fpu.c revision 51498
1/*- 2 * Copyright (c) 1990 William Jolitz. 3 * Copyright (c) 1991 The Regents of the University of California. 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 3. All advertising materials mentioning features or use of this software 15 * must display the following acknowledgement: 16 * This product includes software developed by the University of 17 * California, Berkeley and its contributors. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from: @(#)npx.c 7.2 (Berkeley) 5/12/91 35 * $FreeBSD: head/sys/amd64/amd64/fpu.c 51498 1999-09-21 10:51:47Z phk $ 36 */ 37 38#include "npx.h" 39#if NNPX > 0 40 41#include "opt_debug_npx.h" 42#include "opt_math_emulate.h" 43 44#include <sys/param.h> 45#include <sys/systm.h> 46#include <sys/bus.h> 47#include <sys/kernel.h> 48#include <sys/malloc.h> 49#include <sys/module.h> 50#include <sys/sysctl.h> 51#include <sys/proc.h> 52#include <machine/bus.h> 53#include <sys/rman.h> 54#ifdef NPX_DEBUG 55#include <sys/syslog.h> 56#endif 57#include <sys/signalvar.h> 58 59#ifndef SMP 60#include <machine/asmacros.h> 61#endif 62#include <machine/cputypes.h> 63#include <machine/frame.h> 64#include <machine/ipl.h> 65#include <machine/md_var.h> 66#include <machine/pcb.h> 67#include <machine/psl.h> 68#ifndef SMP 69#include <machine/clock.h> 70#endif 71#include <machine/resource.h> 72#include <machine/specialreg.h> 73#include <machine/segments.h> 74 75#ifndef SMP 76#include <i386/isa/icu.h> 77#include <i386/isa/intr_machdep.h> 78#include <i386/isa/isa.h> 79#endif 80 81/* 82 * 387 and 287 Numeric Coprocessor Extension (NPX) Driver. 83 */ 84 85/* Configuration flags. */ 86#define NPX_DISABLE_I586_OPTIMIZED_BCOPY (1 << 0) 87#define NPX_DISABLE_I586_OPTIMIZED_BZERO (1 << 1) 88#define NPX_DISABLE_I586_OPTIMIZED_COPYIO (1 << 2) 89#define NPX_PREFER_EMULATOR (1 << 3) 90 91#ifdef __GNUC__ 92 93#define fldcw(addr) __asm("fldcw %0" : : "m" (*(addr))) 94#define fnclex() __asm("fnclex") 95#define fninit() __asm("fninit") 96#define fnop() __asm("fnop") 97#define fnsave(addr) __asm __volatile("fnsave %0" : "=m" (*(addr))) 98#define fnstcw(addr) __asm __volatile("fnstcw %0" : "=m" (*(addr))) 99#define fnstsw(addr) __asm __volatile("fnstsw %0" : "=m" (*(addr))) 100#define fp_divide_by_0() __asm("fldz; fld1; fdiv %st,%st(1); fnop") 101#define frstor(addr) __asm("frstor %0" : : "m" (*(addr))) 102#define start_emulating() __asm("smsw %%ax; orb %0,%%al; lmsw %%ax" \ 103 : : "n" (CR0_TS) : "ax") 104#define stop_emulating() __asm("clts") 105 106#else /* not __GNUC__ */ 107 108void fldcw __P((caddr_t addr)); 109void fnclex __P((void)); 110void fninit __P((void)); 111void fnop __P((void)); 112void fnsave __P((caddr_t addr)); 113void fnstcw __P((caddr_t addr)); 114void fnstsw __P((caddr_t addr)); 115void fp_divide_by_0 __P((void)); 116void frstor __P((caddr_t addr)); 117void start_emulating __P((void)); 118void stop_emulating __P((void)); 119 120#endif /* __GNUC__ */ 121 122typedef u_char bool_t; 123 124static int npx_attach __P((device_t dev)); 125 void npx_intr __P((void *)); 126static void npx_identify __P((driver_t *driver, device_t parent)); 127static int npx_probe __P((device_t dev)); 128static int npx_probe1 __P((device_t dev)); 129#ifdef I586_CPU 130static long timezero __P((const char *funcname, 131 void (*func)(void *buf, size_t len))); 132#endif /* I586_CPU */ 133 134int hw_float; /* XXX currently just alias for npx_exists */ 135 136SYSCTL_INT(_hw,HW_FLOATINGPT, floatingpoint, 137 CTLFLAG_RD, &hw_float, 0, 138 "Floatingpoint instructions executed in hardware"); 139 140#ifndef SMP 141static u_int npx0_imask = SWI_CLOCK_MASK; 142static struct gate_descriptor npx_idt_probeintr; 143static int npx_intrno; 144static volatile u_int npx_intrs_while_probing; 145static volatile u_int npx_traps_while_probing; 146#endif 147 148static bool_t npx_ex16; 149static bool_t npx_exists; 150static bool_t npx_irq13; 151static int npx_irq; /* irq number */ 152 153#ifndef SMP 154/* 155 * Special interrupt handlers. Someday intr0-intr15 will be used to count 156 * interrupts. We'll still need a special exception 16 handler. The busy 157 * latch stuff in probeintr() can be moved to npxprobe(). 158 */ 159inthand_t probeintr; 160__asm(" \n\ 161 .text \n\ 162 .p2align 2,0x90 \n\ 163 .type " __XSTRING(CNAME(probeintr)) ",@function \n\ 164" __XSTRING(CNAME(probeintr)) ": \n\ 165 ss \n\ 166 incl " __XSTRING(CNAME(npx_intrs_while_probing)) " \n\ 167 pushl %eax \n\ 168 movb $0x20,%al # EOI (asm in strings loses cpp features) \n\ 169 outb %al,$0xa0 # IO_ICU2 \n\ 170 outb %al,$0x20 # IO_ICU1 \n\ 171 movb $0,%al \n\ 172 outb %al,$0xf0 # clear BUSY# latch \n\ 173 popl %eax \n\ 174 iret \n\ 175"); 176 177inthand_t probetrap; 178__asm(" \n\ 179 .text \n\ 180 .p2align 2,0x90 \n\ 181 .type " __XSTRING(CNAME(probetrap)) ",@function \n\ 182" __XSTRING(CNAME(probetrap)) ": \n\ 183 ss \n\ 184 incl " __XSTRING(CNAME(npx_traps_while_probing)) " \n\ 185 fnclex \n\ 186 iret \n\ 187"); 188#endif /* SMP */ 189 190/* 191 * Identify routine. Create a connection point on our parent for probing. 192 */ 193static void 194npx_identify(driver, parent) 195 driver_t *driver; 196 device_t parent; 197{ 198 device_t child; 199 200 child = BUS_ADD_CHILD(parent, 0, "npx", 0); 201 if (child == NULL) 202 panic("npx_identify"); 203} 204 205/* 206 * Probe routine. Initialize cr0 to give correct behaviour for [f]wait 207 * whether the device exists or not (XXX should be elsewhere). Set flags 208 * to tell npxattach() what to do. Modify device struct if npx doesn't 209 * need to use interrupts. Return 1 if device exists. 210 */ 211static int 212npx_probe(dev) 213 device_t dev; 214{ 215#ifdef SMP 216 217 if (resource_int_value("npx", 0, "irq", &npx_irq) != 0) 218 npx_irq = 13; 219 return npx_probe1(dev); 220 221#else /* SMP */ 222 223 int result; 224 u_long save_eflags; 225 u_char save_icu1_mask; 226 u_char save_icu2_mask; 227 struct gate_descriptor save_idt_npxintr; 228 struct gate_descriptor save_idt_npxtrap; 229 /* 230 * This routine is now just a wrapper for npxprobe1(), to install 231 * special npx interrupt and trap handlers, to enable npx interrupts 232 * and to disable other interrupts. Someday isa_configure() will 233 * install suitable handlers and run with interrupts enabled so we 234 * won't need to do so much here. 235 */ 236 if (resource_int_value("npx", 0, "irq", &npx_irq) != 0) 237 npx_irq = 13; 238 npx_intrno = NRSVIDT + npx_irq; 239 save_eflags = read_eflags(); 240 disable_intr(); 241 save_icu1_mask = inb(IO_ICU1 + 1); 242 save_icu2_mask = inb(IO_ICU2 + 1); 243 save_idt_npxintr = idt[npx_intrno]; 244 save_idt_npxtrap = idt[16]; 245 outb(IO_ICU1 + 1, ~IRQ_SLAVE); 246 outb(IO_ICU2 + 1, ~(1 << (npx_irq - 8))); 247 setidt(16, probetrap, SDT_SYS386TGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 248 setidt(npx_intrno, probeintr, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 249 npx_idt_probeintr = idt[npx_intrno]; 250 enable_intr(); 251 result = npx_probe1(dev); 252 disable_intr(); 253 outb(IO_ICU1 + 1, save_icu1_mask); 254 outb(IO_ICU2 + 1, save_icu2_mask); 255 idt[npx_intrno] = save_idt_npxintr; 256 idt[16] = save_idt_npxtrap; 257 write_eflags(save_eflags); 258 return (result); 259 260#endif /* SMP */ 261} 262 263static int 264npx_probe1(dev) 265 device_t dev; 266{ 267#ifndef SMP 268 u_short control; 269 u_short status; 270#endif 271 272 /* 273 * Partially reset the coprocessor, if any. Some BIOS's don't reset 274 * it after a warm boot. 275 */ 276 outb(0xf1, 0); /* full reset on some systems, NOP on others */ 277 outb(0xf0, 0); /* clear BUSY# latch */ 278 /* 279 * Prepare to trap all ESC (i.e., NPX) instructions and all WAIT 280 * instructions. We must set the CR0_MP bit and use the CR0_TS 281 * bit to control the trap, because setting the CR0_EM bit does 282 * not cause WAIT instructions to trap. It's important to trap 283 * WAIT instructions - otherwise the "wait" variants of no-wait 284 * control instructions would degenerate to the "no-wait" variants 285 * after FP context switches but work correctly otherwise. It's 286 * particularly important to trap WAITs when there is no NPX - 287 * otherwise the "wait" variants would always degenerate. 288 * 289 * Try setting CR0_NE to get correct error reporting on 486DX's. 290 * Setting it should fail or do nothing on lesser processors. 291 */ 292 load_cr0(rcr0() | CR0_MP | CR0_NE); 293 /* 294 * But don't trap while we're probing. 295 */ 296 stop_emulating(); 297 /* 298 * Finish resetting the coprocessor, if any. If there is an error 299 * pending, then we may get a bogus IRQ13, but probeintr() will handle 300 * it OK. Bogus halts have never been observed, but we enabled 301 * IRQ13 and cleared the BUSY# latch early to handle them anyway. 302 */ 303 fninit(); 304 305#ifdef SMP 306 /* 307 * Exception 16 MUST work for SMP. 308 */ 309 npx_irq13 = 0; 310 npx_ex16 = hw_float = npx_exists = 1; 311 device_set_desc(dev, "math processor"); 312 return (0); 313 314#else /* !SMP */ 315 device_set_desc(dev, "math processor"); 316 317 /* 318 * Don't use fwait here because it might hang. 319 * Don't use fnop here because it usually hangs if there is no FPU. 320 */ 321 DELAY(1000); /* wait for any IRQ13 */ 322#ifdef DIAGNOSTIC 323 if (npx_intrs_while_probing != 0) 324 printf("fninit caused %u bogus npx interrupt(s)\n", 325 npx_intrs_while_probing); 326 if (npx_traps_while_probing != 0) 327 printf("fninit caused %u bogus npx trap(s)\n", 328 npx_traps_while_probing); 329#endif 330 /* 331 * Check for a status of mostly zero. 332 */ 333 status = 0x5a5a; 334 fnstsw(&status); 335 if ((status & 0xb8ff) == 0) { 336 /* 337 * Good, now check for a proper control word. 338 */ 339 control = 0x5a5a; 340 fnstcw(&control); 341 if ((control & 0x1f3f) == 0x033f) { 342 hw_float = npx_exists = 1; 343 /* 344 * We have an npx, now divide by 0 to see if exception 345 * 16 works. 346 */ 347 control &= ~(1 << 2); /* enable divide by 0 trap */ 348 fldcw(&control); 349 npx_traps_while_probing = npx_intrs_while_probing = 0; 350 fp_divide_by_0(); 351 if (npx_traps_while_probing != 0) { 352 /* 353 * Good, exception 16 works. 354 */ 355 npx_ex16 = 1; 356 return (0); 357 } 358 if (npx_intrs_while_probing != 0) { 359 int rid; 360 struct resource *r; 361 void *intr; 362 /* 363 * Bad, we are stuck with IRQ13. 364 */ 365 npx_irq13 = 1; 366 /* 367 * npxattach would be too late to set npx0_imask 368 */ 369 npx0_imask |= (1 << npx_irq); 370 371 /* 372 * We allocate these resources permanently, 373 * so there is no need to keep track of them. 374 */ 375 rid = 0; 376 r = bus_alloc_resource(dev, SYS_RES_IOPORT, 377 &rid, IO_NPX, IO_NPX, 378 IO_NPXSIZE, RF_ACTIVE); 379 if (r == 0) 380 panic("npx: can't get ports"); 381 rid = 0; 382 r = bus_alloc_resource(dev, SYS_RES_IRQ, 383 &rid, npx_irq, npx_irq, 384 1, RF_ACTIVE); 385 if (r == 0) 386 panic("npx: can't get IRQ"); 387 BUS_SETUP_INTR(device_get_parent(dev), 388 dev, r, INTR_TYPE_MISC, 389 npx_intr, 0, &intr); 390 if (intr == 0) 391 panic("npx: can't create intr"); 392 393 return (0); 394 } 395 /* 396 * Worse, even IRQ13 is broken. Use emulator. 397 */ 398 } 399 } 400 /* 401 * Probe failed, but we want to get to npxattach to initialize the 402 * emulator and say that it has been installed. XXX handle devices 403 * that aren't really devices better. 404 */ 405 return (0); 406#endif /* SMP */ 407} 408 409/* 410 * Attach routine - announce which it is, and wire into system 411 */ 412int 413npx_attach(dev) 414 device_t dev; 415{ 416 int flags; 417 418 if (resource_int_value("npx", 0, "flags", &flags) != 0) 419 flags = 0; 420 421 device_print_prettyname(dev); 422 if (flags) 423 printf("flags 0x%x ", flags); 424 if (npx_irq13) { 425 printf("using IRQ 13 interface\n"); 426 } else { 427#if defined(MATH_EMULATE) || defined(GPL_MATH_EMULATE) 428 if (npx_ex16) { 429 if (!(flags & NPX_PREFER_EMULATOR)) 430 printf("INT 16 interface\n"); 431 else { 432 printf("FPU exists, but flags request " 433 "emulator\n"); 434 hw_float = npx_exists = 0; 435 } 436 } else if (npx_exists) { 437 printf("error reporting broken; using 387 emulator\n"); 438 hw_float = npx_exists = 0; 439 } else 440 printf("387 emulator\n"); 441#else 442 if (npx_ex16) { 443 printf("INT 16 interface\n"); 444 if (flags & NPX_PREFER_EMULATOR) { 445 printf("emulator requested, but none compiled " 446 "into kernel, using FPU\n"); 447 } 448 } else 449 printf("no 387 emulator in kernel and no FPU!\n"); 450#endif 451 } 452 npxinit(__INITIAL_NPXCW__); 453 454#ifdef I586_CPU 455 if (cpu_class == CPUCLASS_586 && npx_ex16 && npx_exists && 456 timezero("i586_bzero()", i586_bzero) < 457 timezero("bzero()", bzero) * 4 / 5) { 458 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_BCOPY)) { 459 bcopy_vector = i586_bcopy; 460 ovbcopy_vector = i586_bcopy; 461 } 462 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_BZERO)) 463 bzero = i586_bzero; 464 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_COPYIO)) { 465 copyin_vector = i586_copyin; 466 copyout_vector = i586_copyout; 467 } 468 } 469#endif 470 471 return (0); /* XXX unused */ 472} 473 474/* 475 * Initialize floating point unit. 476 */ 477void 478npxinit(control) 479 u_short control; 480{ 481 struct save87 dummy; 482 483 if (!npx_exists) 484 return; 485 /* 486 * fninit has the same h/w bugs as fnsave. Use the detoxified 487 * fnsave to throw away any junk in the fpu. npxsave() initializes 488 * the fpu and sets npxproc = NULL as important side effects. 489 */ 490 npxsave(&dummy); 491 stop_emulating(); 492 fldcw(&control); 493 if (curpcb != NULL) 494 fnsave(&curpcb->pcb_savefpu); 495 start_emulating(); 496} 497 498/* 499 * Free coprocessor (if we have it). 500 */ 501void 502npxexit(p) 503 struct proc *p; 504{ 505 506 if (p == npxproc) 507 npxsave(&curpcb->pcb_savefpu); 508#ifdef NPX_DEBUG 509 if (npx_exists) { 510 u_int masked_exceptions; 511 512 masked_exceptions = curpcb->pcb_savefpu.sv_env.en_cw 513 & curpcb->pcb_savefpu.sv_env.en_sw & 0x7f; 514 /* 515 * Log exceptions that would have trapped with the old 516 * control word (overflow, divide by 0, and invalid operand). 517 */ 518 if (masked_exceptions & 0x0d) 519 log(LOG_ERR, 520 "pid %d (%s) exited with masked floating point exceptions 0x%02x\n", 521 p->p_pid, p->p_comm, masked_exceptions); 522 } 523#endif 524} 525 526/* 527 * The following mechanism is used to ensure that the FPE_... value 528 * that is passed as a trapcode to the signal handler of the user 529 * process does not have more than one bit set. 530 * 531 * Multiple bits may be set if the user process modifies the control 532 * word while a status word bit is already set. While this is a sign 533 * of bad coding, we have no choise than to narrow them down to one 534 * bit, since we must not send a trapcode that is not exactly one of 535 * the FPE_ macros. 536 * 537 * The mechanism has a static table with 127 entries. Each combination 538 * of the 7 FPU status word exception bits directly translates to a 539 * position in this table, where a single FPE_... value is stored. 540 * This FPE_... value stored there is considered the "most important" 541 * of the exception bits and will be sent as the signal code. The 542 * precedence of the bits is based upon Intel Document "Numerical 543 * Applications", Chapter "Special Computational Situations". 544 * 545 * The macro to choose one of these values does these steps: 1) Throw 546 * away status word bits that cannot be masked. 2) Throw away the bits 547 * currently masked in the control word, assuming the user isn't 548 * interested in them anymore. 3) Reinsert status word bit 7 (stack 549 * fault) if it is set, which cannot be masked but must be presered. 550 * 4) Use the remaining bits to point into the trapcode table. 551 * 552 * The 6 maskable bits in order of their preference, as stated in the 553 * above referenced Intel manual: 554 * 1 Invalid operation (FP_X_INV) 555 * 1a Stack underflow 556 * 1b Stack overflow 557 * 1c Operand of unsupported format 558 * 1d SNaN operand. 559 * 2 QNaN operand (not an exception, irrelavant here) 560 * 3 Any other invalid-operation not mentioned above or zero divide 561 * (FP_X_INV, FP_X_DZ) 562 * 4 Denormal operand (FP_X_DNML) 563 * 5 Numeric over/underflow (FP_X_OFL, FP_X_UFL) 564 * 6 Inexact result (FP_X_IMP) 565 */ 566static char fpetable[128] = { 567 0, 568 FPE_FLTINV, /* 1 - INV */ 569 FPE_FLTUND, /* 2 - DNML */ 570 FPE_FLTINV, /* 3 - INV | DNML */ 571 FPE_FLTDIV, /* 4 - DZ */ 572 FPE_FLTINV, /* 5 - INV | DZ */ 573 FPE_FLTDIV, /* 6 - DNML | DZ */ 574 FPE_FLTINV, /* 7 - INV | DNML | DZ */ 575 FPE_FLTOVF, /* 8 - OFL */ 576 FPE_FLTINV, /* 9 - INV | OFL */ 577 FPE_FLTUND, /* A - DNML | OFL */ 578 FPE_FLTINV, /* B - INV | DNML | OFL */ 579 FPE_FLTDIV, /* C - DZ | OFL */ 580 FPE_FLTINV, /* D - INV | DZ | OFL */ 581 FPE_FLTDIV, /* E - DNML | DZ | OFL */ 582 FPE_FLTINV, /* F - INV | DNML | DZ | OFL */ 583 FPE_FLTUND, /* 10 - UFL */ 584 FPE_FLTINV, /* 11 - INV | UFL */ 585 FPE_FLTUND, /* 12 - DNML | UFL */ 586 FPE_FLTINV, /* 13 - INV | DNML | UFL */ 587 FPE_FLTDIV, /* 14 - DZ | UFL */ 588 FPE_FLTINV, /* 15 - INV | DZ | UFL */ 589 FPE_FLTDIV, /* 16 - DNML | DZ | UFL */ 590 FPE_FLTINV, /* 17 - INV | DNML | DZ | UFL */ 591 FPE_FLTOVF, /* 18 - OFL | UFL */ 592 FPE_FLTINV, /* 19 - INV | OFL | UFL */ 593 FPE_FLTUND, /* 1A - DNML | OFL | UFL */ 594 FPE_FLTINV, /* 1B - INV | DNML | OFL | UFL */ 595 FPE_FLTDIV, /* 1C - DZ | OFL | UFL */ 596 FPE_FLTINV, /* 1D - INV | DZ | OFL | UFL */ 597 FPE_FLTDIV, /* 1E - DNML | DZ | OFL | UFL */ 598 FPE_FLTINV, /* 1F - INV | DNML | DZ | OFL | UFL */ 599 FPE_FLTRES, /* 20 - IMP */ 600 FPE_FLTINV, /* 21 - INV | IMP */ 601 FPE_FLTUND, /* 22 - DNML | IMP */ 602 FPE_FLTINV, /* 23 - INV | DNML | IMP */ 603 FPE_FLTDIV, /* 24 - DZ | IMP */ 604 FPE_FLTINV, /* 25 - INV | DZ | IMP */ 605 FPE_FLTDIV, /* 26 - DNML | DZ | IMP */ 606 FPE_FLTINV, /* 27 - INV | DNML | DZ | IMP */ 607 FPE_FLTOVF, /* 28 - OFL | IMP */ 608 FPE_FLTINV, /* 29 - INV | OFL | IMP */ 609 FPE_FLTUND, /* 2A - DNML | OFL | IMP */ 610 FPE_FLTINV, /* 2B - INV | DNML | OFL | IMP */ 611 FPE_FLTDIV, /* 2C - DZ | OFL | IMP */ 612 FPE_FLTINV, /* 2D - INV | DZ | OFL | IMP */ 613 FPE_FLTDIV, /* 2E - DNML | DZ | OFL | IMP */ 614 FPE_FLTINV, /* 2F - INV | DNML | DZ | OFL | IMP */ 615 FPE_FLTUND, /* 30 - UFL | IMP */ 616 FPE_FLTINV, /* 31 - INV | UFL | IMP */ 617 FPE_FLTUND, /* 32 - DNML | UFL | IMP */ 618 FPE_FLTINV, /* 33 - INV | DNML | UFL | IMP */ 619 FPE_FLTDIV, /* 34 - DZ | UFL | IMP */ 620 FPE_FLTINV, /* 35 - INV | DZ | UFL | IMP */ 621 FPE_FLTDIV, /* 36 - DNML | DZ | UFL | IMP */ 622 FPE_FLTINV, /* 37 - INV | DNML | DZ | UFL | IMP */ 623 FPE_FLTOVF, /* 38 - OFL | UFL | IMP */ 624 FPE_FLTINV, /* 39 - INV | OFL | UFL | IMP */ 625 FPE_FLTUND, /* 3A - DNML | OFL | UFL | IMP */ 626 FPE_FLTINV, /* 3B - INV | DNML | OFL | UFL | IMP */ 627 FPE_FLTDIV, /* 3C - DZ | OFL | UFL | IMP */ 628 FPE_FLTINV, /* 3D - INV | DZ | OFL | UFL | IMP */ 629 FPE_FLTDIV, /* 3E - DNML | DZ | OFL | UFL | IMP */ 630 FPE_FLTINV, /* 3F - INV | DNML | DZ | OFL | UFL | IMP */ 631 FPE_FLTSUB, /* 40 - STK */ 632 FPE_FLTSUB, /* 41 - INV | STK */ 633 FPE_FLTUND, /* 42 - DNML | STK */ 634 FPE_FLTSUB, /* 43 - INV | DNML | STK */ 635 FPE_FLTDIV, /* 44 - DZ | STK */ 636 FPE_FLTSUB, /* 45 - INV | DZ | STK */ 637 FPE_FLTDIV, /* 46 - DNML | DZ | STK */ 638 FPE_FLTSUB, /* 47 - INV | DNML | DZ | STK */ 639 FPE_FLTOVF, /* 48 - OFL | STK */ 640 FPE_FLTSUB, /* 49 - INV | OFL | STK */ 641 FPE_FLTUND, /* 4A - DNML | OFL | STK */ 642 FPE_FLTSUB, /* 4B - INV | DNML | OFL | STK */ 643 FPE_FLTDIV, /* 4C - DZ | OFL | STK */ 644 FPE_FLTSUB, /* 4D - INV | DZ | OFL | STK */ 645 FPE_FLTDIV, /* 4E - DNML | DZ | OFL | STK */ 646 FPE_FLTSUB, /* 4F - INV | DNML | DZ | OFL | STK */ 647 FPE_FLTUND, /* 50 - UFL | STK */ 648 FPE_FLTSUB, /* 51 - INV | UFL | STK */ 649 FPE_FLTUND, /* 52 - DNML | UFL | STK */ 650 FPE_FLTSUB, /* 53 - INV | DNML | UFL | STK */ 651 FPE_FLTDIV, /* 54 - DZ | UFL | STK */ 652 FPE_FLTSUB, /* 55 - INV | DZ | UFL | STK */ 653 FPE_FLTDIV, /* 56 - DNML | DZ | UFL | STK */ 654 FPE_FLTSUB, /* 57 - INV | DNML | DZ | UFL | STK */ 655 FPE_FLTOVF, /* 58 - OFL | UFL | STK */ 656 FPE_FLTSUB, /* 59 - INV | OFL | UFL | STK */ 657 FPE_FLTUND, /* 5A - DNML | OFL | UFL | STK */ 658 FPE_FLTSUB, /* 5B - INV | DNML | OFL | UFL | STK */ 659 FPE_FLTDIV, /* 5C - DZ | OFL | UFL | STK */ 660 FPE_FLTSUB, /* 5D - INV | DZ | OFL | UFL | STK */ 661 FPE_FLTDIV, /* 5E - DNML | DZ | OFL | UFL | STK */ 662 FPE_FLTSUB, /* 5F - INV | DNML | DZ | OFL | UFL | STK */ 663 FPE_FLTRES, /* 60 - IMP | STK */ 664 FPE_FLTSUB, /* 61 - INV | IMP | STK */ 665 FPE_FLTUND, /* 62 - DNML | IMP | STK */ 666 FPE_FLTSUB, /* 63 - INV | DNML | IMP | STK */ 667 FPE_FLTDIV, /* 64 - DZ | IMP | STK */ 668 FPE_FLTSUB, /* 65 - INV | DZ | IMP | STK */ 669 FPE_FLTDIV, /* 66 - DNML | DZ | IMP | STK */ 670 FPE_FLTSUB, /* 67 - INV | DNML | DZ | IMP | STK */ 671 FPE_FLTOVF, /* 68 - OFL | IMP | STK */ 672 FPE_FLTSUB, /* 69 - INV | OFL | IMP | STK */ 673 FPE_FLTUND, /* 6A - DNML | OFL | IMP | STK */ 674 FPE_FLTSUB, /* 6B - INV | DNML | OFL | IMP | STK */ 675 FPE_FLTDIV, /* 6C - DZ | OFL | IMP | STK */ 676 FPE_FLTSUB, /* 6D - INV | DZ | OFL | IMP | STK */ 677 FPE_FLTDIV, /* 6E - DNML | DZ | OFL | IMP | STK */ 678 FPE_FLTSUB, /* 6F - INV | DNML | DZ | OFL | IMP | STK */ 679 FPE_FLTUND, /* 70 - UFL | IMP | STK */ 680 FPE_FLTSUB, /* 71 - INV | UFL | IMP | STK */ 681 FPE_FLTUND, /* 72 - DNML | UFL | IMP | STK */ 682 FPE_FLTSUB, /* 73 - INV | DNML | UFL | IMP | STK */ 683 FPE_FLTDIV, /* 74 - DZ | UFL | IMP | STK */ 684 FPE_FLTSUB, /* 75 - INV | DZ | UFL | IMP | STK */ 685 FPE_FLTDIV, /* 76 - DNML | DZ | UFL | IMP | STK */ 686 FPE_FLTSUB, /* 77 - INV | DNML | DZ | UFL | IMP | STK */ 687 FPE_FLTOVF, /* 78 - OFL | UFL | IMP | STK */ 688 FPE_FLTSUB, /* 79 - INV | OFL | UFL | IMP | STK */ 689 FPE_FLTUND, /* 7A - DNML | OFL | UFL | IMP | STK */ 690 FPE_FLTSUB, /* 7B - INV | DNML | OFL | UFL | IMP | STK */ 691 FPE_FLTDIV, /* 7C - DZ | OFL | UFL | IMP | STK */ 692 FPE_FLTSUB, /* 7D - INV | DZ | OFL | UFL | IMP | STK */ 693 FPE_FLTDIV, /* 7E - DNML | DZ | OFL | UFL | IMP | STK */ 694 FPE_FLTSUB, /* 7F - INV | DNML | DZ | OFL | UFL | IMP | STK */ 695}; 696 697/* 698 * Preserve the FP status word, clear FP exceptions, then generate a SIGFPE. 699 * 700 * Clearing exceptions is necessary mainly to avoid IRQ13 bugs. We now 701 * depend on longjmp() restoring a usable state. Restoring the state 702 * or examining it might fail if we didn't clear exceptions. 703 * 704 * The error code chosen will be one of the FPE_... macros. It will be 705 * sent as the second argument to old BSD-style signal handlers and as 706 * "siginfo_t->si_code" (second argument) to SA_SIGINFO signal handlers. 707 * 708 * XXX the FP state is not preserved across signal handlers. So signal 709 * handlers cannot afford to do FP unless they preserve the state or 710 * longjmp() out. Both preserving the state and longjmp()ing may be 711 * destroyed by IRQ13 bugs. Clearing FP exceptions is not an acceptable 712 * solution for signals other than SIGFPE. 713 */ 714void 715npx_intr(dummy) 716 void *dummy; 717{ 718 int code; 719 u_short control; 720 struct intrframe *frame; 721 722 if (npxproc == NULL || !npx_exists) { 723 printf("npxintr: npxproc = %p, curproc = %p, npx_exists = %d\n", 724 npxproc, curproc, npx_exists); 725 panic("npxintr from nowhere"); 726 } 727 if (npxproc != curproc) { 728 printf("npxintr: npxproc = %p, curproc = %p, npx_exists = %d\n", 729 npxproc, curproc, npx_exists); 730 panic("npxintr from non-current process"); 731 } 732 733 outb(0xf0, 0); 734 fnstsw(&curpcb->pcb_savefpu.sv_ex_sw); 735 fnstcw(&control); 736 fnclex(); 737 738 /* 739 * Pass exception to process. 740 */ 741 frame = (struct intrframe *)&dummy; /* XXX */ 742 if ((ISPL(frame->if_cs) == SEL_UPL) || (frame->if_eflags & PSL_VM)) { 743 /* 744 * Interrupt is essentially a trap, so we can afford to call 745 * the SIGFPE handler (if any) as soon as the interrupt 746 * returns. 747 * 748 * XXX little or nothing is gained from this, and plenty is 749 * lost - the interrupt frame has to contain the trap frame 750 * (this is otherwise only necessary for the rescheduling trap 751 * in doreti, and the frame for that could easily be set up 752 * just before it is used). 753 */ 754 curproc->p_md.md_regs = INTR_TO_TRAPFRAME(frame); 755 /* 756 * Encode the appropriate code for detailed information on 757 * this exception. 758 */ 759 code = 760 fpetable[(curpcb->pcb_savefpu.sv_ex_sw & ~control & 0x3f) | 761 (curpcb->pcb_savefpu.sv_ex_sw & 0x40)]; 762 trapsignal(curproc, SIGFPE, code); 763 } else { 764 /* 765 * Nested interrupt. These losers occur when: 766 * o an IRQ13 is bogusly generated at a bogus time, e.g.: 767 * o immediately after an fnsave or frstor of an 768 * error state. 769 * o a couple of 386 instructions after 770 * "fstpl _memvar" causes a stack overflow. 771 * These are especially nasty when combined with a 772 * trace trap. 773 * o an IRQ13 occurs at the same time as another higher- 774 * priority interrupt. 775 * 776 * Treat them like a true async interrupt. 777 */ 778 psignal(curproc, SIGFPE); 779 } 780} 781 782/* 783 * Implement device not available (DNA) exception 784 * 785 * It would be better to switch FP context here (if curproc != npxproc) 786 * and not necessarily for every context switch, but it is too hard to 787 * access foreign pcb's. 788 */ 789int 790npxdna() 791{ 792 if (!npx_exists) 793 return (0); 794 if (npxproc != NULL) { 795 printf("npxdna: npxproc = %p, curproc = %p\n", 796 npxproc, curproc); 797 panic("npxdna"); 798 } 799 stop_emulating(); 800 /* 801 * Record new context early in case frstor causes an IRQ13. 802 */ 803 npxproc = curproc; 804 curpcb->pcb_savefpu.sv_ex_sw = 0; 805 /* 806 * The following frstor may cause an IRQ13 when the state being 807 * restored has a pending error. The error will appear to have been 808 * triggered by the current (npx) user instruction even when that 809 * instruction is a no-wait instruction that should not trigger an 810 * error (e.g., fnclex). On at least one 486 system all of the 811 * no-wait instructions are broken the same as frstor, so our 812 * treatment does not amplify the breakage. On at least one 813 * 386/Cyrix 387 system, fnclex works correctly while frstor and 814 * fnsave are broken, so our treatment breaks fnclex if it is the 815 * first FPU instruction after a context switch. 816 */ 817 frstor(&curpcb->pcb_savefpu); 818 819 return (1); 820} 821 822/* 823 * Wrapper for fnsave instruction to handle h/w bugs. If there is an error 824 * pending, then fnsave generates a bogus IRQ13 on some systems. Force 825 * any IRQ13 to be handled immediately, and then ignore it. This routine is 826 * often called at splhigh so it must not use many system services. In 827 * particular, it's much easier to install a special handler than to 828 * guarantee that it's safe to use npxintr() and its supporting code. 829 */ 830void 831npxsave(addr) 832 struct save87 *addr; 833{ 834#ifdef SMP 835 836 stop_emulating(); 837 fnsave(addr); 838 /* fnop(); */ 839 start_emulating(); 840 npxproc = NULL; 841 842#else /* SMP */ 843 844 u_char icu1_mask; 845 u_char icu2_mask; 846 u_char old_icu1_mask; 847 u_char old_icu2_mask; 848 struct gate_descriptor save_idt_npxintr; 849 850 disable_intr(); 851 old_icu1_mask = inb(IO_ICU1 + 1); 852 old_icu2_mask = inb(IO_ICU2 + 1); 853 save_idt_npxintr = idt[npx_intrno]; 854 outb(IO_ICU1 + 1, old_icu1_mask & ~(IRQ_SLAVE | npx0_imask)); 855 outb(IO_ICU2 + 1, old_icu2_mask & ~(npx0_imask >> 8)); 856 idt[npx_intrno] = npx_idt_probeintr; 857 enable_intr(); 858 stop_emulating(); 859 fnsave(addr); 860 fnop(); 861 start_emulating(); 862 npxproc = NULL; 863 disable_intr(); 864 icu1_mask = inb(IO_ICU1 + 1); /* masks may have changed */ 865 icu2_mask = inb(IO_ICU2 + 1); 866 outb(IO_ICU1 + 1, 867 (icu1_mask & ~npx0_imask) | (old_icu1_mask & npx0_imask)); 868 outb(IO_ICU2 + 1, 869 (icu2_mask & ~(npx0_imask >> 8)) 870 | (old_icu2_mask & (npx0_imask >> 8))); 871 idt[npx_intrno] = save_idt_npxintr; 872 enable_intr(); /* back to usual state */ 873 874#endif /* SMP */ 875} 876 877#ifdef I586_CPU 878static long 879timezero(funcname, func) 880 const char *funcname; 881 void (*func) __P((void *buf, size_t len)); 882 883{ 884 void *buf; 885#define BUFSIZE 1000000 886 long usec; 887 struct timeval finish, start; 888 889 buf = malloc(BUFSIZE, M_TEMP, M_NOWAIT); 890 if (buf == NULL) 891 return (BUFSIZE); 892 microtime(&start); 893 (*func)(buf, BUFSIZE); 894 microtime(&finish); 895 usec = 1000000 * (finish.tv_sec - start.tv_sec) + 896 finish.tv_usec - start.tv_usec; 897 if (usec <= 0) 898 usec = 1; 899 if (bootverbose) 900 printf("%s bandwidth = %ld bytes/sec\n", 901 funcname, (long)(BUFSIZE * (int64_t)1000000 / usec)); 902 free(buf, M_TEMP); 903 return (usec); 904} 905#endif /* I586_CPU */ 906 907static device_method_t npx_methods[] = { 908 /* Device interface */ 909 DEVMETHOD(device_identify, npx_identify), 910 DEVMETHOD(device_probe, npx_probe), 911 DEVMETHOD(device_attach, npx_attach), 912 DEVMETHOD(device_detach, bus_generic_detach), 913 DEVMETHOD(device_shutdown, bus_generic_shutdown), 914 DEVMETHOD(device_suspend, bus_generic_suspend), 915 DEVMETHOD(device_resume, bus_generic_resume), 916 917 { 0, 0 } 918}; 919 920static driver_t npx_driver = { 921 "npx", 922 npx_methods, 923 1, /* no softc */ 924}; 925 926static devclass_t npx_devclass; 927 928/* 929 * We prefer to attach to the root nexus so that the usual case (exception 16) 930 * doesn't describe the processor as being `on isa'. 931 */ 932DRIVER_MODULE(npx, nexus, npx_driver, npx_devclass, 0, 0); 933 934#endif /* NNPX > 0 */ 935