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 * $FreeBSD$ 36 */ 37 38 #include "opt_cpu.h" 39 #include "opt_debug_npx.h" 40 #include "opt_math_emulate.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/bus.h> 45 #include <sys/kernel.h> 46 #include <sys/lock.h> 47 #include <sys/malloc.h> 48 #include <sys/module.h> 49 #include <sys/mutex.h> 50 #include <sys/mutex.h> 51 #include <sys/proc.h> 52 #include <sys/sysctl.h> 53 #include <machine/bus.h> 54 #include <sys/rman.h> 55 #ifdef NPX_DEBUG 56 #include <sys/syslog.h> 57 #endif 58 #include <sys/signalvar.h> 59 #include <sys/user.h> 60 61 #ifndef SMP 62 #include <machine/asmacros.h> 63 #endif 64 #include <machine/cputypes.h> 65 #include <machine/frame.h> 66 #include <machine/md_var.h> 67 #include <machine/pcb.h> 68 #include <machine/psl.h> 69 #ifndef SMP 70 #include <machine/clock.h> 71 #endif 72 #include <machine/resource.h> 73 #include <machine/specialreg.h> 74 #include <machine/segments.h> 75 76 #ifndef SMP 77 #include <i386/isa/icu.h> 78 #include <i386/isa/intr_machdep.h> 79 #include <i386/isa/isa.h> 80 #endif 81 #include <isa/isavar.h> 82 83 /* 84 * 387 and 287 Numeric Coprocessor Extension (NPX) Driver. 85 */ 86 87 /* Configuration flags. */ 88 #define NPX_DISABLE_I586_OPTIMIZED_BCOPY (1 << 0) 89 #define NPX_DISABLE_I586_OPTIMIZED_BZERO (1 << 1) 90 #define NPX_DISABLE_I586_OPTIMIZED_COPYIO (1 << 2) 91 #define NPX_PREFER_EMULATOR (1 << 3) 92 93 #ifdef __GNUC__ 94 95 #define fldcw(addr) __asm("fldcw %0" : : "m" (*(addr))) 96 #define fnclex() __asm("fnclex") 97 #define fninit() __asm("fninit") 98 #define fnsave(addr) __asm __volatile("fnsave %0" : "=m" (*(addr))) 99 #define fnstcw(addr) __asm __volatile("fnstcw %0" : "=m" (*(addr))) 100 #define fnstsw(addr) __asm __volatile("fnstsw %0" : "=m" (*(addr))) 101 #define fp_divide_by_0() __asm("fldz; fld1; fdiv %st,%st(1); fnop") 102 #define frstor(addr) __asm("frstor %0" : : "m" (*(addr))) 103 #ifdef CPU_ENABLE_SSE 104 #define fxrstor(addr) __asm("fxrstor %0" : : "m" (*(addr))) 105 #define fxsave(addr) __asm __volatile("fxsave %0" : "=m" (*(addr))) 106 #endif 107 #define start_emulating() __asm("smsw %%ax; orb %0,%%al; lmsw %%ax" \ 108 : : "n" (CR0_TS) : "ax") 109 #define stop_emulating() __asm("clts") 110 111 #else /* not __GNUC__ */ 112 113 void fldcw __P((caddr_t addr)); 114 void fnclex __P((void)); 115 void fninit __P((void)); 116 void fnsave __P((caddr_t addr)); 117 void fnstcw __P((caddr_t addr)); 118 void fnstsw __P((caddr_t addr)); 119 void fp_divide_by_0 __P((void)); 120 void frstor __P((caddr_t addr)); 121 #ifdef CPU_ENABLE_SSE 122 void fxsave __P((caddr_t addr)); 123 void fxrstor __P((caddr_t addr)); 124 #endif 125 void start_emulating __P((void)); 126 void stop_emulating __P((void)); 127 128 #endif /* __GNUC__ */ 129 130 #ifdef CPU_ENABLE_SSE 131 #define GET_FPU_CW(proc) \ 132 (cpu_fxsr ? \ 133 (proc)->p_addr->u_pcb.pcb_save.sv_xmm.sv_env.en_cw : \ 134 (proc)->p_addr->u_pcb.pcb_save.sv_87.sv_env.en_cw) 135 #define GET_FPU_SW(proc) \ 136 (cpu_fxsr ? \ 137 (proc)->p_addr->u_pcb.pcb_save.sv_xmm.sv_env.en_sw : \ 138 (proc)->p_addr->u_pcb.pcb_save.sv_87.sv_env.en_sw) 139 #define GET_FPU_EXSW_PTR(pcb) \ 140 (cpu_fxsr ? \ 141 &(pcb)->pcb_save.sv_xmm.sv_ex_sw : \ 142 &(pcb)->pcb_save.sv_87.sv_ex_sw) 143 #else /* CPU_ENABLE_SSE */ 144 #define GET_FPU_CW(proc) \ 145 (proc->p_addr->u_pcb.pcb_save.sv_87.sv_env.en_cw) 146 #define GET_FPU_SW(proc) \ 147 (proc->p_addr->u_pcb.pcb_save.sv_87.sv_env.en_sw) 148 #define GET_FPU_EXSW_PTR(pcb) \ 149 (&(pcb)->pcb_save.sv_87.sv_ex_sw) 150 #endif /* CPU_ENABLE_SSE */ 151 152 typedef u_char bool_t; 153 154 static int npx_attach __P((device_t dev)); 155 static void npx_identify __P((driver_t *driver, device_t parent)); 156 #ifndef SMP 157 static void npx_intr __P((void *)); 158 #endif 159 static int npx_probe __P((device_t dev)); 160 static int npx_probe1 __P((device_t dev)); 161 static void fpusave __P((union savefpu *)); 162 static void fpurstor __P((union savefpu *)); 163 #ifdef I586_CPU_XXX 164 static long timezero __P((const char *funcname, 165 void (*func)(void *buf, size_t len))); 166 #endif /* I586_CPU */ 167 168 int hw_float; /* XXX currently just alias for npx_exists */ 169 170 SYSCTL_INT(_hw,HW_FLOATINGPT, floatingpoint, 171 CTLFLAG_RD, &hw_float, 0, 172 "Floatingpoint instructions executed in hardware"); 173 174 #ifndef SMP 175 static volatile u_int npx_intrs_while_probing; 176 static volatile u_int npx_traps_while_probing; 177 #endif 178 179 static bool_t npx_ex16; 180 static bool_t npx_exists; 181 static bool_t npx_irq13; 182 static int npx_irq; /* irq number */ 183 184 #ifndef SMP 185 /* 186 * Special interrupt handlers. Someday intr0-intr15 will be used to count 187 * interrupts. We'll still need a special exception 16 handler. The busy 188 * latch stuff in probeintr() can be moved to npxprobe(). 189 */ 190 inthand_t probeintr; 191 __asm(" \n\ 192 .text \n\ 193 .p2align 2,0x90 \n\ 194 .type " __XSTRING(CNAME(probeintr)) ",@function \n\ 195 " __XSTRING(CNAME(probeintr)) ": \n\ 196 ss \n\ 197 incl " __XSTRING(CNAME(npx_intrs_while_probing)) " \n\ 198 pushl %eax \n\ 199 movb $0x20,%al # EOI (asm in strings loses cpp features) \n\ 200 outb %al,$0xa0 # IO_ICU2 \n\ 201 outb %al,$0x20 # IO_ICU1 \n\ 202 movb $0,%al \n\ 203 outb %al,$0xf0 # clear BUSY# latch \n\ 204 popl %eax \n\ 205 iret \n\ 206 "); 207 208 inthand_t probetrap; 209 __asm(" \n\ 210 .text \n\ 211 .p2align 2,0x90 \n\ 212 .type " __XSTRING(CNAME(probetrap)) ",@function \n\ 213 " __XSTRING(CNAME(probetrap)) ": \n\ 214 ss \n\ 215 incl " __XSTRING(CNAME(npx_traps_while_probing)) " \n\ 216 fnclex \n\ 217 iret \n\ 218 "); 219 #endif /* SMP */ 220 221 /* 222 * Identify routine. Create a connection point on our parent for probing. 223 */ 224 static void 225 npx_identify(driver, parent) 226 driver_t *driver; 227 device_t parent; 228 { 229 device_t child; 230 231 child = BUS_ADD_CHILD(parent, 0, "npx", 0); 232 if (child == NULL) 233 panic("npx_identify"); 234 } 235 236 #ifndef SMP 237 /* 238 * Do minimal handling of npx interrupts to convert them to traps. 239 */ 240 static void 241 npx_intr(dummy) 242 void *dummy; 243 { 244 struct proc *p; 245 246 /* 247 * The BUSY# latch must be cleared in all cases so that the next 248 * unmasked npx exception causes an interrupt. 249 */ 250 outb(0xf0, 0); 251 252 /* 253 * npxproc is normally non-null here. In that case, schedule an 254 * AST to finish the exception handling in the correct context 255 * (this interrupt may occur after the process has entered the 256 * kernel via a syscall or an interrupt). Otherwise, the npx 257 * state of the process that caused this interrupt must have been 258 * pushed to the process' pcb, and clearing of the busy latch 259 * above has finished the (essentially null) handling of this 260 * interrupt. Control will eventually return to the instruction 261 * that caused it and it will repeat. We will eventually (usually 262 * soon) win the race to handle the interrupt properly. 263 */ 264 p = PCPU_GET(npxproc); 265 if (p != NULL) { 266 p->p_addr->u_pcb.pcb_flags |= PCB_NPXTRAP; 267 mtx_lock_spin(&sched_lock); 268 p->p_sflag |= PS_ASTPENDING; 269 mtx_unlock_spin(&sched_lock); 270 } 271 } 272 273 /* 274 * XXX these "local" variables of npx_probe() are non-local so that 275 * npxprobe1() can abuse them. 276 */ 277 static int npx_intrno; 278 static struct gate_descriptor save_idt_npxintr; 279 #endif /* !SMP */ 280 281 /* 282 * Probe routine. Initialize cr0 to give correct behaviour for [f]wait 283 * whether the device exists or not (XXX should be elsewhere). Set flags 284 * to tell npxattach() what to do. Modify device struct if npx doesn't 285 * need to use interrupts. Return 1 if device exists. 286 */ 287 static int 288 npx_probe(dev) 289 device_t dev; 290 { 291 #ifdef SMP 292 293 if (resource_int_value("npx", 0, "irq", &npx_irq) != 0) 294 npx_irq = 13; 295 return npx_probe1(dev); 296 297 #else /* SMP */ 298 299 int result; 300 critical_t savecrit; 301 u_char save_icu1_mask; 302 u_char save_icu2_mask; 303 struct gate_descriptor save_idt_npxtrap; 304 /* 305 * This routine is now just a wrapper for npxprobe1(), to install 306 * special npx interrupt and trap handlers, to enable npx interrupts 307 * and to disable other interrupts. Someday isa_configure() will 308 * install suitable handlers and run with interrupts enabled so we 309 * won't need to do so much here. 310 */ 311 if (resource_int_value("npx", 0, "irq", &npx_irq) != 0) 312 npx_irq = 13; 313 npx_intrno = NRSVIDT + npx_irq; 314 savecrit = critical_enter(); 315 save_icu1_mask = inb(IO_ICU1 + 1); 316 save_icu2_mask = inb(IO_ICU2 + 1); 317 save_idt_npxintr = idt[npx_intrno]; 318 save_idt_npxtrap = idt[16]; 319 outb(IO_ICU1 + 1, ~IRQ_SLAVE); 320 outb(IO_ICU2 + 1, ~(1 << (npx_irq - 8))); 321 setidt(16, probetrap, SDT_SYS386TGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 322 setidt(npx_intrno, probeintr, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 323 324 /* 325 * XXX This looks highly bogus, but it appears that npc_probe1 326 * needs interrupts enabled. Does this make any difference 327 * here? 328 */ 329 critical_exit(savecrit); 330 result = npx_probe1(dev); 331 savecrit = critical_enter(); 332 outb(IO_ICU1 + 1, save_icu1_mask); 333 outb(IO_ICU2 + 1, save_icu2_mask); 334 idt[npx_intrno] = save_idt_npxintr; 335 idt[16] = save_idt_npxtrap; 336 critical_exit(savecrit); 337 return (result); 338 339 #endif /* SMP */ 340 } 341 342 static int 343 npx_probe1(dev) 344 device_t dev; 345 { 346 #ifndef SMP 347 u_short control; 348 u_short status; 349 #endif 350 351 /* 352 * Partially reset the coprocessor, if any. Some BIOS's don't reset 353 * it after a warm boot. 354 */ 355 outb(0xf1, 0); /* full reset on some systems, NOP on others */ 356 outb(0xf0, 0); /* clear BUSY# latch */ 357 /* 358 * Prepare to trap all ESC (i.e., NPX) instructions and all WAIT 359 * instructions. We must set the CR0_MP bit and use the CR0_TS 360 * bit to control the trap, because setting the CR0_EM bit does 361 * not cause WAIT instructions to trap. It's important to trap 362 * WAIT instructions - otherwise the "wait" variants of no-wait 363 * control instructions would degenerate to the "no-wait" variants 364 * after FP context switches but work correctly otherwise. It's 365 * particularly important to trap WAITs when there is no NPX - 366 * otherwise the "wait" variants would always degenerate. 367 * 368 * Try setting CR0_NE to get correct error reporting on 486DX's. 369 * Setting it should fail or do nothing on lesser processors. 370 */ 371 load_cr0(rcr0() | CR0_MP | CR0_NE); 372 /* 373 * But don't trap while we're probing. 374 */ 375 stop_emulating(); 376 /* 377 * Finish resetting the coprocessor, if any. If there is an error 378 * pending, then we may get a bogus IRQ13, but probeintr() will handle 379 * it OK. Bogus halts have never been observed, but we enabled 380 * IRQ13 and cleared the BUSY# latch early to handle them anyway. 381 */ 382 fninit(); 383 384 #ifdef SMP 385 /* 386 * Exception 16 MUST work for SMP. 387 */ 388 npx_irq13 = 0; 389 npx_ex16 = hw_float = npx_exists = 1; 390 device_set_desc(dev, "math processor"); 391 return (0); 392 393 #else /* !SMP */ 394 device_set_desc(dev, "math processor"); 395 396 /* 397 * Don't use fwait here because it might hang. 398 * Don't use fnop here because it usually hangs if there is no FPU. 399 */ 400 DELAY(1000); /* wait for any IRQ13 */ 401 #ifdef DIAGNOSTIC 402 if (npx_intrs_while_probing != 0) 403 printf("fninit caused %u bogus npx interrupt(s)\n", 404 npx_intrs_while_probing); 405 if (npx_traps_while_probing != 0) 406 printf("fninit caused %u bogus npx trap(s)\n", 407 npx_traps_while_probing); 408 #endif 409 /* 410 * Check for a status of mostly zero. 411 */ 412 status = 0x5a5a; 413 fnstsw(&status); 414 if ((status & 0xb8ff) == 0) { 415 /* 416 * Good, now check for a proper control word. 417 */ 418 control = 0x5a5a; 419 fnstcw(&control); 420 if ((control & 0x1f3f) == 0x033f) { 421 hw_float = npx_exists = 1; 422 /* 423 * We have an npx, now divide by 0 to see if exception 424 * 16 works. 425 */ 426 control &= ~(1 << 2); /* enable divide by 0 trap */ 427 fldcw(&control); 428 npx_traps_while_probing = npx_intrs_while_probing = 0; 429 fp_divide_by_0(); 430 if (npx_traps_while_probing != 0) { 431 /* 432 * Good, exception 16 works. 433 */ 434 npx_ex16 = 1; 435 return (0); 436 } 437 if (npx_intrs_while_probing != 0) { 438 int rid; 439 struct resource *r; 440 void *intr; 441 /* 442 * Bad, we are stuck with IRQ13. 443 */ 444 npx_irq13 = 1; 445 446 /* 447 * We allocate these resources permanently, 448 * so there is no need to keep track of them. 449 */ 450 rid = 0; 451 r = bus_alloc_resource(dev, SYS_RES_IOPORT, 452 &rid, IO_NPX, IO_NPX, 453 IO_NPXSIZE, RF_ACTIVE); 454 if (r == 0) 455 panic("npx: can't get ports"); 456 rid = 0; 457 r = bus_alloc_resource(dev, SYS_RES_IRQ, 458 &rid, npx_irq, npx_irq, 459 1, RF_ACTIVE); 460 if (r == 0) 461 panic("npx: can't get IRQ"); 462 BUS_SETUP_INTR(device_get_parent(dev), 463 dev, r, 464 INTR_TYPE_MISC | INTR_FAST, 465 npx_intr, 0, &intr); 466 if (intr == 0) 467 panic("npx: can't create intr"); 468 469 /* 470 * XXX BUS_SETUP_INTR() has changed 471 * idt[npx_intrno] to point to Xfastintr0 472 * instead of Xfastintr0. Adjust 473 * save_idt_npxintr so that npxprobe() 474 * doesn't undo this. 475 */ 476 save_idt_npxintr = idt[npx_intrno]; 477 478 return (0); 479 } 480 /* 481 * Worse, even IRQ13 is broken. Use emulator. 482 */ 483 } 484 } 485 /* 486 * Probe failed, but we want to get to npxattach to initialize the 487 * emulator and say that it has been installed. XXX handle devices 488 * that aren't really devices better. 489 */ 490 return (0); 491 #endif /* SMP */ 492 } 493 494 /* 495 * Attach routine - announce which it is, and wire into system 496 */ 497 int 498 npx_attach(dev) 499 device_t dev; 500 { 501 int flags; 502 503 if (resource_int_value("npx", 0, "flags", &flags) != 0) 504 flags = 0; 505 506 if (flags) 507 device_printf(dev, "flags 0x%x ", flags); 508 if (npx_irq13) { 509 device_printf(dev, "using IRQ 13 interface\n"); 510 } else { 511 #if defined(MATH_EMULATE) || defined(GPL_MATH_EMULATE) 512 if (npx_ex16) { 513 if (!(flags & NPX_PREFER_EMULATOR)) 514 device_printf(dev, "INT 16 interface\n"); 515 else { 516 device_printf(dev, "FPU exists, but flags request " 517 "emulator\n"); 518 hw_float = npx_exists = 0; 519 } 520 } else if (npx_exists) { 521 device_printf(dev, "error reporting broken; using 387 emulator\n"); 522 hw_float = npx_exists = 0; 523 } else 524 device_printf(dev, "387 emulator\n"); 525 #else 526 if (npx_ex16) { 527 device_printf(dev, "INT 16 interface\n"); 528 if (flags & NPX_PREFER_EMULATOR) { 529 device_printf(dev, "emulator requested, but none compiled " 530 "into kernel, using FPU\n"); 531 } 532 } else 533 device_printf(dev, "no 387 emulator in kernel and no FPU!\n"); 534 #endif 535 } 536 npxinit(__INITIAL_NPXCW__); 537 538 #ifdef I586_CPU_XXX 539 if (cpu_class == CPUCLASS_586 && npx_ex16 && npx_exists && 540 timezero("i586_bzero()", i586_bzero) < 541 timezero("bzero()", bzero) * 4 / 5) { 542 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_BCOPY)) { 543 bcopy_vector = i586_bcopy; 544 ovbcopy_vector = i586_bcopy; 545 } 546 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_BZERO)) 547 bzero = i586_bzero; 548 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_COPYIO)) { 549 copyin_vector = i586_copyin; 550 copyout_vector = i586_copyout; 551 } 552 } 553 #endif 554 555 return (0); /* XXX unused */ 556 } 557 558 /* 559 * Initialize floating point unit. 560 */ 561 void 562 npxinit(control) 563 u_short control; 564 { 565 static union savefpu dummy; 566 critical_t savecrit; 567 568 if (!npx_exists) 569 return; 570 /* 571 * fninit has the same h/w bugs as fnsave. Use the detoxified 572 * fnsave to throw away any junk in the fpu. npxsave() initializes 573 * the fpu and sets npxproc = NULL as important side effects. 574 */ 575 savecrit = critical_enter(); 576 npxsave(&dummy); 577 stop_emulating(); 578 fldcw(&control); 579 if (PCPU_GET(curpcb) != NULL) 580 fpusave(&PCPU_GET(curpcb)->pcb_save); 581 start_emulating(); 582 critical_exit(savecrit); 583 } 584 585 /* 586 * Free coprocessor (if we have it). 587 */ 588 void 589 npxexit(p) 590 struct proc *p; 591 { 592 critical_t savecrit; 593 594 savecrit = critical_enter(); 595 if (p == PCPU_GET(npxproc)) 596 npxsave(&PCPU_GET(curpcb)->pcb_save); 597 critical_exit(savecrit); 598 #ifdef NPX_DEBUG 599 if (npx_exists) { 600 u_int masked_exceptions; 601 602 masked_exceptions = PCPU_GET(curpcb)->pcb_save.sv_87.sv_env.en_cw 603 & PCPU_GET(curpcb)->pcb_save.sv_87.sv_env.en_sw & 0x7f; 604 /* 605 * Log exceptions that would have trapped with the old 606 * control word (overflow, divide by 0, and invalid operand). 607 */ 608 if (masked_exceptions & 0x0d) 609 log(LOG_ERR, 610 "pid %d (%s) exited with masked floating point exceptions 0x%02x\n", 611 p->p_pid, p->p_comm, masked_exceptions); 612 } 613 #endif 614 } 615 616 /* 617 * The following mechanism is used to ensure that the FPE_... value 618 * that is passed as a trapcode to the signal handler of the user 619 * process does not have more than one bit set. 620 * 621 * Multiple bits may be set if the user process modifies the control 622 * word while a status word bit is already set. While this is a sign 623 * of bad coding, we have no choise than to narrow them down to one 624 * bit, since we must not send a trapcode that is not exactly one of 625 * the FPE_ macros. 626 * 627 * The mechanism has a static table with 127 entries. Each combination 628 * of the 7 FPU status word exception bits directly translates to a 629 * position in this table, where a single FPE_... value is stored. 630 * This FPE_... value stored there is considered the "most important" 631 * of the exception bits and will be sent as the signal code. The 632 * precedence of the bits is based upon Intel Document "Numerical 633 * Applications", Chapter "Special Computational Situations". 634 * 635 * The macro to choose one of these values does these steps: 1) Throw 636 * away status word bits that cannot be masked. 2) Throw away the bits 637 * currently masked in the control word, assuming the user isn't 638 * interested in them anymore. 3) Reinsert status word bit 7 (stack 639 * fault) if it is set, which cannot be masked but must be presered. 640 * 4) Use the remaining bits to point into the trapcode table. 641 * 642 * The 6 maskable bits in order of their preference, as stated in the 643 * above referenced Intel manual: 644 * 1 Invalid operation (FP_X_INV) 645 * 1a Stack underflow 646 * 1b Stack overflow 647 * 1c Operand of unsupported format 648 * 1d SNaN operand. 649 * 2 QNaN operand (not an exception, irrelavant here) 650 * 3 Any other invalid-operation not mentioned above or zero divide 651 * (FP_X_INV, FP_X_DZ) 652 * 4 Denormal operand (FP_X_DNML) 653 * 5 Numeric over/underflow (FP_X_OFL, FP_X_UFL) 654 * 6 Inexact result (FP_X_IMP) 655 */ 656 static char fpetable[128] = { 657 0, 658 FPE_FLTINV, /* 1 - INV */ 659 FPE_FLTUND, /* 2 - DNML */ 660 FPE_FLTINV, /* 3 - INV | DNML */ 661 FPE_FLTDIV, /* 4 - DZ */ 662 FPE_FLTINV, /* 5 - INV | DZ */ 663 FPE_FLTDIV, /* 6 - DNML | DZ */ 664 FPE_FLTINV, /* 7 - INV | DNML | DZ */ 665 FPE_FLTOVF, /* 8 - OFL */ 666 FPE_FLTINV, /* 9 - INV | OFL */ 667 FPE_FLTUND, /* A - DNML | OFL */ 668 FPE_FLTINV, /* B - INV | DNML | OFL */ 669 FPE_FLTDIV, /* C - DZ | OFL */ 670 FPE_FLTINV, /* D - INV | DZ | OFL */ 671 FPE_FLTDIV, /* E - DNML | DZ | OFL */ 672 FPE_FLTINV, /* F - INV | DNML | DZ | OFL */ 673 FPE_FLTUND, /* 10 - UFL */ 674 FPE_FLTINV, /* 11 - INV | UFL */ 675 FPE_FLTUND, /* 12 - DNML | UFL */ 676 FPE_FLTINV, /* 13 - INV | DNML | UFL */ 677 FPE_FLTDIV, /* 14 - DZ | UFL */ 678 FPE_FLTINV, /* 15 - INV | DZ | UFL */ 679 FPE_FLTDIV, /* 16 - DNML | DZ | UFL */ 680 FPE_FLTINV, /* 17 - INV | DNML | DZ | UFL */ 681 FPE_FLTOVF, /* 18 - OFL | UFL */ 682 FPE_FLTINV, /* 19 - INV | OFL | UFL */ 683 FPE_FLTUND, /* 1A - DNML | OFL | UFL */ 684 FPE_FLTINV, /* 1B - INV | DNML | OFL | UFL */ 685 FPE_FLTDIV, /* 1C - DZ | OFL | UFL */ 686 FPE_FLTINV, /* 1D - INV | DZ | OFL | UFL */ 687 FPE_FLTDIV, /* 1E - DNML | DZ | OFL | UFL */ 688 FPE_FLTINV, /* 1F - INV | DNML | DZ | OFL | UFL */ 689 FPE_FLTRES, /* 20 - IMP */ 690 FPE_FLTINV, /* 21 - INV | IMP */ 691 FPE_FLTUND, /* 22 - DNML | IMP */ 692 FPE_FLTINV, /* 23 - INV | DNML | IMP */ 693 FPE_FLTDIV, /* 24 - DZ | IMP */ 694 FPE_FLTINV, /* 25 - INV | DZ | IMP */ 695 FPE_FLTDIV, /* 26 - DNML | DZ | IMP */ 696 FPE_FLTINV, /* 27 - INV | DNML | DZ | IMP */ 697 FPE_FLTOVF, /* 28 - OFL | IMP */ 698 FPE_FLTINV, /* 29 - INV | OFL | IMP */ 699 FPE_FLTUND, /* 2A - DNML | OFL | IMP */ 700 FPE_FLTINV, /* 2B - INV | DNML | OFL | IMP */ 701 FPE_FLTDIV, /* 2C - DZ | OFL | IMP */ 702 FPE_FLTINV, /* 2D - INV | DZ | OFL | IMP */ 703 FPE_FLTDIV, /* 2E - DNML | DZ | OFL | IMP */ 704 FPE_FLTINV, /* 2F - INV | DNML | DZ | OFL | IMP */ 705 FPE_FLTUND, /* 30 - UFL | IMP */ 706 FPE_FLTINV, /* 31 - INV | UFL | IMP */ 707 FPE_FLTUND, /* 32 - DNML | UFL | IMP */ 708 FPE_FLTINV, /* 33 - INV | DNML | UFL | IMP */ 709 FPE_FLTDIV, /* 34 - DZ | UFL | IMP */ 710 FPE_FLTINV, /* 35 - INV | DZ | UFL | IMP */ 711 FPE_FLTDIV, /* 36 - DNML | DZ | UFL | IMP */ 712 FPE_FLTINV, /* 37 - INV | DNML | DZ | UFL | IMP */ 713 FPE_FLTOVF, /* 38 - OFL | UFL | IMP */ 714 FPE_FLTINV, /* 39 - INV | OFL | UFL | IMP */ 715 FPE_FLTUND, /* 3A - DNML | OFL | UFL | IMP */ 716 FPE_FLTINV, /* 3B - INV | DNML | OFL | UFL | IMP */ 717 FPE_FLTDIV, /* 3C - DZ | OFL | UFL | IMP */ 718 FPE_FLTINV, /* 3D - INV | DZ | OFL | UFL | IMP */ 719 FPE_FLTDIV, /* 3E - DNML | DZ | OFL | UFL | IMP */ 720 FPE_FLTINV, /* 3F - INV | DNML | DZ | OFL | UFL | IMP */ 721 FPE_FLTSUB, /* 40 - STK */ 722 FPE_FLTSUB, /* 41 - INV | STK */ 723 FPE_FLTUND, /* 42 - DNML | STK */ 724 FPE_FLTSUB, /* 43 - INV | DNML | STK */ 725 FPE_FLTDIV, /* 44 - DZ | STK */ 726 FPE_FLTSUB, /* 45 - INV | DZ | STK */ 727 FPE_FLTDIV, /* 46 - DNML | DZ | STK */ 728 FPE_FLTSUB, /* 47 - INV | DNML | DZ | STK */ 729 FPE_FLTOVF, /* 48 - OFL | STK */ 730 FPE_FLTSUB, /* 49 - INV | OFL | STK */ 731 FPE_FLTUND, /* 4A - DNML | OFL | STK */ 732 FPE_FLTSUB, /* 4B - INV | DNML | OFL | STK */ 733 FPE_FLTDIV, /* 4C - DZ | OFL | STK */ 734 FPE_FLTSUB, /* 4D - INV | DZ | OFL | STK */ 735 FPE_FLTDIV, /* 4E - DNML | DZ | OFL | STK */ 736 FPE_FLTSUB, /* 4F - INV | DNML | DZ | OFL | STK */ 737 FPE_FLTUND, /* 50 - UFL | STK */ 738 FPE_FLTSUB, /* 51 - INV | UFL | STK */ 739 FPE_FLTUND, /* 52 - DNML | UFL | STK */ 740 FPE_FLTSUB, /* 53 - INV | DNML | UFL | STK */ 741 FPE_FLTDIV, /* 54 - DZ | UFL | STK */ 742 FPE_FLTSUB, /* 55 - INV | DZ | UFL | STK */ 743 FPE_FLTDIV, /* 56 - DNML | DZ | UFL | STK */ 744 FPE_FLTSUB, /* 57 - INV | DNML | DZ | UFL | STK */ 745 FPE_FLTOVF, /* 58 - OFL | UFL | STK */ 746 FPE_FLTSUB, /* 59 - INV | OFL | UFL | STK */ 747 FPE_FLTUND, /* 5A - DNML | OFL | UFL | STK */ 748 FPE_FLTSUB, /* 5B - INV | DNML | OFL | UFL | STK */ 749 FPE_FLTDIV, /* 5C - DZ | OFL | UFL | STK */ 750 FPE_FLTSUB, /* 5D - INV | DZ | OFL | UFL | STK */ 751 FPE_FLTDIV, /* 5E - DNML | DZ | OFL | UFL | STK */ 752 FPE_FLTSUB, /* 5F - INV | DNML | DZ | OFL | UFL | STK */ 753 FPE_FLTRES, /* 60 - IMP | STK */ 754 FPE_FLTSUB, /* 61 - INV | IMP | STK */ 755 FPE_FLTUND, /* 62 - DNML | IMP | STK */ 756 FPE_FLTSUB, /* 63 - INV | DNML | IMP | STK */ 757 FPE_FLTDIV, /* 64 - DZ | IMP | STK */ 758 FPE_FLTSUB, /* 65 - INV | DZ | IMP | STK */ 759 FPE_FLTDIV, /* 66 - DNML | DZ | IMP | STK */ 760 FPE_FLTSUB, /* 67 - INV | DNML | DZ | IMP | STK */ 761 FPE_FLTOVF, /* 68 - OFL | IMP | STK */ 762 FPE_FLTSUB, /* 69 - INV | OFL | IMP | STK */ 763 FPE_FLTUND, /* 6A - DNML | OFL | IMP | STK */ 764 FPE_FLTSUB, /* 6B - INV | DNML | OFL | IMP | STK */ 765 FPE_FLTDIV, /* 6C - DZ | OFL | IMP | STK */ 766 FPE_FLTSUB, /* 6D - INV | DZ | OFL | IMP | STK */ 767 FPE_FLTDIV, /* 6E - DNML | DZ | OFL | IMP | STK */ 768 FPE_FLTSUB, /* 6F - INV | DNML | DZ | OFL | IMP | STK */ 769 FPE_FLTUND, /* 70 - UFL | IMP | STK */ 770 FPE_FLTSUB, /* 71 - INV | UFL | IMP | STK */ 771 FPE_FLTUND, /* 72 - DNML | UFL | IMP | STK */ 772 FPE_FLTSUB, /* 73 - INV | DNML | UFL | IMP | STK */ 773 FPE_FLTDIV, /* 74 - DZ | UFL | IMP | STK */ 774 FPE_FLTSUB, /* 75 - INV | DZ | UFL | IMP | STK */ 775 FPE_FLTDIV, /* 76 - DNML | DZ | UFL | IMP | STK */ 776 FPE_FLTSUB, /* 77 - INV | DNML | DZ | UFL | IMP | STK */ 777 FPE_FLTOVF, /* 78 - OFL | UFL | IMP | STK */ 778 FPE_FLTSUB, /* 79 - INV | OFL | UFL | IMP | STK */ 779 FPE_FLTUND, /* 7A - DNML | OFL | UFL | IMP | STK */ 780 FPE_FLTSUB, /* 7B - INV | DNML | OFL | UFL | IMP | STK */ 781 FPE_FLTDIV, /* 7C - DZ | OFL | UFL | IMP | STK */ 782 FPE_FLTSUB, /* 7D - INV | DZ | OFL | UFL | IMP | STK */ 783 FPE_FLTDIV, /* 7E - DNML | DZ | OFL | UFL | IMP | STK */ 784 FPE_FLTSUB, /* 7F - INV | DNML | DZ | OFL | UFL | IMP | STK */ 785 }; 786 787 /* 788 * Preserve the FP status word, clear FP exceptions, then generate a SIGFPE. 789 * 790 * Clearing exceptions is necessary mainly to avoid IRQ13 bugs. We now 791 * depend on longjmp() restoring a usable state. Restoring the state 792 * or examining it might fail if we didn't clear exceptions. 793 * 794 * The error code chosen will be one of the FPE_... macros. It will be 795 * sent as the second argument to old BSD-style signal handlers and as 796 * "siginfo_t->si_code" (second argument) to SA_SIGINFO signal handlers. 797 * 798 * XXX the FP state is not preserved across signal handlers. So signal 799 * handlers cannot afford to do FP unless they preserve the state or 800 * longjmp() out. Both preserving the state and longjmp()ing may be 801 * destroyed by IRQ13 bugs. Clearing FP exceptions is not an acceptable 802 * solution for signals other than SIGFPE. 803 */ 804 int 805 npxtrap() 806 { 807 critical_t savecrit; 808 u_short control, status; 809 u_long *exstat; 810 811 if (!npx_exists) { 812 printf("npxtrap: npxproc = %p, curproc = %p, npx_exists = %d\n", 813 PCPU_GET(npxproc), curproc, npx_exists); 814 panic("npxtrap from nowhere"); 815 } 816 savecrit = critical_enter(); 817 818 /* 819 * Interrupt handling (for another interrupt) may have pushed the 820 * state to memory. Fetch the relevant parts of the state from 821 * wherever they are. 822 */ 823 if (PCPU_GET(npxproc) != curproc) { 824 control = GET_FPU_CW(curproc); 825 status = GET_FPU_SW(curproc); 826 } else { 827 fnstcw(&control); 828 fnstsw(&status); 829 } 830 831 exstat = GET_FPU_EXSW_PTR(&curproc->p_addr->u_pcb); 832 *exstat = status; 833 if (PCPU_GET(npxproc) != curproc) 834 GET_FPU_SW(curproc) &= ~0x80bf; 835 else 836 fnclex(); 837 critical_exit(savecrit); 838 return (fpetable[status & ((~control & 0x3f) | 0x40)]); 839 } 840 841 /* 842 * Implement device not available (DNA) exception 843 * 844 * It would be better to switch FP context here (if curproc != npxproc) 845 * and not necessarily for every context switch, but it is too hard to 846 * access foreign pcb's. 847 */ 848 int 849 npxdna() 850 { 851 u_long *exstat; 852 critical_t s; 853 854 if (!npx_exists) 855 return (0); 856 if (PCPU_GET(npxproc) != NULL) { 857 printf("npxdna: npxproc = %p, curproc = %p\n", 858 PCPU_GET(npxproc), curproc); 859 panic("npxdna"); 860 } 861 s = critical_enter(); 862 stop_emulating(); 863 /* 864 * Record new context early in case frstor causes an IRQ13. 865 */ 866 PCPU_SET(npxproc, CURPROC); 867 868 exstat = GET_FPU_EXSW_PTR(PCPU_GET(curpcb)); 869 *exstat = 0; 870 /* 871 * The following frstor may cause an IRQ13 when the state being 872 * restored has a pending error. The error will appear to have been 873 * triggered by the current (npx) user instruction even when that 874 * instruction is a no-wait instruction that should not trigger an 875 * error (e.g., fnclex). On at least one 486 system all of the 876 * no-wait instructions are broken the same as frstor, so our 877 * treatment does not amplify the breakage. On at least one 878 * 386/Cyrix 387 system, fnclex works correctly while frstor and 879 * fnsave are broken, so our treatment breaks fnclex if it is the 880 * first FPU instruction after a context switch. 881 */ 882 fpurstor(&PCPU_GET(curpcb)->pcb_save); 883 critical_exit(s); 884 885 return (1); 886 } 887 888 /* 889 * Wrapper for fnsave instruction, partly to handle hardware bugs. When npx 890 * exceptions are reported via IRQ13, spurious IRQ13's may be triggered by 891 * no-wait npx instructions. See the Intel application note AP-578 for 892 * details. This doesn't cause any additional complications here. IRQ13's 893 * are inherently asynchronous unless the CPU is frozen to deliver them -- 894 * one that started in userland may be delivered many instructions later, 895 * after the process has entered the kernel. It may even be delivered after 896 * the fnsave here completes. A spurious IRQ13 for the fnsave is handled in 897 * the same way as a very-late-arriving non-spurious IRQ13 from user mode: 898 * it is normally ignored at first because we set npxproc to NULL; it is 899 * normally retriggered in npxdna() after return to user mode. 900 * 901 * npxsave() must be called with interrupts disabled, so that it clears 902 * npxproc atomically with saving the state. We require callers to do the 903 * disabling, since most callers need to disable interrupts anyway to call 904 * npxsave() atomically with checking npxproc. 905 * 906 * A previous version of npxsave() went to great lengths to excecute fnsave 907 * with interrupts enabled in case executing it froze the CPU. This case 908 * can't happen, at least for Intel CPU/NPX's. Spurious IRQ13's don't imply 909 * spurious freezes. 910 */ 911 void 912 npxsave(addr) 913 union savefpu *addr; 914 { 915 916 stop_emulating(); 917 fpusave(addr); 918 919 start_emulating(); 920 PCPU_SET(npxproc, NULL); 921 } 922 923 static void 924 fpusave(addr) 925 union savefpu *addr; 926 { 927 928 #ifdef CPU_ENABLE_SSE 929 if (cpu_fxsr) 930 fxsave(addr); 931 else 932 #endif 933 fnsave(addr); 934 } 935 936 static void 937 fpurstor(addr) 938 union savefpu *addr; 939 { 940 941 #ifdef CPU_ENABLE_SSE 942 if (cpu_fxsr) 943 fxrstor(addr); 944 else 945 #endif 946 frstor(addr); 947 } 948 949 #ifdef I586_CPU_XXX 950 static long 951 timezero(funcname, func) 952 const char *funcname; 953 void (*func) __P((void *buf, size_t len)); 954 955 { 956 void *buf; 957 #define BUFSIZE 1048576 958 long usec; 959 struct timeval finish, start; 960 961 buf = malloc(BUFSIZE, M_TEMP, M_NOWAIT); 962 if (buf == NULL) 963 return (BUFSIZE); 964 microtime(&start); 965 (*func)(buf, BUFSIZE); 966 microtime(&finish); 967 usec = 1000000 * (finish.tv_sec - start.tv_sec) + 968 finish.tv_usec - start.tv_usec; 969 if (usec <= 0) 970 usec = 1; 971 if (bootverbose) 972 printf("%s bandwidth = %u kBps\n", funcname, 973 (u_int32_t)(((BUFSIZE >> 10) * 1000000) / usec)); 974 free(buf, M_TEMP); 975 return (usec); 976 } 977 #endif /* I586_CPU */ 978 979 static device_method_t npx_methods[] = { 980 /* Device interface */ 981 DEVMETHOD(device_identify, npx_identify), 982 DEVMETHOD(device_probe, npx_probe), 983 DEVMETHOD(device_attach, npx_attach), 984 DEVMETHOD(device_detach, bus_generic_detach), 985 DEVMETHOD(device_shutdown, bus_generic_shutdown), 986 DEVMETHOD(device_suspend, bus_generic_suspend), 987 DEVMETHOD(device_resume, bus_generic_resume), 988 989 { 0, 0 } 990 }; 991 992 static driver_t npx_driver = { 993 "npx", 994 npx_methods, 995 1, /* no softc */ 996 }; 997 998 static devclass_t npx_devclass; 999 1000 /* 1001 * We prefer to attach to the root nexus so that the usual case (exception 16) 1002 * doesn't describe the processor as being `on isa'. 1003 */ 1004 DRIVER_MODULE(npx, nexus, npx_driver, npx_devclass, 0, 0); 1005 1006 /* 1007 * This sucks up the legacy ISA support assignments from PNPBIOS/ACPI. 1008 */ 1009 static struct isa_pnp_id npxisa_ids[] = { 1010 { 0x040cd041, "Legacy ISA coprocessor support" }, /* PNP0C04 */ 1011 { 0 } 1012 }; 1013 1014 static int 1015 npxisa_probe(device_t dev) 1016 { 1017 int result; 1018 if ((result = ISA_PNP_PROBE(device_get_parent(dev), dev, npxisa_ids)) <= 0) { 1019 device_quiet(dev); 1020 } 1021 return(result); 1022 } 1023 1024 static int 1025 npxisa_attach(device_t dev) 1026 { 1027 return (0); 1028 } 1029 1030 static device_method_t npxisa_methods[] = { 1031 /* Device interface */ 1032 DEVMETHOD(device_probe, npxisa_probe), 1033 DEVMETHOD(device_attach, npxisa_attach), 1034 DEVMETHOD(device_detach, bus_generic_detach), 1035 DEVMETHOD(device_shutdown, bus_generic_shutdown), 1036 DEVMETHOD(device_suspend, bus_generic_suspend), 1037 DEVMETHOD(device_resume, bus_generic_resume), 1038 1039 { 0, 0 } 1040 }; 1041 1042 static driver_t npxisa_driver = { 1043 "npxisa", 1044 npxisa_methods, 1045 1, /* no softc */ 1046 }; 1047 1048 static devclass_t npxisa_devclass; 1049 1050 DRIVER_MODULE(npxisa, isa, npxisa_driver, npxisa_devclass, 0, 0); 1051 DRIVER_MODULE(npxisa, acpi, npxisa_driver, npxisa_devclass, 0, 0); 1052 1053