fpu.c revision 91315
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 91315 2002-02-26 17:06:21Z dillon $ 36 */ 37 38#include "opt_cpu.h" 39#include "opt_debug_npx.h" 40#include "opt_isa.h" 41#include "opt_math_emulate.h" 42#include "opt_npx.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/lock.h> 49#include <sys/malloc.h> 50#include <sys/module.h> 51#include <sys/mutex.h> 52#include <sys/mutex.h> 53#include <sys/proc.h> 54#include <sys/sysctl.h> 55#include <machine/bus.h> 56#include <sys/rman.h> 57#ifdef NPX_DEBUG 58#include <sys/syslog.h> 59#endif 60#include <sys/signalvar.h> 61#include <sys/user.h> 62 63#ifndef SMP 64#include <machine/asmacros.h> 65#endif 66#include <machine/cputypes.h> 67#include <machine/frame.h> 68#include <machine/md_var.h> 69#include <machine/pcb.h> 70#include <machine/psl.h> 71#ifndef SMP 72#include <machine/clock.h> 73#endif 74#include <machine/resource.h> 75#include <machine/specialreg.h> 76#include <machine/segments.h> 77 78#ifndef SMP 79#include <i386/isa/icu.h> 80#ifdef PC98 81#include <pc98/pc98/pc98.h> 82#else 83#include <i386/isa/isa.h> 84#endif 85#endif 86#include <i386/isa/intr_machdep.h> 87#ifdef DEV_ISA 88#include <isa/isavar.h> 89#endif 90 91/* 92 * 387 and 287 Numeric Coprocessor Extension (NPX) Driver. 93 */ 94 95/* Configuration flags. */ 96#define NPX_DISABLE_I586_OPTIMIZED_BCOPY (1 << 0) 97#define NPX_DISABLE_I586_OPTIMIZED_BZERO (1 << 1) 98#define NPX_DISABLE_I586_OPTIMIZED_COPYIO (1 << 2) 99#define NPX_PREFER_EMULATOR (1 << 3) 100 101#ifdef __GNUC__ 102 103#define fldcw(addr) __asm("fldcw %0" : : "m" (*(addr))) 104#define fnclex() __asm("fnclex") 105#define fninit() __asm("fninit") 106#define fnsave(addr) __asm __volatile("fnsave %0" : "=m" (*(addr))) 107#define fnstcw(addr) __asm __volatile("fnstcw %0" : "=m" (*(addr))) 108#define fnstsw(addr) __asm __volatile("fnstsw %0" : "=m" (*(addr))) 109#define fp_divide_by_0() __asm("fldz; fld1; fdiv %st,%st(1); fnop") 110#define frstor(addr) __asm("frstor %0" : : "m" (*(addr))) 111#ifdef CPU_ENABLE_SSE 112#define fxrstor(addr) __asm("fxrstor %0" : : "m" (*(addr))) 113#define fxsave(addr) __asm __volatile("fxsave %0" : "=m" (*(addr))) 114#endif 115#define start_emulating() __asm("smsw %%ax; orb %0,%%al; lmsw %%ax" \ 116 : : "n" (CR0_TS) : "ax") 117#define stop_emulating() __asm("clts") 118 119#else /* not __GNUC__ */ 120 121void fldcw __P((caddr_t addr)); 122void fnclex __P((void)); 123void fninit __P((void)); 124void fnsave __P((caddr_t addr)); 125void fnstcw __P((caddr_t addr)); 126void fnstsw __P((caddr_t addr)); 127void fp_divide_by_0 __P((void)); 128void frstor __P((caddr_t addr)); 129#ifdef CPU_ENABLE_SSE 130void fxsave __P((caddr_t addr)); 131void fxrstor __P((caddr_t addr)); 132#endif 133void start_emulating __P((void)); 134void stop_emulating __P((void)); 135 136#endif /* __GNUC__ */ 137 138#ifdef CPU_ENABLE_SSE 139#define GET_FPU_CW(thread) \ 140 (cpu_fxsr ? \ 141 (thread)->td_pcb->pcb_save.sv_xmm.sv_env.en_cw : \ 142 (thread)->td_pcb->pcb_save.sv_87.sv_env.en_cw) 143#define GET_FPU_SW(thread) \ 144 (cpu_fxsr ? \ 145 (thread)->td_pcb->pcb_save.sv_xmm.sv_env.en_sw : \ 146 (thread)->td_pcb->pcb_save.sv_87.sv_env.en_sw) 147#define GET_FPU_EXSW_PTR(pcb) \ 148 (cpu_fxsr ? \ 149 &(pcb)->pcb_save.sv_xmm.sv_ex_sw : \ 150 &(pcb)->pcb_save.sv_87.sv_ex_sw) 151#else /* CPU_ENABLE_SSE */ 152#define GET_FPU_CW(thread) \ 153 (thread->td_pcb->pcb_save.sv_87.sv_env.en_cw) 154#define GET_FPU_SW(thread) \ 155 (thread->td_pcb->pcb_save.sv_87.sv_env.en_sw) 156#define GET_FPU_EXSW_PTR(pcb) \ 157 (&(pcb)->pcb_save.sv_87.sv_ex_sw) 158#endif /* CPU_ENABLE_SSE */ 159 160typedef u_char bool_t; 161 162static int npx_attach __P((device_t dev)); 163static void npx_identify __P((driver_t *driver, device_t parent)); 164#ifndef SMP 165static void npx_intr __P((void *)); 166#endif 167static int npx_probe __P((device_t dev)); 168static void fpusave __P((union savefpu *)); 169static void fpurstor __P((union savefpu *)); 170#ifdef I586_CPU_XXX 171static long timezero __P((const char *funcname, 172 void (*func)(void *buf, size_t len))); 173#endif /* I586_CPU */ 174 175int hw_float; /* XXX currently just alias for npx_exists */ 176 177SYSCTL_INT(_hw,HW_FLOATINGPT, floatingpoint, 178 CTLFLAG_RD, &hw_float, 0, 179 "Floatingpoint instructions executed in hardware"); 180 181#ifndef SMP 182static volatile u_int npx_intrs_while_probing; 183static volatile u_int npx_traps_while_probing; 184#endif 185 186static bool_t npx_ex16; 187static bool_t npx_exists; 188static bool_t npx_irq13; 189 190#ifndef SMP 191alias_for_inthand_t probetrap; 192__asm(" \n\ 193 .text \n\ 194 .p2align 2,0x90 \n\ 195 .type " __XSTRING(CNAME(probetrap)) ",@function \n\ 196" __XSTRING(CNAME(probetrap)) ": \n\ 197 ss \n\ 198 incl " __XSTRING(CNAME(npx_traps_while_probing)) " \n\ 199 fnclex \n\ 200 iret \n\ 201"); 202#endif /* SMP */ 203 204/* 205 * Identify routine. Create a connection point on our parent for probing. 206 */ 207static void 208npx_identify(driver, parent) 209 driver_t *driver; 210 device_t parent; 211{ 212 device_t child; 213 214 child = BUS_ADD_CHILD(parent, 0, "npx", 0); 215 if (child == NULL) 216 panic("npx_identify"); 217} 218 219#ifndef SMP 220/* 221 * Do minimal handling of npx interrupts to convert them to traps. 222 */ 223static void 224npx_intr(dummy) 225 void *dummy; 226{ 227 struct thread *td; 228 229#ifndef SMP 230 npx_intrs_while_probing++; 231#endif 232 233 /* 234 * The BUSY# latch must be cleared in all cases so that the next 235 * unmasked npx exception causes an interrupt. 236 */ 237#ifdef PC98 238 outb(0xf8, 0); 239#else 240 outb(0xf0, 0); 241#endif 242 243 /* 244 * fpcurthread is normally non-null here. In that case, schedule an 245 * AST to finish the exception handling in the correct context 246 * (this interrupt may occur after the thread has entered the 247 * kernel via a syscall or an interrupt). Otherwise, the npx 248 * state of the thread that caused this interrupt must have been 249 * pushed to the thread's pcb, and clearing of the busy latch 250 * above has finished the (essentially null) handling of this 251 * interrupt. Control will eventually return to the instruction 252 * that caused it and it will repeat. We will eventually (usually 253 * soon) win the race to handle the interrupt properly. 254 */ 255 td = PCPU_GET(fpcurthread); 256 if (td != NULL) { 257 td->td_pcb->pcb_flags |= PCB_NPXTRAP; 258 mtx_lock_spin(&sched_lock); 259 td->td_kse->ke_flags |= KEF_ASTPENDING; 260 mtx_unlock_spin(&sched_lock); 261 } 262} 263#endif /* !SMP */ 264 265/* 266 * Probe routine. Initialize cr0 to give correct behaviour for [f]wait 267 * whether the device exists or not (XXX should be elsewhere). Set flags 268 * to tell npxattach() what to do. Modify device struct if npx doesn't 269 * need to use interrupts. Return 0 if device exists. 270 */ 271static int 272npx_probe(dev) 273 device_t dev; 274{ 275#ifndef SMP 276 struct gate_descriptor save_idt_npxtrap; 277 struct resource *ioport_res, *irq_res; 278 void *irq_cookie; 279 int ioport_rid, irq_num, irq_rid; 280 u_short control; 281 u_short status; 282 283 save_idt_npxtrap = idt[16]; 284 setidt(16, probetrap, SDT_SYS386TGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 285 ioport_rid = 0; 286 ioport_res = bus_alloc_resource(dev, SYS_RES_IOPORT, &ioport_rid, 287 IO_NPX, IO_NPX, IO_NPXSIZE, RF_ACTIVE); 288 if (ioport_res == NULL) 289 panic("npx: can't get ports"); 290#ifdef PC98 291 if (resource_int_value("npx", 0, "irq", &irq_num) != 0) 292 irq_num = 8; 293#else 294 if (resource_int_value("npx", 0, "irq", &irq_num) != 0) 295 irq_num = 13; 296#endif 297 irq_rid = 0; 298 irq_res = bus_alloc_resource(dev, SYS_RES_IRQ, &ioport_rid, irq_num, 299 irq_num, 1, RF_ACTIVE); 300 if (irq_res == NULL) 301 panic("npx: can't get IRQ"); 302 if (bus_setup_intr(dev, irq_res, INTR_TYPE_MISC | INTR_FAST, npx_intr, 303 NULL, &irq_cookie) != 0) 304 panic("npx: can't create intr"); 305#endif /* !SMP */ 306 307 /* 308 * Partially reset the coprocessor, if any. Some BIOS's don't reset 309 * it after a warm boot. 310 */ 311#ifdef PC98 312 outb(0xf8,0); 313#else 314 outb(0xf1, 0); /* full reset on some systems, NOP on others */ 315 outb(0xf0, 0); /* clear BUSY# latch */ 316#endif 317 /* 318 * Prepare to trap all ESC (i.e., NPX) instructions and all WAIT 319 * instructions. We must set the CR0_MP bit and use the CR0_TS 320 * bit to control the trap, because setting the CR0_EM bit does 321 * not cause WAIT instructions to trap. It's important to trap 322 * WAIT instructions - otherwise the "wait" variants of no-wait 323 * control instructions would degenerate to the "no-wait" variants 324 * after FP context switches but work correctly otherwise. It's 325 * particularly important to trap WAITs when there is no NPX - 326 * otherwise the "wait" variants would always degenerate. 327 * 328 * Try setting CR0_NE to get correct error reporting on 486DX's. 329 * Setting it should fail or do nothing on lesser processors. 330 */ 331 load_cr0(rcr0() | CR0_MP | CR0_NE); 332 /* 333 * But don't trap while we're probing. 334 */ 335 stop_emulating(); 336 /* 337 * Finish resetting the coprocessor, if any. If there is an error 338 * pending, then we may get a bogus IRQ13, but npx_intr() will handle 339 * it OK. Bogus halts have never been observed, but we enabled 340 * IRQ13 and cleared the BUSY# latch early to handle them anyway. 341 */ 342 fninit(); 343 344 device_set_desc(dev, "math processor"); 345 346#ifdef SMP 347 348 /* 349 * Exception 16 MUST work for SMP. 350 */ 351 npx_ex16 = hw_float = npx_exists = 1; 352 return (0); 353 354#else /* !SMP */ 355 356 /* 357 * Don't use fwait here because it might hang. 358 * Don't use fnop here because it usually hangs if there is no FPU. 359 */ 360 DELAY(1000); /* wait for any IRQ13 */ 361#ifdef DIAGNOSTIC 362 if (npx_intrs_while_probing != 0) 363 printf("fninit caused %u bogus npx interrupt(s)\n", 364 npx_intrs_while_probing); 365 if (npx_traps_while_probing != 0) 366 printf("fninit caused %u bogus npx trap(s)\n", 367 npx_traps_while_probing); 368#endif 369 /* 370 * Check for a status of mostly zero. 371 */ 372 status = 0x5a5a; 373 fnstsw(&status); 374 if ((status & 0xb8ff) == 0) { 375 /* 376 * Good, now check for a proper control word. 377 */ 378 control = 0x5a5a; 379 fnstcw(&control); 380 if ((control & 0x1f3f) == 0x033f) { 381 hw_float = npx_exists = 1; 382 /* 383 * We have an npx, now divide by 0 to see if exception 384 * 16 works. 385 */ 386 control &= ~(1 << 2); /* enable divide by 0 trap */ 387 fldcw(&control); 388#ifdef FPU_ERROR_BROKEN 389 /* 390 * FPU error signal doesn't work on some CPU 391 * accelerator board. 392 */ 393 npx_ex16 = 1; 394 return (0); 395#endif 396 npx_traps_while_probing = npx_intrs_while_probing = 0; 397 fp_divide_by_0(); 398 if (npx_traps_while_probing != 0) { 399 /* 400 * Good, exception 16 works. 401 */ 402 npx_ex16 = 1; 403 goto no_irq13; 404 } 405 if (npx_intrs_while_probing != 0) { 406 /* 407 * Bad, we are stuck with IRQ13. 408 */ 409 npx_irq13 = 1; 410 idt[16] = save_idt_npxtrap; 411 return (0); 412 } 413 /* 414 * Worse, even IRQ13 is broken. Use emulator. 415 */ 416 } 417 } 418 /* 419 * Probe failed, but we want to get to npxattach to initialize the 420 * emulator and say that it has been installed. XXX handle devices 421 * that aren't really devices better. 422 */ 423 /* FALLTHROUGH */ 424no_irq13: 425 idt[16] = save_idt_npxtrap; 426 bus_teardown_intr(dev, irq_res, irq_cookie); 427 428 /* 429 * XXX hack around brokenness of bus_teardown_intr(). If we left the 430 * irq active then we would get it instead of exception 16. 431 */ 432 { 433 critical_t crit; 434 435 crit = cpu_critical_enter(); 436 mtx_lock_spin(&icu_lock); 437 INTRDIS(1 << irq_num); 438 mtx_unlock_spin(&icu_lock); 439 cpu_critical_exit(crit); 440 } 441 442 bus_release_resource(dev, SYS_RES_IRQ, irq_rid, irq_res); 443 bus_release_resource(dev, SYS_RES_IOPORT, ioport_rid, ioport_res); 444 return (0); 445 446#endif /* SMP */ 447} 448 449/* 450 * Attach routine - announce which it is, and wire into system 451 */ 452int 453npx_attach(dev) 454 device_t dev; 455{ 456 int flags; 457 458 if (resource_int_value("npx", 0, "flags", &flags) != 0) 459 flags = 0; 460 461 if (flags) 462 device_printf(dev, "flags 0x%x ", flags); 463 if (npx_irq13) { 464 device_printf(dev, "using IRQ 13 interface\n"); 465 } else { 466#if defined(MATH_EMULATE) || defined(GPL_MATH_EMULATE) 467 if (npx_ex16) { 468 if (!(flags & NPX_PREFER_EMULATOR)) 469 device_printf(dev, "INT 16 interface\n"); 470 else { 471 device_printf(dev, "FPU exists, but flags request " 472 "emulator\n"); 473 hw_float = npx_exists = 0; 474 } 475 } else if (npx_exists) { 476 device_printf(dev, "error reporting broken; using 387 emulator\n"); 477 hw_float = npx_exists = 0; 478 } else 479 device_printf(dev, "387 emulator\n"); 480#else 481 if (npx_ex16) { 482 device_printf(dev, "INT 16 interface\n"); 483 if (flags & NPX_PREFER_EMULATOR) { 484 device_printf(dev, "emulator requested, but none compiled " 485 "into kernel, using FPU\n"); 486 } 487 } else 488 device_printf(dev, "no 387 emulator in kernel and no FPU!\n"); 489#endif 490 } 491 npxinit(__INITIAL_NPXCW__); 492 493#ifdef I586_CPU_XXX 494 if (cpu_class == CPUCLASS_586 && npx_ex16 && npx_exists && 495 timezero("i586_bzero()", i586_bzero) < 496 timezero("bzero()", bzero) * 4 / 5) { 497 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_BCOPY)) { 498 bcopy_vector = i586_bcopy; 499 ovbcopy_vector = i586_bcopy; 500 } 501 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_BZERO)) 502 bzero = i586_bzero; 503 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_COPYIO)) { 504 copyin_vector = i586_copyin; 505 copyout_vector = i586_copyout; 506 } 507 } 508#endif 509 510 return (0); /* XXX unused */ 511} 512 513/* 514 * Initialize floating point unit. 515 */ 516void 517npxinit(control) 518 u_short control; 519{ 520 static union savefpu dummy; 521 critical_t savecrit; 522 523 if (!npx_exists) 524 return; 525 /* 526 * fninit has the same h/w bugs as fnsave. Use the detoxified 527 * fnsave to throw away any junk in the fpu. npxsave() initializes 528 * the fpu and sets fpcurthread = NULL as important side effects. 529 */ 530 savecrit = cpu_critical_enter(); 531 npxsave(&dummy); 532 stop_emulating(); 533#ifdef CPU_ENABLE_SSE 534 /* XXX npxsave() doesn't actually initialize the fpu in the SSE case. */ 535 if (cpu_fxsr) 536 fninit(); 537#endif 538 fldcw(&control); 539 if (PCPU_GET(curpcb) != NULL) 540 fpusave(&PCPU_GET(curpcb)->pcb_save); 541 start_emulating(); 542 cpu_critical_exit(savecrit); 543} 544 545/* 546 * Free coprocessor (if we have it). 547 */ 548void 549npxexit(td) 550 struct thread *td; 551{ 552 critical_t savecrit; 553 554 savecrit = cpu_critical_enter(); 555 if (td == PCPU_GET(fpcurthread)) 556 npxsave(&PCPU_GET(curpcb)->pcb_save); 557 cpu_critical_exit(savecrit); 558#ifdef NPX_DEBUG 559 if (npx_exists) { 560 u_int masked_exceptions; 561 562 masked_exceptions = PCPU_GET(curpcb)->pcb_save.sv_87.sv_env.en_cw 563 & PCPU_GET(curpcb)->pcb_save.sv_87.sv_env.en_sw & 0x7f; 564 /* 565 * Log exceptions that would have trapped with the old 566 * control word (overflow, divide by 0, and invalid operand). 567 */ 568 if (masked_exceptions & 0x0d) 569 log(LOG_ERR, 570 "pid %d (%s) exited with masked floating point exceptions 0x%02x\n", 571 td->td_proc->p_pid, td->td_proc->p_comm, 572 masked_exceptions); 573 } 574#endif 575} 576 577/* 578 * The following mechanism is used to ensure that the FPE_... value 579 * that is passed as a trapcode to the signal handler of the user 580 * process does not have more than one bit set. 581 * 582 * Multiple bits may be set if the user process modifies the control 583 * word while a status word bit is already set. While this is a sign 584 * of bad coding, we have no choise than to narrow them down to one 585 * bit, since we must not send a trapcode that is not exactly one of 586 * the FPE_ macros. 587 * 588 * The mechanism has a static table with 127 entries. Each combination 589 * of the 7 FPU status word exception bits directly translates to a 590 * position in this table, where a single FPE_... value is stored. 591 * This FPE_... value stored there is considered the "most important" 592 * of the exception bits and will be sent as the signal code. The 593 * precedence of the bits is based upon Intel Document "Numerical 594 * Applications", Chapter "Special Computational Situations". 595 * 596 * The macro to choose one of these values does these steps: 1) Throw 597 * away status word bits that cannot be masked. 2) Throw away the bits 598 * currently masked in the control word, assuming the user isn't 599 * interested in them anymore. 3) Reinsert status word bit 7 (stack 600 * fault) if it is set, which cannot be masked but must be presered. 601 * 4) Use the remaining bits to point into the trapcode table. 602 * 603 * The 6 maskable bits in order of their preference, as stated in the 604 * above referenced Intel manual: 605 * 1 Invalid operation (FP_X_INV) 606 * 1a Stack underflow 607 * 1b Stack overflow 608 * 1c Operand of unsupported format 609 * 1d SNaN operand. 610 * 2 QNaN operand (not an exception, irrelavant here) 611 * 3 Any other invalid-operation not mentioned above or zero divide 612 * (FP_X_INV, FP_X_DZ) 613 * 4 Denormal operand (FP_X_DNML) 614 * 5 Numeric over/underflow (FP_X_OFL, FP_X_UFL) 615 * 6 Inexact result (FP_X_IMP) 616 */ 617static char fpetable[128] = { 618 0, 619 FPE_FLTINV, /* 1 - INV */ 620 FPE_FLTUND, /* 2 - DNML */ 621 FPE_FLTINV, /* 3 - INV | DNML */ 622 FPE_FLTDIV, /* 4 - DZ */ 623 FPE_FLTINV, /* 5 - INV | DZ */ 624 FPE_FLTDIV, /* 6 - DNML | DZ */ 625 FPE_FLTINV, /* 7 - INV | DNML | DZ */ 626 FPE_FLTOVF, /* 8 - OFL */ 627 FPE_FLTINV, /* 9 - INV | OFL */ 628 FPE_FLTUND, /* A - DNML | OFL */ 629 FPE_FLTINV, /* B - INV | DNML | OFL */ 630 FPE_FLTDIV, /* C - DZ | OFL */ 631 FPE_FLTINV, /* D - INV | DZ | OFL */ 632 FPE_FLTDIV, /* E - DNML | DZ | OFL */ 633 FPE_FLTINV, /* F - INV | DNML | DZ | OFL */ 634 FPE_FLTUND, /* 10 - UFL */ 635 FPE_FLTINV, /* 11 - INV | UFL */ 636 FPE_FLTUND, /* 12 - DNML | UFL */ 637 FPE_FLTINV, /* 13 - INV | DNML | UFL */ 638 FPE_FLTDIV, /* 14 - DZ | UFL */ 639 FPE_FLTINV, /* 15 - INV | DZ | UFL */ 640 FPE_FLTDIV, /* 16 - DNML | DZ | UFL */ 641 FPE_FLTINV, /* 17 - INV | DNML | DZ | UFL */ 642 FPE_FLTOVF, /* 18 - OFL | UFL */ 643 FPE_FLTINV, /* 19 - INV | OFL | UFL */ 644 FPE_FLTUND, /* 1A - DNML | OFL | UFL */ 645 FPE_FLTINV, /* 1B - INV | DNML | OFL | UFL */ 646 FPE_FLTDIV, /* 1C - DZ | OFL | UFL */ 647 FPE_FLTINV, /* 1D - INV | DZ | OFL | UFL */ 648 FPE_FLTDIV, /* 1E - DNML | DZ | OFL | UFL */ 649 FPE_FLTINV, /* 1F - INV | DNML | DZ | OFL | UFL */ 650 FPE_FLTRES, /* 20 - IMP */ 651 FPE_FLTINV, /* 21 - INV | IMP */ 652 FPE_FLTUND, /* 22 - DNML | IMP */ 653 FPE_FLTINV, /* 23 - INV | DNML | IMP */ 654 FPE_FLTDIV, /* 24 - DZ | IMP */ 655 FPE_FLTINV, /* 25 - INV | DZ | IMP */ 656 FPE_FLTDIV, /* 26 - DNML | DZ | IMP */ 657 FPE_FLTINV, /* 27 - INV | DNML | DZ | IMP */ 658 FPE_FLTOVF, /* 28 - OFL | IMP */ 659 FPE_FLTINV, /* 29 - INV | OFL | IMP */ 660 FPE_FLTUND, /* 2A - DNML | OFL | IMP */ 661 FPE_FLTINV, /* 2B - INV | DNML | OFL | IMP */ 662 FPE_FLTDIV, /* 2C - DZ | OFL | IMP */ 663 FPE_FLTINV, /* 2D - INV | DZ | OFL | IMP */ 664 FPE_FLTDIV, /* 2E - DNML | DZ | OFL | IMP */ 665 FPE_FLTINV, /* 2F - INV | DNML | DZ | OFL | IMP */ 666 FPE_FLTUND, /* 30 - UFL | IMP */ 667 FPE_FLTINV, /* 31 - INV | UFL | IMP */ 668 FPE_FLTUND, /* 32 - DNML | UFL | IMP */ 669 FPE_FLTINV, /* 33 - INV | DNML | UFL | IMP */ 670 FPE_FLTDIV, /* 34 - DZ | UFL | IMP */ 671 FPE_FLTINV, /* 35 - INV | DZ | UFL | IMP */ 672 FPE_FLTDIV, /* 36 - DNML | DZ | UFL | IMP */ 673 FPE_FLTINV, /* 37 - INV | DNML | DZ | UFL | IMP */ 674 FPE_FLTOVF, /* 38 - OFL | UFL | IMP */ 675 FPE_FLTINV, /* 39 - INV | OFL | UFL | IMP */ 676 FPE_FLTUND, /* 3A - DNML | OFL | UFL | IMP */ 677 FPE_FLTINV, /* 3B - INV | DNML | OFL | UFL | IMP */ 678 FPE_FLTDIV, /* 3C - DZ | OFL | UFL | IMP */ 679 FPE_FLTINV, /* 3D - INV | DZ | OFL | UFL | IMP */ 680 FPE_FLTDIV, /* 3E - DNML | DZ | OFL | UFL | IMP */ 681 FPE_FLTINV, /* 3F - INV | DNML | DZ | OFL | UFL | IMP */ 682 FPE_FLTSUB, /* 40 - STK */ 683 FPE_FLTSUB, /* 41 - INV | STK */ 684 FPE_FLTUND, /* 42 - DNML | STK */ 685 FPE_FLTSUB, /* 43 - INV | DNML | STK */ 686 FPE_FLTDIV, /* 44 - DZ | STK */ 687 FPE_FLTSUB, /* 45 - INV | DZ | STK */ 688 FPE_FLTDIV, /* 46 - DNML | DZ | STK */ 689 FPE_FLTSUB, /* 47 - INV | DNML | DZ | STK */ 690 FPE_FLTOVF, /* 48 - OFL | STK */ 691 FPE_FLTSUB, /* 49 - INV | OFL | STK */ 692 FPE_FLTUND, /* 4A - DNML | OFL | STK */ 693 FPE_FLTSUB, /* 4B - INV | DNML | OFL | STK */ 694 FPE_FLTDIV, /* 4C - DZ | OFL | STK */ 695 FPE_FLTSUB, /* 4D - INV | DZ | OFL | STK */ 696 FPE_FLTDIV, /* 4E - DNML | DZ | OFL | STK */ 697 FPE_FLTSUB, /* 4F - INV | DNML | DZ | OFL | STK */ 698 FPE_FLTUND, /* 50 - UFL | STK */ 699 FPE_FLTSUB, /* 51 - INV | UFL | STK */ 700 FPE_FLTUND, /* 52 - DNML | UFL | STK */ 701 FPE_FLTSUB, /* 53 - INV | DNML | UFL | STK */ 702 FPE_FLTDIV, /* 54 - DZ | UFL | STK */ 703 FPE_FLTSUB, /* 55 - INV | DZ | UFL | STK */ 704 FPE_FLTDIV, /* 56 - DNML | DZ | UFL | STK */ 705 FPE_FLTSUB, /* 57 - INV | DNML | DZ | UFL | STK */ 706 FPE_FLTOVF, /* 58 - OFL | UFL | STK */ 707 FPE_FLTSUB, /* 59 - INV | OFL | UFL | STK */ 708 FPE_FLTUND, /* 5A - DNML | OFL | UFL | STK */ 709 FPE_FLTSUB, /* 5B - INV | DNML | OFL | UFL | STK */ 710 FPE_FLTDIV, /* 5C - DZ | OFL | UFL | STK */ 711 FPE_FLTSUB, /* 5D - INV | DZ | OFL | UFL | STK */ 712 FPE_FLTDIV, /* 5E - DNML | DZ | OFL | UFL | STK */ 713 FPE_FLTSUB, /* 5F - INV | DNML | DZ | OFL | UFL | STK */ 714 FPE_FLTRES, /* 60 - IMP | STK */ 715 FPE_FLTSUB, /* 61 - INV | IMP | STK */ 716 FPE_FLTUND, /* 62 - DNML | IMP | STK */ 717 FPE_FLTSUB, /* 63 - INV | DNML | IMP | STK */ 718 FPE_FLTDIV, /* 64 - DZ | IMP | STK */ 719 FPE_FLTSUB, /* 65 - INV | DZ | IMP | STK */ 720 FPE_FLTDIV, /* 66 - DNML | DZ | IMP | STK */ 721 FPE_FLTSUB, /* 67 - INV | DNML | DZ | IMP | STK */ 722 FPE_FLTOVF, /* 68 - OFL | IMP | STK */ 723 FPE_FLTSUB, /* 69 - INV | OFL | IMP | STK */ 724 FPE_FLTUND, /* 6A - DNML | OFL | IMP | STK */ 725 FPE_FLTSUB, /* 6B - INV | DNML | OFL | IMP | STK */ 726 FPE_FLTDIV, /* 6C - DZ | OFL | IMP | STK */ 727 FPE_FLTSUB, /* 6D - INV | DZ | OFL | IMP | STK */ 728 FPE_FLTDIV, /* 6E - DNML | DZ | OFL | IMP | STK */ 729 FPE_FLTSUB, /* 6F - INV | DNML | DZ | OFL | IMP | STK */ 730 FPE_FLTUND, /* 70 - UFL | IMP | STK */ 731 FPE_FLTSUB, /* 71 - INV | UFL | IMP | STK */ 732 FPE_FLTUND, /* 72 - DNML | UFL | IMP | STK */ 733 FPE_FLTSUB, /* 73 - INV | DNML | UFL | IMP | STK */ 734 FPE_FLTDIV, /* 74 - DZ | UFL | IMP | STK */ 735 FPE_FLTSUB, /* 75 - INV | DZ | UFL | IMP | STK */ 736 FPE_FLTDIV, /* 76 - DNML | DZ | UFL | IMP | STK */ 737 FPE_FLTSUB, /* 77 - INV | DNML | DZ | UFL | IMP | STK */ 738 FPE_FLTOVF, /* 78 - OFL | UFL | IMP | STK */ 739 FPE_FLTSUB, /* 79 - INV | OFL | UFL | IMP | STK */ 740 FPE_FLTUND, /* 7A - DNML | OFL | UFL | IMP | STK */ 741 FPE_FLTSUB, /* 7B - INV | DNML | OFL | UFL | IMP | STK */ 742 FPE_FLTDIV, /* 7C - DZ | OFL | UFL | IMP | STK */ 743 FPE_FLTSUB, /* 7D - INV | DZ | OFL | UFL | IMP | STK */ 744 FPE_FLTDIV, /* 7E - DNML | DZ | OFL | UFL | IMP | STK */ 745 FPE_FLTSUB, /* 7F - INV | DNML | DZ | OFL | UFL | IMP | STK */ 746}; 747 748/* 749 * Preserve the FP status word, clear FP exceptions, then generate a SIGFPE. 750 * 751 * Clearing exceptions is necessary mainly to avoid IRQ13 bugs. We now 752 * depend on longjmp() restoring a usable state. Restoring the state 753 * or examining it might fail if we didn't clear exceptions. 754 * 755 * The error code chosen will be one of the FPE_... macros. It will be 756 * sent as the second argument to old BSD-style signal handlers and as 757 * "siginfo_t->si_code" (second argument) to SA_SIGINFO signal handlers. 758 * 759 * XXX the FP state is not preserved across signal handlers. So signal 760 * handlers cannot afford to do FP unless they preserve the state or 761 * longjmp() out. Both preserving the state and longjmp()ing may be 762 * destroyed by IRQ13 bugs. Clearing FP exceptions is not an acceptable 763 * solution for signals other than SIGFPE. 764 */ 765int 766npxtrap() 767{ 768 critical_t savecrit; 769 u_short control, status; 770 u_long *exstat; 771 772 if (!npx_exists) { 773 printf("npxtrap: fpcurthread = %p, curthread = %p, npx_exists = %d\n", 774 PCPU_GET(fpcurthread), curthread, npx_exists); 775 panic("npxtrap from nowhere"); 776 } 777 savecrit = cpu_critical_enter(); 778 779 /* 780 * Interrupt handling (for another interrupt) may have pushed the 781 * state to memory. Fetch the relevant parts of the state from 782 * wherever they are. 783 */ 784 if (PCPU_GET(fpcurthread) != curthread) { 785 control = GET_FPU_CW(curthread); 786 status = GET_FPU_SW(curthread); 787 } else { 788 fnstcw(&control); 789 fnstsw(&status); 790 } 791 792 exstat = GET_FPU_EXSW_PTR(curthread->td_pcb); 793 *exstat = status; 794 if (PCPU_GET(fpcurthread) != curthread) 795 GET_FPU_SW(curthread) &= ~0x80bf; 796 else 797 fnclex(); 798 cpu_critical_exit(savecrit); 799 return (fpetable[status & ((~control & 0x3f) | 0x40)]); 800} 801 802/* 803 * Implement device not available (DNA) exception 804 * 805 * It would be better to switch FP context here (if curthread != fpcurthread) 806 * and not necessarily for every context switch, but it is too hard to 807 * access foreign pcb's. 808 */ 809int 810npxdna() 811{ 812 u_long *exstat; 813 critical_t s; 814 815 if (!npx_exists) 816 return (0); 817 if (PCPU_GET(fpcurthread) != NULL) { 818 printf("npxdna: fpcurthread = %p, curthread = %p\n", 819 PCPU_GET(fpcurthread), curthread); 820 panic("npxdna"); 821 } 822 s = cpu_critical_enter(); 823 stop_emulating(); 824 /* 825 * Record new context early in case frstor causes an IRQ13. 826 */ 827 PCPU_SET(fpcurthread, curthread); 828 829 exstat = GET_FPU_EXSW_PTR(PCPU_GET(curpcb)); 830 *exstat = 0; 831 /* 832 * The following frstor may cause an IRQ13 when the state being 833 * restored has a pending error. The error will appear to have been 834 * triggered by the current (npx) user instruction even when that 835 * instruction is a no-wait instruction that should not trigger an 836 * error (e.g., fnclex). On at least one 486 system all of the 837 * no-wait instructions are broken the same as frstor, so our 838 * treatment does not amplify the breakage. On at least one 839 * 386/Cyrix 387 system, fnclex works correctly while frstor and 840 * fnsave are broken, so our treatment breaks fnclex if it is the 841 * first FPU instruction after a context switch. 842 */ 843 fpurstor(&PCPU_GET(curpcb)->pcb_save); 844 cpu_critical_exit(s); 845 846 return (1); 847} 848 849/* 850 * Wrapper for fnsave instruction, partly to handle hardware bugs. When npx 851 * exceptions are reported via IRQ13, spurious IRQ13's may be triggered by 852 * no-wait npx instructions. See the Intel application note AP-578 for 853 * details. This doesn't cause any additional complications here. IRQ13's 854 * are inherently asynchronous unless the CPU is frozen to deliver them -- 855 * one that started in userland may be delivered many instructions later, 856 * after the process has entered the kernel. It may even be delivered after 857 * the fnsave here completes. A spurious IRQ13 for the fnsave is handled in 858 * the same way as a very-late-arriving non-spurious IRQ13 from user mode: 859 * it is normally ignored at first because we set fpcurthread to NULL; it is 860 * normally retriggered in npxdna() after return to user mode. 861 * 862 * npxsave() must be called with interrupts disabled, so that it clears 863 * fpcurthread atomically with saving the state. We require callers to do the 864 * disabling, since most callers need to disable interrupts anyway to call 865 * npxsave() atomically with checking fpcurthread. 866 * 867 * A previous version of npxsave() went to great lengths to excecute fnsave 868 * with interrupts enabled in case executing it froze the CPU. This case 869 * can't happen, at least for Intel CPU/NPX's. Spurious IRQ13's don't imply 870 * spurious freezes. 871 */ 872void 873npxsave(addr) 874 union savefpu *addr; 875{ 876 877 stop_emulating(); 878 fpusave(addr); 879 880 start_emulating(); 881 PCPU_SET(fpcurthread, NULL); 882} 883 884static void 885fpusave(addr) 886 union savefpu *addr; 887{ 888 889#ifdef CPU_ENABLE_SSE 890 if (cpu_fxsr) 891 fxsave(addr); 892 else 893#endif 894 fnsave(addr); 895} 896 897static void 898fpurstor(addr) 899 union savefpu *addr; 900{ 901 902#ifdef CPU_ENABLE_SSE 903 if (cpu_fxsr) 904 fxrstor(addr); 905 else 906#endif 907 frstor(addr); 908} 909 910#ifdef I586_CPU_XXX 911static long 912timezero(funcname, func) 913 const char *funcname; 914 void (*func) __P((void *buf, size_t len)); 915 916{ 917 void *buf; 918#define BUFSIZE 1048576 919 long usec; 920 struct timeval finish, start; 921 922 buf = malloc(BUFSIZE, M_TEMP, M_NOWAIT); 923 if (buf == NULL) 924 return (BUFSIZE); 925 microtime(&start); 926 (*func)(buf, BUFSIZE); 927 microtime(&finish); 928 usec = 1000000 * (finish.tv_sec - start.tv_sec) + 929 finish.tv_usec - start.tv_usec; 930 if (usec <= 0) 931 usec = 1; 932 if (bootverbose) 933 printf("%s bandwidth = %u kBps\n", funcname, 934 (u_int32_t)(((BUFSIZE >> 10) * 1000000) / usec)); 935 free(buf, M_TEMP); 936 return (usec); 937} 938#endif /* I586_CPU */ 939 940static device_method_t npx_methods[] = { 941 /* Device interface */ 942 DEVMETHOD(device_identify, npx_identify), 943 DEVMETHOD(device_probe, npx_probe), 944 DEVMETHOD(device_attach, npx_attach), 945 DEVMETHOD(device_detach, bus_generic_detach), 946 DEVMETHOD(device_shutdown, bus_generic_shutdown), 947 DEVMETHOD(device_suspend, bus_generic_suspend), 948 DEVMETHOD(device_resume, bus_generic_resume), 949 950 { 0, 0 } 951}; 952 953static driver_t npx_driver = { 954 "npx", 955 npx_methods, 956 1, /* no softc */ 957}; 958 959static devclass_t npx_devclass; 960 961#ifdef DEV_ISA 962/* 963 * We prefer to attach to the root nexus so that the usual case (exception 16) 964 * doesn't describe the processor as being `on isa'. 965 */ 966DRIVER_MODULE(npx, nexus, npx_driver, npx_devclass, 0, 0); 967 968/* 969 * This sucks up the legacy ISA support assignments from PNPBIOS/ACPI. 970 */ 971static struct isa_pnp_id npxisa_ids[] = { 972 { 0x040cd041, "Legacy ISA coprocessor support" }, /* PNP0C04 */ 973 { 0 } 974}; 975 976static int 977npxisa_probe(device_t dev) 978{ 979 int result; 980 if ((result = ISA_PNP_PROBE(device_get_parent(dev), dev, npxisa_ids)) <= 0) { 981 device_quiet(dev); 982 } 983 return(result); 984} 985 986static int 987npxisa_attach(device_t dev) 988{ 989 return (0); 990} 991 992static device_method_t npxisa_methods[] = { 993 /* Device interface */ 994 DEVMETHOD(device_probe, npxisa_probe), 995 DEVMETHOD(device_attach, npxisa_attach), 996 DEVMETHOD(device_detach, bus_generic_detach), 997 DEVMETHOD(device_shutdown, bus_generic_shutdown), 998 DEVMETHOD(device_suspend, bus_generic_suspend), 999 DEVMETHOD(device_resume, bus_generic_resume), 1000 1001 { 0, 0 } 1002}; 1003 1004static driver_t npxisa_driver = { 1005 "npxisa", 1006 npxisa_methods, 1007 1, /* no softc */ 1008}; 1009 1010static devclass_t npxisa_devclass; 1011 1012DRIVER_MODULE(npxisa, isa, npxisa_driver, npxisa_devclass, 0, 0); 1013#ifndef PC98 1014DRIVER_MODULE(npxisa, acpi, npxisa_driver, npxisa_devclass, 0, 0); 1015#endif 1016#endif /* DEV_ISA */ 1017