMTLX " CC=%04lx\n", 570 "imulq", s0, s1, res, flags & CC_MASK); 571 } 572 #endif 573 574 #define TEST_IMUL_IM(size, rsize, op0, op1)\ 575 {\ 576 long res, flags, s1;\ 577 flags = 0;\ 578 res = 0;\ 579 s1 = op1;\ 580 asm volatile ("push %3\n\t"\ 581 "popf\n\t"\ 582 "imul" size " $" #op0 ", %" rsize "2, %" rsize "0\n\t" \ 583 "pushf\n\t"\ 584 "pop %1\n\t"\ 585 : "=r" (res), "=g" (flags)\ 586 : "r" (s1), "1" (flags), "0" (res));\ 587 printf("%-10s A=" FMTLX " B=" FMTLX " R=" FMTLX " CC=%04lx\n",\ 588 "imul" size " im", (long)op0, (long)op1, res, flags & CC_MASK);\ 589 } 590 591 592 #undef CC_MASK 593 #define CC_MASK (0) 594 595 #define OP div 596 #include "test-i386-muldiv.h" 597 598 #define OP idiv 599 #include "test-i386-muldiv.h" 600 601 void test_mul(void) 602 { 603 test_imulb(0x1234561d, 4); 604 test_imulb(3, -4); 605 test_imulb(0x80, 0x80); 606 test_imulb(0x10, 0x10); 607 608 test_imulw(0, 0x1234001d, 45); 609 test_imulw(0, 23, -45); 610 test_imulw(0, 0x8000, 0x8000); 611 test_imulw(0, 0x100, 0x100); 612 613 test_imull(0, 0x1234001d, 45); 614 test_imull(0, 23, -45); 615 test_imull(0, 0x80000000, 0x80000000); 616 test_imull(0, 0x10000, 0x10000); 617 618 test_mulb(0x1234561d, 4); 619 test_mulb(3, -4); 620 test_mulb(0x80, 0x80); 621 test_mulb(0x10, 0x10); 622 623 test_mulw(0, 0x1234001d, 45); 624 test_mulw(0, 23, -45); 625 test_mulw(0, 0x8000, 0x8000); 626 test_mulw(0, 0x100, 0x100); 627 628 test_mull(0, 0x1234001d, 45); 629 test_mull(0, 23, -45); 630 test_mull(0, 0x80000000, 0x80000000); 631 test_mull(0, 0x10000, 0x10000); 632 633 test_imulw2(0x1234001d, 45); 634 test_imulw2(23, -45); 635 test_imulw2(0x8000, 0x8000); 636 test_imulw2(0x100, 0x100); 637 638 test_imull2(0x1234001d, 45); 639 test_imull2(23, -45); 640 test_imull2(0x80000000, 0x80000000); 641 test_imull2(0x10000, 0x10000); 642 643 TEST_IMUL_IM("w", "w", 45, 0x1234); 644 TEST_IMUL_IM("w", "w", -45, 23); 645 TEST_IMUL_IM("w", "w", 0x8000, 0x80000000); 646 TEST_IMUL_IM("w", "w", 0x7fff, 0x1000); 647 648 TEST_IMUL_IM("l", "k", 45, 0x1234); 649 TEST_IMUL_IM("l", "k", -45, 23); 650 TEST_IMUL_IM("l", "k", 0x8000, 0x80000000); 651 TEST_IMUL_IM("l", "k", 0x7fff, 0x1000); 652 653 test_idivb(0x12341678, 0x127e); 654 test_idivb(0x43210123, -5); 655 test_idivb(0x12340004, -1); 656 657 test_idivw(0, 0x12345678, 12347); 658 test_idivw(0, -23223, -45); 659 test_idivw(0, 0x12348000, -1); 660 test_idivw(0x12343, 0x12345678, 0x81238567); 661 662 test_idivl(0, 0x12345678, 12347); 663 test_idivl(0, -233223, -45); 664 test_idivl(0, 0x80000000, -1); 665 test_idivl(0x12343, 0x12345678, 0x81234567); 666 667 test_divb(0x12341678, 0x127e); 668 test_divb(0x43210123, -5); 669 test_divb(0x12340004, -1); 670 671 test_divw(0, 0x12345678, 12347); 672 test_divw(0, -23223, -45); 673 test_divw(0, 0x12348000, -1); 674 test_divw(0x12343, 0x12345678, 0x81238567); 675 676 test_divl(0, 0x12345678, 12347); 677 test_divl(0, -233223, -45); 678 test_divl(0, 0x80000000, -1); 679 test_divl(0x12343, 0x12345678, 0x81234567); 680 681 #if defined(__x86_64__) 682 test_imulq(0, 0x1234001d1234001d, 45); 683 test_imulq(0, 23, -45); 684 test_imulq(0, 0x8000000000000000, 0x8000000000000000); 685 test_imulq(0, 0x100000000, 0x100000000); 686 687 test_mulq(0, 0x1234001d1234001d, 45); 688 test_mulq(0, 23, -45); 689 test_mulq(0, 0x8000000000000000, 0x8000000000000000); 690 test_mulq(0, 0x100000000, 0x100000000); 691 692 test_imulq2(0x1234001d1234001d, 45); 693 test_imulq2(23, -45); 694 test_imulq2(0x8000000000000000, 0x8000000000000000); 695 test_imulq2(0x100000000, 0x100000000); 696 697 TEST_IMUL_IM("q", "", 45, 0x12341234); 698 TEST_IMUL_IM("q", "", -45, 23); 699 TEST_IMUL_IM("q", "", 0x8000, 0x8000000000000000); 700 TEST_IMUL_IM("q", "", 0x7fff, 0x10000000); 701 702 test_idivq(0, 0x12345678abcdef, 12347); 703 test_idivq(0, -233223, -45); 704 test_idivq(0, 0x8000000000000000, -1); 705 test_idivq(0x12343, 0x12345678, 0x81234567); 706 707 test_divq(0, 0x12345678abcdef, 12347); 708 test_divq(0, -233223, -45); 709 test_divq(0, 0x8000000000000000, -1); 710 test_divq(0x12343, 0x12345678, 0x81234567); 711 #endif 712 } 713 714 #define TEST_BSX(op, size, op0)\ 715 {\ 716 long res, val, resz;\ 717 val = op0;\ 718 asm("xor %1, %1\n"\ 719 "mov $0x12345678, %0\n"\ 720 #op " %" size "2, %" size "0 ; setz %b1" \ 721 : "=r" (res), "=q" (resz)\ 722 : "g" (val));\ 723 printf("%-10s A=" FMTLX " R=" FMTLX " %ld\n", #op, val, res, resz);\ 724 } 725 726 void test_bsx(void) 727 { 728 TEST_BSX(bsrw, "w", 0); 729 TEST_BSX(bsrw, "w", 0x12340128); 730 TEST_BSX(bsfw, "w", 0); 731 TEST_BSX(bsfw, "w", 0x12340128); 732 TEST_BSX(bsrl, "k", 0); 733 TEST_BSX(bsrl, "k", 0x00340128); 734 TEST_BSX(bsfl, "k", 0); 735 TEST_BSX(bsfl, "k", 0x00340128); 736 #if defined(__x86_64__) 737 TEST_BSX(bsrq, "", 0); 738 TEST_BSX(bsrq, "", 0x003401281234); 739 TEST_BSX(bsfq, "", 0); 740 TEST_BSX(bsfq, "", 0x003401281234); 741 #endif 742 } 743 744 /**********************************************/ 745 746 union float64u { 747 double d; 748 uint64_t l; 749 }; 750 751 union float64u q_nan = { .l = 0xFFF8000000000000LL }; 752 union float64u s_nan = { .l = 0xFFF0000000000000LL }; 753 754 void test_fops(double a, double b) 755 { 756 printf("a=%f b=%f a+b=%f\n", a, b, a + b); 757 printf("a=%f b=%f a-b=%f\n", a, b, a - b); 758 printf("a=%f b=%f a*b=%f\n", a, b, a * b); 759 printf("a=%f b=%f a/b=%f\n", a, b, a / b); 760 printf("a=%f b=%f fmod(a, b)=%f\n", a, b, fmod(a, b)); 761 printf("a=%f sqrt(a)=%f\n", a, sqrt(a)); 762 printf("a=%f sin(a)=%f\n", a, sin(a)); 763 printf("a=%f cos(a)=%f\n", a, cos(a)); 764 printf("a=%f tan(a)=%f\n", a, tan(a)); 765 printf("a=%f log(a)=%f\n", a, log(a)); 766 printf("a=%f exp(a)=%f\n", a, exp(a)); 767 printf("a=%f b=%f atan2(a, b)=%f\n", a, b, atan2(a, b)); 768 /* just to test some op combining */ 769 printf("a=%f asin(sin(a))=%f\n", a, asin(sin(a))); 770 printf("a=%f acos(cos(a))=%f\n", a, acos(cos(a))); 771 printf("a=%f atan(tan(a))=%f\n", a, atan(tan(a))); 772 773 } 774 775 void fpu_clear_exceptions(void) 776 { 777 struct __attribute__((packed)) { 778 uint16_t fpuc; 779 uint16_t dummy1; 780 uint16_t fpus; 781 uint16_t dummy2; 782 uint16_t fptag; 783 uint16_t dummy3; 784 uint32_t ignored[4]; 785 long double fpregs[8]; 786 } float_env32; 787 788 asm volatile ("fnstenv %0\n" : : "m" (float_env32)); 789 float_env32.fpus &= ~0x7f; 790 asm volatile ("fldenv %0\n" : : "m" (float_env32)); 791 } 792 793 /* XXX: display exception bits when supported */ 794 #define FPUS_EMASK 0x0000 795 //#define FPUS_EMASK 0x007f 796 797 void test_fcmp(double a, double b) 798 { 799 long eflags, fpus; 800 801 fpu_clear_exceptions(); 802 asm("fcom %2\n" 803 "fstsw %%ax\n" 804 : "=a" (fpus) 805 : "t" (a), "u" (b)); 806 printf("fcom(%f %f)=%04lx \n", 807 a, b, fpus & (0x4500 | FPUS_EMASK)); 808 fpu_clear_exceptions(); 809 asm("fucom %2\n" 810 "fstsw %%ax\n" 811 : "=a" (fpus) 812 : "t" (a), "u" (b)); 813 printf("fucom(%f %f)=%04lx\n", 814 a, b, fpus & (0x4500 | FPUS_EMASK)); 815 if (TEST_FCOMI) { 816 /* test f(u)comi instruction */ 817 fpu_clear_exceptions(); 818 asm("fcomi %3, %2\n" 819 "fstsw %%ax\n" 820 "pushf\n" 821 "pop %0\n" 822 : "=r" (eflags), "=a" (fpus) 823 : "t" (a), "u" (b)); 824 printf("fcomi(%f %f)=%04lx %02lx\n", 825 a, b, fpus & FPUS_EMASK, eflags & (CC_Z | CC_P | CC_C)); 826 fpu_clear_exceptions(); 827 asm("fucomi %3, %2\n" 828 "fstsw %%ax\n" 829 "pushf\n" 830 "pop %0\n" 831 : "=r" (eflags), "=a" (fpus) 832 : "t" (a), "u" (b)); 833 printf("fucomi(%f %f)=%04lx %02lx\n", 834 a, b, fpus & FPUS_EMASK, eflags & (CC_Z | CC_P | CC_C)); 835 } 836 fpu_clear_exceptions(); 837 asm volatile("fxam\n" 838 "fstsw %%ax\n" 839 : "=a" (fpus) 840 : "t" (a)); 841 printf("fxam(%f)=%04lx\n", a, fpus & 0x4700); 842 fpu_clear_exceptions(); 843 } 844 845 void test_fcvt(double a) 846 { 847 float fa; 848 long double la; 849 int16_t fpuc; 850 int i; 851 int64_t lla; 852 int ia; 853 int16_t wa; 854 double ra; 855 856 fa = a; 857 la = a; 858 printf("(float)%f = %f\n", a, fa); 859 printf("(long double)%f = %Lf\n", a, la); 860 printf("a=" FMT64X "\n", *(uint64_t *)&a); 861 printf("la=" FMT64X " %04x\n", *(uint64_t *)&la, 862 *(unsigned short *)((char *)(&la) + 8)); 863 864 /* test all roundings */ 865 asm volatile ("fstcw %0" : "=m" (fpuc)); 866 for(i=0;i<4;i++) { 867 uint16_t val16; 868 val16 = (fpuc & ~0x0c00) | (i << 10); 869 asm volatile ("fldcw %0" : : "m" (val16)); 870 asm volatile ("fist %0" : "=m" (wa) : "t" (a)); 871 asm volatile ("fistl %0" : "=m" (ia) : "t" (a)); 872 asm volatile ("fistpll %0" : "=m" (lla) : "t" (a) : "st"); 873 asm volatile ("frndint ; fstl %0" : "=m" (ra) : "t" (a)); 874 asm volatile ("fldcw %0" : : "m" (fpuc)); 875 printf("(short)a = %d\n", wa); 876 printf("(int)a = %d\n", ia); 877 printf("(int64_t)a = " FMT64X "\n", lla); 878 printf("rint(a) = %f\n", ra); 879 } 880 } 881 882 #define TEST(N) \ 883 asm("fld" #N : "=t" (a)); \ 884 printf("fld" #N "= %f\n", a); 885 886 void test_fconst(void) 887 { 888 double a; 889 TEST(1); 890 TEST(l2t); 891 TEST(l2e); 892 TEST(pi); 893 TEST(lg2); 894 TEST(ln2); 895 TEST(z); 896 } 897 898 void test_fbcd(double a) 899 { 900 unsigned short bcd[5]; 901 double b; 902 903 asm("fbstp %0" : "=m" (bcd[0]) : "t" (a) : "st"); 904 asm("fbld %1" : "=t" (b) : "m" (bcd[0])); 905 printf("a=%f bcd=%04x%04x%04x%04x%04x b=%f\n", 906 a, bcd[4], bcd[3], bcd[2], bcd[1], bcd[0], b); 907 } 908 909 #define TEST_ENV(env, save, restore)\ 910 {\ 911 memset((env), 0xaa, sizeof(*(env)));\ 912 for(i=0;i<5;i++)\ 913 asm volatile ("fldl %0" : : "m" (dtab[i]));\ 914 asm volatile (save " %0\n" : : "m" (*(env)));\ 915 asm volatile (restore " %0\n": : "m" (*(env)));\ 916 for(i=0;i<5;i++)\ 917 asm volatile ("fstpl %0" : "=m" (rtab[i]));\ 918 for(i=0;i<5;i++)\ 919 printf("res[%d]=%f\n", i, rtab[i]);\ 920 printf("fpuc=%04x fpus=%04x fptag=%04x\n",\ 921 (env)->fpuc,\ 922 (env)->fpus & 0xff00,\ 923 (env)->fptag);\ 924 } 925 926 void test_fenv(void) 927 { 928 struct __attribute__((packed)) { 929 uint16_t fpuc; 930 uint16_t dummy1; 931 uint16_t fpus; 932 uint16_t dummy2; 933 uint16_t fptag; 934 uint16_t dummy3; 935 uint32_t ignored[4]; 936 long double fpregs[8]; 937 } float_env32; 938 struct __attribute__((packed)) { 939 uint16_t fpuc; 940 uint16_t fpus; 941 uint16_t fptag; 942 uint16_t ignored[4]; 943 long double fpregs[8]; 944 } float_env16; 945 double dtab[8]; 946 double rtab[8]; 947 int i; 948 949 for(i=0;i<8;i++) 950 dtab[i] = i + 1; 951 952 TEST_ENV(&float_env16, "data16 fnstenv", "data16 fldenv"); 953 TEST_ENV(&float_env16, "data16 fnsave", "data16 frstor"); 954 TEST_ENV(&float_env32, "fnstenv", "fldenv"); 955 TEST_ENV(&float_env32, "fnsave", "frstor"); 956 957 /* test for ffree */ 958 for(i=0;i<5;i++) 959 asm volatile ("fldl %0" : : "m" (dtab[i])); 960 asm volatile("ffree %st(2)"); 961 asm volatile ("fnstenv %0\n" : : "m" (float_env32)); 962 asm volatile ("fninit"); 963 printf("fptag=%04x\n", float_env32.fptag); 964 } 965 966 967 #define TEST_FCMOV(a, b, eflags, CC)\ 968 {\ 969 double res;\ 970 asm("push %3\n"\ 971 "popf\n"\ 972 "fcmov" CC " %2, %0\n"\ 973 : "=t" (res)\ 974 : "0" (a), "u" (b), "g" (eflags));\ 975 printf("fcmov%s eflags=0x%04lx-> %f\n", \ 976 CC, (long)eflags, res);\ 977 } 978 979 void test_fcmov(void) 980 { 981 double a, b; 982 long eflags, i; 983 984 a = 1.0; 985 b = 2.0; 986 for(i = 0; i < 4; i++) { 987 eflags = 0; 988 if (i & 1) 989 eflags |= CC_C; 990 if (i & 2) 991 eflags |= CC_Z; 992 TEST_FCMOV(a, b, eflags, "b"); 993 TEST_FCMOV(a, b, eflags, "e"); 994 TEST_FCMOV(a, b, eflags, "be"); 995 TEST_FCMOV(a, b, eflags, "nb"); 996 TEST_FCMOV(a, b, eflags, "ne"); 997 TEST_FCMOV(a, b, eflags, "nbe"); 998 } 999 TEST_FCMOV(a, b, 0, "u"); 1000 TEST_FCMOV(a, b, CC_P, "u"); 1001 TEST_FCMOV(a, b, 0, "nu"); 1002 TEST_FCMOV(a, b, CC_P, "nu"); 1003 } 1004 1005 void test_floats(void) 1006 { 1007 test_fops(2, 3); 1008 test_fops(1.4, -5); 1009 test_fcmp(2, -1); 1010 test_fcmp(2, 2); 1011 test_fcmp(2, 3); 1012 test_fcmp(2, q_nan.d); 1013 test_fcmp(q_nan.d, -1); 1014 test_fcmp(-1.0/0.0, -1); 1015 test_fcmp(1.0/0.0, -1); 1016 test_fcvt(0.5); 1017 test_fcvt(-0.5); 1018 test_fcvt(1.0/7.0); 1019 test_fcvt(-1.0/9.0); 1020 test_fcvt(32768); 1021 test_fcvt(-1e20); 1022 test_fcvt(-1.0/0.0); 1023 test_fcvt(1.0/0.0); 1024 test_fcvt(q_nan.d); 1025 test_fconst(); 1026 test_fbcd(1234567890123456.0); 1027 test_fbcd(-123451234567890.0); 1028 test_fenv(); 1029 if (TEST_CMOV) { 1030 test_fcmov(); 1031 } 1032 } 1033 1034 /**********************************************/ 1035 #if !defined(__x86_64__) 1036 1037 #define TEST_BCD(op, op0, cc_in, cc_mask)\ 1038 {\ 1039 int res, flags;\ 1040 res = op0;\ 1041 flags = cc_in;\ 1042 asm ("push %3\n\t"\ 1043 "popf\n\t"\ 1044 #op "\n\t"\ 1045 "pushf\n\t"\ 1046 "pop %1\n\t"\ 1047 : "=a" (res), "=g" (flags)\ 1048 : "0" (res), "1" (flags));\ 1049 printf("%-10s A=%08x R=%08x CCIN=%04x CC=%04x\n",\ 1050 #op, op0, res, cc_in, flags & cc_mask);\ 1051 } 1052 1053 void test_bcd(void) 1054 { 1055 TEST_BCD(daa, 0x12340503, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1056 TEST_BCD(daa, 0x12340506, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1057 TEST_BCD(daa, 0x12340507, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1058 TEST_BCD(daa, 0x12340559, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1059 TEST_BCD(daa, 0x12340560, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1060 TEST_BCD(daa, 0x1234059f, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1061 TEST_BCD(daa, 0x123405a0, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1062 TEST_BCD(daa, 0x12340503, 0, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1063 TEST_BCD(daa, 0x12340506, 0, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1064 TEST_BCD(daa, 0x12340503, CC_C, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1065 TEST_BCD(daa, 0x12340506, CC_C, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1066 TEST_BCD(daa, 0x12340503, CC_C | CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1067 TEST_BCD(daa, 0x12340506, CC_C | CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1068 1069 TEST_BCD(das, 0x12340503, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1070 TEST_BCD(das, 0x12340506, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1071 TEST_BCD(das, 0x12340507, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1072 TEST_BCD(das, 0x12340559, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1073 TEST_BCD(das, 0x12340560, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1074 TEST_BCD(das, 0x1234059f, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1075 TEST_BCD(das, 0x123405a0, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1076 TEST_BCD(das, 0x12340503, 0, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1077 TEST_BCD(das, 0x12340506, 0, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1078 TEST_BCD(das, 0x12340503, CC_C, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1079 TEST_BCD(das, 0x12340506, CC_C, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1080 TEST_BCD(das, 0x12340503, CC_C | CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1081 TEST_BCD(das, 0x12340506, CC_C | CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_A)); 1082 1083 TEST_BCD(aaa, 0x12340205, CC_A, (CC_C | CC_A)); 1084 TEST_BCD(aaa, 0x12340306, CC_A, (CC_C | CC_A)); 1085 TEST_BCD(aaa, 0x1234040a, CC_A, (CC_C | CC_A)); 1086 TEST_BCD(aaa, 0x123405fa, CC_A, (CC_C | CC_A)); 1087 TEST_BCD(aaa, 0x12340205, 0, (CC_C | CC_A)); 1088 TEST_BCD(aaa, 0x12340306, 0, (CC_C | CC_A)); 1089 TEST_BCD(aaa, 0x1234040a, 0, (CC_C | CC_A)); 1090 TEST_BCD(aaa, 0x123405fa, 0, (CC_C | CC_A)); 1091 1092 TEST_BCD(aas, 0x12340205, CC_A, (CC_C | CC_A)); 1093 TEST_BCD(aas, 0x12340306, CC_A, (CC_C | CC_A)); 1094 TEST_BCD(aas, 0x1234040a, CC_A, (CC_C | CC_A)); 1095 TEST_BCD(aas, 0x123405fa, CC_A, (CC_C | CC_A)); 1096 TEST_BCD(aas, 0x12340205, 0, (CC_C | CC_A)); 1097 TEST_BCD(aas, 0x12340306, 0, (CC_C | CC_A)); 1098 TEST_BCD(aas, 0x1234040a, 0, (CC_C | CC_A)); 1099 TEST_BCD(aas, 0x123405fa, 0, (CC_C | CC_A)); 1100 1101 TEST_BCD(aam, 0x12340547, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_O | CC_A)); 1102 TEST_BCD(aad, 0x12340407, CC_A, (CC_C | CC_P | CC_Z | CC_S | CC_O | CC_A)); 1103 } 1104 #endif 1105 1106 #define TEST_XCHG(op, size, opconst)\ 1107 {\ 1108 long op0, op1;\ 1109 op0 = i2l(0x12345678);\ 1110 op1 = i2l(0xfbca7654);\ 1111 asm(#op " %" size "0, %" size "1" \ 1112 : "=q" (op0), opconst (op1) \ 1113 : "0" (op0));\ 1114 printf("%-10s A=" FMTLX " B=" FMTLX "\n",\ 1115 #op, op0, op1);\ 1116 } 1117 1118 #define TEST_CMPXCHG(op, size, opconst, eax)\ 1119 {\ 1120 long op0, op1, op2;\ 1121 op0 = i2l(0x12345678);\ 1122 op1 = i2l(0xfbca7654);\ 1123 op2 = i2l(eax);\ 1124 asm(#op " %" size "0, %" size "1" \ 1125 : "=q" (op0), opconst (op1) \ 1126 : "0" (op0), "a" (op2));\ 1127 printf("%-10s EAX=" FMTLX " A=" FMTLX " C=" FMTLX "\n",\ 1128 #op, op2, op0, op1);\ 1129 } 1130 1131 void test_xchg(void) 1132 { 1133 #if defined(__x86_64__) 1134 TEST_XCHG(xchgq, "", "+q"); 1135 #endif 1136 TEST_XCHG(xchgl, "k", "+q"); 1137 TEST_XCHG(xchgw, "w", "+q"); 1138 TEST_XCHG(xchgb, "b", "+q"); 1139 1140 #if defined(__x86_64__) 1141 TEST_XCHG(xchgq, "", "=m"); 1142 #endif 1143 TEST_XCHG(xchgl, "k", "+m"); 1144 TEST_XCHG(xchgw, "w", "+m"); 1145 TEST_XCHG(xchgb, "b", "+m"); 1146 1147 #if defined(__x86_64__) 1148 TEST_XCHG(xaddq, "", "+q"); 1149 #endif 1150 TEST_XCHG(xaddl, "k", "+q"); 1151 TEST_XCHG(xaddw, "w", "+q"); 1152 TEST_XCHG(xaddb, "b", "+q"); 1153 1154 { 1155 int res; 1156 res = 0x12345678; 1157 asm("xaddl %1, %0" : "=r" (res) : "0" (res)); 1158 printf("xaddl same res=%08x\n", res); 1159 } 1160 1161 #if defined(__x86_64__) 1162 TEST_XCHG(xaddq, "", "+m"); 1163 #endif 1164 TEST_XCHG(xaddl, "k", "+m"); 1165 TEST_XCHG(xaddw, "w", "+m"); 1166 TEST_XCHG(xaddb, "b", "+m"); 1167 1168 #if defined(__x86_64__) 1169 TEST_CMPXCHG(cmpxchgq, "", "+q", 0xfbca7654); 1170 #endif 1171 TEST_CMPXCHG(cmpxchgl, "k", "+q", 0xfbca7654); 1172 TEST_CMPXCHG(cmpxchgw, "w", "+q", 0xfbca7654); 1173 TEST_CMPXCHG(cmpxchgb, "b", "+q", 0xfbca7654); 1174 1175 #if defined(__x86_64__) 1176 TEST_CMPXCHG(cmpxchgq, "", "+q", 0xfffefdfc); 1177 #endif 1178 TEST_CMPXCHG(cmpxchgl, "k", "+q", 0xfffefdfc); 1179 TEST_CMPXCHG(cmpxchgw, "w", "+q", 0xfffefdfc); 1180 TEST_CMPXCHG(cmpxchgb, "b", "+q", 0xfffefdfc); 1181 1182 #if defined(__x86_64__) 1183 TEST_CMPXCHG(cmpxchgq, "", "+m", 0xfbca7654); 1184 #endif 1185 TEST_CMPXCHG(cmpxchgl, "k", "+m", 0xfbca7654); 1186 TEST_CMPXCHG(cmpxchgw, "w", "+m", 0xfbca7654); 1187 TEST_CMPXCHG(cmpxchgb, "b", "+m", 0xfbca7654); 1188 1189 #if defined(__x86_64__) 1190 TEST_CMPXCHG(cmpxchgq, "", "+m", 0xfffefdfc); 1191 #endif 1192 TEST_CMPXCHG(cmpxchgl, "k", "+m", 0xfffefdfc); 1193 TEST_CMPXCHG(cmpxchgw, "w", "+m", 0xfffefdfc); 1194 TEST_CMPXCHG(cmpxchgb, "b", "+m", 0xfffefdfc); 1195 1196 { 1197 uint64_t op0, op1, op2; 1198 long eax, edx; 1199 long i, eflags; 1200 1201 for(i = 0; i < 2; i++) { 1202 op0 = 0x123456789abcdLL; 1203 eax = i2l(op0 & 0xffffffff); 1204 edx = i2l(op0 >> 32); 1205 if (i == 0) 1206 op1 = 0xfbca765423456LL; 1207 else 1208 op1 = op0; 1209 op2 = 0x6532432432434LL; 1210 asm("cmpxchg8b %2\n" 1211 "pushf\n" 1212 "pop %3\n" 1213 : "=a" (eax), "=d" (edx), "=m" (op1), "=g" (eflags) 1214 : "0" (eax), "1" (edx), "m" (op1), "b" ((int)op2), "c" ((int)(op2 >> 32))); 1215 printf("cmpxchg8b: eax=" FMTLX " edx=" FMTLX " op1=" FMT64X " CC=%02lx\n", 1216 eax, edx, op1, eflags & CC_Z); 1217 } 1218 } 1219 } 1220 1221 #ifdef TEST_SEGS 1222 /**********************************************/ 1223 /* segmentation tests */ 1224 1225 #include <sys/syscall.h> 1226 #include <unistd.h> 1227 #include <asm/ldt.h> 1228 #include <linux/version.h> 1229 1230 static inline int modify_ldt(int func, void * ptr, unsigned long bytecount) 1231 { 1232 return syscall(__NR_modify_ldt, func, ptr, bytecount); 1233 } 1234 1235 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 66) 1236 #define modify_ldt_ldt_s user_desc 1237 #endif 1238 1239 #define MK_SEL(n) (((n) << 3) | 7) 1240 1241 uint8_t seg_data1[4096]; 1242 uint8_t seg_data2[4096]; 1243 1244 #define TEST_LR(op, size, seg, mask)\ 1245 {\ 1246 int res, res2;\ 1247 uint16_t mseg = seg;\ 1248 res = 0x12345678;\ 1249 asm (op " %" size "2, %" size "0\n" \ 1250 "movl $0, %1\n"\ 1251 "jnz 1f\n"\ 1252 "movl $1, %1\n"\ 1253 "1:\n"\ 1254 : "=r" (res), "=r" (res2) : "m" (mseg), "0" (res));\ 1255 printf(op ": Z=%d %08x\n", res2, res & ~(mask));\ 1256 } 1257 1258 #define TEST_ARPL(op, size, op1, op2)\ 1259 {\ 1260 long a, b, c; \ 1261 a = (op1); \ 1262 b = (op2); \ 1263 asm volatile(op " %" size "3, %" size "0\n"\ 1264 "movl $0,%1\n"\ 1265 "jnz 1f\n"\ 1266 "movl $1,%1\n"\ 1267 "1:\n"\ 1268 : "=r" (a), "=r" (c) : "0" (a), "r" (b)); \ 1269 printf(op size " A=" FMTLX " B=" FMTLX " R=" FMTLX " z=%ld\n",\ 1270 (long)(op1), (long)(op2), a, c);\ 1271 } 1272 1273 /* NOTE: we use Linux modify_ldt syscall */ 1274 void test_segs(void) 1275 { 1276 struct modify_ldt_ldt_s ldt; 1277 long long ldt_table[3]; 1278 int res, res2; 1279 char tmp; 1280 struct { 1281 uint32_t offset; 1282 uint16_t seg; 1283 } __attribute__((packed)) segoff; 1284 1285 ldt.entry_number = 1; 1286 ldt.base_addr = (unsigned long)&seg_data1; 1287 ldt.limit = (sizeof(seg_data1) + 0xfff) >> 12; 1288 ldt.seg_32bit = 1; 1289 ldt.contents = MODIFY_LDT_CONTENTS_DATA; 1290 ldt.read_exec_only = 0; 1291 ldt.limit_in_pages = 1; 1292 ldt.seg_not_present = 0; 1293 ldt.useable = 1; 1294 modify_ldt(1, &ldt, sizeof(ldt)); /* write ldt entry */ 1295 1296 ldt.entry_number = 2; 1297 ldt.base_addr = (unsigned long)&seg_data2; 1298 ldt.limit = (sizeof(seg_data2) + 0xfff) >> 12; 1299 ldt.seg_32bit = 1; 1300 ldt.contents = MODIFY_LDT_CONTENTS_DATA; 1301 ldt.read_exec_only = 0; 1302 ldt.limit_in_pages = 1; 1303 ldt.seg_not_present = 0; 1304 ldt.useable = 1; 1305 modify_ldt(1, &ldt, sizeof(ldt)); /* write ldt entry */ 1306 1307 modify_ldt(0, &ldt_table, sizeof(ldt_table)); /* read ldt entries */ 1308 #if 0 1309 { 1310 int i; 1311 for(i=0;i<3;i++) 1312 printf("%d: %016Lx\n", i, ldt_table[i]); 1313 } 1314 #endif 1315 /* do some tests with fs or gs */ 1316 asm volatile ("movl %0, %%fs" : : "r" (MK_SEL(1))); 1317 1318 seg_data1[1] = 0xaa; 1319 seg_data2[1] = 0x55; 1320 1321 asm volatile ("fs movzbl 0x1, %0" : "=r" (res)); 1322 printf("FS[1] = %02x\n", res); 1323 1324 asm volatile ("pushl %%gs\n" 1325 "movl %1, %%gs\n" 1326 "gs movzbl 0x1, %0\n" 1327 "popl %%gs\n" 1328 : "=r" (res) 1329 : "r" (MK_SEL(2))); 1330 printf("GS[1] = %02x\n", res); 1331 1332 /* tests with ds/ss (implicit segment case) */ 1333 tmp = 0xa5; 1334 asm volatile ("pushl %%ebp\n\t" 1335 "pushl %%ds\n\t" 1336 "movl %2, %%ds\n\t" 1337 "movl %3, %%ebp\n\t" 1338 "movzbl 0x1, %0\n\t" 1339 "movzbl (%%ebp), %1\n\t" 1340 "popl %%ds\n\t" 1341 "popl %%ebp\n\t" 1342 : "=r" (res), "=r" (res2) 1343 : "r" (MK_SEL(1)), "r" (&tmp)); 1344 printf("DS[1] = %02x\n", res); 1345 printf("SS[tmp] = %02x\n", res2); 1346 1347 segoff.seg = MK_SEL(2); 1348 segoff.offset = 0xabcdef12; 1349 asm volatile("lfs %2, %0\n\t" 1350 "movl %%fs, %1\n\t" 1351 : "=r" (res), "=g" (res2) 1352 : "m" (segoff)); 1353 printf("FS:reg = %04x:%08x\n", res2, res); 1354 1355 TEST_LR("larw", "w", MK_SEL(2), 0x0100); 1356 TEST_LR("larl", "", MK_SEL(2), 0x0100); 1357 TEST_LR("lslw", "w", MK_SEL(2), 0); 1358 TEST_LR("lsll", "", MK_SEL(2), 0); 1359 1360 TEST_LR("larw", "w", 0xfff8, 0); 1361 TEST_LR("larl", "", 0xfff8, 0); 1362 TEST_LR("lslw", "w", 0xfff8, 0); 1363 TEST_LR("lsll", "", 0xfff8, 0); 1364 1365 TEST_ARPL("arpl", "w", 0x12345678 | 3, 0x762123c | 1); 1366 TEST_ARPL("arpl", "w", 0x12345678 | 1, 0x762123c | 3); 1367 TEST_ARPL("arpl", "w", 0x12345678 | 1, 0x762123c | 1); 1368 } 1369 1370 /* 16 bit code test */ 1371 extern char code16_start, code16_end; 1372 extern char code16_func1; 1373 extern char code16_func2; 1374 extern char code16_func3; 1375 1376 void test_code16(void) 1377 { 1378 struct modify_ldt_ldt_s ldt; 1379 int res, res2; 1380 1381 /* build a code segment */ 1382 ldt.entry_number = 1; 1383 ldt.base_addr = (unsigned long)&code16_start; 1384 ldt.limit = &code16_end - &code16_start; 1385 ldt.seg_32bit = 0; 1386 ldt.contents = MODIFY_LDT_CONTENTS_CODE; 1387 ldt.read_exec_only = 0; 1388 ldt.limit_in_pages = 0; 1389 ldt.seg_not_present = 0; 1390 ldt.useable = 1; 1391 modify_ldt(1, &ldt, sizeof(ldt)); /* write ldt entry */ 1392 1393 /* call the first function */ 1394 asm volatile ("lcall %1, %2" 1395 : "=a" (res) 1396 : "i" (MK_SEL(1)), "i" (&code16_func1): "memory", "cc"); 1397 printf("func1() = 0x%08x\n", res); 1398 asm volatile ("lcall %2, %3" 1399 : "=a" (res), "=c" (res2) 1400 : "i" (MK_SEL(1)), "i" (&code16_func2): "memory", "cc"); 1401 printf("func2() = 0x%08x spdec=%d\n", res, res2); 1402 asm volatile ("lcall %1, %2" 1403 : "=a" (res) 1404 : "i" (MK_SEL(1)), "i" (&code16_func3): "memory", "cc"); 1405 printf("func3() = 0x%08x\n", res); 1406 } 1407 #endif 1408 1409 #if defined(__x86_64__) 1410 asm(".globl func_lret\n" 1411 "func_lret:\n" 1412 "movl $0x87654641, %eax\n" 1413 "lretq\n"); 1414 #else 1415 asm(".globl func_lret\n" 1416 "func_lret:\n" 1417 "movl $0x87654321, %eax\n" 1418 "lret\n" 1419 1420 ".globl func_iret\n" 1421 "func_iret:\n" 1422 "movl $0xabcd4321, %eax\n" 1423 "iret\n"); 1424 #endif 1425 1426 extern char func_lret; 1427 extern char func_iret; 1428 1429 void test_misc(void) 1430 { 1431 char table[256]; 1432 long res, i; 1433 1434 for(i=0;i<256;i++) table[i] = 256 - i; 1435 res = 0x12345678; 1436 asm ("xlat" : "=a" (res) : "b" (table), "0" (res)); 1437 printf("xlat: EAX=" FMTLX "\n", res); 1438 1439 #if defined(__x86_64__) 1440 #if 0 1441 { 1442 /* XXX: see if Intel Core2 and AMD64 behavior really 1443 differ. Here we implemented the Intel way which is not 1444 compatible yet with QEMU. */ 1445 static struct __attribute__((packed)) { 1446 uint64_t offset; 1447 uint16_t seg; 1448 } desc; 1449 long cs_sel; 1450 1451 asm volatile ("mov %%cs, %0" : "=r" (cs_sel)); 1452 1453 asm volatile ("push %1\n" 1454 "call func_lret\n" 1455 : "=a" (res) 1456 : "r" (cs_sel) : "memory", "cc"); 1457 printf("func_lret=" FMTLX "\n", res); 1458 1459 desc.offset = (long)&func_lret; 1460 desc.seg = cs_sel; 1461 1462 asm volatile ("xor %%rax, %%rax\n" 1463 "rex64 lcall *(%%rcx)\n" 1464 : "=a" (res) 1465 : "c" (&desc) 1466 : "memory", "cc"); 1467 printf("func_lret2=" FMTLX "\n", res); 1468 1469 asm volatile ("push %2\n" 1470 "mov $ 1f, %%rax\n" 1471 "push %%rax\n" 1472 "rex64 ljmp *(%%rcx)\n" 1473 "1:\n" 1474 : "=a" (res) 1475 : "c" (&desc), "b" (cs_sel) 1476 : "memory", "cc"); 1477 printf("func_lret3=" FMTLX "\n", res); 1478 } 1479 #endif 1480 #else 1481 asm volatile ("push %%cs ; call %1" 1482 : "=a" (res) 1483 : "m" (func_lret): "memory", "cc"); 1484 printf("func_lret=" FMTLX "\n", res); 1485 1486 asm volatile ("pushf ; push %%cs ; call %1" 1487 : "=a" (res) 1488 : "m" (func_iret): "memory", "cc"); 1489 printf("func_iret=" FMTLX "\n", res); 1490 #endif 1491 1492 #if defined(__x86_64__) 1493 /* specific popl test */ 1494 asm volatile ("push $12345432 ; push $0x9abcdef ; pop (%%rsp) ; pop %0" 1495 : "=g" (res)); 1496 printf("popl esp=" FMTLX "\n", res); 1497 #else 1498 /* specific popl test */ 1499 asm volatile ("pushl $12345432 ; pushl $0x9abcdef ; popl (%%esp) ; popl %0" 1500 : "=g" (res)); 1501 printf("popl esp=" FMTLX "\n", res); 1502 1503 /* specific popw test */ 1504 asm volatile ("pushl $12345432 ; pushl $0x9abcdef ; popw (%%esp) ; addl $2, %%esp ; popl %0" 1505 : "=g" (res)); 1506 printf("popw esp=" FMTLX "\n", res); 1507 #endif 1508 } 1509 1510 uint8_t str_buffer[4096]; 1511 1512 #define TEST_STRING1(OP, size, DF, REP)\ 1513 {\ 1514 long esi, edi, eax, ecx, eflags;\ 1515 \ 1516 esi = (long)(str_buffer + sizeof(str_buffer) / 2);\ 1517 edi = (long)(str_buffer + sizeof(str_buffer) / 2) + 16;\ 1518 eax = i2l(0x12345678);\ 1519 ecx = 17;\ 1520 \ 1521 asm volatile ("push $0\n\t"\ 1522 "popf\n\t"\ 1523 DF "\n\t"\ 1524 REP #OP size "\n\t"\ 1525 "cld\n\t"\ 1526 "pushf\n\t"\ 1527 "pop %4\n\t"\ 1528 : "=S" (esi), "=D" (edi), "=a" (eax), "=c" (ecx), "=g" (eflags)\ 1529 : "0" (esi), "1" (edi), "2" (eax), "3" (ecx));\ 1530 printf("%-10s ESI=" FMTLX " EDI=" FMTLX " EAX=" FMTLX " ECX=" FMTLX " EFL=%04x\n",\ 1531 REP #OP size, esi, edi, eax, ecx,\ 1532 (int)(eflags & (CC_C | CC_P | CC_Z | CC_S | CC_O | CC_A)));\ 1533 } 1534 1535 #define TEST_STRING(OP, REP)\ 1536 TEST_STRING1(OP, "b", "", REP);\ 1537 TEST_STRING1(OP, "w", "", REP);\ 1538 TEST_STRING1(OP, "l", "", REP);\ 1539 X86_64_ONLY(TEST_STRING1(OP, "q", "", REP));\ 1540 TEST_STRING1(OP, "b", "std", REP);\ 1541 TEST_STRING1(OP, "w", "std", REP);\ 1542 TEST_STRING1(OP, "l", "std", REP);\ 1543 X86_64_ONLY(TEST_STRING1(OP, "q", "std", REP)) 1544 1545 void test_string(void) 1546 { 1547 int i; 1548 for(i = 0;i < sizeof(str_buffer); i++) 1549 str_buffer[i] = i + 0x56; 1550 TEST_STRING(stos, ""); 1551 TEST_STRING(stos, "rep "); 1552 TEST_STRING(lods, ""); /* to verify stos */ 1553 TEST_STRING(lods, "rep "); 1554 TEST_STRING(movs, ""); 1555 TEST_STRING(movs, "rep "); 1556 TEST_STRING(lods, ""); /* to verify stos */ 1557 1558 /* XXX: better tests */ 1559 TEST_STRING(scas, ""); 1560 TEST_STRING(scas, "repz "); 1561 TEST_STRING(scas, "repnz "); 1562 TEST_STRING(cmps, ""); 1563 TEST_STRING(cmps, "repz "); 1564 TEST_STRING(cmps, "repnz "); 1565 } 1566 1567 #ifdef TEST_VM86 1568 /* VM86 test */ 1569 1570 static inline void set_bit(uint8_t *a, unsigned int bit) 1571 { 1572 a[bit / 8] |= (1 << (bit % 8)); 1573 } 1574 1575 static inline uint8_t *seg_to_linear(unsigned int seg, unsigned int reg) 1576 { 1577 return (uint8_t *)((seg << 4) + (reg & 0xffff)); 1578 } 1579 1580 static inline void pushw(struct vm86_regs *r, int val) 1581 { 1582 r->esp = (r->esp & ~0xffff) | ((r->esp - 2) & 0xffff); 1583 *(uint16_t *)seg_to_linear(r->ss, r->esp) = val; 1584 } 1585 1586 static inline int vm86(int func, struct vm86plus_struct *v86) 1587 { 1588 return syscall(__NR_vm86, func, v86); 1589 } 1590 1591 extern char vm86_code_start; 1592 extern char vm86_code_end; 1593 1594 #define VM86_CODE_CS 0x100 1595 #define VM86_CODE_IP 0x100 1596 1597 void test_vm86(void) 1598 { 1599 struct vm86plus_struct ctx; 1600 struct vm86_regs *r; 1601 uint8_t *vm86_mem; 1602 int seg, ret; 1603 1604 vm86_mem = mmap((void *)0x00000000, 0x110000, 1605 PROT_WRITE | PROT_READ | PROT_EXEC, 1606 MAP_FIXED | MAP_ANON | MAP_PRIVATE, -1, 0); 1607 if (vm86_mem == MAP_FAILED) { 1608 printf("ERROR: could not map vm86 memory"); 1609 return; 1610 } 1611 memset(&ctx, 0, sizeof(ctx)); 1612 1613 /* init basic registers */ 1614 r = &ctx.regs; 1615 r->eip = VM86_CODE_IP; 1616 r->esp = 0xfffe; 1617 seg = VM86_CODE_CS; 1618 r->cs = seg; 1619 r->ss = seg; 1620 r->ds = seg; 1621 r->es = seg; 1622 r->fs = seg; 1623 r->gs = seg; 1624 r->eflags = VIF_MASK; 1625 1626 /* move code to proper address. We use the same layout as a .com 1627 dos program. */ 1628 memcpy(vm86_mem + (VM86_CODE_CS << 4) + VM86_CODE_IP, 1629 &vm86_code_start, &vm86_code_end - &vm86_code_start); 1630 1631 /* mark int 0x21 as being emulated */ 1632 set_bit((uint8_t *)&ctx.int_revectored, 0x21); 1633 1634 for(;;) { 1635 ret = vm86(VM86_ENTER, &ctx); 1636 switch(VM86_TYPE(ret)) { 1637 case VM86_INTx: 1638 { 1639 int int_num, ah, v; 1640 1641 int_num = VM86_ARG(ret); 1642 if (int_num != 0x21) 1643 goto unknown_int; 1644 ah = (r->eax >> 8) & 0xff; 1645 switch(ah) { 1646 case 0x00: /* exit */ 1647 goto the_end; 1648 case 0x02: /* write char */ 1649 { 1650 uint8_t c = r->edx; 1651 putchar(c); 1652 } 1653 break; 1654 case 0x09: /* write string */ 1655 { 1656 uint8_t c, *ptr; 1657 ptr = seg_to_linear(r->ds, r->edx); 1658 for(;;) { 1659 c = *ptr++; 1660 if (c == '$') 1661 break; 1662 putchar(c); 1663 } 1664 r->eax = (r->eax & ~0xff) | '$'; 1665 } 1666 break; 1667 case 0xff: /* extension: write eflags number in edx */ 1668 v = (int)r->edx; 1669 #ifndef LINUX_VM86_IOPL_FIX 1670 v &= ~0x3000; 1671 #endif 1672 printf("%08x\n", v); 1673 break; 1674 default: 1675 unknown_int: 1676 printf("unsupported int 0x%02x\n", int_num); 1677 goto the_end; 1678 } 1679 } 1680 break; 1681 case VM86_SIGNAL: 1682 /* a signal came, we just ignore that */ 1683 break; 1684 case VM86_STI: 1685 break; 1686 default: 1687 printf("ERROR: unhandled vm86 return code (0x%x)\n", ret); 1688 goto the_end; 1689 } 1690 } 1691 the_end: 1692 printf("VM86 end\n"); 1693 munmap(vm86_mem, 0x110000); 1694 } 1695 #endif 1696 1697 /* exception tests */ 1698 #if defined(__i386__) && !defined(REG_EAX) 1699 #define REG_EAX EAX 1700 #define REG_EBX EBX 1701 #define REG_ECX ECX 1702 #define REG_EDX EDX 1703 #define REG_ESI ESI 1704 #define REG_EDI EDI 1705 #define REG_EBP EBP 1706 #define REG_ESP ESP 1707 #define REG_EIP EIP 1708 #define REG_EFL EFL 1709 #define REG_TRAPNO TRAPNO 1710 #define REG_ERR ERR 1711 #endif 1712 1713 #if defined(__x86_64__) 1714 #define REG_EIP REG_RIP 1715 #endif 1716 1717 jmp_buf jmp_env; 1718 int v1; 1719 int tab[2]; 1720 1721 void sig_handler(int sig, siginfo_t *info, void *puc) 1722 { 1723 struct ucontext *uc = puc; 1724 1725 printf("si_signo=%d si_errno=%d si_code=%d", 1726 info->si_signo, info->si_errno, info->si_code); 1727 printf(" si_addr=0x%08lx", 1728 (unsigned long)info->si_addr); 1729 printf("\n"); 1730 1731 printf("trapno=" FMTLX " err=" FMTLX, 1732 (long)uc->uc_mcontext.gregs[REG_TRAPNO], 1733 (long)uc->uc_mcontext.gregs[REG_ERR]); 1734 printf(" EIP=" FMTLX, (long)uc->uc_mcontext.gregs[REG_EIP]); 1735 printf("\n"); 1736 longjmp(jmp_env, 1); 1737 } 1738 1739 void test_exceptions(void) 1740 { 1741 struct sigaction act; 1742 volatile int val; 1743 1744 act.sa_sigaction = sig_handler; 1745 sigemptyset(&act.sa_mask); 1746 act.sa_flags = SA_SIGINFO | SA_NODEFER; 1747 sigaction(SIGFPE, &act, NULL); 1748 sigaction(SIGILL, &act, NULL); 1749 sigaction(SIGSEGV, &act, NULL); 1750 sigaction(SIGBUS, &act, NULL); 1751 sigaction(SIGTRAP, &act, NULL); 1752 1753 /* test division by zero reporting */ 1754 printf("DIVZ exception:\n"); 1755 if (setjmp(jmp_env) == 0) { 1756 /* now divide by zero */ 1757 v1 = 0; 1758 v1 = 2 / v1; 1759 } 1760 1761 #if !defined(__x86_64__) 1762 printf("BOUND exception:\n"); 1763 if (setjmp(jmp_env) == 0) { 1764 /* bound exception */ 1765 tab[0] = 1; 1766 tab[1] = 10; 1767 asm volatile ("bound %0, %1" : : "r" (11), "m" (tab[0])); 1768 } 1769 #endif 1770 1771 #ifdef TEST_SEGS 1772 printf("segment exceptions:\n"); 1773 if (setjmp(jmp_env) == 0) { 1774 /* load an invalid segment */ 1775 asm volatile ("movl %0, %%fs" : : "r" ((0x1234 << 3) | 1)); 1776 } 1777 if (setjmp(jmp_env) == 0) { 1778 /* null data segment is valid */ 1779 asm volatile ("movl %0, %%fs" : : "r" (3)); 1780 /* null stack segment */ 1781 asm volatile ("movl %0, %%ss" : : "r" (3)); 1782 } 1783 1784 { 1785 struct modify_ldt_ldt_s ldt; 1786 ldt.entry_number = 1; 1787 ldt.base_addr = (unsigned long)&seg_data1; 1788 ldt.limit = (sizeof(seg_data1) + 0xfff) >> 12; 1789 ldt.seg_32bit = 1; 1790 ldt.contents = MODIFY_LDT_CONTENTS_DATA; 1791 ldt.read_exec_only = 0; 1792 ldt.limit_in_pages = 1; 1793 ldt.seg_not_present = 1; 1794 ldt.useable = 1; 1795 modify_ldt(1, &ldt, sizeof(ldt)); /* write ldt entry */ 1796 1797 if (setjmp(jmp_env) == 0) { 1798 /* segment not present */ 1799 asm volatile ("movl %0, %%fs" : : "r" (MK_SEL(1))); 1800 } 1801 } 1802 #endif 1803 1804 /* test SEGV reporting */ 1805 printf("PF exception:\n"); 1806 if (setjmp(jmp_env) == 0) { 1807 val = 1; 1808 /* we add a nop to test a weird PC retrieval case */ 1809 asm volatile ("nop"); 1810 /* now store in an invalid address */ 1811 *(char *)0x1234 = 1; 1812 } 1813 1814 /* test SEGV reporting */ 1815 printf("PF exception:\n"); 1816 if (setjmp(jmp_env) == 0) { 1817 val = 1; 1818 /* read from an invalid address */ 1819 v1 = *(char *)0x1234; 1820 } 1821 1822 /* test illegal instruction reporting */ 1823 printf("UD2 exception:\n"); 1824 if (setjmp(jmp_env) == 0) { 1825 /* now execute an invalid instruction */ 1826 asm volatile("ud2"); 1827 } 1828 printf("lock nop exception:\n"); 1829 if (setjmp(jmp_env) == 0) { 1830 /* now execute an invalid instruction */ 1831 asm volatile("lock nop"); 1832 } 1833 1834 printf("INT exception:\n"); 1835 if (setjmp(jmp_env) == 0) { 1836 asm volatile ("int $0xfd"); 1837 } 1838 if (setjmp(jmp_env) == 0) { 1839 asm volatile ("int $0x01"); 1840 } 1841 if (setjmp(jmp_env) == 0) { 1842 asm volatile (".byte 0xcd, 0x03"); 1843 } 1844 if (setjmp(jmp_env) == 0) { 1845 asm volatile ("int $0x04"); 1846 } 1847 if (setjmp(jmp_env) == 0) { 1848 asm volatile ("int $0x05"); 1849 } 1850 1851 printf("INT3 exception:\n"); 1852 if (setjmp(jmp_env) == 0) { 1853 asm volatile ("int3"); 1854 } 1855 1856 printf("CLI exception:\n"); 1857 if (setjmp(jmp_env) == 0) { 1858 asm volatile ("cli"); 1859 } 1860 1861 printf("STI exception:\n"); 1862 if (setjmp(jmp_env) == 0) { 1863 asm volatile ("cli"); 1864 } 1865 1866 #if !defined(__x86_64__) 1867 printf("INTO exception:\n"); 1868 if (setjmp(jmp_env) == 0) { 1869 /* overflow exception */ 1870 asm volatile ("addl $1, %0 ; into" : : "r" (0x7fffffff)); 1871 } 1872 #endif 1873 1874 printf("OUTB exception:\n"); 1875 if (setjmp(jmp_env) == 0) { 1876 asm volatile ("outb %%al, %%dx" : : "d" (0x4321), "a" (0)); 1877 } 1878 1879 printf("INB exception:\n"); 1880 if (setjmp(jmp_env) == 0) { 1881 asm volatile ("inb %%dx, %%al" : "=a" (val) : "d" (0x4321)); 1882 } 1883 1884 printf("REP OUTSB exception:\n"); 1885 if (setjmp(jmp_env) == 0) { 1886 asm volatile ("rep outsb" : : "d" (0x4321), "S" (tab), "c" (1)); 1887 } 1888 1889 printf("REP INSB exception:\n"); 1890 if (setjmp(jmp_env) == 0) { 1891 asm volatile ("rep insb" : : "d" (0x4321), "D" (tab), "c" (1)); 1892 } 1893 1894 printf("HLT exception:\n"); 1895 if (setjmp(jmp_env) == 0) { 1896 asm volatile ("hlt"); 1897 } 1898 1899 printf("single step exception:\n"); 1900 val = 0; 1901 if (setjmp(jmp_env) == 0) { 1902 asm volatile ("pushf\n" 1903 "orl $0x00100, (%%esp)\n" 1904 "popf\n" 1905 "movl $0xabcd, %0\n" 1906 "movl $0x0, %0\n" : "=m" (val) : : "cc", "memory"); 1907 } 1908 printf("val=0x%x\n", val); 1909 } 1910 1911 #if !defined(__x86_64__) 1912 /* specific precise single step test */ 1913 void sig_trap_handler(int sig, siginfo_t *info, void *puc) 1914 { 1915 struct ucontext *uc = puc; 1916 printf("EIP=" FMTLX "\n", (long)uc->uc_mcontext.gregs[REG_EIP]); 1917 } 1918 1919 const uint8_t sstep_buf1[4] = { 1, 2, 3, 4}; 1920 uint8_t sstep_buf2[4]; 1921 1922 void test_single_step(void) 1923 { 1924 struct sigaction act; 1925 volatile int val; 1926 int i; 1927 1928 val = 0; 1929 act.sa_sigaction = sig_trap_handler; 1930 sigemptyset(&act.sa_mask); 1931 act.sa_flags = SA_SIGINFO; 1932 sigaction(SIGTRAP, &act, NULL); 1933 asm volatile ("pushf\n" 1934 "orl $0x00100, (%%esp)\n" 1935 "popf\n" 1936 "movl $0xabcd, %0\n" 1937 1938 /* jmp test */ 1939 "movl $3, %%ecx\n" 1940 "1:\n" 1941 "addl $1, %0\n" 1942 "decl %%ecx\n" 1943 "jnz 1b\n" 1944 1945 /* movsb: the single step should stop at each movsb iteration */ 1946 "movl $sstep_buf1, %%esi\n" 1947 "movl $sstep_buf2, %%edi\n" 1948 "movl $0, %%ecx\n" 1949 "rep movsb\n" 1950 "movl $3, %%ecx\n" 1951 "rep movsb\n" 1952 "movl $1, %%ecx\n" 1953 "rep movsb\n" 1954 1955 /* cmpsb: the single step should stop at each cmpsb iteration */ 1956 "movl $sstep_buf1, %%esi\n" 1957 "movl $sstep_buf2, %%edi\n" 1958 "movl $0, %%ecx\n" 1959 "rep cmpsb\n" 1960 "movl $4, %%ecx\n" 1961 "rep cmpsb\n" 1962 1963 /* getpid() syscall: single step should skip one 1964 instruction */ 1965 "movl $20, %%eax\n" 1966 "int $0x80\n" 1967 "movl $0, %%eax\n" 1968 1969 /* when modifying SS, trace is not done on the next 1970 instruction */ 1971 "movl %%ss, %%ecx\n" 1972 "movl %%ecx, %%ss\n" 1973 "addl $1, %0\n" 1974 "movl $1, %%eax\n" 1975 "movl %%ecx, %%ss\n" 1976 "jmp 1f\n" 1977 "addl $1, %0\n" 1978 "1:\n" 1979 "movl $1, %%eax\n" 1980 "pushl %%ecx\n" 1981 "popl %%ss\n" 1982 "addl $1, %0\n" 1983 "movl $1, %%eax\n" 1984 1985 "pushf\n" 1986 "andl $~0x00100, (%%esp)\n" 1987 "popf\n" 1988 : "=m" (val) 1989 : 1990 : "cc", "memory", "eax", "ecx", "esi", "edi"); 1991 printf("val=%d\n", val); 1992 for(i = 0; i < 4; i++) 1993 printf("sstep_buf2[%d] = %d\n", i, sstep_buf2[i]); 1994 } 1995 1996 /* self modifying code test */ 1997 uint8_t code[] = { 1998 0xb8, 0x1, 0x00, 0x00, 0x00, /* movl $1, %eax */ 1999 0xc3, /* ret */ 2000 }; 2001 2002 asm(".section \".data\"\n" 2003 "smc_code2:\n" 2004 "movl 4(%esp), %eax\n" 2005 "movl %eax, smc_patch_addr2 + 1\n" 2006 "nop\n" 2007 "nop\n" 2008 "nop\n" 2009 "nop\n" 2010 "nop\n" 2011 "nop\n" 2012 "nop\n" 2013 "nop\n" 2014 "smc_patch_addr2:\n" 2015 "movl $1, %eax\n" 2016 "ret\n" 2017 ".previous\n" 2018 ); 2019 2020 typedef int FuncType(void); 2021 extern int smc_code2(int); 2022 void test_self_modifying_code(void) 2023 { 2024 int i; 2025 printf("self modifying code:\n"); 2026 printf("func1 = 0x%x\n", ((FuncType *)code)()); 2027 for(i = 2; i <= 4; i++) { 2028 code[1] = i; 2029 printf("func%d = 0x%x\n", i, ((FuncType *)code)()); 2030 } 2031 2032 /* more difficult test : the modified code is just after the 2033 modifying instruction. It is forbidden in Intel specs, but it 2034 is used by old DOS programs */ 2035 for(i = 2; i <= 4; i++) { 2036 printf("smc_code2(%d) = %d\n", i, smc_code2(i)); 2037 } 2038 } 2039 #endif 2040 2041 long enter_stack[4096]; 2042 2043 #if defined(__x86_64__) 2044 #define RSP "%%rsp" 2045 #define RBP "%%rbp" 2046 #else 2047 #define RSP "%%esp" 2048 #define RBP "%%ebp" 2049 #endif 2050 2051 #define TEST_ENTER(size, stack_type, level)\ 2052 {\ 2053 long esp_save, esp_val, ebp_val, ebp_save, i;\ 2054 stack_type *ptr, *stack_end, *stack_ptr;\ 2055 memset(enter_stack, 0, sizeof(enter_stack));\ 2056 stack_end = stack_ptr = (stack_type *)(enter_stack + 4096);\ 2057 ebp_val = (long)stack_ptr;\ 2058 for(i=1;i<=32;i++)\ 2059 *--stack_ptr = i;\ 2060 esp_val = (long)stack_ptr;\ 2061 asm("mov " RSP ", %[esp_save]\n"\ 2062 "mov " RBP ", %[ebp_save]\n"\ 2063 "mov %[esp_val], " RSP "\n"\ 2064 "mov %[ebp_val], " RBP "\n"\ 2065 "enter" size " $8, $" #level "\n"\ 2066 "mov " RSP ", %[esp_val]\n"\ 2067 "mov " RBP ", %[ebp_val]\n"\ 2068 "mov %[esp_save], " RSP "\n"\ 2069 "mov %[ebp_save], " RBP "\n"\ 2070 : [esp_save] "=r" (esp_save),\ 2071 [ebp_save] "=r" (ebp_save),\ 2072 [esp_val] "=r" (esp_val),\ 2073 [ebp_val] "=r" (ebp_val)\ 2074 : "[esp_val]" (esp_val),\ 2075 "[ebp_val]" (ebp_val));\ 2076 printf("level=%d:\n", level);\ 2077 printf("esp_val=" FMTLX "\n", esp_val - (long)stack_end);\ 2078 printf("ebp_val=" FMTLX "\n", ebp_val - (long)stack_end);\ 2079 for(ptr = (stack_type *)esp_val; ptr < stack_end; ptr++)\ 2080 printf(FMTLX "\n", (long)ptr[0]);\ 2081 } 2082 2083 static void test_enter(void) 2084 { 2085 #if defined(__x86_64__) 2086 TEST_ENTER("q", uint64_t, 0); 2087 TEST_ENTER("q", uint64_t, 1); 2088 TEST_ENTER("q", uint64_t, 2); 2089 TEST_ENTER("q", uint64_t, 31); 2090 #else 2091 TEST_ENTER("l", uint32_t, 0); 2092 TEST_ENTER("l", uint32_t, 1); 2093 TEST_ENTER("l", uint32_t, 2); 2094 TEST_ENTER("l", uint32_t, 31); 2095 #endif 2096 2097 TEST_ENTER("w", uint16_t, 0); 2098 TEST_ENTER("w", uint16_t, 1); 2099 TEST_ENTER("w", uint16_t, 2); 2100 TEST_ENTER("w", uint16_t, 31); 2101 } 2102 2103 #ifdef TEST_SSE 2104 2105 typedef int __m64 __attribute__ ((__mode__ (__V2SI__))); 2106 typedef float __m128 __attribute__ ((__mode__(__V4SF__))); 2107 2108 typedef union { 2109 double d[2]; 2110 float s[4]; 2111 uint32_t l[4]; 2112 uint64_t q[2]; 2113 __m128 dq; 2114 } XMMReg; 2115 2116 static uint64_t __attribute__((aligned(16))) test_values[4][2] = { 2117 { 0x456723c698694873, 0xdc515cff944a58ec }, 2118 { 0x1f297ccd58bad7ab, 0x41f21efba9e3e146 }, 2119 { 0x007c62c2085427f8, 0x231be9e8cde7438d }, 2120 { 0x0f76255a085427f8, 0xc233e9e8c4c9439a }, 2121 }; 2122 2123 #define SSE_OP(op)\ 2124 {\ 2125 asm volatile (#op " %2, %0" : "=x" (r.dq) : "0" (a.dq), "x" (b.dq));\ 2126 printf("%-9s: a=" FMT64X "" FMT64X " b=" FMT64X "" FMT64X " r=" FMT64X "" FMT64X "\n",\ 2127 #op,\ 2128 a.q[1], a.q[0],\ 2129 b.q[1], b.q[0],\ 2130 r.q[1], r.q[0]);\ 2131 } 2132 2133 #define SSE_OP2(op)\ 2134 {\ 2135 int i;\ 2136 for(i=0;i<2;i++) {\ 2137 a.q[0] = test_values[2*i][0];\ 2138 a.q[1] = test_values[2*i][1];\ 2139 b.q[0] = test_values[2*i+1][0];\ 2140 b.q[1] = test_values[2*i+1][1];\ 2141 SSE_OP(op);\ 2142 }\ 2143 } 2144 2145 #define MMX_OP2(op)\ 2146 {\ 2147 int i;\ 2148 for(i=0;i<2;i++) {\ 2149 a.q[0] = test_values[2*i][0];\ 2150 b.q[0] = test_values[2*i+1][0];\ 2151 asm volatile (#op " %2, %0" : "=y" (r.q[0]) : "0" (a.q[0]), "y" (b.q[0]));\ 2152 printf("%-9s: a=" FMT64X " b=" FMT64X " r=" FMT64X "\n",\ 2153 #op,\ 2154 a.q[0],\ 2155 b.q[0],\ 2156 r.q[0]);\ 2157 }\ 2158 SSE_OP2(op);\ 2159 } 2160 2161 #define SHUF_OP(op, ib)\ 2162 {\ 2163 a.q[0] = test_values[0][0];\ 2164 a.q[1] = test_values[0][1];\ 2165 b.q[0] = test_values[1][0];\ 2166 b.q[1] = test_values[1][1];\ 2167 asm volatile (#op " $" #ib ", %2, %0" : "=x" (r.dq) : "0" (a.dq), "x" (b.dq));\ 2168 printf("%-9s: a=" FMT64X "" FMT64X " b=" FMT64X "" FMT64X " ib=%02x r=" FMT64X "" FMT64X "\n",\ 2169 #op,\ 2170 a.q[1], a.q[0],\ 2171 b.q[1], b.q[0],\ 2172 ib,\ 2173 r.q[1], r.q[0]);\ 2174 } 2175 2176 #define PSHUF_OP(op, ib)\ 2177 {\ 2178 int i;\ 2179 for(i=0;i<2;i++) {\ 2180 a.q[0] = test_values[2*i][0];\ 2181 a.q[1] = test_values[2*i][1];\ 2182 asm volatile (#op " $" #ib ", %1, %0" : "=x" (r.dq) : "x" (a.dq));\ 2183 printf("%-9s: a=" FMT64X "" FMT64X " ib=%02x r=" FMT64X "" FMT64X "\n",\ 2184 #op,\ 2185 a.q[1], a.q[0],\ 2186 ib,\ 2187 r.q[1], r.q[0]);\ 2188 }\ 2189 } 2190 2191 #define SHIFT_IM(op, ib)\ 2192 {\ 2193 int i;\ 2194 for(i=0;i<2;i++) {\ 2195 a.q[0] = test_values[2*i][0];\ 2196 a.q[1] = test_values[2*i][1];\ 2197 asm volatile (#op " $" #ib ", %0" : "=x" (r.dq) : "0" (a.dq));\ 2198 printf("%-9s: a=" FMT64X "" FMT64X " ib=%02x r=" FMT64X "" FMT64X "\n",\ 2199 #op,\ 2200 a.q[1], a.q[0],\ 2201 ib,\ 2202 r.q[1], r.q[0]);\ 2203 }\ 2204 } 2205 2206 #define SHIFT_OP(op, ib)\ 2207 {\ 2208 int i;\ 2209 SHIFT_IM(op, ib);\ 2210 for(i=0;i<2;i++) {\ 2211 a.q[0] = test_values[2*i][0];\ 2212 a.q[1] = test_values[2*i][1];\ 2213 b.q[0] = ib;\ 2214 b.q[1] = 0;\ 2215 asm volatile (#op " %2, %0" : "=x" (r.dq) : "0" (a.dq), "x" (b.dq));\ 2216 printf("%-9s: a=" FMT64X "" FMT64X " b=" FMT64X "" FMT64X " r=" FMT64X "" FMT64X "\n",\ 2217 #op,\ 2218 a.q[1], a.q[0],\ 2219 b.q[1], b.q[0],\ 2220 r.q[1], r.q[0]);\ 2221 }\ 2222 } 2223 2224 #define MOVMSK(op)\ 2225 {\ 2226 int i, reg;\ 2227 for(i=0;i<2;i++) {\ 2228 a.q[0] = test_values[2*i][0];\ 2229 a.q[1] = test_values[2*i][1];\ 2230 asm volatile (#op " %1, %0" : "=r" (reg) : "x" (a.dq));\ 2231 printf("%-9s: a=" FMT64X "" FMT64X " r=%08x\n",\ 2232 #op,\ 2233 a.q[1], a.q[0],\ 2234 reg);\ 2235 }\ 2236 } 2237 2238 #define SSE_OPS(a) \ 2239 SSE_OP(a ## ps);\ 2240 SSE_OP(a ## ss); 2241 2242 #define SSE_OPD(a) \ 2243 SSE_OP(a ## pd);\ 2244 SSE_OP(a ## sd); 2245 2246 #define SSE_COMI(op, field)\ 2247 {\ 2248 unsigned int eflags;\ 2249 XMMReg a, b;\ 2250 a.field[0] = a1;\ 2251 b.field[0] = b1;\ 2252 asm volatile (#op " %2, %1\n"\ 2253 "pushf\n"\ 2254 "pop %0\n"\ 2255 : "=m" (eflags)\ 2256 : "x" (a.dq), "x" (b.dq));\ 2257 printf("%-9s: a=%f b=%f cc=%04x\n",\ 2258 #op, a1, b1,\ 2259 eflags & (CC_C | CC_P | CC_Z | CC_S | CC_O | CC_A));\ 2260 } 2261 2262 void test_sse_comi(double a1, double b1) 2263 { 2264 SSE_COMI(ucomiss, s); 2265 SSE_COMI(ucomisd, d); 2266 SSE_COMI(comiss, s); 2267 SSE_COMI(comisd, d); 2268 } 2269 2270 #define CVT_OP_XMM(op)\ 2271 {\ 2272 asm volatile (#op " %1, %0" : "=x" (r.dq) : "x" (a.dq));\ 2273 printf("%-9s: a=" FMT64X "" FMT64X " r=" FMT64X "" FMT64X "\n",\ 2274 #op,\ 2275 a.q[1], a.q[0],\ 2276 r.q[1], r.q[0]);\ 2277 } 2278 2279 /* Force %xmm0 usage to avoid the case where both register index are 0 2280 to test intruction decoding more extensively */ 2281 #define CVT_OP_XMM2MMX(op)\ 2282 {\ 2283 asm volatile (#op " %1, %0" : "=y" (r.q[0]) : "x" (a.dq) \ 2284 : "%xmm0"); \ 2285 asm volatile("emms\n"); \ 2286 printf("%-9s: a=" FMT64X "" FMT64X " r=" FMT64X "\n",\ 2287 #op,\ 2288 a.q[1], a.q[0],\ 2289 r.q[0]);\ 2290 } 2291 2292 #define CVT_OP_MMX2XMM(op)\ 2293 {\ 2294 asm volatile (#op " %1, %0" : "=x" (r.dq) : "y" (a.q[0]));\ 2295 asm volatile("emms\n"); \ 2296 printf("%-9s: a=" FMT64X " r=" FMT64X "" FMT64X "\n",\ 2297 #op,\ 2298 a.q[0],\ 2299 r.q[1], r.q[0]);\ 2300 } 2301 2302 #define CVT_OP_REG2XMM(op)\ 2303 {\ 2304 asm volatile (#op " %1, %0" : "=x" (r.dq) : "r" (a.l[0]));\ 2305 printf("%-9s: a=%08x r=" FMT64X "" FMT64X "\n",\ 2306 #op,\ 2307 a.l[0],\ 2308 r.q[1], r.q[0]);\ 2309 } 2310 2311 #define CVT_OP_XMM2REG(op)\ 2312 {\ 2313 asm volatile (#op " %1, %0" : "=r" (r.l[0]) : "x" (a.dq));\ 2314 printf("%-9s: a=" FMT64X "" FMT64X " r=%08x\n",\ 2315 #op,\ 2316 a.q[1], a.q[0],\ 2317 r.l[0]);\ 2318 } 2319 2320 struct fpxstate { 2321 uint16_t fpuc; 2322 uint16_t fpus; 2323 uint16_t fptag; 2324 uint16_t fop; 2325 uint32_t fpuip; 2326 uint16_t cs_sel; 2327 uint16_t dummy0; 2328 uint32_t fpudp; 2329 uint16_t ds_sel; 2330 uint16_t dummy1; 2331 uint32_t mxcsr; 2332 uint32_t mxcsr_mask; 2333 uint8_t fpregs1[8 * 16]; 2334 uint8_t xmm_regs[8 * 16]; 2335 uint8_t dummy2[224]; 2336 }; 2337 2338 static struct fpxstate fpx_state __attribute__((aligned(16))); 2339 static struct fpxstate fpx_state2 __attribute__((aligned(16))); 2340 2341 void test_fxsave(void) 2342 { 2343 struct fpxstate *fp = &fpx_state; 2344 struct fpxstate *fp2 = &fpx_state2; 2345 int i, nb_xmm; 2346 XMMReg a, b; 2347 a.q[0] = test_values[0][0]; 2348 a.q[1] = test_values[0][1]; 2349 b.q[0] = test_values[1][0]; 2350 b.q[1] = test_values[1][1]; 2351 2352 asm("movdqa %2, %%xmm0\n" 2353 "movdqa %3, %%xmm7\n" 2354 #if defined(__x86_64__) 2355 "movdqa %2, %%xmm15\n" 2356 #endif 2357 " fld1\n" 2358 " fldpi\n" 2359 " fldln2\n" 2360 " fxsave %0\n" 2361 " fxrstor %0\n" 2362 " fxsave %1\n" 2363 " fninit\n" 2364 : "=m" (*(uint32_t *)fp2), "=m" (*(uint32_t *)fp) 2365 : "m" (a), "m" (b)); 2366 printf("fpuc=%04x\n", fp->fpuc); 2367 printf("fpus=%04x\n", fp->fpus); 2368 printf("fptag=%04x\n", fp->fptag); 2369 for(i = 0; i < 3; i++) { 2370 printf("ST%d: " FMT64X " %04x\n", 2371 i, 2372 *(uint64_t *)&fp->fpregs1[i * 16], 2373 *(uint16_t *)&fp->fpregs1[i * 16 + 8]); 2374 } 2375 printf("mxcsr=%08x\n", fp->mxcsr & 0x1f80); 2376 #if defined(__x86_64__) 2377 nb_xmm = 16; 2378 #else 2379 nb_xmm = 8; 2380 #endif 2381 for(i = 0; i < nb_xmm; i++) { 2382 printf("xmm%d: " FMT64X "" FMT64X "\n", 2383 i, 2384 *(uint64_t *)&fp->xmm_regs[i * 16], 2385 *(uint64_t *)&fp->xmm_regs[i * 16 + 8]); 2386 } 2387 } 2388 2389 void test_sse(void) 2390 { 2391 XMMReg r, a, b; 2392 int i; 2393 2394 MMX_OP2(punpcklbw); 2395 MMX_OP2(punpcklwd); 2396 MMX_OP2(punpckldq); 2397 MMX_OP2(packsswb); 2398 MMX_OP2(pcmpgtb); 2399 MMX_OP2(pcmpgtw); 2400 MMX_OP2(pcmpgtd); 2401 MMX_OP2(packuswb); 2402 MMX_OP2(punpckhbw); 2403 MMX_OP2(punpckhwd); 2404 MMX_OP2(punpckhdq); 2405 MMX_OP2(packssdw); 2406 MMX_OP2(pcmpeqb); 2407 MMX_OP2(pcmpeqw); 2408 MMX_OP2(pcmpeqd); 2409 2410 MMX_OP2(paddq); 2411 MMX_OP2(pmullw); 2412 MMX_OP2(psubusb); 2413 MMX_OP2(psubusw); 2414 MMX_OP2(pminub); 2415 MMX_OP2(pand); 2416 MMX_OP2(paddusb); 2417 MMX_OP2(paddusw); 2418 MMX_OP2(pmaxub); 2419 MMX_OP2(pandn); 2420 2421 MMX_OP2(pmulhuw); 2422 MMX_OP2(pmulhw); 2423 2424 MMX_OP2(psubsb); 2425 MMX_OP2(psubsw); 2426 MMX_OP2(pminsw); 2427 MMX_OP2(por); 2428 MMX_OP2(paddsb); 2429 MMX_OP2(paddsw); 2430 MMX_OP2(pmaxsw); 2431 MMX_OP2(pxor); 2432 MMX_OP2(pmuludq); 2433 MMX_OP2(pmaddwd); 2434 MMX_OP2(psadbw); 2435 MMX_OP2(psubb); 2436 MMX_OP2(psubw); 2437 MMX_OP2(psubd); 2438 MMX_OP2(psubq); 2439 MMX_OP2(paddb); 2440 MMX_OP2(paddw); 2441 MMX_OP2(paddd); 2442 2443 MMX_OP2(pavgb); 2444 MMX_OP2(pavgw); 2445 2446 asm volatile ("pinsrw $1, %1, %0" : "=y" (r.q[0]) : "r" (0x12345678)); 2447 printf("%-9s: r=" FMT64X "\n", "pinsrw", r.q[0]); 2448 2449 asm volatile ("pinsrw $5, %1, %0" : "=x" (r.dq) : "r" (0x12345678)); 2450 printf("%-9s: r=" FMT64X "" FMT64X "\n", "pinsrw", r.q[1], r.q[0]); 2451 2452 a.q[0] = test_values[0][0]; 2453 a.q[1] = test_values[0][1]; 2454 asm volatile ("pextrw $1, %1, %0" : "=r" (r.l[0]) : "y" (a.q[0])); 2455 printf("%-9s: r=%08x\n", "pextrw", r.l[0]); 2456 2457 asm volatile ("pextrw $5, %1, %0" : "=r" (r.l[0]) : "x" (a.dq)); 2458 printf("%-9s: r=%08x\n", "pextrw", r.l[0]); 2459 2460 asm volatile ("pmovmskb %1, %0" : "=r" (r.l[0]) : "y" (a.q[0])); 2461 printf("%-9s: r=%08x\n", "pmovmskb", r.l[0]); 2462 2463 asm volatile ("pmovmskb %1, %0" : "=r" (r.l[0]) : "x" (a.dq)); 2464 printf("%-9s: r=%08x\n", "pmovmskb", r.l[0]); 2465 2466 { 2467 r.q[0] = -1; 2468 r.q[1] = -1; 2469 2470 a.q[0] = test_values[0][0]; 2471 a.q[1] = test_values[0][1]; 2472 b.q[0] = test_values[1][0]; 2473 b.q[1] = test_values[1][1]; 2474 asm volatile("maskmovq %1, %0" : 2475 : "y" (a.q[0]), "y" (b.q[0]), "D" (&r) 2476 : "memory"); 2477 printf("%-9s: r=" FMT64X " a=" FMT64X " b=" FMT64X "\n", 2478 "maskmov", 2479 r.q[0], 2480 a.q[0], 2481 b.q[0]); 2482 asm volatile("maskmovdqu %1, %0" : 2483 : "x" (a.dq), "x" (b.dq), "D" (&r) 2484 : "memory"); 2485 printf("%-9s: r=" FMT64X "" FMT64X " a=" FMT64X "" FMT64X " b=" FMT64X "" FMT64X "\n", 2486 "maskmov", 2487 r.q[1], r.q[0], 2488 a.q[1], a.q[0], 2489 b.q[1], b.q[0]); 2490 } 2491 2492 asm volatile ("emms"); 2493 2494 SSE_OP2(punpcklqdq); 2495 SSE_OP2(punpckhqdq); 2496 SSE_OP2(andps); 2497 SSE_OP2(andpd); 2498 SSE_OP2(andnps); 2499 SSE_OP2(andnpd); 2500 SSE_OP2(orps); 2501 SSE_OP2(orpd); 2502 SSE_OP2(xorps); 2503 SSE_OP2(xorpd); 2504 2505 SSE_OP2(unpcklps); 2506 SSE_OP2(unpcklpd); 2507 SSE_OP2(unpckhps); 2508 SSE_OP2(unpckhpd); 2509 2510 SHUF_OP(shufps, 0x78); 2511 SHUF_OP(shufpd, 0x02); 2512 2513 PSHUF_OP(pshufd, 0x78); 2514 PSHUF_OP(pshuflw, 0x78); 2515 PSHUF_OP(pshufhw, 0x78); 2516 2517 SHIFT_OP(psrlw, 7); 2518 SHIFT_OP(psrlw, 16); 2519 SHIFT_OP(psraw, 7); 2520 SHIFT_OP(psraw, 16); 2521 SHIFT_OP(psllw, 7); 2522 SHIFT_OP(psllw, 16); 2523 2524 SHIFT_OP(psrld, 7); 2525 SHIFT_OP(psrld, 32); 2526 SHIFT_OP(psrad, 7); 2527 SHIFT_OP(psrad, 32); 2528 SHIFT_OP(pslld, 7); 2529 SHIFT_OP(pslld, 32); 2530 2531 SHIFT_OP(psrlq, 7); 2532 SHIFT_OP(psrlq, 32); 2533 SHIFT_OP(psllq, 7); 2534 SHIFT_OP(psllq, 32); 2535 2536 SHIFT_IM(psrldq, 16); 2537 SHIFT_IM(psrldq, 7); 2538 SHIFT_IM(pslldq, 16); 2539 SHIFT_IM(pslldq, 7); 2540 2541 MOVMSK(movmskps); 2542 MOVMSK(movmskpd); 2543 2544 /* FPU specific ops */ 2545 2546 { 2547 uint32_t mxcsr; 2548 asm volatile("stmxcsr %0" : "=m" (mxcsr)); 2549 printf("mxcsr=%08x\n", mxcsr & 0x1f80); 2550 asm volatile("ldmxcsr %0" : : "m" (mxcsr)); 2551 } 2552 2553 test_sse_comi(2, -1); 2554 test_sse_comi(2, 2); 2555 test_sse_comi(2, 3); 2556 test_sse_comi(2, q_nan.d); 2557 test_sse_comi(q_nan.d, -1); 2558 2559 for(i = 0; i < 2; i++) { 2560 a.s[0] = 2.7; 2561 a.s[1] = 3.4; 2562 a.s[2] = 4; 2563 a.s[3] = -6.3; 2564 b.s[0] = 45.7; 2565 b.s[1] = 353.4; 2566 b.s[2] = 4; 2567 b.s[3] = 56.3; 2568 if (i == 1) { 2569 a.s[0] = q_nan.d; 2570 b.s[3] = q_nan.d; 2571 } 2572 2573 SSE_OPS(add); 2574 SSE_OPS(mul); 2575 SSE_OPS(sub); 2576 SSE_OPS(min); 2577 SSE_OPS(div); 2578 SSE_OPS(max); 2579 SSE_OPS(sqrt); 2580 SSE_OPS(cmpeq); 2581 SSE_OPS(cmplt); 2582 SSE_OPS(cmple); 2583 SSE_OPS(cmpunord); 2584 SSE_OPS(cmpneq); 2585 SSE_OPS(cmpnlt); 2586 SSE_OPS(cmpnle); 2587 SSE_OPS(cmpord); 2588 2589 2590 a.d[0] = 2.7; 2591 a.d[1] = -3.4; 2592 b.d[0] = 45.7; 2593 b.d[1] = -53.4; 2594 if (i == 1) { 2595 a.d[0] = q_nan.d; 2596 b.d[1] = q_nan.d; 2597 } 2598 SSE_OPD(add); 2599 SSE_OPD(mul); 2600 SSE_OPD(sub); 2601 SSE_OPD(min); 2602 SSE_OPD(div); 2603 SSE_OPD(max); 2604 SSE_OPD(sqrt); 2605 SSE_OPD(cmpeq); 2606 SSE_OPD(cmplt); 2607 SSE_OPD(cmple); 2608 SSE_OPD(cmpunord); 2609 SSE_OPD(cmpneq); 2610 SSE_OPD(cmpnlt); 2611 SSE_OPD(cmpnle); 2612 SSE_OPD(cmpord); 2613 } 2614 2615 /* float to float/int */ 2616 a.s[0] = 2.7; 2617 a.s[1] = 3.4; 2618 a.s[2] = 4; 2619 a.s[3] = -6.3; 2620 CVT_OP_XMM(cvtps2pd); 2621 CVT_OP_XMM(cvtss2sd); 2622 CVT_OP_XMM2MMX(cvtps2pi); 2623 CVT_OP_XMM2MMX(cvttps2pi); 2624 CVT_OP_XMM2REG(cvtss2si); 2625 CVT_OP_XMM2REG(cvttss2si); 2626 CVT_OP_XMM(cvtps2dq); 2627 CVT_OP_XMM(cvttps2dq); 2628 2629 a.d[0] = 2.6; 2630 a.d[1] = -3.4; 2631 CVT_OP_XMM(cvtpd2ps); 2632 CVT_OP_XMM(cvtsd2ss); 2633 CVT_OP_XMM2MMX(cvtpd2pi); 2634 CVT_OP_XMM2MMX(cvttpd2pi); 2635 CVT_OP_XMM2REG(cvtsd2si); 2636 CVT_OP_XMM2REG(cvttsd2si); 2637 CVT_OP_XMM(cvtpd2dq); 2638 CVT_OP_XMM(cvttpd2dq); 2639 2640 /* sse/mmx moves */ 2641 CVT_OP_XMM2MMX(movdq2q); 2642 CVT_OP_MMX2XMM(movq2dq); 2643 2644 /* int to float */ 2645 a.l[0] = -6; 2646 a.l[1] = 2; 2647 a.l[2] = 100; 2648 a.l[3] = -60000; 2649 CVT_OP_MMX2XMM(cvtpi2ps); 2650 CVT_OP_MMX2XMM(cvtpi2pd); 2651 CVT_OP_REG2XMM(cvtsi2ss); 2652 CVT_OP_REG2XMM(cvtsi2sd); 2653 CVT_OP_XMM(cvtdq2ps); 2654 CVT_OP_XMM(cvtdq2pd); 2655 2656 /* XXX: test PNI insns */ 2657 #if 0 2658 SSE_OP2(movshdup); 2659 #endif 2660 asm volatile ("emms"); 2661 } 2662 2663 #endif 2664 2665 #define TEST_CONV_RAX(op)\ 2666 {\ 2667 unsigned long a, r;\ 2668 a = i2l(0x8234a6f8);\ 2669 r = a;\ 2670 asm volatile(#op : "=a" (r) : "0" (r));\ 2671 printf("%-10s A=" FMTLX " R=" FMTLX "\n", #op, a, r);\ 2672 } 2673 2674 #define TEST_CONV_RAX_RDX(op)\ 2675 {\ 2676 unsigned long a, d, r, rh; \ 2677 a = i2l(0x8234a6f8);\ 2678 d = i2l(0x8345a1f2);\ 2679 r = a;\ 2680 rh = d;\ 2681 asm volatile(#op : "=a" (r), "=d" (rh) : "0" (r), "1" (rh)); \ 2682 printf("%-10s A=" FMTLX " R=" FMTLX ":" FMTLX "\n", #op, a, r, rh); \ 2683 } 2684 2685 void test_conv(void) 2686 { 2687 TEST_CONV_RAX(cbw); 2688 TEST_CONV_RAX(cwde); 2689 #if defined(__x86_64__) 2690 TEST_CONV_RAX(cdqe); 2691 #endif 2692 2693 TEST_CONV_RAX_RDX(cwd); 2694 TEST_CONV_RAX_RDX(cdq); 2695 #if defined(__x86_64__) 2696 TEST_CONV_RAX_RDX(cqo); 2697 #endif 2698 2699 { 2700 unsigned long a, r; 2701 a = i2l(0x12345678); 2702 asm volatile("bswapl %k0" : "=r" (r) : "0" (a)); 2703 printf("%-10s: A=" FMTLX " R=" FMTLX "\n", "bswapl", a, r); 2704 } 2705 #if defined(__x86_64__) 2706 { 2707 unsigned long a, r; 2708 a = i2l(0x12345678); 2709 asm volatile("bswapq %0" : "=r" (r) : "0" (a)); 2710 printf("%-10s: A=" FMTLX " R=" FMTLX "\n", "bswapq", a, r); 2711 } 2712 #endif 2713 } 2714 2715 extern void *__start_initcall; 2716 extern void *__stop_initcall; 2717 2718 2719 int main(int argc, char **argv) 2720 { 2721 void **ptr; 2722 void (*func)(void); 2723 2724 ptr = &__start_initcall; 2725 while (ptr != &__stop_initcall) { 2726 func = *ptr++; 2727 func(); 2728 } 2729 test_bsx(); 2730 test_mul(); 2731 test_jcc(); 2732 test_loop(); 2733 test_floats(); 2734 #if !defined(__x86_64__) 2735 test_bcd(); 2736 #endif 2737 test_xchg(); 2738 test_string(); 2739 test_misc(); 2740 test_lea(); 2741 #ifdef TEST_SEGS 2742 test_segs(); 2743 test_code16(); 2744 #endif 2745 #ifdef TEST_VM86 2746 test_vm86(); 2747 #endif 2748 #if !defined(__x86_64__) 2749 test_exceptions(); 2750 test_self_modifying_code(); 2751 test_single_step(); 2752 #endif 2753 test_enter(); 2754 test_conv(); 2755 #ifdef TEST_SSE 2756 test_sse(); 2757 test_fxsave(); 2758 #endif 2759 return 0; 2760 } 2761