FreeBSD/Linux Kernel Cross Reference
sys/i386/isa/npx.c
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/malloc.h>
47 #include <sys/module.h>
48 #include <sys/sysctl.h>
49 #include <sys/proc.h>
50 #include <machine/bus.h>
51 #include <sys/rman.h>
52 #ifdef NPX_DEBUG
53 #include <sys/syslog.h>
54 #endif
55 #include <sys/signalvar.h>
56
57 #ifndef SMP
58 #include <machine/asmacros.h>
59 #endif
60 #include <machine/cputypes.h>
61 #include <machine/frame.h>
62 #include <machine/ipl.h>
63 #include <machine/md_var.h>
64 #include <machine/pcb.h>
65 #include <machine/psl.h>
66 #ifndef SMP
67 #include <machine/clock.h>
68 #endif
69 #include <machine/resource.h>
70 #include <machine/specialreg.h>
71 #include <machine/segments.h>
72
73 #ifndef SMP
74 #include <i386/isa/icu.h>
75 #include <i386/isa/intr_machdep.h>
76 #include <i386/isa/isa.h>
77 #endif
78
79 /*
80 * 387 and 287 Numeric Coprocessor Extension (NPX) Driver.
81 */
82
83 /* Configuration flags. */
84 #define NPX_DISABLE_I586_OPTIMIZED_BCOPY (1 << 0)
85 #define NPX_DISABLE_I586_OPTIMIZED_BZERO (1 << 1)
86 #define NPX_DISABLE_I586_OPTIMIZED_COPYIO (1 << 2)
87 #define NPX_PREFER_EMULATOR (1 << 3)
88
89 #ifdef __GNUC__
90
91 #define fldcw(addr) __asm("fldcw %0" : : "m" (*(addr)))
92 #define fnclex() __asm("fnclex")
93 #define fninit() __asm("fninit")
94 #define fnop() __asm("fnop")
95 #define fnsave(addr) __asm __volatile("fnsave %0" : "=m" (*(addr)))
96 #define fnstcw(addr) __asm __volatile("fnstcw %0" : "=m" (*(addr)))
97 #define fnstsw(addr) __asm __volatile("fnstsw %0" : "=m" (*(addr)))
98 #define fp_divide_by_0() __asm("fldz; fld1; fdiv %st,%st(1); fnop")
99 #define frstor(addr) __asm("frstor %0" : : "m" (*(addr)))
100 #ifdef CPU_ENABLE_SSE
101 #define fxrstor(addr) __asm("fxrstor %0" : : "m" (*(addr)))
102 #define fxsave(addr) __asm __volatile("fxsave %0" : "=m" (*(addr)))
103 #endif
104 #define start_emulating() __asm("smsw %%ax; orb %0,%%al; lmsw %%ax" \
105 : : "n" (CR0_TS) : "ax")
106 #define stop_emulating() __asm("clts")
107
108 #else /* not __GNUC__ */
109
110 void fldcw __P((caddr_t addr));
111 void fnclex __P((void));
112 void fninit __P((void));
113 void fnop __P((void));
114 void fnsave __P((caddr_t addr));
115 void fnstcw __P((caddr_t addr));
116 void fnstsw __P((caddr_t addr));
117 void fp_divide_by_0 __P((void));
118 void frstor __P((caddr_t addr));
119 #ifdef CPU_ENABLE_SSE
120 void fxsave __P((caddr_t addr));
121 void fxrstor __P((caddr_t addr));
122 #endif
123 void start_emulating __P((void));
124 void stop_emulating __P((void));
125
126 #endif /* __GNUC__ */
127
128 #ifdef CPU_ENABLE_SSE
129 #define GET_FPU_EXSW_PTR(pcb) \
130 (cpu_fxsr ? \
131 &(pcb)->pcb_save.sv_xmm.sv_ex_sw : \
132 &(pcb)->pcb_save.sv_87.sv_ex_sw)
133 #else /* CPU_ENABLE_SSE */
134 #define GET_FPU_EXSW_PTR(pcb) \
135 (&(pcb)->pcb_save.sv_87.sv_ex_sw)
136 #endif /* CPU_ENABLE_SSE */
137
138 typedef u_char bool_t;
139
140 #ifdef CPU_ENABLE_SSE
141 static void fpu_clean_state(void);
142 #endif
143
144 static int npx_attach __P((device_t dev));
145 void npx_intr __P((void *));
146 static void npx_identify __P((driver_t *driver, device_t parent));
147 static int npx_probe __P((device_t dev));
148 static int npx_probe1 __P((device_t dev));
149 static void fpusave __P((union savefpu *));
150 static void fpurstor __P((union savefpu *));
151 #ifdef I586_CPU
152 static long timezero __P((const char *funcname,
153 void (*func)(void *buf, size_t len)));
154 #endif /* I586_CPU */
155
156 int hw_float; /* XXX currently just alias for npx_exists */
157
158 SYSCTL_INT(_hw,HW_FLOATINGPT, floatingpoint,
159 CTLFLAG_RD, &hw_float, 0,
160 "Floatingpoint instructions executed in hardware");
161
162 #ifndef SMP
163 static u_int npx0_imask = SWI_CLOCK_MASK;
164 static struct gate_descriptor npx_idt_probeintr;
165 static int npx_intrno;
166 static volatile u_int npx_intrs_while_probing;
167 static volatile u_int npx_traps_while_probing;
168 #endif
169
170 static bool_t npx_ex16;
171 static bool_t npx_exists;
172 static bool_t npx_irq13;
173 static int npx_irq; /* irq number */
174
175 #ifndef SMP
176 /*
177 * Special interrupt handlers. Someday intr0-intr15 will be used to count
178 * interrupts. We'll still need a special exception 16 handler. The busy
179 * latch stuff in probeintr() can be moved to npxprobe().
180 */
181 inthand_t probeintr;
182 __asm(" \n\
183 .text \n\
184 .p2align 2,0x90 \n\
185 .type " __XSTRING(CNAME(probeintr)) ",@function \n\
186 " __XSTRING(CNAME(probeintr)) ": \n\
187 ss \n\
188 incl " __XSTRING(CNAME(npx_intrs_while_probing)) " \n\
189 pushl %eax \n\
190 movb $0x20,%al # EOI (asm in strings loses cpp features) \n\
191 outb %al,$0xa0 # IO_ICU2 \n\
192 outb %al,$0x20 # IO_ICU1 \n\
193 movb $0,%al \n\
194 outb %al,$0xf0 # clear BUSY# latch \n\
195 popl %eax \n\
196 iret \n\
197 ");
198
199 inthand_t probetrap;
200 __asm(" \n\
201 .text \n\
202 .p2align 2,0x90 \n\
203 .type " __XSTRING(CNAME(probetrap)) ",@function \n\
204 " __XSTRING(CNAME(probetrap)) ": \n\
205 ss \n\
206 incl " __XSTRING(CNAME(npx_traps_while_probing)) " \n\
207 fnclex \n\
208 iret \n\
209 ");
210 #endif /* SMP */
211
212 /*
213 * Identify routine. Create a connection point on our parent for probing.
214 */
215 static void
216 npx_identify(driver, parent)
217 driver_t *driver;
218 device_t parent;
219 {
220 device_t child;
221
222 child = BUS_ADD_CHILD(parent, 0, "npx", 0);
223 if (child == NULL)
224 panic("npx_identify");
225 }
226
227 /*
228 * Probe routine. Initialize cr0 to give correct behaviour for [f]wait
229 * whether the device exists or not (XXX should be elsewhere). Set flags
230 * to tell npxattach() what to do. Modify device struct if npx doesn't
231 * need to use interrupts. Return 1 if device exists.
232 */
233 static int
234 npx_probe(dev)
235 device_t dev;
236 {
237 #ifdef SMP
238
239 if (resource_int_value("npx", 0, "irq", &npx_irq) != 0)
240 npx_irq = 13;
241 return npx_probe1(dev);
242
243 #else /* SMP */
244
245 int result;
246 u_long save_eflags;
247 u_char save_icu1_mask;
248 u_char save_icu2_mask;
249 struct gate_descriptor save_idt_npxintr;
250 struct gate_descriptor save_idt_npxtrap;
251 /*
252 * This routine is now just a wrapper for npxprobe1(), to install
253 * special npx interrupt and trap handlers, to enable npx interrupts
254 * and to disable other interrupts. Someday isa_configure() will
255 * install suitable handlers and run with interrupts enabled so we
256 * won't need to do so much here.
257 */
258 if (resource_int_value("npx", 0, "irq", &npx_irq) != 0)
259 npx_irq = 13;
260 npx_intrno = NRSVIDT + npx_irq;
261 save_eflags = read_eflags();
262 disable_intr();
263 save_icu1_mask = inb(IO_ICU1 + 1);
264 save_icu2_mask = inb(IO_ICU2 + 1);
265 save_idt_npxintr = idt[npx_intrno];
266 save_idt_npxtrap = idt[16];
267 outb(IO_ICU1 + 1, ~IRQ_SLAVE);
268 outb(IO_ICU2 + 1, ~(1 << (npx_irq - 8)));
269 setidt(16, probetrap, SDT_SYS386TGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
270 setidt(npx_intrno, probeintr, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
271 npx_idt_probeintr = idt[npx_intrno];
272 enable_intr();
273 result = npx_probe1(dev);
274 disable_intr();
275 outb(IO_ICU1 + 1, save_icu1_mask);
276 outb(IO_ICU2 + 1, save_icu2_mask);
277 idt[npx_intrno] = save_idt_npxintr;
278 idt[16] = save_idt_npxtrap;
279 write_eflags(save_eflags);
280 return (result);
281
282 #endif /* SMP */
283 }
284
285 static int
286 npx_probe1(dev)
287 device_t dev;
288 {
289 #ifndef SMP
290 u_short control;
291 u_short status;
292 #endif
293
294 /*
295 * Partially reset the coprocessor, if any. Some BIOS's don't reset
296 * it after a warm boot.
297 */
298 outb(0xf1, 0); /* full reset on some systems, NOP on others */
299 outb(0xf0, 0); /* clear BUSY# latch */
300 /*
301 * Prepare to trap all ESC (i.e., NPX) instructions and all WAIT
302 * instructions. We must set the CR0_MP bit and use the CR0_TS
303 * bit to control the trap, because setting the CR0_EM bit does
304 * not cause WAIT instructions to trap. It's important to trap
305 * WAIT instructions - otherwise the "wait" variants of no-wait
306 * control instructions would degenerate to the "no-wait" variants
307 * after FP context switches but work correctly otherwise. It's
308 * particularly important to trap WAITs when there is no NPX -
309 * otherwise the "wait" variants would always degenerate.
310 *
311 * Try setting CR0_NE to get correct error reporting on 486DX's.
312 * Setting it should fail or do nothing on lesser processors.
313 */
314 load_cr0(rcr0() | CR0_MP | CR0_NE);
315 /*
316 * But don't trap while we're probing.
317 */
318 stop_emulating();
319 /*
320 * Finish resetting the coprocessor, if any. If there is an error
321 * pending, then we may get a bogus IRQ13, but probeintr() will handle
322 * it OK. Bogus halts have never been observed, but we enabled
323 * IRQ13 and cleared the BUSY# latch early to handle them anyway.
324 */
325 fninit();
326
327 #ifdef SMP
328 /*
329 * Exception 16 MUST work for SMP.
330 */
331 npx_irq13 = 0;
332 npx_ex16 = hw_float = npx_exists = 1;
333 device_set_desc(dev, "math processor");
334 return (0);
335
336 #else /* !SMP */
337 device_set_desc(dev, "math processor");
338
339 /*
340 * Don't use fwait here because it might hang.
341 * Don't use fnop here because it usually hangs if there is no FPU.
342 */
343 DELAY(1000); /* wait for any IRQ13 */
344 #ifdef DIAGNOSTIC
345 if (npx_intrs_while_probing != 0)
346 printf("fninit caused %u bogus npx interrupt(s)\n",
347 npx_intrs_while_probing);
348 if (npx_traps_while_probing != 0)
349 printf("fninit caused %u bogus npx trap(s)\n",
350 npx_traps_while_probing);
351 #endif
352 /*
353 * Check for a status of mostly zero.
354 */
355 status = 0x5a5a;
356 fnstsw(&status);
357 if ((status & 0xb8ff) == 0) {
358 /*
359 * Good, now check for a proper control word.
360 */
361 control = 0x5a5a;
362 fnstcw(&control);
363 if ((control & 0x1f3f) == 0x033f) {
364 hw_float = npx_exists = 1;
365 /*
366 * We have an npx, now divide by 0 to see if exception
367 * 16 works.
368 */
369 control &= ~(1 << 2); /* enable divide by 0 trap */
370 fldcw(&control);
371 npx_traps_while_probing = npx_intrs_while_probing = 0;
372 fp_divide_by_0();
373 if (npx_traps_while_probing != 0) {
374 /*
375 * Good, exception 16 works.
376 */
377 npx_ex16 = 1;
378 return (0);
379 }
380 if (npx_intrs_while_probing != 0) {
381 int rid;
382 struct resource *r;
383 void *intr;
384 /*
385 * Bad, we are stuck with IRQ13.
386 */
387 npx_irq13 = 1;
388 /*
389 * npxattach would be too late to set npx0_imask
390 */
391 npx0_imask |= (1 << npx_irq);
392
393 /*
394 * We allocate these resources permanently,
395 * so there is no need to keep track of them.
396 */
397 rid = 0;
398 r = bus_alloc_resource(dev, SYS_RES_IOPORT,
399 &rid, IO_NPX, IO_NPX,
400 IO_NPXSIZE, RF_ACTIVE);
401 if (r == 0)
402 panic("npx: can't get ports");
403 rid = 0;
404 r = bus_alloc_resource(dev, SYS_RES_IRQ,
405 &rid, npx_irq, npx_irq,
406 1, RF_ACTIVE);
407 if (r == 0)
408 panic("npx: can't get IRQ");
409 BUS_SETUP_INTR(device_get_parent(dev),
410 dev, r, INTR_TYPE_MISC,
411 npx_intr, 0, &intr);
412 if (intr == 0)
413 panic("npx: can't create intr");
414
415 return (0);
416 }
417 /*
418 * Worse, even IRQ13 is broken. Use emulator.
419 */
420 }
421 }
422 /*
423 * Probe failed, but we want to get to npxattach to initialize the
424 * emulator and say that it has been installed. XXX handle devices
425 * that aren't really devices better.
426 */
427 return (0);
428 #endif /* SMP */
429 }
430
431 /*
432 * Attach routine - announce which it is, and wire into system
433 */
434 int
435 npx_attach(dev)
436 device_t dev;
437 {
438 int flags;
439
440 if (resource_int_value("npx", 0, "flags", &flags) != 0)
441 flags = 0;
442
443 if (flags)
444 device_printf(dev, "flags 0x%x ", flags);
445 if (npx_irq13) {
446 device_printf(dev, "using IRQ 13 interface\n");
447 } else {
448 #if defined(MATH_EMULATE) || defined(GPL_MATH_EMULATE)
449 if (npx_ex16) {
450 if (!(flags & NPX_PREFER_EMULATOR))
451 device_printf(dev, "INT 16 interface\n");
452 else {
453 device_printf(dev, "FPU exists, but flags request "
454 "emulator\n");
455 hw_float = npx_exists = 0;
456 }
457 } else if (npx_exists) {
458 device_printf(dev, "error reporting broken; using 387 emulator\n");
459 hw_float = npx_exists = 0;
460 } else
461 device_printf(dev, "387 emulator\n");
462 #else
463 if (npx_ex16) {
464 device_printf(dev, "INT 16 interface\n");
465 if (flags & NPX_PREFER_EMULATOR) {
466 device_printf(dev, "emulator requested, but none compiled "
467 "into kernel, using FPU\n");
468 }
469 } else
470 device_printf(dev, "no 387 emulator in kernel and no FPU!\n");
471 #endif
472 }
473 npxinit(__INITIAL_NPXCW__);
474
475 #ifdef I586_CPU
476 if (cpu_class == CPUCLASS_586 && npx_ex16 && npx_exists &&
477 timezero("i586_bzero()", i586_bzero) <
478 timezero("bzero()", bzero) * 4 / 5) {
479 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_BCOPY)) {
480 bcopy_vector = i586_bcopy;
481 ovbcopy_vector = i586_bcopy;
482 }
483 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_BZERO))
484 bzero = i586_bzero;
485 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_COPYIO)) {
486 copyin_vector = i586_copyin;
487 copyout_vector = i586_copyout;
488 }
489 }
490 #endif
491
492 return (0); /* XXX unused */
493 }
494
495 /*
496 * Initialize floating point unit.
497 */
498 void
499 npxinit(control)
500 u_short control;
501 {
502 static union savefpu dummy;
503
504 if (!npx_exists)
505 return;
506 /*
507 * fninit has the same h/w bugs as fnsave. Use the detoxified
508 * fnsave to throw away any junk in the fpu. npxsave() initializes
509 * the fpu and sets npxproc = NULL as important side effects.
510 */
511 npxsave(&dummy);
512 stop_emulating();
513 #ifdef CPU_ENABLE_SSE
514 /* XXX npxsave() doesn't actually initialize the fpu in the SSE case. */
515 if (cpu_fxsr)
516 fninit();
517 #endif
518 fldcw(&control);
519 if (curpcb != NULL)
520 fpusave(&curpcb->pcb_save);
521 start_emulating();
522 }
523
524 /*
525 * Free coprocessor (if we have it).
526 */
527 void
528 npxexit(p)
529 struct proc *p;
530 {
531
532 if (p == npxproc)
533 npxsave(&curpcb->pcb_save);
534 #ifdef NPX_DEBUG
535 if (npx_exists) {
536 u_int masked_exceptions;
537
538 masked_exceptions = curpcb->pcb_save.sv_87.sv_env.en_cw
539 & curpcb->pcb_save.sv_87.sv_env.en_sw & 0x7f;
540 /*
541 * Log exceptions that would have trapped with the old
542 * control word (overflow, divide by 0, and invalid operand).
543 */
544 if (masked_exceptions & 0x0d)
545 log(LOG_ERR,
546 "pid %d (%s) exited with masked floating point exceptions 0x%02x\n",
547 p->p_pid, p->p_comm, masked_exceptions);
548 }
549 #endif
550 }
551
552 /*
553 * The following mechanism is used to ensure that the FPE_... value
554 * that is passed as a trapcode to the signal handler of the user
555 * process does not have more than one bit set.
556 *
557 * Multiple bits may be set if the user process modifies the control
558 * word while a status word bit is already set. While this is a sign
559 * of bad coding, we have no choise than to narrow them down to one
560 * bit, since we must not send a trapcode that is not exactly one of
561 * the FPE_ macros.
562 *
563 * The mechanism has a static table with 127 entries. Each combination
564 * of the 7 FPU status word exception bits directly translates to a
565 * position in this table, where a single FPE_... value is stored.
566 * This FPE_... value stored there is considered the "most important"
567 * of the exception bits and will be sent as the signal code. The
568 * precedence of the bits is based upon Intel Document "Numerical
569 * Applications", Chapter "Special Computational Situations".
570 *
571 * The macro to choose one of these values does these steps: 1) Throw
572 * away status word bits that cannot be masked. 2) Throw away the bits
573 * currently masked in the control word, assuming the user isn't
574 * interested in them anymore. 3) Reinsert status word bit 7 (stack
575 * fault) if it is set, which cannot be masked but must be presered.
576 * 4) Use the remaining bits to point into the trapcode table.
577 *
578 * The 6 maskable bits in order of their preference, as stated in the
579 * above referenced Intel manual:
580 * 1 Invalid operation (FP_X_INV)
581 * 1a Stack underflow
582 * 1b Stack overflow
583 * 1c Operand of unsupported format
584 * 1d SNaN operand.
585 * 2 QNaN operand (not an exception, irrelavant here)
586 * 3 Any other invalid-operation not mentioned above or zero divide
587 * (FP_X_INV, FP_X_DZ)
588 * 4 Denormal operand (FP_X_DNML)
589 * 5 Numeric over/underflow (FP_X_OFL, FP_X_UFL)
590 * 6 Inexact result (FP_X_IMP)
591 */
592 static char fpetable[128] = {
593 0,
594 FPE_FLTINV, /* 1 - INV */
595 FPE_FLTUND, /* 2 - DNML */
596 FPE_FLTINV, /* 3 - INV | DNML */
597 FPE_FLTDIV, /* 4 - DZ */
598 FPE_FLTINV, /* 5 - INV | DZ */
599 FPE_FLTDIV, /* 6 - DNML | DZ */
600 FPE_FLTINV, /* 7 - INV | DNML | DZ */
601 FPE_FLTOVF, /* 8 - OFL */
602 FPE_FLTINV, /* 9 - INV | OFL */
603 FPE_FLTUND, /* A - DNML | OFL */
604 FPE_FLTINV, /* B - INV | DNML | OFL */
605 FPE_FLTDIV, /* C - DZ | OFL */
606 FPE_FLTINV, /* D - INV | DZ | OFL */
607 FPE_FLTDIV, /* E - DNML | DZ | OFL */
608 FPE_FLTINV, /* F - INV | DNML | DZ | OFL */
609 FPE_FLTUND, /* 10 - UFL */
610 FPE_FLTINV, /* 11 - INV | UFL */
611 FPE_FLTUND, /* 12 - DNML | UFL */
612 FPE_FLTINV, /* 13 - INV | DNML | UFL */
613 FPE_FLTDIV, /* 14 - DZ | UFL */
614 FPE_FLTINV, /* 15 - INV | DZ | UFL */
615 FPE_FLTDIV, /* 16 - DNML | DZ | UFL */
616 FPE_FLTINV, /* 17 - INV | DNML | DZ | UFL */
617 FPE_FLTOVF, /* 18 - OFL | UFL */
618 FPE_FLTINV, /* 19 - INV | OFL | UFL */
619 FPE_FLTUND, /* 1A - DNML | OFL | UFL */
620 FPE_FLTINV, /* 1B - INV | DNML | OFL | UFL */
621 FPE_FLTDIV, /* 1C - DZ | OFL | UFL */
622 FPE_FLTINV, /* 1D - INV | DZ | OFL | UFL */
623 FPE_FLTDIV, /* 1E - DNML | DZ | OFL | UFL */
624 FPE_FLTINV, /* 1F - INV | DNML | DZ | OFL | UFL */
625 FPE_FLTRES, /* 20 - IMP */
626 FPE_FLTINV, /* 21 - INV | IMP */
627 FPE_FLTUND, /* 22 - DNML | IMP */
628 FPE_FLTINV, /* 23 - INV | DNML | IMP */
629 FPE_FLTDIV, /* 24 - DZ | IMP */
630 FPE_FLTINV, /* 25 - INV | DZ | IMP */
631 FPE_FLTDIV, /* 26 - DNML | DZ | IMP */
632 FPE_FLTINV, /* 27 - INV | DNML | DZ | IMP */
633 FPE_FLTOVF, /* 28 - OFL | IMP */
634 FPE_FLTINV, /* 29 - INV | OFL | IMP */
635 FPE_FLTUND, /* 2A - DNML | OFL | IMP */
636 FPE_FLTINV, /* 2B - INV | DNML | OFL | IMP */
637 FPE_FLTDIV, /* 2C - DZ | OFL | IMP */
638 FPE_FLTINV, /* 2D - INV | DZ | OFL | IMP */
639 FPE_FLTDIV, /* 2E - DNML | DZ | OFL | IMP */
640 FPE_FLTINV, /* 2F - INV | DNML | DZ | OFL | IMP */
641 FPE_FLTUND, /* 30 - UFL | IMP */
642 FPE_FLTINV, /* 31 - INV | UFL | IMP */
643 FPE_FLTUND, /* 32 - DNML | UFL | IMP */
644 FPE_FLTINV, /* 33 - INV | DNML | UFL | IMP */
645 FPE_FLTDIV, /* 34 - DZ | UFL | IMP */
646 FPE_FLTINV, /* 35 - INV | DZ | UFL | IMP */
647 FPE_FLTDIV, /* 36 - DNML | DZ | UFL | IMP */
648 FPE_FLTINV, /* 37 - INV | DNML | DZ | UFL | IMP */
649 FPE_FLTOVF, /* 38 - OFL | UFL | IMP */
650 FPE_FLTINV, /* 39 - INV | OFL | UFL | IMP */
651 FPE_FLTUND, /* 3A - DNML | OFL | UFL | IMP */
652 FPE_FLTINV, /* 3B - INV | DNML | OFL | UFL | IMP */
653 FPE_FLTDIV, /* 3C - DZ | OFL | UFL | IMP */
654 FPE_FLTINV, /* 3D - INV | DZ | OFL | UFL | IMP */
655 FPE_FLTDIV, /* 3E - DNML | DZ | OFL | UFL | IMP */
656 FPE_FLTINV, /* 3F - INV | DNML | DZ | OFL | UFL | IMP */
657 FPE_FLTSUB, /* 40 - STK */
658 FPE_FLTSUB, /* 41 - INV | STK */
659 FPE_FLTUND, /* 42 - DNML | STK */
660 FPE_FLTSUB, /* 43 - INV | DNML | STK */
661 FPE_FLTDIV, /* 44 - DZ | STK */
662 FPE_FLTSUB, /* 45 - INV | DZ | STK */
663 FPE_FLTDIV, /* 46 - DNML | DZ | STK */
664 FPE_FLTSUB, /* 47 - INV | DNML | DZ | STK */
665 FPE_FLTOVF, /* 48 - OFL | STK */
666 FPE_FLTSUB, /* 49 - INV | OFL | STK */
667 FPE_FLTUND, /* 4A - DNML | OFL | STK */
668 FPE_FLTSUB, /* 4B - INV | DNML | OFL | STK */
669 FPE_FLTDIV, /* 4C - DZ | OFL | STK */
670 FPE_FLTSUB, /* 4D - INV | DZ | OFL | STK */
671 FPE_FLTDIV, /* 4E - DNML | DZ | OFL | STK */
672 FPE_FLTSUB, /* 4F - INV | DNML | DZ | OFL | STK */
673 FPE_FLTUND, /* 50 - UFL | STK */
674 FPE_FLTSUB, /* 51 - INV | UFL | STK */
675 FPE_FLTUND, /* 52 - DNML | UFL | STK */
676 FPE_FLTSUB, /* 53 - INV | DNML | UFL | STK */
677 FPE_FLTDIV, /* 54 - DZ | UFL | STK */
678 FPE_FLTSUB, /* 55 - INV | DZ | UFL | STK */
679 FPE_FLTDIV, /* 56 - DNML | DZ | UFL | STK */
680 FPE_FLTSUB, /* 57 - INV | DNML | DZ | UFL | STK */
681 FPE_FLTOVF, /* 58 - OFL | UFL | STK */
682 FPE_FLTSUB, /* 59 - INV | OFL | UFL | STK */
683 FPE_FLTUND, /* 5A - DNML | OFL | UFL | STK */
684 FPE_FLTSUB, /* 5B - INV | DNML | OFL | UFL | STK */
685 FPE_FLTDIV, /* 5C - DZ | OFL | UFL | STK */
686 FPE_FLTSUB, /* 5D - INV | DZ | OFL | UFL | STK */
687 FPE_FLTDIV, /* 5E - DNML | DZ | OFL | UFL | STK */
688 FPE_FLTSUB, /* 5F - INV | DNML | DZ | OFL | UFL | STK */
689 FPE_FLTRES, /* 60 - IMP | STK */
690 FPE_FLTSUB, /* 61 - INV | IMP | STK */
691 FPE_FLTUND, /* 62 - DNML | IMP | STK */
692 FPE_FLTSUB, /* 63 - INV | DNML | IMP | STK */
693 FPE_FLTDIV, /* 64 - DZ | IMP | STK */
694 FPE_FLTSUB, /* 65 - INV | DZ | IMP | STK */
695 FPE_FLTDIV, /* 66 - DNML | DZ | IMP | STK */
696 FPE_FLTSUB, /* 67 - INV | DNML | DZ | IMP | STK */
697 FPE_FLTOVF, /* 68 - OFL | IMP | STK */
698 FPE_FLTSUB, /* 69 - INV | OFL | IMP | STK */
699 FPE_FLTUND, /* 6A - DNML | OFL | IMP | STK */
700 FPE_FLTSUB, /* 6B - INV | DNML | OFL | IMP | STK */
701 FPE_FLTDIV, /* 6C - DZ | OFL | IMP | STK */
702 FPE_FLTSUB, /* 6D - INV | DZ | OFL | IMP | STK */
703 FPE_FLTDIV, /* 6E - DNML | DZ | OFL | IMP | STK */
704 FPE_FLTSUB, /* 6F - INV | DNML | DZ | OFL | IMP | STK */
705 FPE_FLTUND, /* 70 - UFL | IMP | STK */
706 FPE_FLTSUB, /* 71 - INV | UFL | IMP | STK */
707 FPE_FLTUND, /* 72 - DNML | UFL | IMP | STK */
708 FPE_FLTSUB, /* 73 - INV | DNML | UFL | IMP | STK */
709 FPE_FLTDIV, /* 74 - DZ | UFL | IMP | STK */
710 FPE_FLTSUB, /* 75 - INV | DZ | UFL | IMP | STK */
711 FPE_FLTDIV, /* 76 - DNML | DZ | UFL | IMP | STK */
712 FPE_FLTSUB, /* 77 - INV | DNML | DZ | UFL | IMP | STK */
713 FPE_FLTOVF, /* 78 - OFL | UFL | IMP | STK */
714 FPE_FLTSUB, /* 79 - INV | OFL | UFL | IMP | STK */
715 FPE_FLTUND, /* 7A - DNML | OFL | UFL | IMP | STK */
716 FPE_FLTSUB, /* 7B - INV | DNML | OFL | UFL | IMP | STK */
717 FPE_FLTDIV, /* 7C - DZ | OFL | UFL | IMP | STK */
718 FPE_FLTSUB, /* 7D - INV | DZ | OFL | UFL | IMP | STK */
719 FPE_FLTDIV, /* 7E - DNML | DZ | OFL | UFL | IMP | STK */
720 FPE_FLTSUB, /* 7F - INV | DNML | DZ | OFL | UFL | IMP | STK */
721 };
722
723 /*
724 * Preserve the FP status word, clear FP exceptions, then generate a SIGFPE.
725 *
726 * Clearing exceptions is necessary mainly to avoid IRQ13 bugs. We now
727 * depend on longjmp() restoring a usable state. Restoring the state
728 * or examining it might fail if we didn't clear exceptions.
729 *
730 * The error code chosen will be one of the FPE_... macros. It will be
731 * sent as the second argument to old BSD-style signal handlers and as
732 * "siginfo_t->si_code" (second argument) to SA_SIGINFO signal handlers.
733 *
734 * XXX the FP state is not preserved across signal handlers. So signal
735 * handlers cannot afford to do FP unless they preserve the state or
736 * longjmp() out. Both preserving the state and longjmp()ing may be
737 * destroyed by IRQ13 bugs. Clearing FP exceptions is not an acceptable
738 * solution for signals other than SIGFPE.
739 */
740 void
741 npx_intr(dummy)
742 void *dummy;
743 {
744 int code;
745 u_short control;
746 struct intrframe *frame;
747 u_long *exstat;
748
749 if (npxproc == NULL || !npx_exists) {
750 printf("npxintr: npxproc = %p, curproc = %p, npx_exists = %d\n",
751 npxproc, curproc, npx_exists);
752 panic("npxintr from nowhere");
753 }
754 if (npxproc != curproc) {
755 printf("npxintr: npxproc = %p, curproc = %p, npx_exists = %d\n",
756 npxproc, curproc, npx_exists);
757 panic("npxintr from non-current process");
758 }
759
760 exstat = GET_FPU_EXSW_PTR(curpcb);
761 outb(0xf0, 0);
762 fnstsw(exstat);
763 fnstcw(&control);
764 fnclex();
765
766 /*
767 * Pass exception to process.
768 */
769 frame = (struct intrframe *)&dummy; /* XXX */
770 if ((ISPL(frame->if_cs) == SEL_UPL) || (frame->if_eflags & PSL_VM)) {
771 /*
772 * Interrupt is essentially a trap, so we can afford to call
773 * the SIGFPE handler (if any) as soon as the interrupt
774 * returns.
775 *
776 * XXX little or nothing is gained from this, and plenty is
777 * lost - the interrupt frame has to contain the trap frame
778 * (this is otherwise only necessary for the rescheduling trap
779 * in doreti, and the frame for that could easily be set up
780 * just before it is used).
781 */
782 curproc->p_md.md_regs = INTR_TO_TRAPFRAME(frame);
783 /*
784 * Encode the appropriate code for detailed information on
785 * this exception.
786 */
787 code =
788 fpetable[(*exstat & ~control & 0x3f) | (*exstat & 0x40)];
789 trapsignal(curproc, SIGFPE, code);
790 } else {
791 /*
792 * Nested interrupt. These losers occur when:
793 * o an IRQ13 is bogusly generated at a bogus time, e.g.:
794 * o immediately after an fnsave or frstor of an
795 * error state.
796 * o a couple of 386 instructions after
797 * "fstpl _memvar" causes a stack overflow.
798 * These are especially nasty when combined with a
799 * trace trap.
800 * o an IRQ13 occurs at the same time as another higher-
801 * priority interrupt.
802 *
803 * Treat them like a true async interrupt.
804 */
805 psignal(curproc, SIGFPE);
806 }
807 }
808
809 /*
810 * Implement device not available (DNA) exception
811 *
812 * It would be better to switch FP context here (if curproc != npxproc)
813 * and not necessarily for every context switch, but it is too hard to
814 * access foreign pcb's.
815 */
816 int
817 npxdna()
818 {
819 u_long *exstat;
820
821 if (!npx_exists)
822 return (0);
823 if (npxproc != NULL) {
824 printf("npxdna: npxproc = %p, curproc = %p\n",
825 npxproc, curproc);
826 panic("npxdna");
827 }
828 stop_emulating();
829 /*
830 * Record new context early in case frstor causes an IRQ13.
831 */
832 npxproc = curproc;
833 exstat = GET_FPU_EXSW_PTR(curpcb);
834 *exstat = 0;
835 /*
836 * The following frstor may cause an IRQ13 when the state being
837 * restored has a pending error. The error will appear to have been
838 * triggered by the current (npx) user instruction even when that
839 * instruction is a no-wait instruction that should not trigger an
840 * error (e.g., fnclex). On at least one 486 system all of the
841 * no-wait instructions are broken the same as frstor, so our
842 * treatment does not amplify the breakage. On at least one
843 * 386/Cyrix 387 system, fnclex works correctly while frstor and
844 * fnsave are broken, so our treatment breaks fnclex if it is the
845 * first FPU instruction after a context switch.
846 */
847 fpurstor(&curpcb->pcb_save);
848
849 return (1);
850 }
851
852 /*
853 * Wrapper for fnsave instruction to handle h/w bugs. If there is an error
854 * pending, then fnsave generates a bogus IRQ13 on some systems. Force
855 * any IRQ13 to be handled immediately, and then ignore it. This routine is
856 * often called at splhigh so it must not use many system services. In
857 * particular, it's much easier to install a special handler than to
858 * guarantee that it's safe to use npxintr() and its supporting code.
859 */
860 void
861 npxsave(addr)
862 union savefpu *addr;
863 {
864 #if defined(SMP) || defined(CPU_ENABLE_SSE)
865
866 stop_emulating();
867 fpusave(addr);
868
869 /* fnop(); */
870 start_emulating();
871 npxproc = NULL;
872
873 #else /* SMP or CPU_ENABLE_SSE */
874
875 u_char icu1_mask;
876 u_char icu2_mask;
877 u_char old_icu1_mask;
878 u_char old_icu2_mask;
879 struct gate_descriptor save_idt_npxintr;
880
881 disable_intr();
882 old_icu1_mask = inb(IO_ICU1 + 1);
883 old_icu2_mask = inb(IO_ICU2 + 1);
884 save_idt_npxintr = idt[npx_intrno];
885 outb(IO_ICU1 + 1, old_icu1_mask & ~(IRQ_SLAVE | npx0_imask));
886 outb(IO_ICU2 + 1, old_icu2_mask & ~(npx0_imask >> 8));
887 idt[npx_intrno] = npx_idt_probeintr;
888 enable_intr();
889 stop_emulating();
890 fnsave(addr);
891 fnop();
892 start_emulating();
893 npxproc = NULL;
894 disable_intr();
895 icu1_mask = inb(IO_ICU1 + 1); /* masks may have changed */
896 icu2_mask = inb(IO_ICU2 + 1);
897 outb(IO_ICU1 + 1,
898 (icu1_mask & ~npx0_imask) | (old_icu1_mask & npx0_imask));
899 outb(IO_ICU2 + 1,
900 (icu2_mask & ~(npx0_imask >> 8))
901 | (old_icu2_mask & (npx0_imask >> 8)));
902 idt[npx_intrno] = save_idt_npxintr;
903 enable_intr(); /* back to usual state */
904
905 #endif /* SMP */
906 }
907
908 static void
909 fpusave(addr)
910 union savefpu *addr;
911 {
912
913 #ifdef CPU_ENABLE_SSE
914 if (cpu_fxsr)
915 fxsave(addr);
916 else
917 #endif
918 fnsave(addr);
919 }
920
921 #ifdef CPU_ENABLE_SSE
922 /*
923 * On AuthenticAMD processors, the fxrstor instruction does not restore
924 * the x87's stored last instruction pointer, last data pointer, and last
925 * opcode values, except in the rare case in which the exception summary
926 * (ES) bit in the x87 status word is set to 1.
927 *
928 * In order to avoid leaking this information across processes, we clean
929 * these values by performing a dummy load before executing fxrstor().
930 */
931 static double dummy_variable = 0.0;
932 static void
933 fpu_clean_state(void)
934 {
935 u_short status;
936
937 /*
938 * Clear the ES bit in the x87 status word if it is currently
939 * set, in order to avoid causing a fault in the upcoming load.
940 */
941 fnstsw(&status);
942 if (status & 0x80)
943 fnclex();
944
945 /*
946 * Load the dummy variable into the x87 stack. This mangles
947 * the x87 stack, but we don't care since we're about to call
948 * fxrstor() anyway.
949 */
950 __asm __volatile("ffree %%st(7); fld %0" : : "m" (dummy_variable));
951 }
952 #endif /* CPU_ENABLE_SSE */
953
954 static void
955 fpurstor(addr)
956 union savefpu *addr;
957 {
958
959 #ifdef CPU_ENABLE_SSE
960 if (cpu_fxsr) {
961 fpu_clean_state();
962 fxrstor(addr);
963 } else
964 #endif
965 frstor(addr);
966 }
967
968 #ifdef I586_CPU
969 static long
970 timezero(funcname, func)
971 const char *funcname;
972 void (*func) __P((void *buf, size_t len));
973
974 {
975 void *buf;
976 #define BUFSIZE 1000000
977 long usec;
978 struct timeval finish, start;
979
980 buf = malloc(BUFSIZE, M_TEMP, M_NOWAIT);
981 if (buf == NULL)
982 return (BUFSIZE);
983 microtime(&start);
984 (*func)(buf, BUFSIZE);
985 microtime(&finish);
986 usec = 1000000 * (finish.tv_sec - start.tv_sec) +
987 finish.tv_usec - start.tv_usec;
988 if (usec <= 0)
989 usec = 1;
990 if (bootverbose)
991 printf("%s bandwidth = %ld bytes/sec\n",
992 funcname, (long)(BUFSIZE * (int64_t)1000000 / usec));
993 free(buf, M_TEMP);
994 return (usec);
995 }
996 #endif /* I586_CPU */
997
998 static device_method_t npx_methods[] = {
999 /* Device interface */
1000 DEVMETHOD(device_identify, npx_identify),
1001 DEVMETHOD(device_probe, npx_probe),
1002 DEVMETHOD(device_attach, npx_attach),
1003 DEVMETHOD(device_detach, bus_generic_detach),
1004 DEVMETHOD(device_shutdown, bus_generic_shutdown),
1005 DEVMETHOD(device_suspend, bus_generic_suspend),
1006 DEVMETHOD(device_resume, bus_generic_resume),
1007
1008 { 0, 0 }
1009 };
1010
1011 static driver_t npx_driver = {
1012 "npx",
1013 npx_methods,
1014 1, /* no softc */
1015 };
1016
1017 static devclass_t npx_devclass;
1018
1019 /*
1020 * We prefer to attach to the root nexus so that the usual case (exception 16)
1021 * doesn't describe the processor as being `on isa'.
1022 */
1023 DRIVER_MODULE(npx, nexus, npx_driver, npx_devclass, 0, 0);
Cache object: 10c8acff91d844e8b2034a27c7df9322
|