| 1 | /* |
| 2 | * linux/kernel/math/math_emulate.c |
| 3 | * |
| 4 | * (C) 1991 Linus Torvalds |
| 5 | * |
| 6 | * [expediant "port" of linux 8087 emulator to 386BSD, with apologies -wfj] |
| 7 | * |
| 8 | * from: 386BSD 0.1 |
| 9 | * $FreeBSD: src/sys/i386/i386/math_emulate.c,v 1.35 1999/08/28 00:43:47 peter Exp $ |
| 10 | * $DragonFly: src/sys/platform/pc32/i386/math_emulate.c,v 1.9 2007/02/03 17:05:58 corecode Exp $ |
| 11 | */ |
| 12 | |
| 13 | /* |
| 14 | * Limited emulation 27.12.91 - mostly loads/stores, which gcc wants |
| 15 | * even for soft-float, unless you use bruce evans' patches. The patches |
| 16 | * are great, but they have to be re-applied for every version, and the |
| 17 | * library is different for soft-float and 80387. So emulation is more |
| 18 | * practical, even though it's slower. |
| 19 | * |
| 20 | * 28.12.91 - loads/stores work, even BCD. I'll have to start thinking |
| 21 | * about add/sub/mul/div. Urgel. I should find some good source, but I'll |
| 22 | * just fake up something. |
| 23 | * |
| 24 | * 30.12.91 - add/sub/mul/div/com seem to work mostly. I should really |
| 25 | * test every possible combination. |
| 26 | */ |
| 27 | |
| 28 | /* |
| 29 | * This file is full of ugly macros etc: one problem was that gcc simply |
| 30 | * didn't want to make the structures as they should be: it has to try to |
| 31 | * align them. Sickening code, but at least I've hidden the ugly things |
| 32 | * in this one file: the other files don't need to know about these things. |
| 33 | * |
| 34 | * The other files also don't care about ST(x) etc - they just get addresses |
| 35 | * to 80-bit temporary reals, and do with them as they please. I wanted to |
| 36 | * hide most of the 387-specific things here. |
| 37 | */ |
| 38 | |
| 39 | #include <sys/param.h> |
| 40 | #include <sys/systm.h> |
| 41 | #include <sys/reg.h> |
| 42 | |
| 43 | #include <machine/frame.h> |
| 44 | |
| 45 | #include <sys/proc.h> |
| 46 | #include <sys/kernel.h> |
| 47 | |
| 48 | #include <vm/vm.h> |
| 49 | #include <sys/lock.h> |
| 50 | #include <vm/pmap.h> |
| 51 | #include <vm/vm_map.h> |
| 52 | #include <sys/user.h> |
| 53 | |
| 54 | #define __ALIGNED_TEMP_REAL 1 |
| 55 | #include "math_emu.h" |
| 56 | |
| 57 | #define bswapw(x) __asm__("xchgb %%al,%%ah":"=a" (x):"0" ((short)x)) |
| 58 | #define ST(x) (*__st((x))) |
| 59 | #define PST(x) ((const temp_real *) __st((x))) |
| 60 | #define math_abort(tfp, signo) tfp->tf_eip = oldeip; return (signo); |
| 61 | |
| 62 | /* |
| 63 | * We don't want these inlined - it gets too messy in the machine-code. |
| 64 | */ |
| 65 | static void fpop(void); |
| 66 | static void fpush(void); |
| 67 | static void fxchg(temp_real_unaligned *a, temp_real_unaligned *b); |
| 68 | static temp_real_unaligned *__st(int i); |
| 69 | |
| 70 | static unsigned char |
| 71 | get_fs_byte(char *adr) |
| 72 | { return(fubyte(adr)); } |
| 73 | |
| 74 | static unsigned short |
| 75 | get_fs_word(unsigned short *adr) |
| 76 | { return(fuword(adr)); } |
| 77 | |
| 78 | static u_int32_t |
| 79 | get_fs_long(u_int32_t *adr) |
| 80 | { return(fuword(adr)); } |
| 81 | |
| 82 | static void |
| 83 | put_fs_byte(unsigned char val, char *adr) |
| 84 | { (void)subyte(adr,val); } |
| 85 | |
| 86 | static void |
| 87 | put_fs_word(unsigned short val, short *adr) |
| 88 | { (void)susword(adr,val); } |
| 89 | |
| 90 | static void |
| 91 | put_fs_long(u_long val, u_int32_t *adr) |
| 92 | { (void)suword(adr,val); } |
| 93 | |
| 94 | static int |
| 95 | math_emulate(struct trapframe *info) |
| 96 | { |
| 97 | unsigned short code; |
| 98 | temp_real tmp; |
| 99 | char *address; |
| 100 | u_int32_t oldeip; |
| 101 | |
| 102 | /* ever used fp? */ |
| 103 | if ((curthread->td_pcb->pcb_flags & FP_SOFTFP) == 0) { |
| 104 | curthread->td_pcb->pcb_flags |= FP_SOFTFP; |
| 105 | I387.cwd = 0x037f; |
| 106 | I387.swd = 0x0000; |
| 107 | I387.twd = 0x0000; |
| 108 | } |
| 109 | |
| 110 | if (I387.cwd & I387.swd & 0x3f) |
| 111 | I387.swd |= 0x8000; |
| 112 | else |
| 113 | I387.swd &= 0x7fff; |
| 114 | oldeip = info->tf_eip; |
| 115 | /* 0x001f means user code space */ |
| 116 | if ((u_short)info->tf_cs != 0x001F) { |
| 117 | kprintf("math_emulate: %04x:%08lx\n", (u_short)info->tf_cs, |
| 118 | (u_long)oldeip); |
| 119 | panic("?Math emulation needed in kernel?"); |
| 120 | } |
| 121 | /* completely ignore an operand-size prefix */ |
| 122 | if (get_fs_byte((char *) info->tf_eip) == 0x66) |
| 123 | info->tf_eip++; |
| 124 | code = get_fs_word((unsigned short *) info->tf_eip); |
| 125 | bswapw(code); |
| 126 | code &= 0x7ff; |
| 127 | I387.fip = oldeip; |
| 128 | *(unsigned short *) &I387.fcs = (u_short) info->tf_cs; |
| 129 | *(1+(unsigned short *) &I387.fcs) = code; |
| 130 | info->tf_eip += 2; |
| 131 | switch (code) { |
| 132 | case 0x1d0: /* fnop */ |
| 133 | return(0); |
| 134 | case 0x1d1: case 0x1d2: case 0x1d3: /* fst to 32-bit mem */ |
| 135 | case 0x1d4: case 0x1d5: case 0x1d6: case 0x1d7: |
| 136 | math_abort(info,SIGILL); |
| 137 | case 0x1e0: /* fchs */ |
| 138 | ST(0).exponent ^= 0x8000; |
| 139 | return(0); |
| 140 | case 0x1e1: /* fabs */ |
| 141 | ST(0).exponent &= 0x7fff; |
| 142 | return(0); |
| 143 | case 0x1e2: case 0x1e3: |
| 144 | math_abort(info,SIGILL); |
| 145 | case 0x1e4: /* ftst */ |
| 146 | ftst(PST(0)); |
| 147 | return(0); |
| 148 | case 0x1e5: /* fxam */ |
| 149 | kprintf("fxam not implemented\n"); |
| 150 | math_abort(info,SIGILL); |
| 151 | case 0x1e6: case 0x1e7: /* fldenv */ |
| 152 | math_abort(info,SIGILL); |
| 153 | case 0x1e8: /* fld1 */ |
| 154 | fpush(); |
| 155 | ST(0) = CONST1; |
| 156 | return(0); |
| 157 | case 0x1e9: /* fld2t */ |
| 158 | fpush(); |
| 159 | ST(0) = CONSTL2T; |
| 160 | return(0); |
| 161 | case 0x1ea: /* fld2e */ |
| 162 | fpush(); |
| 163 | ST(0) = CONSTL2E; |
| 164 | return(0); |
| 165 | case 0x1eb: /* fldpi */ |
| 166 | fpush(); |
| 167 | ST(0) = CONSTPI; |
| 168 | return(0); |
| 169 | case 0x1ec: /* fldlg2 */ |
| 170 | fpush(); |
| 171 | ST(0) = CONSTLG2; |
| 172 | return(0); |
| 173 | case 0x1ed: /* fldln2 */ |
| 174 | fpush(); |
| 175 | ST(0) = CONSTLN2; |
| 176 | return(0); |
| 177 | case 0x1ee: /* fldz */ |
| 178 | fpush(); |
| 179 | ST(0) = CONSTZ; |
| 180 | return(0); |
| 181 | case 0x1ef: |
| 182 | math_abort(info,SIGILL); |
| 183 | case 0x1f0: /* f2xm1 */ |
| 184 | case 0x1f1: /* fyl2x */ |
| 185 | case 0x1f2: /* fptan */ |
| 186 | case 0x1f3: /* fpatan */ |
| 187 | case 0x1f4: /* fxtract */ |
| 188 | case 0x1f5: /* fprem1 */ |
| 189 | case 0x1f6: /* fdecstp */ |
| 190 | case 0x1f7: /* fincstp */ |
| 191 | case 0x1f8: /* fprem */ |
| 192 | case 0x1f9: /* fyl2xp1 */ |
| 193 | case 0x1fa: /* fsqrt */ |
| 194 | case 0x1fb: /* fsincos */ |
| 195 | case 0x1fe: /* fsin */ |
| 196 | case 0x1ff: /* fcos */ |
| 197 | uprintf( |
| 198 | "math_emulate: instruction %04x not implemented\n", |
| 199 | code + 0xd800); |
| 200 | math_abort(info,SIGILL); |
| 201 | case 0x1fc: /* frndint */ |
| 202 | frndint(PST(0),&tmp); |
| 203 | real_to_real(&tmp,&ST(0)); |
| 204 | return(0); |
| 205 | case 0x1fd: /* fscale */ |
| 206 | /* incomplete and totally inadequate -wfj */ |
| 207 | Fscale(PST(0), PST(1), &tmp); |
| 208 | real_to_real(&tmp,&ST(0)); |
| 209 | return(0); /* 19 Sep 92*/ |
| 210 | case 0x2e9: /* ????? */ |
| 211 | /* if this should be a fucomp ST(0),ST(1) , it must be a 0x3e9 ATS */ |
| 212 | fucom(PST(1),PST(0)); |
| 213 | fpop(); fpop(); |
| 214 | return(0); |
| 215 | case 0x3d0: case 0x3d1: /* fist ?? */ |
| 216 | return(0); |
| 217 | case 0x3e2: /* fclex */ |
| 218 | I387.swd &= 0x7f00; |
| 219 | return(0); |
| 220 | case 0x3e3: /* fninit */ |
| 221 | I387.cwd = 0x037f; |
| 222 | I387.swd = 0x0000; |
| 223 | I387.twd = 0x0000; |
| 224 | return(0); |
| 225 | case 0x3e4: |
| 226 | return(0); |
| 227 | case 0x6d9: /* fcompp */ |
| 228 | fcom(PST(1),PST(0)); |
| 229 | fpop(); fpop(); |
| 230 | return(0); |
| 231 | case 0x7e0: /* fstsw ax */ |
| 232 | *(short *) &info->tf_eax = I387.swd; |
| 233 | return(0); |
| 234 | } |
| 235 | switch (code >> 3) { |
| 236 | case 0x18: /* fadd */ |
| 237 | fadd(PST(0),PST(code & 7),&tmp); |
| 238 | real_to_real(&tmp,&ST(0)); |
| 239 | return(0); |
| 240 | case 0x19: /* fmul */ |
| 241 | fmul(PST(0),PST(code & 7),&tmp); |
| 242 | real_to_real(&tmp,&ST(0)); |
| 243 | return(0); |
| 244 | case 0x1a: /* fcom */ |
| 245 | fcom(PST(code & 7),PST(0)); |
| 246 | return(0); |
| 247 | case 0x1b: /* fcomp */ |
| 248 | fcom(PST(code & 7),PST(0)); |
| 249 | fpop(); |
| 250 | return(0); |
| 251 | case 0x1c: /* fsubr */ |
| 252 | real_to_real(&ST(code & 7),&tmp); |
| 253 | tmp.exponent ^= 0x8000; |
| 254 | fadd(PST(0),&tmp,&tmp); |
| 255 | real_to_real(&tmp,&ST(0)); |
| 256 | return(0); |
| 257 | case 0x1d: /* fsub */ |
| 258 | ST(0).exponent ^= 0x8000; |
| 259 | fadd(PST(0),PST(code & 7),&tmp); |
| 260 | real_to_real(&tmp,&ST(0)); |
| 261 | return(0); |
| 262 | case 0x1e: /* fdivr */ |
| 263 | fdiv(PST(0),PST(code & 7),&tmp); |
| 264 | real_to_real(&tmp,&ST(0)); |
| 265 | return(0); |
| 266 | case 0x1f: /* fdiv */ |
| 267 | fdiv(PST(code & 7),PST(0),&tmp); |
| 268 | real_to_real(&tmp,&ST(0)); |
| 269 | return(0); |
| 270 | case 0x38: /* fld */ |
| 271 | fpush(); |
| 272 | ST(0) = ST((code & 7)+1); /* why plus 1 ????? ATS */ |
| 273 | return(0); |
| 274 | case 0x39: /* fxch */ |
| 275 | fxchg(&ST(0),&ST(code & 7)); |
| 276 | return(0); |
| 277 | case 0x3b: /* ??? ??? wrong ???? ATS */ |
| 278 | ST(code & 7) = ST(0); |
| 279 | fpop(); |
| 280 | return(0); |
| 281 | case 0x98: /* fadd */ |
| 282 | fadd(PST(0),PST(code & 7),&tmp); |
| 283 | real_to_real(&tmp,&ST(code & 7)); |
| 284 | return(0); |
| 285 | case 0x99: /* fmul */ |
| 286 | fmul(PST(0),PST(code & 7),&tmp); |
| 287 | real_to_real(&tmp,&ST(code & 7)); |
| 288 | return(0); |
| 289 | case 0x9a: /* ???? , my manual don't list a direction bit |
| 290 | for fcom , ??? ATS */ |
| 291 | fcom(PST(code & 7),PST(0)); |
| 292 | return(0); |
| 293 | case 0x9b: /* same as above , ATS */ |
| 294 | fcom(PST(code & 7),PST(0)); |
| 295 | fpop(); |
| 296 | return(0); |
| 297 | case 0x9c: /* fsubr */ |
| 298 | ST(code & 7).exponent ^= 0x8000; |
| 299 | fadd(PST(0),PST(code & 7),&tmp); |
| 300 | real_to_real(&tmp,&ST(code & 7)); |
| 301 | return(0); |
| 302 | case 0x9d: /* fsub */ |
| 303 | real_to_real(&ST(0),&tmp); |
| 304 | tmp.exponent ^= 0x8000; |
| 305 | fadd(PST(code & 7),&tmp,&tmp); |
| 306 | real_to_real(&tmp,&ST(code & 7)); |
| 307 | return(0); |
| 308 | case 0x9e: /* fdivr */ |
| 309 | fdiv(PST(0),PST(code & 7),&tmp); |
| 310 | real_to_real(&tmp,&ST(code & 7)); |
| 311 | return(0); |
| 312 | case 0x9f: /* fdiv */ |
| 313 | fdiv(PST(code & 7),PST(0),&tmp); |
| 314 | real_to_real(&tmp,&ST(code & 7)); |
| 315 | return(0); |
| 316 | case 0xb8: /* ffree */ |
| 317 | kprintf("ffree not implemented\n"); |
| 318 | math_abort(info,SIGILL); |
| 319 | case 0xb9: /* fstp ???? where is the pop ? ATS */ |
| 320 | fxchg(&ST(0),&ST(code & 7)); |
| 321 | return(0); |
| 322 | case 0xba: /* fst */ |
| 323 | ST(code & 7) = ST(0); |
| 324 | return(0); |
| 325 | case 0xbb: /* ????? encoding of fstp to mem ? ATS */ |
| 326 | ST(code & 7) = ST(0); |
| 327 | fpop(); |
| 328 | return(0); |
| 329 | case 0xbc: /* fucom */ |
| 330 | fucom(PST(code & 7),PST(0)); |
| 331 | return(0); |
| 332 | case 0xbd: /* fucomp */ |
| 333 | fucom(PST(code & 7),PST(0)); |
| 334 | fpop(); |
| 335 | return(0); |
| 336 | case 0xd8: /* faddp */ |
| 337 | fadd(PST(code & 7),PST(0),&tmp); |
| 338 | real_to_real(&tmp,&ST(code & 7)); |
| 339 | fpop(); |
| 340 | return(0); |
| 341 | case 0xd9: /* fmulp */ |
| 342 | fmul(PST(code & 7),PST(0),&tmp); |
| 343 | real_to_real(&tmp,&ST(code & 7)); |
| 344 | fpop(); |
| 345 | return(0); |
| 346 | case 0xda: /* ??? encoding of ficom with 16 bit mem ? ATS */ |
| 347 | fcom(PST(code & 7),PST(0)); |
| 348 | fpop(); |
| 349 | return(0); |
| 350 | case 0xdc: /* fsubrp */ |
| 351 | ST(code & 7).exponent ^= 0x8000; |
| 352 | fadd(PST(0),PST(code & 7),&tmp); |
| 353 | real_to_real(&tmp,&ST(code & 7)); |
| 354 | fpop(); |
| 355 | return(0); |
| 356 | case 0xdd: /* fsubp */ |
| 357 | real_to_real(&ST(0),&tmp); |
| 358 | tmp.exponent ^= 0x8000; |
| 359 | fadd(PST(code & 7),&tmp,&tmp); |
| 360 | real_to_real(&tmp,&ST(code & 7)); |
| 361 | fpop(); |
| 362 | return(0); |
| 363 | case 0xde: /* fdivrp */ |
| 364 | fdiv(PST(0),PST(code & 7),&tmp); |
| 365 | real_to_real(&tmp,&ST(code & 7)); |
| 366 | fpop(); |
| 367 | return(0); |
| 368 | case 0xdf: /* fdivp */ |
| 369 | fdiv(PST(code & 7),PST(0),&tmp); |
| 370 | real_to_real(&tmp,&ST(code & 7)); |
| 371 | fpop(); |
| 372 | return(0); |
| 373 | case 0xf8: /* fild 16-bit mem ???? ATS */ |
| 374 | kprintf("ffree not implemented\n"); |
| 375 | math_abort(info,SIGILL); |
| 376 | fpop(); |
| 377 | return(0); |
| 378 | case 0xf9: /* ????? ATS */ |
| 379 | fxchg(&ST(0),&ST(code & 7)); |
| 380 | return(0); |
| 381 | case 0xfa: /* fist 16-bit mem ? ATS */ |
| 382 | case 0xfb: /* fistp 16-bit mem ? ATS */ |
| 383 | ST(code & 7) = ST(0); |
| 384 | fpop(); |
| 385 | return(0); |
| 386 | } |
| 387 | switch ((code>>3) & 0xe7) { |
| 388 | case 0x22: |
| 389 | put_short_real(PST(0),info,code); |
| 390 | return(0); |
| 391 | case 0x23: |
| 392 | put_short_real(PST(0),info,code); |
| 393 | fpop(); |
| 394 | return(0); |
| 395 | case 0x24: |
| 396 | address = ea(info,code); |
| 397 | for (code = 0 ; code < 7 ; code++) { |
| 398 | ((int32_t *) & I387)[code] = |
| 399 | get_fs_long((u_int32_t *) address); |
| 400 | address += 4; |
| 401 | } |
| 402 | return(0); |
| 403 | case 0x25: |
| 404 | address = ea(info,code); |
| 405 | *(unsigned short *) &I387.cwd = |
| 406 | get_fs_word((unsigned short *) address); |
| 407 | return(0); |
| 408 | case 0x26: |
| 409 | address = ea(info,code); |
| 410 | /*verify_area(address,28);*/ |
| 411 | for (code = 0 ; code < 7 ; code++) { |
| 412 | put_fs_long( ((int32_t *) & I387)[code], |
| 413 | (u_int32_t *) address); |
| 414 | address += 4; |
| 415 | } |
| 416 | return(0); |
| 417 | case 0x27: |
| 418 | address = ea(info,code); |
| 419 | /*verify_area(address,2);*/ |
| 420 | put_fs_word(I387.cwd,(short *) address); |
| 421 | return(0); |
| 422 | case 0x62: |
| 423 | put_long_int(PST(0),info,code); |
| 424 | return(0); |
| 425 | case 0x63: |
| 426 | put_long_int(PST(0),info,code); |
| 427 | fpop(); |
| 428 | return(0); |
| 429 | case 0x65: |
| 430 | fpush(); |
| 431 | get_temp_real(&tmp,info,code); |
| 432 | real_to_real(&tmp,&ST(0)); |
| 433 | return(0); |
| 434 | case 0x67: |
| 435 | put_temp_real(PST(0),info,code); |
| 436 | fpop(); |
| 437 | return(0); |
| 438 | case 0xa2: |
| 439 | put_long_real(PST(0),info,code); |
| 440 | return(0); |
| 441 | case 0xa3: |
| 442 | put_long_real(PST(0),info,code); |
| 443 | fpop(); |
| 444 | return(0); |
| 445 | case 0xa4: |
| 446 | address = ea(info,code); |
| 447 | for (code = 0 ; code < 27 ; code++) { |
| 448 | ((int32_t *) & I387)[code] = |
| 449 | get_fs_long((u_int32_t *) address); |
| 450 | address += 4; |
| 451 | } |
| 452 | return(0); |
| 453 | case 0xa6: |
| 454 | address = ea(info,code); |
| 455 | /*verify_area(address,108);*/ |
| 456 | for (code = 0 ; code < 27 ; code++) { |
| 457 | put_fs_long( ((int32_t *) & I387)[code], |
| 458 | (u_int32_t *) address); |
| 459 | address += 4; |
| 460 | } |
| 461 | I387.cwd = 0x037f; |
| 462 | I387.swd = 0x0000; |
| 463 | I387.twd = 0x0000; |
| 464 | return(0); |
| 465 | case 0xa7: |
| 466 | address = ea(info,code); |
| 467 | /*verify_area(address,2);*/ |
| 468 | put_fs_word(I387.swd,(short *) address); |
| 469 | return(0); |
| 470 | case 0xe2: |
| 471 | put_short_int(PST(0),info,code); |
| 472 | return(0); |
| 473 | case 0xe3: |
| 474 | put_short_int(PST(0),info,code); |
| 475 | fpop(); |
| 476 | return(0); |
| 477 | case 0xe4: |
| 478 | fpush(); |
| 479 | get_BCD(&tmp,info,code); |
| 480 | real_to_real(&tmp,&ST(0)); |
| 481 | return(0); |
| 482 | case 0xe5: |
| 483 | fpush(); |
| 484 | get_longlong_int(&tmp,info,code); |
| 485 | real_to_real(&tmp,&ST(0)); |
| 486 | return(0); |
| 487 | case 0xe6: |
| 488 | put_BCD(PST(0),info,code); |
| 489 | fpop(); |
| 490 | return(0); |
| 491 | case 0xe7: |
| 492 | put_longlong_int(PST(0),info,code); |
| 493 | fpop(); |
| 494 | return(0); |
| 495 | } |
| 496 | switch (code >> 9) { |
| 497 | case 0: |
| 498 | get_short_real(&tmp,info,code); |
| 499 | break; |
| 500 | case 1: |
| 501 | get_long_int(&tmp,info,code); |
| 502 | break; |
| 503 | case 2: |
| 504 | get_long_real(&tmp,info,code); |
| 505 | break; |
| 506 | case 4: |
| 507 | get_short_int(&tmp,info,code); |
| 508 | } |
| 509 | switch ((code>>3) & 0x27) { |
| 510 | case 0: |
| 511 | fadd(&tmp,PST(0),&tmp); |
| 512 | real_to_real(&tmp,&ST(0)); |
| 513 | return(0); |
| 514 | case 1: |
| 515 | fmul(&tmp,PST(0),&tmp); |
| 516 | real_to_real(&tmp,&ST(0)); |
| 517 | return(0); |
| 518 | case 2: |
| 519 | fcom(&tmp,PST(0)); |
| 520 | return(0); |
| 521 | case 3: |
| 522 | fcom(&tmp,PST(0)); |
| 523 | fpop(); |
| 524 | return(0); |
| 525 | case 4: |
| 526 | tmp.exponent ^= 0x8000; |
| 527 | fadd(&tmp,PST(0),&tmp); |
| 528 | real_to_real(&tmp,&ST(0)); |
| 529 | return(0); |
| 530 | case 5: |
| 531 | ST(0).exponent ^= 0x8000; |
| 532 | fadd(&tmp,PST(0),&tmp); |
| 533 | real_to_real(&tmp,&ST(0)); |
| 534 | return(0); |
| 535 | case 6: |
| 536 | fdiv(PST(0),&tmp,&tmp); |
| 537 | real_to_real(&tmp,&ST(0)); |
| 538 | return(0); |
| 539 | case 7: |
| 540 | fdiv(&tmp,PST(0),&tmp); |
| 541 | real_to_real(&tmp,&ST(0)); |
| 542 | return(0); |
| 543 | } |
| 544 | if ((code & 0x138) == 0x100) { |
| 545 | fpush(); |
| 546 | real_to_real(&tmp,&ST(0)); |
| 547 | return(0); |
| 548 | } |
| 549 | kprintf("Unknown math-insns: %04x:%08x %04x\n",(u_short)info->tf_cs, |
| 550 | info->tf_eip,code); |
| 551 | math_abort(info,SIGFPE); |
| 552 | } |
| 553 | |
| 554 | static void |
| 555 | fpop(void) |
| 556 | { |
| 557 | u_int32_t tmp; |
| 558 | |
| 559 | tmp = I387.swd & 0xffffc7ffUL; |
| 560 | I387.swd += 0x00000800; |
| 561 | I387.swd &= 0x00003800; |
| 562 | I387.swd |= tmp; |
| 563 | } |
| 564 | |
| 565 | static void |
| 566 | fpush(void) |
| 567 | { |
| 568 | u_int32_t tmp; |
| 569 | |
| 570 | tmp = I387.swd & 0xffffc7ffUL; |
| 571 | I387.swd += 0x00003800; |
| 572 | I387.swd &= 0x00003800; |
| 573 | I387.swd |= tmp; |
| 574 | } |
| 575 | |
| 576 | static void |
| 577 | fxchg(temp_real_unaligned *a, temp_real_unaligned *b) |
| 578 | { |
| 579 | temp_real_unaligned c; |
| 580 | |
| 581 | c = *a; |
| 582 | *a = *b; |
| 583 | *b = c; |
| 584 | } |
| 585 | |
| 586 | static temp_real_unaligned * |
| 587 | __st(int i) |
| 588 | { |
| 589 | i += I387.swd >> 11; |
| 590 | i &= 7; |
| 591 | return (temp_real_unaligned *) (i*10 + (char *)(I387.st_space)); |
| 592 | } |
| 593 | |
| 594 | /* |
| 595 | * linux/kernel/math/ea.c |
| 596 | * |
| 597 | * (C) 1991 Linus Torvalds |
| 598 | */ |
| 599 | |
| 600 | /* |
| 601 | * Calculate the effective address. |
| 602 | */ |
| 603 | |
| 604 | |
| 605 | static int __regoffset[] = { |
| 606 | tEAX, tECX, tEDX, tEBX, tESP, tEBP, tESI, tEDI |
| 607 | }; |
| 608 | |
| 609 | #define REG(x) (((int *)curthread->td_lwp->lwp_md.md_regs)[__regoffset[(x)]]) |
| 610 | |
| 611 | static char * |
| 612 | sib(struct trapframe *info, int mod) |
| 613 | { |
| 614 | unsigned char ss,index,base; |
| 615 | int32_t offset = 0; |
| 616 | |
| 617 | base = get_fs_byte((char *) info->tf_eip); |
| 618 | info->tf_eip++; |
| 619 | ss = base >> 6; |
| 620 | index = (base >> 3) & 7; |
| 621 | base &= 7; |
| 622 | if (index == 4) |
| 623 | offset = 0; |
| 624 | else |
| 625 | offset = REG(index); |
| 626 | offset <<= ss; |
| 627 | if (mod || base != 5) |
| 628 | offset += REG(base); |
| 629 | if (mod == 1) { |
| 630 | offset += (signed char) get_fs_byte((char *) info->tf_eip); |
| 631 | info->tf_eip++; |
| 632 | } else if (mod == 2 || base == 5) { |
| 633 | offset += (signed) get_fs_long((u_int32_t *) info->tf_eip); |
| 634 | info->tf_eip += 4; |
| 635 | } |
| 636 | I387.foo = offset; |
| 637 | I387.fos = 0x17; |
| 638 | return (char *) offset; |
| 639 | } |
| 640 | |
| 641 | static char * |
| 642 | ea(struct trapframe *info, unsigned short code) |
| 643 | { |
| 644 | unsigned char mod,rm; |
| 645 | int32_t *tmp; |
| 646 | int offset = 0; |
| 647 | |
| 648 | mod = (code >> 6) & 3; |
| 649 | rm = code & 7; |
| 650 | if (rm == 4 && mod != 3) |
| 651 | return sib(info,mod); |
| 652 | if (rm == 5 && !mod) { |
| 653 | offset = get_fs_long((u_int32_t *) info->tf_eip); |
| 654 | info->tf_eip += 4; |
| 655 | I387.foo = offset; |
| 656 | I387.fos = 0x17; |
| 657 | return (char *) offset; |
| 658 | } |
| 659 | tmp = (int32_t *) ®(rm); |
| 660 | switch (mod) { |
| 661 | case 0: offset = 0; break; |
| 662 | case 1: |
| 663 | offset = (signed char) get_fs_byte((char *) info->tf_eip); |
| 664 | info->tf_eip++; |
| 665 | break; |
| 666 | case 2: |
| 667 | offset = (signed) get_fs_long((u_int32_t *) info->tf_eip); |
| 668 | info->tf_eip += 4; |
| 669 | break; |
| 670 | #ifdef notyet |
| 671 | case 3: |
| 672 | math_abort(info,1<<(SIGILL-1)); |
| 673 | #endif |
| 674 | } |
| 675 | I387.foo = offset; |
| 676 | I387.fos = 0x17; |
| 677 | return offset + (char *) *tmp; |
| 678 | } |
| 679 | /* |
| 680 | * linux/kernel/math/get_put.c |
| 681 | * |
| 682 | * (C) 1991 Linus Torvalds |
| 683 | */ |
| 684 | |
| 685 | /* |
| 686 | * This file handles all accesses to user memory: getting and putting |
| 687 | * ints/reals/BCD etc. This is the only part that concerns itself with |
| 688 | * other than temporary real format. All other cals are strictly temp_real. |
| 689 | */ |
| 690 | |
| 691 | static void |
| 692 | get_short_real(temp_real *tmp, struct trapframe *info, unsigned short code) |
| 693 | { |
| 694 | char *addr; |
| 695 | short_real sr; |
| 696 | |
| 697 | addr = ea(info,code); |
| 698 | sr = get_fs_long((u_int32_t *) addr); |
| 699 | short_to_temp(&sr,tmp); |
| 700 | } |
| 701 | |
| 702 | static void |
| 703 | get_long_real(temp_real *tmp, struct trapframe *info, unsigned short code) |
| 704 | { |
| 705 | char *addr; |
| 706 | long_real lr; |
| 707 | |
| 708 | addr = ea(info,code); |
| 709 | lr.a = get_fs_long((u_int32_t *) addr); |
| 710 | lr.b = get_fs_long(1 + (u_int32_t *) addr); |
| 711 | long_to_temp(&lr,tmp); |
| 712 | } |
| 713 | |
| 714 | static void |
| 715 | get_temp_real(temp_real *tmp, struct trapframe *info, unsigned short code) |
| 716 | { |
| 717 | char *addr; |
| 718 | |
| 719 | addr = ea(info,code); |
| 720 | tmp->a = get_fs_long((u_int32_t *) addr); |
| 721 | tmp->b = get_fs_long(1 + (u_int32_t *) addr); |
| 722 | tmp->exponent = get_fs_word(4 + (unsigned short *) addr); |
| 723 | } |
| 724 | |
| 725 | static void |
| 726 | get_short_int(temp_real *tmp, struct trapframe *info, unsigned short code) |
| 727 | { |
| 728 | char *addr; |
| 729 | temp_int ti; |
| 730 | |
| 731 | addr = ea(info,code); |
| 732 | ti.a = (signed short) get_fs_word((unsigned short *) addr); |
| 733 | ti.b = 0; |
| 734 | if ((ti.sign = (ti.a < 0)) != 0) |
| 735 | ti.a = - ti.a; |
| 736 | int_to_real(&ti,tmp); |
| 737 | } |
| 738 | |
| 739 | static void |
| 740 | get_long_int(temp_real *tmp, struct trapframe *info, unsigned short code) |
| 741 | { |
| 742 | char *addr; |
| 743 | temp_int ti; |
| 744 | |
| 745 | addr = ea(info,code); |
| 746 | ti.a = get_fs_long((u_int32_t *) addr); |
| 747 | ti.b = 0; |
| 748 | if ((ti.sign = (ti.a < 0)) != 0) |
| 749 | ti.a = - ti.a; |
| 750 | int_to_real(&ti,tmp); |
| 751 | } |
| 752 | |
| 753 | static void |
| 754 | get_longlong_int(temp_real *tmp, struct trapframe *info, unsigned short code) |
| 755 | { |
| 756 | char *addr; |
| 757 | temp_int ti; |
| 758 | |
| 759 | addr = ea(info,code); |
| 760 | ti.a = get_fs_long((u_int32_t *) addr); |
| 761 | ti.b = get_fs_long(1 + (u_int32_t *) addr); |
| 762 | if ((ti.sign = (ti.b < 0)) != 0) |
| 763 | __asm__("notl %0 ; notl %1\n\t" |
| 764 | "addl $1,%0 ; adcl $0,%1" |
| 765 | :"=r" (ti.a),"=r" (ti.b) |
| 766 | :"0" (ti.a),"1" (ti.b)); |
| 767 | int_to_real(&ti,tmp); |
| 768 | } |
| 769 | |
| 770 | #define MUL10(low,high) \ |
| 771 | __asm__("addl %0,%0 ; adcl %1,%1\n\t" \ |
| 772 | "movl %0,%%ecx ; movl %1,%%ebx\n\t" \ |
| 773 | "addl %0,%0 ; adcl %1,%1\n\t" \ |
| 774 | "addl %0,%0 ; adcl %1,%1\n\t" \ |
| 775 | "addl %%ecx,%0 ; adcl %%ebx,%1" \ |
| 776 | :"=a" (low),"=d" (high) \ |
| 777 | :"0" (low),"1" (high):"cx","bx") |
| 778 | |
| 779 | #define ADD64(val,low,high) \ |
| 780 | __asm__("addl %4,%0 ; adcl $0,%1":"=r" (low),"=r" (high) \ |
| 781 | :"0" (low),"1" (high),"r" ((u_int32_t) (val))) |
| 782 | |
| 783 | static void |
| 784 | get_BCD(temp_real *tmp, struct trapframe *info, unsigned short code) |
| 785 | { |
| 786 | int k; |
| 787 | char *addr; |
| 788 | temp_int i; |
| 789 | unsigned char c; |
| 790 | |
| 791 | addr = ea(info,code); |
| 792 | addr += 9; |
| 793 | i.sign = 0x80 & get_fs_byte(addr--); |
| 794 | i.a = i.b = 0; |
| 795 | for (k = 0; k < 9; k++) { |
| 796 | c = get_fs_byte(addr--); |
| 797 | MUL10(i.a, i.b); |
| 798 | ADD64((c>>4), i.a, i.b); |
| 799 | MUL10(i.a, i.b); |
| 800 | ADD64((c&0xf), i.a, i.b); |
| 801 | } |
| 802 | int_to_real(&i,tmp); |
| 803 | } |
| 804 | |
| 805 | static void |
| 806 | put_short_real(const temp_real *tmp, |
| 807 | struct trapframe *info, unsigned short code) |
| 808 | { |
| 809 | char *addr; |
| 810 | short_real sr; |
| 811 | |
| 812 | addr = ea(info,code); |
| 813 | /*verify_area(addr,4);*/ |
| 814 | temp_to_short(tmp,&sr); |
| 815 | put_fs_long(sr,(u_int32_t *) addr); |
| 816 | } |
| 817 | |
| 818 | static void |
| 819 | put_long_real(const temp_real *tmp, |
| 820 | struct trapframe *info, unsigned short code) |
| 821 | { |
| 822 | char *addr; |
| 823 | long_real lr; |
| 824 | |
| 825 | addr = ea(info,code); |
| 826 | /*verify_area(addr,8);*/ |
| 827 | temp_to_long(tmp,&lr); |
| 828 | put_fs_long(lr.a, (u_int32_t *) addr); |
| 829 | put_fs_long(lr.b, 1 + (u_int32_t *) addr); |
| 830 | } |
| 831 | |
| 832 | static void |
| 833 | put_temp_real(const temp_real *tmp, |
| 834 | struct trapframe *info, unsigned short code) |
| 835 | { |
| 836 | char *addr; |
| 837 | |
| 838 | addr = ea(info,code); |
| 839 | /*verify_area(addr,10);*/ |
| 840 | put_fs_long(tmp->a, (u_int32_t *) addr); |
| 841 | put_fs_long(tmp->b, 1 + (u_int32_t *) addr); |
| 842 | put_fs_word(tmp->exponent, 4 + (short *) addr); |
| 843 | } |
| 844 | |
| 845 | static void |
| 846 | put_short_int(const temp_real *tmp, |
| 847 | struct trapframe *info, unsigned short code) |
| 848 | { |
| 849 | char *addr; |
| 850 | temp_int ti; |
| 851 | |
| 852 | addr = ea(info,code); |
| 853 | real_to_int(tmp,&ti); |
| 854 | /*verify_area(addr,2);*/ |
| 855 | if (ti.sign) |
| 856 | ti.a = -ti.a; |
| 857 | put_fs_word(ti.a,(short *) addr); |
| 858 | } |
| 859 | |
| 860 | static void |
| 861 | put_long_int(const temp_real *tmp, |
| 862 | struct trapframe *info, unsigned short code) |
| 863 | { |
| 864 | char *addr; |
| 865 | temp_int ti; |
| 866 | |
| 867 | addr = ea(info,code); |
| 868 | real_to_int(tmp,&ti); |
| 869 | /*verify_area(addr,4);*/ |
| 870 | if (ti.sign) |
| 871 | ti.a = -ti.a; |
| 872 | put_fs_long(ti.a,(u_int32_t *) addr); |
| 873 | } |
| 874 | |
| 875 | static void |
| 876 | put_longlong_int(const temp_real *tmp, |
| 877 | struct trapframe *info, unsigned short code) |
| 878 | { |
| 879 | char *addr; |
| 880 | temp_int ti; |
| 881 | |
| 882 | addr = ea(info,code); |
| 883 | real_to_int(tmp,&ti); |
| 884 | /*verify_area(addr,8);*/ |
| 885 | if (ti.sign) |
| 886 | __asm__("notl %0 ; notl %1\n\t" |
| 887 | "addl $1,%0 ; adcl $0,%1" |
| 888 | :"=r" (ti.a),"=r" (ti.b) |
| 889 | :"0" (ti.a),"1" (ti.b)); |
| 890 | put_fs_long(ti.a,(u_int32_t *) addr); |
| 891 | put_fs_long(ti.b,1 + (u_int32_t *) addr); |
| 892 | } |
| 893 | |
| 894 | #define DIV10(low,high,rem) \ |
| 895 | __asm__("divl %6 ; xchgl %1,%2 ; divl %6" \ |
| 896 | :"=d" (rem),"=a" (low),"=r" (high) \ |
| 897 | :"0" (0),"1" (high),"2" (low),"c" (10)) |
| 898 | |
| 899 | static void |
| 900 | put_BCD(const temp_real *tmp,struct trapframe *info, unsigned short code) |
| 901 | { |
| 902 | int k,rem; |
| 903 | char *addr; |
| 904 | temp_int i; |
| 905 | unsigned char c; |
| 906 | |
| 907 | addr = ea(info,code); |
| 908 | /*verify_area(addr,10);*/ |
| 909 | real_to_int(tmp,&i); |
| 910 | if (i.sign) |
| 911 | put_fs_byte(0x80, addr+9); |
| 912 | else |
| 913 | put_fs_byte(0, addr+9); |
| 914 | for (k = 0; k < 9; k++) { |
| 915 | DIV10(i.a,i.b,rem); |
| 916 | c = rem; |
| 917 | DIV10(i.a,i.b,rem); |
| 918 | c += rem<<4; |
| 919 | put_fs_byte(c,addr++); |
| 920 | } |
| 921 | } |
| 922 | |
| 923 | /* |
| 924 | * linux/kernel/math/mul.c |
| 925 | * |
| 926 | * (C) 1991 Linus Torvalds |
| 927 | */ |
| 928 | |
| 929 | /* |
| 930 | * temporary real multiplication routine. |
| 931 | */ |
| 932 | |
| 933 | |
| 934 | static void |
| 935 | shift(int *c) |
| 936 | { |
| 937 | __asm__("movl (%0),%%eax ; addl %%eax,(%0)\n\t" |
| 938 | "movl 4(%0),%%eax ; adcl %%eax,4(%0)\n\t" |
| 939 | "movl 8(%0),%%eax ; adcl %%eax,8(%0)\n\t" |
| 940 | "movl 12(%0),%%eax ; adcl %%eax,12(%0)" |
| 941 | ::"r" (c):"ax"); |
| 942 | } |
| 943 | |
| 944 | static void |
| 945 | mul64(const temp_real *a, const temp_real *b, int *c) |
| 946 | { |
| 947 | __asm__("movl (%0),%%eax\n\t" |
| 948 | "mull (%1)\n\t" |
| 949 | "movl %%eax,(%2)\n\t" |
| 950 | "movl %%edx,4(%2)\n\t" |
| 951 | "movl 4(%0),%%eax\n\t" |
| 952 | "mull 4(%1)\n\t" |
| 953 | "movl %%eax,8(%2)\n\t" |
| 954 | "movl %%edx,12(%2)\n\t" |
| 955 | "movl (%0),%%eax\n\t" |
| 956 | "mull 4(%1)\n\t" |
| 957 | "addl %%eax,4(%2)\n\t" |
| 958 | "adcl %%edx,8(%2)\n\t" |
| 959 | "adcl $0,12(%2)\n\t" |
| 960 | "movl 4(%0),%%eax\n\t" |
| 961 | "mull (%1)\n\t" |
| 962 | "addl %%eax,4(%2)\n\t" |
| 963 | "adcl %%edx,8(%2)\n\t" |
| 964 | "adcl $0,12(%2)" |
| 965 | ::"S" (a),"c" (b),"D" (c) |
| 966 | :"ax","dx"); |
| 967 | } |
| 968 | |
| 969 | static void |
| 970 | fmul(const temp_real *src1, const temp_real *src2, temp_real *result) |
| 971 | { |
| 972 | int i,sign; |
| 973 | int tmp[4] = {0,0,0,0}; |
| 974 | |
| 975 | sign = (src1->exponent ^ src2->exponent) & 0x8000; |
| 976 | i = (src1->exponent & 0x7fff) + (src2->exponent & 0x7fff) - 16383 + 1; |
| 977 | if (i<0) { |
| 978 | result->exponent = sign; |
| 979 | result->a = result->b = 0; |
| 980 | return; |
| 981 | } |
| 982 | if (i>0x7fff) { |
| 983 | set_OE(); |
| 984 | return; |
| 985 | } |
| 986 | mul64(src1,src2,tmp); |
| 987 | if (tmp[0] || tmp[1] || tmp[2] || tmp[3]) |
| 988 | while (i && tmp[3] >= 0) { |
| 989 | i--; |
| 990 | shift(tmp); |
| 991 | } |
| 992 | else |
| 993 | i = 0; |
| 994 | result->exponent = i | sign; |
| 995 | result->a = tmp[2]; |
| 996 | result->b = tmp[3]; |
| 997 | } |
| 998 | |
| 999 | /* |
| 1000 | * linux/kernel/math/div.c |
| 1001 | * |
| 1002 | * (C) 1991 Linus Torvalds |
| 1003 | */ |
| 1004 | |
| 1005 | /* |
| 1006 | * temporary real division routine. |
| 1007 | */ |
| 1008 | |
| 1009 | static void |
| 1010 | shift_left(int *c) |
| 1011 | { |
| 1012 | __asm__ __volatile__("movl (%0),%%eax ; addl %%eax,(%0)\n\t" |
| 1013 | "movl 4(%0),%%eax ; adcl %%eax,4(%0)\n\t" |
| 1014 | "movl 8(%0),%%eax ; adcl %%eax,8(%0)\n\t" |
| 1015 | "movl 12(%0),%%eax ; adcl %%eax,12(%0)" |
| 1016 | ::"r" (c):"ax"); |
| 1017 | } |
| 1018 | |
| 1019 | static void |
| 1020 | shift_right(int *c) |
| 1021 | { |
| 1022 | __asm__("shrl $1,12(%0) ; rcrl $1,8(%0) ; rcrl $1,4(%0) ; rcrl $1,(%0)" |
| 1023 | ::"r" (c)); |
| 1024 | } |
| 1025 | |
| 1026 | static int |
| 1027 | try_sub(int *a, int *b) |
| 1028 | { |
| 1029 | char ok; |
| 1030 | |
| 1031 | __asm__ __volatile__("movl (%1),%%eax ; subl %%eax,(%2)\n\t" |
| 1032 | "movl 4(%1),%%eax ; sbbl %%eax,4(%2)\n\t" |
| 1033 | "movl 8(%1),%%eax ; sbbl %%eax,8(%2)\n\t" |
| 1034 | "movl 12(%1),%%eax ; sbbl %%eax,12(%2)\n\t" |
| 1035 | "setae %%al":"=a" (ok):"c" (a),"d" (b)); |
| 1036 | return ok; |
| 1037 | } |
| 1038 | |
| 1039 | static void |
| 1040 | div64(int *a, int *b, int *c) |
| 1041 | { |
| 1042 | int tmp[4]; |
| 1043 | int i; |
| 1044 | unsigned int mask = 0; |
| 1045 | |
| 1046 | c += 4; |
| 1047 | for (i = 0 ; i<64 ; i++) { |
| 1048 | if (!(mask >>= 1)) { |
| 1049 | c--; |
| 1050 | mask = 0x80000000UL; |
| 1051 | } |
| 1052 | tmp[0] = a[0]; tmp[1] = a[1]; |
| 1053 | tmp[2] = a[2]; tmp[3] = a[3]; |
| 1054 | if (try_sub(b,tmp)) { |
| 1055 | *c |= mask; |
| 1056 | a[0] = tmp[0]; a[1] = tmp[1]; |
| 1057 | a[2] = tmp[2]; a[3] = tmp[3]; |
| 1058 | } |
| 1059 | shift_right(b); |
| 1060 | } |
| 1061 | } |
| 1062 | |
| 1063 | static void |
| 1064 | fdiv(const temp_real *src1, const temp_real *src2, temp_real *result) |
| 1065 | { |
| 1066 | int i,sign; |
| 1067 | int a[4],b[4],tmp[4] = {0,0,0,0}; |
| 1068 | |
| 1069 | sign = (src1->exponent ^ src2->exponent) & 0x8000; |
| 1070 | if (!(src2->a || src2->b)) { |
| 1071 | set_ZE(); |
| 1072 | return; |
| 1073 | } |
| 1074 | i = (src1->exponent & 0x7fff) - (src2->exponent & 0x7fff) + 16383; |
| 1075 | if (i<0) { |
| 1076 | set_UE(); |
| 1077 | result->exponent = sign; |
| 1078 | result->a = result->b = 0; |
| 1079 | return; |
| 1080 | } |
| 1081 | a[0] = a[1] = 0; |
| 1082 | a[2] = src1->a; |
| 1083 | a[3] = src1->b; |
| 1084 | b[0] = b[1] = 0; |
| 1085 | b[2] = src2->a; |
| 1086 | b[3] = src2->b; |
| 1087 | while (b[3] >= 0) { |
| 1088 | i++; |
| 1089 | shift_left(b); |
| 1090 | } |
| 1091 | div64(a,b,tmp); |
| 1092 | if (tmp[0] || tmp[1] || tmp[2] || tmp[3]) { |
| 1093 | while (i && tmp[3] >= 0) { |
| 1094 | i--; |
| 1095 | shift_left(tmp); |
| 1096 | } |
| 1097 | if (tmp[3] >= 0) |
| 1098 | set_DE(); |
| 1099 | } else |
| 1100 | i = 0; |
| 1101 | if (i>0x7fff) { |
| 1102 | set_OE(); |
| 1103 | return; |
| 1104 | } |
| 1105 | if (tmp[0] || tmp[1]) |
| 1106 | set_PE(); |
| 1107 | result->exponent = i | sign; |
| 1108 | result->a = tmp[2]; |
| 1109 | result->b = tmp[3]; |
| 1110 | } |
| 1111 | |
| 1112 | /* |
| 1113 | * linux/kernel/math/add.c |
| 1114 | * |
| 1115 | * (C) 1991 Linus Torvalds |
| 1116 | */ |
| 1117 | |
| 1118 | /* |
| 1119 | * temporary real addition routine. |
| 1120 | * |
| 1121 | * NOTE! These aren't exact: they are only 62 bits wide, and don't do |
| 1122 | * correct rounding. Fast hack. The reason is that we shift right the |
| 1123 | * values by two, in order not to have overflow (1 bit), and to be able |
| 1124 | * to move the sign into the mantissa (1 bit). Much simpler algorithms, |
| 1125 | * and 62 bits (61 really - no rounding) accuracy is usually enough. The |
| 1126 | * only time you should notice anything weird is when adding 64-bit |
| 1127 | * integers together. When using doubles (52 bits accuracy), the |
| 1128 | * 61-bit accuracy never shows at all. |
| 1129 | */ |
| 1130 | |
| 1131 | #define NEGINT(a) \ |
| 1132 | __asm__("notl %0 ; notl %1 ; addl $1,%0 ; adcl $0,%1" \ |
| 1133 | :"=r" (a->a),"=r" (a->b) \ |
| 1134 | :"0" (a->a),"1" (a->b)) |
| 1135 | |
| 1136 | static void |
| 1137 | signify(temp_real *a) |
| 1138 | { |
| 1139 | a->exponent += 2; |
| 1140 | __asm__("shrdl $2,%1,%0 ; shrl $2,%1" |
| 1141 | :"=r" (a->a),"=r" (a->b) |
| 1142 | :"0" (a->a),"1" (a->b)); |
| 1143 | if (a->exponent < 0) |
| 1144 | NEGINT(a); |
| 1145 | a->exponent &= 0x7fff; |
| 1146 | } |
| 1147 | |
| 1148 | static void |
| 1149 | unsignify(temp_real *a) |
| 1150 | { |
| 1151 | if (!(a->a || a->b)) { |
| 1152 | a->exponent = 0; |
| 1153 | return; |
| 1154 | } |
| 1155 | a->exponent &= 0x7fff; |
| 1156 | if (a->b < 0) { |
| 1157 | NEGINT(a); |
| 1158 | a->exponent |= 0x8000; |
| 1159 | } |
| 1160 | while (a->b >= 0) { |
| 1161 | a->exponent--; |
| 1162 | __asm__("addl %0,%0 ; adcl %1,%1" |
| 1163 | :"=r" (a->a),"=r" (a->b) |
| 1164 | :"0" (a->a),"1" (a->b)); |
| 1165 | } |
| 1166 | } |
| 1167 | |
| 1168 | static void |
| 1169 | fadd(const temp_real *src1, const temp_real *src2, temp_real *result) |
| 1170 | { |
| 1171 | temp_real a,b; |
| 1172 | int x1,x2,shift; |
| 1173 | |
| 1174 | x1 = src1->exponent & 0x7fff; |
| 1175 | x2 = src2->exponent & 0x7fff; |
| 1176 | if (x1 > x2) { |
| 1177 | a = *src1; |
| 1178 | b = *src2; |
| 1179 | shift = x1-x2; |
| 1180 | } else { |
| 1181 | a = *src2; |
| 1182 | b = *src1; |
| 1183 | shift = x2-x1; |
| 1184 | } |
| 1185 | if (shift >= 64) { |
| 1186 | *result = a; |
| 1187 | return; |
| 1188 | } |
| 1189 | if (shift >= 32) { |
| 1190 | b.a = b.b; |
| 1191 | b.b = 0; |
| 1192 | shift -= 32; |
| 1193 | } |
| 1194 | __asm__("shrdl %4,%1,%0 ; shrl %4,%1" |
| 1195 | :"=r" (b.a),"=r" (b.b) |
| 1196 | :"0" (b.a),"1" (b.b),"c" ((char) shift)); |
| 1197 | signify(&a); |
| 1198 | signify(&b); |
| 1199 | __asm__("addl %4,%0 ; adcl %5,%1" |
| 1200 | :"=r" (a.a),"=r" (a.b) |
| 1201 | :"0" (a.a),"1" (a.b),"g" (b.a),"g" (b.b)); |
| 1202 | unsignify(&a); |
| 1203 | *result = a; |
| 1204 | } |
| 1205 | |
| 1206 | /* |
| 1207 | * linux/kernel/math/compare.c |
| 1208 | * |
| 1209 | * (C) 1991 Linus Torvalds |
| 1210 | */ |
| 1211 | |
| 1212 | /* |
| 1213 | * temporary real comparison routines |
| 1214 | */ |
| 1215 | |
| 1216 | |
| 1217 | #define clear_Cx() (I387.swd &= ~0x4500) |
| 1218 | |
| 1219 | static void |
| 1220 | normalize(temp_real *a) |
| 1221 | { |
| 1222 | int i = a->exponent & 0x7fff; |
| 1223 | int sign = a->exponent & 0x8000; |
| 1224 | |
| 1225 | if (!(a->a || a->b)) { |
| 1226 | a->exponent = 0; |
| 1227 | return; |
| 1228 | } |
| 1229 | while (i && a->b >= 0) { |
| 1230 | i--; |
| 1231 | __asm__("addl %0,%0 ; adcl %1,%1" |
| 1232 | :"=r" (a->a),"=r" (a->b) |
| 1233 | :"0" (a->a),"1" (a->b)); |
| 1234 | } |
| 1235 | a->exponent = i | sign; |
| 1236 | } |
| 1237 | |
| 1238 | static void |
| 1239 | ftst(const temp_real *a) |
| 1240 | { |
| 1241 | temp_real b; |
| 1242 | |
| 1243 | clear_Cx(); |
| 1244 | b = *a; |
| 1245 | normalize(&b); |
| 1246 | if (b.a || b.b || b.exponent) { |
| 1247 | if (b.exponent < 0) |
| 1248 | set_C0(); |
| 1249 | } else |
| 1250 | set_C3(); |
| 1251 | } |
| 1252 | |
| 1253 | static void |
| 1254 | fcom(const temp_real *src1, const temp_real *src2) |
| 1255 | { |
| 1256 | temp_real a; |
| 1257 | |
| 1258 | a = *src1; |
| 1259 | a.exponent ^= 0x8000; |
| 1260 | fadd(&a,src2,&a); |
| 1261 | ftst(&a); |
| 1262 | } |
| 1263 | |
| 1264 | static void |
| 1265 | fucom(const temp_real *src1, const temp_real *src2) |
| 1266 | { |
| 1267 | fcom(src1,src2); |
| 1268 | } |
| 1269 | |
| 1270 | /* |
| 1271 | * linux/kernel/math/convert.c |
| 1272 | * |
| 1273 | * (C) 1991 Linus Torvalds |
| 1274 | */ |
| 1275 | |
| 1276 | |
| 1277 | /* |
| 1278 | * NOTE!!! There is some "non-obvious" optimisations in the temp_to_long |
| 1279 | * and temp_to_short conversion routines: don't touch them if you don't |
| 1280 | * know what's going on. They are the adding of one in the rounding: the |
| 1281 | * overflow bit is also used for adding one into the exponent. Thus it |
| 1282 | * looks like the overflow would be incorrectly handled, but due to the |
| 1283 | * way the IEEE numbers work, things are correct. |
| 1284 | * |
| 1285 | * There is no checking for total overflow in the conversions, though (ie |
| 1286 | * if the temp-real number simply won't fit in a short- or long-real.) |
| 1287 | */ |
| 1288 | |
| 1289 | static void |
| 1290 | short_to_temp(const short_real *a, temp_real *b) |
| 1291 | { |
| 1292 | if (!(*a & 0x7fffffff)) { |
| 1293 | b->a = b->b = 0; |
| 1294 | if (*a) |
| 1295 | b->exponent = 0x8000; |
| 1296 | else |
| 1297 | b->exponent = 0; |
| 1298 | return; |
| 1299 | } |
| 1300 | b->exponent = ((*a>>23) & 0xff)-127+16383; |
| 1301 | if (*a<0) |
| 1302 | b->exponent |= 0x8000; |
| 1303 | b->b = (*a<<8) | 0x80000000UL; |
| 1304 | b->a = 0; |
| 1305 | } |
| 1306 | |
| 1307 | static void |
| 1308 | long_to_temp(const long_real *a, temp_real *b) |
| 1309 | { |
| 1310 | if (!a->a && !(a->b & 0x7fffffff)) { |
| 1311 | b->a = b->b = 0; |
| 1312 | if (a->b) |
| 1313 | b->exponent = 0x8000; |
| 1314 | else |
| 1315 | b->exponent = 0; |
| 1316 | return; |
| 1317 | } |
| 1318 | b->exponent = ((a->b >> 20) & 0x7ff)-1023+16383; |
| 1319 | if (a->b<0) |
| 1320 | b->exponent |= 0x8000; |
| 1321 | b->b = 0x80000000UL | (a->b<<11) | (((u_int32_t)a->a)>>21); |
| 1322 | b->a = a->a<<11; |
| 1323 | } |
| 1324 | |
| 1325 | static void |
| 1326 | temp_to_short(const temp_real *a, short_real *b) |
| 1327 | { |
| 1328 | if (!(a->exponent & 0x7fff)) { |
| 1329 | *b = (a->exponent)?0x80000000UL:0; |
| 1330 | return; |
| 1331 | } |
| 1332 | *b = ((((int32_t) a->exponent)-16383+127) << 23) & 0x7f800000; |
| 1333 | if (a->exponent < 0) |
| 1334 | *b |= 0x80000000UL; |
| 1335 | *b |= (a->b >> 8) & 0x007fffff; |
| 1336 | switch ((int)ROUNDING) { |
| 1337 | case ROUND_NEAREST: |
| 1338 | if ((a->b & 0xff) > 0x80) |
| 1339 | ++*b; |
| 1340 | break; |
| 1341 | case ROUND_DOWN: |
| 1342 | if ((a->exponent & 0x8000) && (a->b & 0xff)) |
| 1343 | ++*b; |
| 1344 | break; |
| 1345 | case ROUND_UP: |
| 1346 | if (!(a->exponent & 0x8000) && (a->b & 0xff)) |
| 1347 | ++*b; |
| 1348 | break; |
| 1349 | } |
| 1350 | } |
| 1351 | |
| 1352 | static void |
| 1353 | temp_to_long(const temp_real *a, long_real *b) |
| 1354 | { |
| 1355 | if (!(a->exponent & 0x7fff)) { |
| 1356 | b->a = 0; |
| 1357 | b->b = (a->exponent)?0x80000000UL:0; |
| 1358 | return; |
| 1359 | } |
| 1360 | b->b = (((0x7fff & (int32_t) a->exponent)-16383+1023) << 20) & |
| 1361 | 0x7ff00000; |
| 1362 | if (a->exponent < 0) |
| 1363 | b->b |= 0x80000000UL; |
| 1364 | b->b |= (a->b >> 11) & 0x000fffff; |
| 1365 | b->a = a->b << 21; |
| 1366 | b->a |= (a->a >> 11) & 0x001fffff; |
| 1367 | switch ((int)ROUNDING) { |
| 1368 | case ROUND_NEAREST: |
| 1369 | if ((a->a & 0x7ff) > 0x400) |
| 1370 | __asm__("addl $1,%0 ; adcl $0,%1" |
| 1371 | :"=r" (b->a),"=r" (b->b) |
| 1372 | :"0" (b->a),"1" (b->b)); |
| 1373 | break; |
| 1374 | case ROUND_DOWN: |
| 1375 | if ((a->exponent & 0x8000) && (a->b & 0xff)) |
| 1376 | __asm__("addl $1,%0 ; adcl $0,%1" |
| 1377 | :"=r" (b->a),"=r" (b->b) |
| 1378 | :"0" (b->a),"1" (b->b)); |
| 1379 | break; |
| 1380 | case ROUND_UP: |
| 1381 | if (!(a->exponent & 0x8000) && (a->b & 0xff)) |
| 1382 | __asm__("addl $1,%0 ; adcl $0,%1" |
| 1383 | :"=r" (b->a),"=r" (b->b) |
| 1384 | :"0" (b->a),"1" (b->b)); |
| 1385 | break; |
| 1386 | } |
| 1387 | } |
| 1388 | |
| 1389 | static void |
| 1390 | frndint(const temp_real *a, temp_real *b) |
| 1391 | { |
| 1392 | int shift = 16383 + 63 - (a->exponent & 0x7fff); |
| 1393 | u_int32_t underflow; |
| 1394 | |
| 1395 | if ((shift < 0) || (shift == 16383+63)) { |
| 1396 | *b = *a; |
| 1397 | return; |
| 1398 | } |
| 1399 | b->a = b->b = underflow = 0; |
| 1400 | b->exponent = a->exponent; |
| 1401 | if (shift < 32) { |
| 1402 | b->b = a->b; b->a = a->a; |
| 1403 | } else if (shift < 64) { |
| 1404 | b->a = a->b; underflow = a->a; |
| 1405 | shift -= 32; |
| 1406 | b->exponent += 32; |
| 1407 | } else if (shift < 96) { |
| 1408 | underflow = a->b; |
| 1409 | shift -= 64; |
| 1410 | b->exponent += 64; |
| 1411 | } else { |
| 1412 | underflow = 1; |
| 1413 | shift = 0; |
| 1414 | } |
| 1415 | b->exponent += shift; |
| 1416 | __asm__("shrdl %2,%1,%0" |
| 1417 | :"=r" (underflow),"=r" (b->a) |
| 1418 | :"c" ((char) shift),"0" (underflow),"1" (b->a)); |
| 1419 | __asm__("shrdl %2,%1,%0" |
| 1420 | :"=r" (b->a),"=r" (b->b) |
| 1421 | :"c" ((char) shift),"0" (b->a),"1" (b->b)); |
| 1422 | __asm__("shrl %1,%0" |
| 1423 | :"=r" (b->b) |
| 1424 | :"c" ((char) shift),"0" (b->b)); |
| 1425 | switch ((int)ROUNDING) { |
| 1426 | case ROUND_NEAREST: |
| 1427 | __asm__("addl %4,%5 ; adcl $0,%0 ; adcl $0,%1" |
| 1428 | :"=r" (b->a),"=r" (b->b) |
| 1429 | :"0" (b->a),"1" (b->b) |
| 1430 | ,"r" (0x7fffffff + (b->a & 1)) |
| 1431 | ,"m" (*&underflow)); |
| 1432 | break; |
| 1433 | case ROUND_UP: |
| 1434 | if ((b->exponent >= 0) && underflow) |
| 1435 | __asm__("addl $1,%0 ; adcl $0,%1" |
| 1436 | :"=r" (b->a),"=r" (b->b) |
| 1437 | :"0" (b->a),"1" (b->b)); |
| 1438 | break; |
| 1439 | case ROUND_DOWN: |
| 1440 | if ((b->exponent < 0) && underflow) |
| 1441 | __asm__("addl $1,%0 ; adcl $0,%1" |
| 1442 | :"=r" (b->a),"=r" (b->b) |
| 1443 | :"0" (b->a),"1" (b->b)); |
| 1444 | break; |
| 1445 | } |
| 1446 | if (b->a || b->b) |
| 1447 | while (b->b >= 0) { |
| 1448 | b->exponent--; |
| 1449 | __asm__("addl %0,%0 ; adcl %1,%1" |
| 1450 | :"=r" (b->a),"=r" (b->b) |
| 1451 | :"0" (b->a),"1" (b->b)); |
| 1452 | } |
| 1453 | else |
| 1454 | b->exponent = 0; |
| 1455 | } |
| 1456 | |
| 1457 | static void |
| 1458 | Fscale(const temp_real *a, const temp_real *b, temp_real *c) |
| 1459 | { |
| 1460 | temp_int ti; |
| 1461 | |
| 1462 | *c = *a; |
| 1463 | if(!c->a && !c->b) { /* 19 Sep 92*/ |
| 1464 | c->exponent = 0; |
| 1465 | return; |
| 1466 | } |
| 1467 | real_to_int(b, &ti); |
| 1468 | if(ti.sign) |
| 1469 | c->exponent -= ti.a; |
| 1470 | else |
| 1471 | c->exponent += ti.a; |
| 1472 | } |
| 1473 | |
| 1474 | static void |
| 1475 | real_to_int(const temp_real *a, temp_int *b) |
| 1476 | { |
| 1477 | int shift = 16383 + 63 - (a->exponent & 0x7fff); |
| 1478 | u_int32_t underflow; |
| 1479 | |
| 1480 | b->a = b->b = underflow = 0; |
| 1481 | b->sign = (a->exponent < 0); |
| 1482 | if (shift < 0) { |
| 1483 | set_OE(); |
| 1484 | return; |
| 1485 | } |
| 1486 | if (shift < 32) { |
| 1487 | b->b = a->b; b->a = a->a; |
| 1488 | } else if (shift < 64) { |
| 1489 | b->a = a->b; underflow = a->a; |
| 1490 | shift -= 32; |
| 1491 | } else if (shift < 96) { |
| 1492 | underflow = a->b; |
| 1493 | shift -= 64; |
| 1494 | } else { |
| 1495 | underflow = 1; |
| 1496 | shift = 0; |
| 1497 | } |
| 1498 | __asm__("shrdl %2,%1,%0" |
| 1499 | :"=r" (underflow),"=r" (b->a) |
| 1500 | :"c" ((char) shift),"0" (underflow),"1" (b->a)); |
| 1501 | __asm__("shrdl %2,%1,%0" |
| 1502 | :"=r" (b->a),"=r" (b->b) |
| 1503 | :"c" ((char) shift),"0" (b->a),"1" (b->b)); |
| 1504 | __asm__("shrl %1,%0" |
| 1505 | :"=r" (b->b) |
| 1506 | :"c" ((char) shift),"0" (b->b)); |
| 1507 | switch ((int)ROUNDING) { |
| 1508 | case ROUND_NEAREST: |
| 1509 | __asm__("addl %4,%5 ; adcl $0,%0 ; adcl $0,%1" |
| 1510 | :"=r" (b->a),"=r" (b->b) |
| 1511 | :"0" (b->a),"1" (b->b) |
| 1512 | ,"r" (0x7fffffff + (b->a & 1)) |
| 1513 | ,"m" (*&underflow)); |
| 1514 | break; |
| 1515 | case ROUND_UP: |
| 1516 | if (!b->sign && underflow) |
| 1517 | __asm__("addl $1,%0 ; adcl $0,%1" |
| 1518 | :"=r" (b->a),"=r" (b->b) |
| 1519 | :"0" (b->a),"1" (b->b)); |
| 1520 | break; |
| 1521 | case ROUND_DOWN: |
| 1522 | if (b->sign && underflow) |
| 1523 | __asm__("addl $1,%0 ; adcl $0,%1" |
| 1524 | :"=r" (b->a),"=r" (b->b) |
| 1525 | :"0" (b->a),"1" (b->b)); |
| 1526 | break; |
| 1527 | } |
| 1528 | } |
| 1529 | |
| 1530 | static void |
| 1531 | int_to_real(const temp_int *a, temp_real *b) |
| 1532 | { |
| 1533 | b->a = a->a; |
| 1534 | b->b = a->b; |
| 1535 | if (b->a || b->b) |
| 1536 | b->exponent = 16383 + 63 + (a->sign? 0x8000:0); |
| 1537 | else { |
| 1538 | b->exponent = 0; |
| 1539 | return; |
| 1540 | } |
| 1541 | while (b->b >= 0) { |
| 1542 | b->exponent--; |
| 1543 | __asm__("addl %0,%0 ; adcl %1,%1" |
| 1544 | :"=r" (b->a),"=r" (b->b) |
| 1545 | :"0" (b->a),"1" (b->b)); |
| 1546 | } |
| 1547 | } |
| 1548 | |
| 1549 | static int |
| 1550 | fpu_modevent(module_t mod, int type, void *unused) |
| 1551 | { |
| 1552 | switch (type) { |
| 1553 | case MOD_LOAD: |
| 1554 | if (pmath_emulate) { |
| 1555 | kprintf("Another Math emulator already present\n"); |
| 1556 | return EBUSY; |
| 1557 | } |
| 1558 | pmath_emulate = math_emulate; |
| 1559 | if (bootverbose) |
| 1560 | kprintf("Math emulator present\n"); |
| 1561 | break; |
| 1562 | case MOD_UNLOAD: |
| 1563 | if (pmath_emulate != math_emulate) { |
| 1564 | kprintf("Cannot unload another math emulator\n"); |
| 1565 | return EACCES; |
| 1566 | } |
| 1567 | pmath_emulate = 0; |
| 1568 | if (bootverbose) |
| 1569 | kprintf("Math emulator unloaded\n"); |
| 1570 | break; |
| 1571 | default: |
| 1572 | break; |
| 1573 | } |
| 1574 | return 0; |
| 1575 | } |
| 1576 | static moduledata_t fpumod = { |
| 1577 | "fpu", |
| 1578 | fpu_modevent, |
| 1579 | 0 |
| 1580 | }; |
| 1581 | DECLARE_MODULE(fpu, fpumod, SI_SUB_DRIVERS, SI_ORDER_ANY); |