2 * Copyright (c) 1990 William Jolitz.
3 * Copyright (c) 1991 The Regents of the University of California.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
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.
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
34 * from: @(#)npx.c 7.2 (Berkeley) 5/12/91
35 * $FreeBSD: src/sys/i386/isa/npx.c,v 1.80.2.3 2001/10/20 19:04:38 tegge Exp $
36 * $DragonFly: src/sys/i386/isa/Attic/npx.c,v 1.3 2003/06/18 07:04:30 dillon Exp $
40 #include "opt_debug_npx.h"
41 #include "opt_math_emulate.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/malloc.h>
48 #include <sys/module.h>
49 #include <sys/sysctl.h>
51 #include <machine/bus.h>
54 #include <sys/syslog.h>
56 #include <sys/signalvar.h>
59 #include <machine/asmacros.h>
61 #include <machine/cputypes.h>
62 #include <machine/frame.h>
63 #include <machine/ipl.h>
64 #include <machine/md_var.h>
65 #include <machine/pcb.h>
66 #include <machine/psl.h>
68 #include <machine/clock.h>
70 #include <machine/resource.h>
71 #include <machine/specialreg.h>
72 #include <machine/segments.h>
75 #include <i386/isa/icu.h>
76 #include <i386/isa/intr_machdep.h>
77 #include <i386/isa/isa.h>
81 * 387 and 287 Numeric Coprocessor Extension (NPX) Driver.
84 /* Configuration flags. */
85 #define NPX_DISABLE_I586_OPTIMIZED_BCOPY (1 << 0)
86 #define NPX_DISABLE_I586_OPTIMIZED_BZERO (1 << 1)
87 #define NPX_DISABLE_I586_OPTIMIZED_COPYIO (1 << 2)
88 #define NPX_PREFER_EMULATOR (1 << 3)
92 #define fldcw(addr) __asm("fldcw %0" : : "m" (*(addr)))
93 #define fnclex() __asm("fnclex")
94 #define fninit() __asm("fninit")
95 #define fnop() __asm("fnop")
96 #define fnsave(addr) __asm __volatile("fnsave %0" : "=m" (*(addr)))
97 #define fnstcw(addr) __asm __volatile("fnstcw %0" : "=m" (*(addr)))
98 #define fnstsw(addr) __asm __volatile("fnstsw %0" : "=m" (*(addr)))
99 #define fp_divide_by_0() __asm("fldz; fld1; fdiv %st,%st(1); fnop")
100 #define frstor(addr) __asm("frstor %0" : : "m" (*(addr)))
101 #ifdef CPU_ENABLE_SSE
102 #define fxrstor(addr) __asm("fxrstor %0" : : "m" (*(addr)))
103 #define fxsave(addr) __asm __volatile("fxsave %0" : "=m" (*(addr)))
105 #define start_emulating() __asm("smsw %%ax; orb %0,%%al; lmsw %%ax" \
106 : : "n" (CR0_TS) : "ax")
107 #define stop_emulating() __asm("clts")
109 #else /* not __GNUC__ */
111 void fldcw __P((caddr_t addr));
112 void fnclex __P((void));
113 void fninit __P((void));
114 void fnop __P((void));
115 void fnsave __P((caddr_t addr));
116 void fnstcw __P((caddr_t addr));
117 void fnstsw __P((caddr_t addr));
118 void fp_divide_by_0 __P((void));
119 void frstor __P((caddr_t addr));
120 #ifdef CPU_ENABLE_SSE
121 void fxsave __P((caddr_t addr));
122 void fxrstor __P((caddr_t addr));
124 void start_emulating __P((void));
125 void stop_emulating __P((void));
127 #endif /* __GNUC__ */
129 #ifdef CPU_ENABLE_SSE
130 #define GET_FPU_EXSW_PTR(pcb) \
132 &(pcb)->pcb_save.sv_xmm.sv_ex_sw : \
133 &(pcb)->pcb_save.sv_87.sv_ex_sw)
134 #else /* CPU_ENABLE_SSE */
135 #define GET_FPU_EXSW_PTR(pcb) \
136 (&(pcb)->pcb_save.sv_87.sv_ex_sw)
137 #endif /* CPU_ENABLE_SSE */
139 typedef u_char bool_t;
141 static int npx_attach __P((device_t dev));
142 void npx_intr __P((void *));
143 static void npx_identify __P((driver_t *driver, device_t parent));
144 static int npx_probe __P((device_t dev));
145 static int npx_probe1 __P((device_t dev));
146 static void fpusave __P((union savefpu *));
147 static void fpurstor __P((union savefpu *));
149 static long timezero __P((const char *funcname,
150 void (*func)(void *buf, size_t len)));
151 #endif /* I586_CPU */
153 int hw_float; /* XXX currently just alias for npx_exists */
155 SYSCTL_INT(_hw,HW_FLOATINGPT, floatingpoint,
156 CTLFLAG_RD, &hw_float, 0,
157 "Floatingpoint instructions executed in hardware");
160 static u_int npx0_imask = SWI_CLOCK_MASK;
161 static struct gate_descriptor npx_idt_probeintr;
162 static int npx_intrno;
163 static volatile u_int npx_intrs_while_probing;
164 static volatile u_int npx_traps_while_probing;
167 static bool_t npx_ex16;
168 static bool_t npx_exists;
169 static bool_t npx_irq13;
170 static int npx_irq; /* irq number */
174 * Special interrupt handlers. Someday intr0-intr15 will be used to count
175 * interrupts. We'll still need a special exception 16 handler. The busy
176 * latch stuff in probeintr() can be moved to npxprobe().
182 .type " __XSTRING(CNAME(probeintr)) ",@function \n\
183 " __XSTRING(CNAME(probeintr)) ": \n\
185 incl " __XSTRING(CNAME(npx_intrs_while_probing)) " \n\
187 movb $0x20,%al # EOI (asm in strings loses cpp features) \n\
188 outb %al,$0xa0 # IO_ICU2 \n\
189 outb %al,$0x20 # IO_ICU1 \n\
191 outb %al,$0xf0 # clear BUSY# latch \n\
200 .type " __XSTRING(CNAME(probetrap)) ",@function \n\
201 " __XSTRING(CNAME(probetrap)) ": \n\
203 incl " __XSTRING(CNAME(npx_traps_while_probing)) " \n\
210 * Identify routine. Create a connection point on our parent for probing.
213 npx_identify(driver, parent)
219 child = BUS_ADD_CHILD(parent, 0, "npx", 0);
221 panic("npx_identify");
225 * Probe routine. Initialize cr0 to give correct behaviour for [f]wait
226 * whether the device exists or not (XXX should be elsewhere). Set flags
227 * to tell npxattach() what to do. Modify device struct if npx doesn't
228 * need to use interrupts. Return 1 if device exists.
236 if (resource_int_value("npx", 0, "irq", &npx_irq) != 0)
238 return npx_probe1(dev);
244 u_char save_icu1_mask;
245 u_char save_icu2_mask;
246 struct gate_descriptor save_idt_npxintr;
247 struct gate_descriptor save_idt_npxtrap;
249 * This routine is now just a wrapper for npxprobe1(), to install
250 * special npx interrupt and trap handlers, to enable npx interrupts
251 * and to disable other interrupts. Someday isa_configure() will
252 * install suitable handlers and run with interrupts enabled so we
253 * won't need to do so much here.
255 if (resource_int_value("npx", 0, "irq", &npx_irq) != 0)
257 npx_intrno = NRSVIDT + npx_irq;
258 save_eflags = read_eflags();
260 save_icu1_mask = inb(IO_ICU1 + 1);
261 save_icu2_mask = inb(IO_ICU2 + 1);
262 save_idt_npxintr = idt[npx_intrno];
263 save_idt_npxtrap = idt[16];
264 outb(IO_ICU1 + 1, ~IRQ_SLAVE);
265 outb(IO_ICU2 + 1, ~(1 << (npx_irq - 8)));
266 setidt(16, probetrap, SDT_SYS386TGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
267 setidt(npx_intrno, probeintr, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
268 npx_idt_probeintr = idt[npx_intrno];
270 result = npx_probe1(dev);
272 outb(IO_ICU1 + 1, save_icu1_mask);
273 outb(IO_ICU2 + 1, save_icu2_mask);
274 idt[npx_intrno] = save_idt_npxintr;
275 idt[16] = save_idt_npxtrap;
276 write_eflags(save_eflags);
292 * Partially reset the coprocessor, if any. Some BIOS's don't reset
293 * it after a warm boot.
295 outb(0xf1, 0); /* full reset on some systems, NOP on others */
296 outb(0xf0, 0); /* clear BUSY# latch */
298 * Prepare to trap all ESC (i.e., NPX) instructions and all WAIT
299 * instructions. We must set the CR0_MP bit and use the CR0_TS
300 * bit to control the trap, because setting the CR0_EM bit does
301 * not cause WAIT instructions to trap. It's important to trap
302 * WAIT instructions - otherwise the "wait" variants of no-wait
303 * control instructions would degenerate to the "no-wait" variants
304 * after FP context switches but work correctly otherwise. It's
305 * particularly important to trap WAITs when there is no NPX -
306 * otherwise the "wait" variants would always degenerate.
308 * Try setting CR0_NE to get correct error reporting on 486DX's.
309 * Setting it should fail or do nothing on lesser processors.
311 load_cr0(rcr0() | CR0_MP | CR0_NE);
313 * But don't trap while we're probing.
317 * Finish resetting the coprocessor, if any. If there is an error
318 * pending, then we may get a bogus IRQ13, but probeintr() will handle
319 * it OK. Bogus halts have never been observed, but we enabled
320 * IRQ13 and cleared the BUSY# latch early to handle them anyway.
326 * Exception 16 MUST work for SMP.
329 npx_ex16 = hw_float = npx_exists = 1;
330 device_set_desc(dev, "math processor");
334 device_set_desc(dev, "math processor");
337 * Don't use fwait here because it might hang.
338 * Don't use fnop here because it usually hangs if there is no FPU.
340 DELAY(1000); /* wait for any IRQ13 */
342 if (npx_intrs_while_probing != 0)
343 printf("fninit caused %u bogus npx interrupt(s)\n",
344 npx_intrs_while_probing);
345 if (npx_traps_while_probing != 0)
346 printf("fninit caused %u bogus npx trap(s)\n",
347 npx_traps_while_probing);
350 * Check for a status of mostly zero.
354 if ((status & 0xb8ff) == 0) {
356 * Good, now check for a proper control word.
360 if ((control & 0x1f3f) == 0x033f) {
361 hw_float = npx_exists = 1;
363 * We have an npx, now divide by 0 to see if exception
366 control &= ~(1 << 2); /* enable divide by 0 trap */
368 npx_traps_while_probing = npx_intrs_while_probing = 0;
370 if (npx_traps_while_probing != 0) {
372 * Good, exception 16 works.
377 if (npx_intrs_while_probing != 0) {
382 * Bad, we are stuck with IRQ13.
386 * npxattach would be too late to set npx0_imask
388 npx0_imask |= (1 << npx_irq);
391 * We allocate these resources permanently,
392 * so there is no need to keep track of them.
395 r = bus_alloc_resource(dev, SYS_RES_IOPORT,
396 &rid, IO_NPX, IO_NPX,
397 IO_NPXSIZE, RF_ACTIVE);
399 panic("npx: can't get ports");
401 r = bus_alloc_resource(dev, SYS_RES_IRQ,
402 &rid, npx_irq, npx_irq,
405 panic("npx: can't get IRQ");
406 BUS_SETUP_INTR(device_get_parent(dev),
407 dev, r, INTR_TYPE_MISC,
410 panic("npx: can't create intr");
415 * Worse, even IRQ13 is broken. Use emulator.
420 * Probe failed, but we want to get to npxattach to initialize the
421 * emulator and say that it has been installed. XXX handle devices
422 * that aren't really devices better.
429 * Attach routine - announce which it is, and wire into system
437 if (resource_int_value("npx", 0, "flags", &flags) != 0)
441 device_printf(dev, "flags 0x%x ", flags);
443 device_printf(dev, "using IRQ 13 interface\n");
445 #if defined(MATH_EMULATE) || defined(GPL_MATH_EMULATE)
447 if (!(flags & NPX_PREFER_EMULATOR))
448 device_printf(dev, "INT 16 interface\n");
450 device_printf(dev, "FPU exists, but flags request "
452 hw_float = npx_exists = 0;
454 } else if (npx_exists) {
455 device_printf(dev, "error reporting broken; using 387 emulator\n");
456 hw_float = npx_exists = 0;
458 device_printf(dev, "387 emulator\n");
461 device_printf(dev, "INT 16 interface\n");
462 if (flags & NPX_PREFER_EMULATOR) {
463 device_printf(dev, "emulator requested, but none compiled "
464 "into kernel, using FPU\n");
467 device_printf(dev, "no 387 emulator in kernel and no FPU!\n");
470 npxinit(__INITIAL_NPXCW__);
473 if (cpu_class == CPUCLASS_586 && npx_ex16 && npx_exists &&
474 timezero("i586_bzero()", i586_bzero) <
475 timezero("bzero()", bzero) * 4 / 5) {
476 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_BCOPY)) {
477 bcopy_vector = i586_bcopy;
478 ovbcopy_vector = i586_bcopy;
480 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_BZERO))
482 if (!(flags & NPX_DISABLE_I586_OPTIMIZED_COPYIO)) {
483 copyin_vector = i586_copyin;
484 copyout_vector = i586_copyout;
489 return (0); /* XXX unused */
493 * Initialize floating point unit.
499 static union savefpu dummy;
504 * fninit has the same h/w bugs as fnsave. Use the detoxified
505 * fnsave to throw away any junk in the fpu. npxsave() initializes
506 * the fpu and sets npxthread = NULL as important side effects.
510 #ifdef CPU_ENABLE_SSE
511 /* XXX npxsave() doesn't actually initialize the fpu in the SSE case. */
517 fpusave(&curpcb->pcb_save);
522 * Free coprocessor (if we have it).
529 if (&p->p_thread == npxthread)
530 npxsave(&curpcb->pcb_save);
533 u_int masked_exceptions;
535 masked_exceptions = curpcb->pcb_save.sv_87.sv_env.en_cw
536 & curpcb->pcb_save.sv_87.sv_env.en_sw & 0x7f;
538 * Log exceptions that would have trapped with the old
539 * control word (overflow, divide by 0, and invalid operand).
541 if (masked_exceptions & 0x0d)
543 "pid %d (%s) exited with masked floating point exceptions 0x%02x\n",
544 p->p_pid, p->p_comm, masked_exceptions);
550 * The following mechanism is used to ensure that the FPE_... value
551 * that is passed as a trapcode to the signal handler of the user
552 * process does not have more than one bit set.
554 * Multiple bits may be set if the user process modifies the control
555 * word while a status word bit is already set. While this is a sign
556 * of bad coding, we have no choise than to narrow them down to one
557 * bit, since we must not send a trapcode that is not exactly one of
560 * The mechanism has a static table with 127 entries. Each combination
561 * of the 7 FPU status word exception bits directly translates to a
562 * position in this table, where a single FPE_... value is stored.
563 * This FPE_... value stored there is considered the "most important"
564 * of the exception bits and will be sent as the signal code. The
565 * precedence of the bits is based upon Intel Document "Numerical
566 * Applications", Chapter "Special Computational Situations".
568 * The macro to choose one of these values does these steps: 1) Throw
569 * away status word bits that cannot be masked. 2) Throw away the bits
570 * currently masked in the control word, assuming the user isn't
571 * interested in them anymore. 3) Reinsert status word bit 7 (stack
572 * fault) if it is set, which cannot be masked but must be presered.
573 * 4) Use the remaining bits to point into the trapcode table.
575 * The 6 maskable bits in order of their preference, as stated in the
576 * above referenced Intel manual:
577 * 1 Invalid operation (FP_X_INV)
580 * 1c Operand of unsupported format
582 * 2 QNaN operand (not an exception, irrelavant here)
583 * 3 Any other invalid-operation not mentioned above or zero divide
584 * (FP_X_INV, FP_X_DZ)
585 * 4 Denormal operand (FP_X_DNML)
586 * 5 Numeric over/underflow (FP_X_OFL, FP_X_UFL)
587 * 6 Inexact result (FP_X_IMP)
589 static char fpetable[128] = {
591 FPE_FLTINV, /* 1 - INV */
592 FPE_FLTUND, /* 2 - DNML */
593 FPE_FLTINV, /* 3 - INV | DNML */
594 FPE_FLTDIV, /* 4 - DZ */
595 FPE_FLTINV, /* 5 - INV | DZ */
596 FPE_FLTDIV, /* 6 - DNML | DZ */
597 FPE_FLTINV, /* 7 - INV | DNML | DZ */
598 FPE_FLTOVF, /* 8 - OFL */
599 FPE_FLTINV, /* 9 - INV | OFL */
600 FPE_FLTUND, /* A - DNML | OFL */
601 FPE_FLTINV, /* B - INV | DNML | OFL */
602 FPE_FLTDIV, /* C - DZ | OFL */
603 FPE_FLTINV, /* D - INV | DZ | OFL */
604 FPE_FLTDIV, /* E - DNML | DZ | OFL */
605 FPE_FLTINV, /* F - INV | DNML | DZ | OFL */
606 FPE_FLTUND, /* 10 - UFL */
607 FPE_FLTINV, /* 11 - INV | UFL */
608 FPE_FLTUND, /* 12 - DNML | UFL */
609 FPE_FLTINV, /* 13 - INV | DNML | UFL */
610 FPE_FLTDIV, /* 14 - DZ | UFL */
611 FPE_FLTINV, /* 15 - INV | DZ | UFL */
612 FPE_FLTDIV, /* 16 - DNML | DZ | UFL */
613 FPE_FLTINV, /* 17 - INV | DNML | DZ | UFL */
614 FPE_FLTOVF, /* 18 - OFL | UFL */
615 FPE_FLTINV, /* 19 - INV | OFL | UFL */
616 FPE_FLTUND, /* 1A - DNML | OFL | UFL */
617 FPE_FLTINV, /* 1B - INV | DNML | OFL | UFL */
618 FPE_FLTDIV, /* 1C - DZ | OFL | UFL */
619 FPE_FLTINV, /* 1D - INV | DZ | OFL | UFL */
620 FPE_FLTDIV, /* 1E - DNML | DZ | OFL | UFL */
621 FPE_FLTINV, /* 1F - INV | DNML | DZ | OFL | UFL */
622 FPE_FLTRES, /* 20 - IMP */
623 FPE_FLTINV, /* 21 - INV | IMP */
624 FPE_FLTUND, /* 22 - DNML | IMP */
625 FPE_FLTINV, /* 23 - INV | DNML | IMP */
626 FPE_FLTDIV, /* 24 - DZ | IMP */
627 FPE_FLTINV, /* 25 - INV | DZ | IMP */
628 FPE_FLTDIV, /* 26 - DNML | DZ | IMP */
629 FPE_FLTINV, /* 27 - INV | DNML | DZ | IMP */
630 FPE_FLTOVF, /* 28 - OFL | IMP */
631 FPE_FLTINV, /* 29 - INV | OFL | IMP */
632 FPE_FLTUND, /* 2A - DNML | OFL | IMP */
633 FPE_FLTINV, /* 2B - INV | DNML | OFL | IMP */
634 FPE_FLTDIV, /* 2C - DZ | OFL | IMP */
635 FPE_FLTINV, /* 2D - INV | DZ | OFL | IMP */
636 FPE_FLTDIV, /* 2E - DNML | DZ | OFL | IMP */
637 FPE_FLTINV, /* 2F - INV | DNML | DZ | OFL | IMP */
638 FPE_FLTUND, /* 30 - UFL | IMP */
639 FPE_FLTINV, /* 31 - INV | UFL | IMP */
640 FPE_FLTUND, /* 32 - DNML | UFL | IMP */
641 FPE_FLTINV, /* 33 - INV | DNML | UFL | IMP */
642 FPE_FLTDIV, /* 34 - DZ | UFL | IMP */
643 FPE_FLTINV, /* 35 - INV | DZ | UFL | IMP */
644 FPE_FLTDIV, /* 36 - DNML | DZ | UFL | IMP */
645 FPE_FLTINV, /* 37 - INV | DNML | DZ | UFL | IMP */
646 FPE_FLTOVF, /* 38 - OFL | UFL | IMP */
647 FPE_FLTINV, /* 39 - INV | OFL | UFL | IMP */
648 FPE_FLTUND, /* 3A - DNML | OFL | UFL | IMP */
649 FPE_FLTINV, /* 3B - INV | DNML | OFL | UFL | IMP */
650 FPE_FLTDIV, /* 3C - DZ | OFL | UFL | IMP */
651 FPE_FLTINV, /* 3D - INV | DZ | OFL | UFL | IMP */
652 FPE_FLTDIV, /* 3E - DNML | DZ | OFL | UFL | IMP */
653 FPE_FLTINV, /* 3F - INV | DNML | DZ | OFL | UFL | IMP */
654 FPE_FLTSUB, /* 40 - STK */
655 FPE_FLTSUB, /* 41 - INV | STK */
656 FPE_FLTUND, /* 42 - DNML | STK */
657 FPE_FLTSUB, /* 43 - INV | DNML | STK */
658 FPE_FLTDIV, /* 44 - DZ | STK */
659 FPE_FLTSUB, /* 45 - INV | DZ | STK */
660 FPE_FLTDIV, /* 46 - DNML | DZ | STK */
661 FPE_FLTSUB, /* 47 - INV | DNML | DZ | STK */
662 FPE_FLTOVF, /* 48 - OFL | STK */
663 FPE_FLTSUB, /* 49 - INV | OFL | STK */
664 FPE_FLTUND, /* 4A - DNML | OFL | STK */
665 FPE_FLTSUB, /* 4B - INV | DNML | OFL | STK */
666 FPE_FLTDIV, /* 4C - DZ | OFL | STK */
667 FPE_FLTSUB, /* 4D - INV | DZ | OFL | STK */
668 FPE_FLTDIV, /* 4E - DNML | DZ | OFL | STK */
669 FPE_FLTSUB, /* 4F - INV | DNML | DZ | OFL | STK */
670 FPE_FLTUND, /* 50 - UFL | STK */
671 FPE_FLTSUB, /* 51 - INV | UFL | STK */
672 FPE_FLTUND, /* 52 - DNML | UFL | STK */
673 FPE_FLTSUB, /* 53 - INV | DNML | UFL | STK */
674 FPE_FLTDIV, /* 54 - DZ | UFL | STK */
675 FPE_FLTSUB, /* 55 - INV | DZ | UFL | STK */
676 FPE_FLTDIV, /* 56 - DNML | DZ | UFL | STK */
677 FPE_FLTSUB, /* 57 - INV | DNML | DZ | UFL | STK */
678 FPE_FLTOVF, /* 58 - OFL | UFL | STK */
679 FPE_FLTSUB, /* 59 - INV | OFL | UFL | STK */
680 FPE_FLTUND, /* 5A - DNML | OFL | UFL | STK */
681 FPE_FLTSUB, /* 5B - INV | DNML | OFL | UFL | STK */
682 FPE_FLTDIV, /* 5C - DZ | OFL | UFL | STK */
683 FPE_FLTSUB, /* 5D - INV | DZ | OFL | UFL | STK */
684 FPE_FLTDIV, /* 5E - DNML | DZ | OFL | UFL | STK */
685 FPE_FLTSUB, /* 5F - INV | DNML | DZ | OFL | UFL | STK */
686 FPE_FLTRES, /* 60 - IMP | STK */
687 FPE_FLTSUB, /* 61 - INV | IMP | STK */
688 FPE_FLTUND, /* 62 - DNML | IMP | STK */
689 FPE_FLTSUB, /* 63 - INV | DNML | IMP | STK */
690 FPE_FLTDIV, /* 64 - DZ | IMP | STK */
691 FPE_FLTSUB, /* 65 - INV | DZ | IMP | STK */
692 FPE_FLTDIV, /* 66 - DNML | DZ | IMP | STK */
693 FPE_FLTSUB, /* 67 - INV | DNML | DZ | IMP | STK */
694 FPE_FLTOVF, /* 68 - OFL | IMP | STK */
695 FPE_FLTSUB, /* 69 - INV | OFL | IMP | STK */
696 FPE_FLTUND, /* 6A - DNML | OFL | IMP | STK */
697 FPE_FLTSUB, /* 6B - INV | DNML | OFL | IMP | STK */
698 FPE_FLTDIV, /* 6C - DZ | OFL | IMP | STK */
699 FPE_FLTSUB, /* 6D - INV | DZ | OFL | IMP | STK */
700 FPE_FLTDIV, /* 6E - DNML | DZ | OFL | IMP | STK */
701 FPE_FLTSUB, /* 6F - INV | DNML | DZ | OFL | IMP | STK */
702 FPE_FLTUND, /* 70 - UFL | IMP | STK */
703 FPE_FLTSUB, /* 71 - INV | UFL | IMP | STK */
704 FPE_FLTUND, /* 72 - DNML | UFL | IMP | STK */
705 FPE_FLTSUB, /* 73 - INV | DNML | UFL | IMP | STK */
706 FPE_FLTDIV, /* 74 - DZ | UFL | IMP | STK */
707 FPE_FLTSUB, /* 75 - INV | DZ | UFL | IMP | STK */
708 FPE_FLTDIV, /* 76 - DNML | DZ | UFL | IMP | STK */
709 FPE_FLTSUB, /* 77 - INV | DNML | DZ | UFL | IMP | STK */
710 FPE_FLTOVF, /* 78 - OFL | UFL | IMP | STK */
711 FPE_FLTSUB, /* 79 - INV | OFL | UFL | IMP | STK */
712 FPE_FLTUND, /* 7A - DNML | OFL | UFL | IMP | STK */
713 FPE_FLTSUB, /* 7B - INV | DNML | OFL | UFL | IMP | STK */
714 FPE_FLTDIV, /* 7C - DZ | OFL | UFL | IMP | STK */
715 FPE_FLTSUB, /* 7D - INV | DZ | OFL | UFL | IMP | STK */
716 FPE_FLTDIV, /* 7E - DNML | DZ | OFL | UFL | IMP | STK */
717 FPE_FLTSUB, /* 7F - INV | DNML | DZ | OFL | UFL | IMP | STK */
721 * Preserve the FP status word, clear FP exceptions, then generate a SIGFPE.
723 * Clearing exceptions is necessary mainly to avoid IRQ13 bugs. We now
724 * depend on longjmp() restoring a usable state. Restoring the state
725 * or examining it might fail if we didn't clear exceptions.
727 * The error code chosen will be one of the FPE_... macros. It will be
728 * sent as the second argument to old BSD-style signal handlers and as
729 * "siginfo_t->si_code" (second argument) to SA_SIGINFO signal handlers.
731 * XXX the FP state is not preserved across signal handlers. So signal
732 * handlers cannot afford to do FP unless they preserve the state or
733 * longjmp() out. Both preserving the state and longjmp()ing may be
734 * destroyed by IRQ13 bugs. Clearing FP exceptions is not an acceptable
735 * solution for signals other than SIGFPE.
743 struct intrframe *frame;
746 if (npxthread == NULL || !npx_exists) {
747 printf("npxintr: npxthread = %p, curthread = %p, npx_exists = %d\n",
748 npxthread, curthread, npx_exists);
749 panic("npxintr from nowhere");
751 if (npxthread != curthread) {
752 printf("npxintr: npxthread = %p, curthread = %p, npx_exists = %d\n",
753 npxthread, curthread, npx_exists);
754 panic("npxintr from non-current process");
757 exstat = GET_FPU_EXSW_PTR(curpcb);
764 * Pass exception to process.
766 frame = (struct intrframe *)&dummy; /* XXX */
767 if ((ISPL(frame->if_cs) == SEL_UPL) || (frame->if_eflags & PSL_VM)) {
769 * Interrupt is essentially a trap, so we can afford to call
770 * the SIGFPE handler (if any) as soon as the interrupt
773 * XXX little or nothing is gained from this, and plenty is
774 * lost - the interrupt frame has to contain the trap frame
775 * (this is otherwise only necessary for the rescheduling trap
776 * in doreti, and the frame for that could easily be set up
777 * just before it is used).
779 curproc->p_md.md_regs = INTR_TO_TRAPFRAME(frame);
781 * Encode the appropriate code for detailed information on
785 fpetable[(*exstat & ~control & 0x3f) | (*exstat & 0x40)];
786 trapsignal(curproc, SIGFPE, code);
789 * Nested interrupt. These losers occur when:
790 * o an IRQ13 is bogusly generated at a bogus time, e.g.:
791 * o immediately after an fnsave or frstor of an
793 * o a couple of 386 instructions after
794 * "fstpl _memvar" causes a stack overflow.
795 * These are especially nasty when combined with a
797 * o an IRQ13 occurs at the same time as another higher-
798 * priority interrupt.
800 * Treat them like a true async interrupt.
802 psignal(curproc, SIGFPE);
807 * Implement device not available (DNA) exception
809 * It would be better to switch FP context here (if curthread != npxthread)
810 * and not necessarily for every context switch, but it is too hard to
811 * access foreign pcb's.
820 if (npxthread != NULL) {
821 printf("npxdna: npxthread = %p, curthread = %p\n",
822 npxthread, curthread);
827 * Record new context early in case frstor causes an IRQ13.
829 npxthread = curthread;
830 exstat = GET_FPU_EXSW_PTR(curpcb);
833 * The following frstor may cause an IRQ13 when the state being
834 * restored has a pending error. The error will appear to have been
835 * triggered by the current (npx) user instruction even when that
836 * instruction is a no-wait instruction that should not trigger an
837 * error (e.g., fnclex). On at least one 486 system all of the
838 * no-wait instructions are broken the same as frstor, so our
839 * treatment does not amplify the breakage. On at least one
840 * 386/Cyrix 387 system, fnclex works correctly while frstor and
841 * fnsave are broken, so our treatment breaks fnclex if it is the
842 * first FPU instruction after a context switch.
844 fpurstor(&curpcb->pcb_save);
850 * Wrapper for fnsave instruction to handle h/w bugs. If there is an error
851 * pending, then fnsave generates a bogus IRQ13 on some systems. Force
852 * any IRQ13 to be handled immediately, and then ignore it. This routine is
853 * often called at splhigh so it must not use many system services. In
854 * particular, it's much easier to install a special handler than to
855 * guarantee that it's safe to use npxintr() and its supporting code.
861 #if defined(SMP) || defined(CPU_ENABLE_SSE)
870 #else /* SMP or CPU_ENABLE_SSE */
874 u_char old_icu1_mask;
875 u_char old_icu2_mask;
876 struct gate_descriptor save_idt_npxintr;
879 old_icu1_mask = inb(IO_ICU1 + 1);
880 old_icu2_mask = inb(IO_ICU2 + 1);
881 save_idt_npxintr = idt[npx_intrno];
882 outb(IO_ICU1 + 1, old_icu1_mask & ~(IRQ_SLAVE | npx0_imask));
883 outb(IO_ICU2 + 1, old_icu2_mask & ~(npx0_imask >> 8));
884 idt[npx_intrno] = npx_idt_probeintr;
892 icu1_mask = inb(IO_ICU1 + 1); /* masks may have changed */
893 icu2_mask = inb(IO_ICU2 + 1);
895 (icu1_mask & ~npx0_imask) | (old_icu1_mask & npx0_imask));
897 (icu2_mask & ~(npx0_imask >> 8))
898 | (old_icu2_mask & (npx0_imask >> 8)));
899 idt[npx_intrno] = save_idt_npxintr;
900 enable_intr(); /* back to usual state */
910 #ifdef CPU_ENABLE_SSE
923 #ifdef CPU_ENABLE_SSE
933 timezero(funcname, func)
934 const char *funcname;
935 void (*func) __P((void *buf, size_t len));
939 #define BUFSIZE 1000000
941 struct timeval finish, start;
943 buf = malloc(BUFSIZE, M_TEMP, M_NOWAIT);
947 (*func)(buf, BUFSIZE);
949 usec = 1000000 * (finish.tv_sec - start.tv_sec) +
950 finish.tv_usec - start.tv_usec;
954 printf("%s bandwidth = %ld bytes/sec\n",
955 funcname, (long)(BUFSIZE * (int64_t)1000000 / usec));
959 #endif /* I586_CPU */
961 static device_method_t npx_methods[] = {
962 /* Device interface */
963 DEVMETHOD(device_identify, npx_identify),
964 DEVMETHOD(device_probe, npx_probe),
965 DEVMETHOD(device_attach, npx_attach),
966 DEVMETHOD(device_detach, bus_generic_detach),
967 DEVMETHOD(device_shutdown, bus_generic_shutdown),
968 DEVMETHOD(device_suspend, bus_generic_suspend),
969 DEVMETHOD(device_resume, bus_generic_resume),
974 static driver_t npx_driver = {
980 static devclass_t npx_devclass;
983 * We prefer to attach to the root nexus so that the usual case (exception 16)
984 * doesn't describe the processor as being `on isa'.
986 DRIVER_MODULE(npx, nexus, npx_driver, npx_devclass, 0, 0);