xref: /freebsd/sys/i386/i386/vm_machdep.c (revision 5ebc7e6281887681c3a348a5a4c902e262ccd656)
1 /*-
2  * Copyright (c) 1982, 1986 The Regents of the University of California.
3  * Copyright (c) 1989, 1990 William Jolitz
4  * Copyright (c) 1994 John Dyson
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * the Systems Programming Group of the University of Utah Computer
9  * Science Department, and William Jolitz.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the University of
22  *	California, Berkeley and its contributors.
23  * 4. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  *
39  *	from: @(#)vm_machdep.c	7.3 (Berkeley) 5/13/91
40  *	Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
41  *	$Id: vm_machdep.c,v 1.37 1995/05/01 23:32:30 dyson Exp $
42  */
43 
44 #include "npx.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/proc.h>
48 #include <sys/malloc.h>
49 #include <sys/buf.h>
50 #include <sys/vnode.h>
51 #include <sys/user.h>
52 
53 #include <machine/clock.h>
54 #include <machine/cpu.h>
55 #include <machine/md_var.h>
56 
57 #include <vm/vm.h>
58 #include <vm/vm_kern.h>
59 
60 #include <i386/isa/isa.h>
61 
62 #ifdef BOUNCE_BUFFERS
63 vm_map_t	io_map;
64 volatile int	kvasfreecnt;
65 
66 
67 caddr_t		bouncememory;
68 int		bouncepages, bpwait;
69 vm_offset_t	*bouncepa;
70 int		bmwait, bmfreeing;
71 
72 #define BITS_IN_UNSIGNED (8*sizeof(unsigned))
73 int		bounceallocarraysize;
74 unsigned	*bounceallocarray;
75 int		bouncefree;
76 
77 #define SIXTEENMEG (4096*4096)
78 #define MAXBKVA 1024
79 int		maxbkva = MAXBKVA*NBPG;
80 
81 /* special list that can be used at interrupt time for eventual kva free */
82 struct kvasfree {
83 	vm_offset_t addr;
84 	vm_offset_t size;
85 } kvaf[MAXBKVA];
86 
87 
88 vm_offset_t vm_bounce_kva();
89 /*
90  * get bounce buffer pages (count physically contiguous)
91  * (only 1 inplemented now)
92  */
93 vm_offset_t
94 vm_bounce_page_find(count)
95 	int count;
96 {
97 	int bit;
98 	int s,i;
99 
100 	if (count != 1)
101 		panic("vm_bounce_page_find -- no support for > 1 page yet!!!");
102 
103 	s = splbio();
104 retry:
105 	for (i = 0; i < bounceallocarraysize; i++) {
106 		if (bounceallocarray[i] != 0xffffffff) {
107 			bit = ffs(~bounceallocarray[i]);
108 			if (bit) {
109 				bounceallocarray[i] |= 1 << (bit - 1) ;
110 				bouncefree -= count;
111 				splx(s);
112 				return bouncepa[(i * BITS_IN_UNSIGNED + (bit - 1))];
113 			}
114 		}
115 	}
116 	bpwait = 1;
117 	tsleep((caddr_t) &bounceallocarray, PRIBIO, "bncwai", 0);
118 	goto retry;
119 }
120 
121 void
122 vm_bounce_kva_free(addr, size, now)
123 	vm_offset_t addr;
124 	vm_offset_t size;
125 	int now;
126 {
127 	int s = splbio();
128 	kvaf[kvasfreecnt].addr = addr;
129 	kvaf[kvasfreecnt].size = size;
130 	++kvasfreecnt;
131 	if( now) {
132 		/*
133 		 * this will do wakeups
134 		 */
135 		vm_bounce_kva(0,0);
136 	} else {
137 		if (bmwait) {
138 		/*
139 		 * if anyone is waiting on the bounce-map, then wakeup
140 		 */
141 			wakeup((caddr_t) io_map);
142 			bmwait = 0;
143 		}
144 	}
145 	splx(s);
146 }
147 
148 /*
149  * free count bounce buffer pages
150  */
151 void
152 vm_bounce_page_free(pa, count)
153 	vm_offset_t pa;
154 	int count;
155 {
156 	int allocindex;
157 	int index;
158 	int bit;
159 
160 	if (count != 1)
161 		panic("vm_bounce_page_free -- no support for > 1 page yet!!!");
162 
163 	for(index=0;index<bouncepages;index++) {
164 		if( pa == bouncepa[index])
165 			break;
166 	}
167 
168 	if( index == bouncepages)
169 		panic("vm_bounce_page_free: invalid bounce buffer");
170 
171 	allocindex = index / BITS_IN_UNSIGNED;
172 	bit = index % BITS_IN_UNSIGNED;
173 
174 	bounceallocarray[allocindex] &= ~(1 << bit);
175 
176 	bouncefree += count;
177 	if (bpwait) {
178 		bpwait = 0;
179 		wakeup((caddr_t) &bounceallocarray);
180 	}
181 }
182 
183 /*
184  * allocate count bounce buffer kva pages
185  */
186 vm_offset_t
187 vm_bounce_kva(size, waitok)
188 	int size;
189 	int waitok;
190 {
191 	int i;
192 	vm_offset_t kva = 0;
193 	vm_offset_t off;
194 	int s = splbio();
195 more:
196 	if (!bmfreeing && kvasfreecnt) {
197 		bmfreeing = 1;
198 		for (i = 0; i < kvasfreecnt; i++) {
199 			for(off=0;off<kvaf[i].size;off+=NBPG) {
200 				pmap_kremove( kvaf[i].addr + off);
201 			}
202 			kmem_free_wakeup(io_map, kvaf[i].addr,
203 				kvaf[i].size);
204 		}
205 		kvasfreecnt = 0;
206 		bmfreeing = 0;
207 		if( bmwait) {
208 			bmwait = 0;
209 			wakeup( (caddr_t) io_map);
210 		}
211 	}
212 
213 	if( size == 0) {
214 		splx(s);
215 		return NULL;
216 	}
217 
218 	if ((kva = kmem_alloc_pageable(io_map, size)) == 0) {
219 		if( !waitok) {
220 			splx(s);
221 			return NULL;
222 		}
223 		bmwait = 1;
224 		tsleep((caddr_t) io_map, PRIBIO, "bmwait", 0);
225 		goto more;
226 	}
227 	splx(s);
228 	return kva;
229 }
230 
231 /*
232  * same as vm_bounce_kva -- but really allocate (but takes pages as arg)
233  */
234 vm_offset_t
235 vm_bounce_kva_alloc(count)
236 int count;
237 {
238 	int i;
239 	vm_offset_t kva;
240 	vm_offset_t pa;
241 	if( bouncepages == 0) {
242 		kva = (vm_offset_t) malloc(count*NBPG, M_TEMP, M_WAITOK);
243 		return kva;
244 	}
245 	kva = vm_bounce_kva(count*NBPG, 1);
246 	for(i=0;i<count;i++) {
247 		pa = vm_bounce_page_find(1);
248 		pmap_kenter(kva + i * NBPG, pa);
249 	}
250 	return kva;
251 }
252 
253 /*
254  * same as vm_bounce_kva_free -- but really free
255  */
256 void
257 vm_bounce_kva_alloc_free(kva, count)
258 	vm_offset_t kva;
259 	int count;
260 {
261 	int i;
262 	vm_offset_t pa;
263 	if( bouncepages == 0) {
264 		free((caddr_t) kva, M_TEMP);
265 		return;
266 	}
267 	for(i = 0; i < count; i++) {
268 		pa = pmap_kextract(kva + i * NBPG);
269 		vm_bounce_page_free(pa, 1);
270 	}
271 	vm_bounce_kva_free(kva, count*NBPG, 0);
272 }
273 
274 /*
275  * do the things necessary to the struct buf to implement
276  * bounce buffers...  inserted before the disk sort
277  */
278 void
279 vm_bounce_alloc(bp)
280 	struct buf *bp;
281 {
282 	int countvmpg;
283 	vm_offset_t vastart, vaend;
284 	vm_offset_t vapstart, vapend;
285 	vm_offset_t va, kva;
286 	vm_offset_t pa;
287 	int dobounceflag = 0;
288 	int i;
289 
290 	if (bouncepages == 0)
291 		return;
292 
293 	if (bp->b_flags & B_BOUNCE) {
294 		printf("vm_bounce_alloc: called recursively???\n");
295 		return;
296 	}
297 
298 	if (bp->b_bufsize < bp->b_bcount) {
299 		printf(
300 		    "vm_bounce_alloc: b_bufsize(0x%lx) < b_bcount(0x%lx) !!\n",
301 			bp->b_bufsize, bp->b_bcount);
302 		panic("vm_bounce_alloc");
303 	}
304 
305 /*
306  *  This is not really necessary
307  *	if( bp->b_bufsize != bp->b_bcount) {
308  *		printf("size: %d, count: %d\n", bp->b_bufsize, bp->b_bcount);
309  *	}
310  */
311 
312 
313 	vastart = (vm_offset_t) bp->b_data;
314 	vaend = (vm_offset_t) bp->b_data + bp->b_bufsize;
315 
316 	vapstart = i386_trunc_page(vastart);
317 	vapend = i386_round_page(vaend);
318 	countvmpg = (vapend - vapstart) / NBPG;
319 
320 /*
321  * if any page is above 16MB, then go into bounce-buffer mode
322  */
323 	va = vapstart;
324 	for (i = 0; i < countvmpg; i++) {
325 		pa = pmap_kextract(va);
326 		if (pa >= SIXTEENMEG)
327 			++dobounceflag;
328 		if( pa == 0)
329 			panic("vm_bounce_alloc: Unmapped page");
330 		va += NBPG;
331 	}
332 	if (dobounceflag == 0)
333 		return;
334 
335 	if (bouncepages < dobounceflag)
336 		panic("Not enough bounce buffers!!!");
337 
338 /*
339  * allocate a replacement kva for b_addr
340  */
341 	kva = vm_bounce_kva(countvmpg*NBPG, 1);
342 #if 0
343 	printf("%s: vapstart: %x, vapend: %x, countvmpg: %d, kva: %x ",
344 		(bp->b_flags & B_READ) ? "read":"write",
345 			vapstart, vapend, countvmpg, kva);
346 #endif
347 	va = vapstart;
348 	for (i = 0; i < countvmpg; i++) {
349 		pa = pmap_kextract(va);
350 		if (pa >= SIXTEENMEG) {
351 			/*
352 			 * allocate a replacement page
353 			 */
354 			vm_offset_t bpa = vm_bounce_page_find(1);
355 			pmap_kenter(kva + (NBPG * i), bpa);
356 #if 0
357 			printf("r(%d): (%x,%x,%x) ", i, va, pa, bpa);
358 #endif
359 			/*
360 			 * if we are writing, the copy the data into the page
361 			 */
362 			if ((bp->b_flags & B_READ) == 0) {
363 				bcopy((caddr_t) va, (caddr_t) kva + (NBPG * i), NBPG);
364 			}
365 		} else {
366 			/*
367 			 * use original page
368 			 */
369 			pmap_kenter(kva + (NBPG * i), pa);
370 		}
371 		va += NBPG;
372 	}
373 
374 /*
375  * flag the buffer as being bounced
376  */
377 	bp->b_flags |= B_BOUNCE;
378 /*
379  * save the original buffer kva
380  */
381 	bp->b_savekva = bp->b_data;
382 /*
383  * put our new kva into the buffer (offset by original offset)
384  */
385 	bp->b_data = (caddr_t) (((vm_offset_t) kva) |
386 				((vm_offset_t) bp->b_savekva & (NBPG - 1)));
387 #if 0
388 	printf("b_savekva: %x, newva: %x\n", bp->b_savekva, bp->b_data);
389 #endif
390 	return;
391 }
392 
393 /*
394  * hook into biodone to free bounce buffer
395  */
396 void
397 vm_bounce_free(bp)
398 	struct buf *bp;
399 {
400 	int i;
401 	vm_offset_t origkva, bouncekva, bouncekvaend;
402 
403 /*
404  * if this isn't a bounced buffer, then just return
405  */
406 	if ((bp->b_flags & B_BOUNCE) == 0)
407 		return;
408 
409 /*
410  *  This check is not necessary
411  *	if (bp->b_bufsize != bp->b_bcount) {
412  *		printf("vm_bounce_free: b_bufsize=%d, b_bcount=%d\n",
413  *			bp->b_bufsize, bp->b_bcount);
414  *	}
415  */
416 
417 	origkva = (vm_offset_t) bp->b_savekva;
418 	bouncekva = (vm_offset_t) bp->b_data;
419 /*
420 	printf("free: %d ", bp->b_bufsize);
421 */
422 
423 /*
424  * check every page in the kva space for b_addr
425  */
426 	for (i = 0; i < bp->b_bufsize; ) {
427 		vm_offset_t mybouncepa;
428 		vm_offset_t copycount;
429 
430 		copycount = i386_round_page(bouncekva + 1) - bouncekva;
431 		mybouncepa = pmap_kextract(i386_trunc_page(bouncekva));
432 
433 /*
434  * if this is a bounced pa, then process as one
435  */
436 		if ( mybouncepa != pmap_kextract( i386_trunc_page( origkva))) {
437 			vm_offset_t tocopy = copycount;
438 			if (i + tocopy > bp->b_bufsize)
439 				tocopy = bp->b_bufsize - i;
440 /*
441  * if this is a read, then copy from bounce buffer into original buffer
442  */
443 			if (bp->b_flags & B_READ)
444 				bcopy((caddr_t) bouncekva, (caddr_t) origkva, tocopy);
445 /*
446  * free the bounce allocation
447  */
448 
449 /*
450 			printf("(kva: %x, pa: %x)", bouncekva, mybouncepa);
451 */
452 			vm_bounce_page_free(mybouncepa, 1);
453 		}
454 
455 		origkva += copycount;
456 		bouncekva += copycount;
457 		i += copycount;
458 	}
459 
460 /*
461 	printf("\n");
462 */
463 /*
464  * add the old kva into the "to free" list
465  */
466 
467 	bouncekva= i386_trunc_page((vm_offset_t) bp->b_data);
468 	bouncekvaend= i386_round_page((vm_offset_t)bp->b_data + bp->b_bufsize);
469 
470 /*
471 	printf("freeva: %d\n", (bouncekvaend - bouncekva) / NBPG);
472 */
473 	vm_bounce_kva_free( bouncekva, (bouncekvaend - bouncekva), 0);
474 	bp->b_data = bp->b_savekva;
475 	bp->b_savekva = 0;
476 	bp->b_flags &= ~B_BOUNCE;
477 
478 	return;
479 }
480 
481 
482 /*
483  * init the bounce buffer system
484  */
485 void
486 vm_bounce_init()
487 {
488 	int i;
489 
490 	kvasfreecnt = 0;
491 
492 	if (bouncepages == 0)
493 		return;
494 
495 	bounceallocarraysize = (bouncepages + BITS_IN_UNSIGNED - 1) / BITS_IN_UNSIGNED;
496 	bounceallocarray = malloc(bounceallocarraysize * sizeof(unsigned), M_TEMP, M_NOWAIT);
497 
498 	if (!bounceallocarray)
499 		panic("Cannot allocate bounce resource array");
500 
501 	bouncepa = malloc(bouncepages * sizeof(vm_offset_t), M_TEMP, M_NOWAIT);
502 	if (!bouncepa)
503 		panic("Cannot allocate physical memory array");
504 
505 	for(i=0;i<bounceallocarraysize;i++) {
506 		bounceallocarray[i] = 0xffffffff;
507 	}
508 
509 	for(i=0;i<bouncepages;i++) {
510 		vm_offset_t pa;
511 		if( (pa = pmap_kextract((vm_offset_t) bouncememory + i * NBPG)) >= SIXTEENMEG)
512 			panic("bounce memory out of range");
513 		if( pa == 0)
514 			panic("bounce memory not resident");
515 		bouncepa[i] = pa;
516 		bounceallocarray[i/(8*sizeof(int))] &= ~(1<<(i%(8*sizeof(int))));
517 	}
518 	bouncefree = bouncepages;
519 
520 }
521 #endif /* BOUNCE_BUFFERS */
522 /*
523  * quick version of vm_fault
524  */
525 
526 void
527 vm_fault_quick( v, prot)
528 	vm_offset_t v;
529 	int prot;
530 {
531 	if (prot & VM_PROT_WRITE)
532 		subyte((char *)v, fubyte((char *)v));
533 	else
534 		(void) fubyte((char *)v);
535 }
536 
537 
538 /*
539  * Finish a fork operation, with process p2 nearly set up.
540  * Copy and update the kernel stack and pcb, making the child
541  * ready to run, and marking it so that it can return differently
542  * than the parent.  Returns 1 in the child process, 0 in the parent.
543  * We currently double-map the user area so that the stack is at the same
544  * address in each process; in the future we will probably relocate
545  * the frame pointers on the stack after copying.
546  */
547 int
548 cpu_fork(p1, p2)
549 	register struct proc *p1, *p2;
550 {
551 	register struct user *up = p2->p_addr;
552 	int offset;
553 
554 	/*
555 	 * Copy pcb and stack from proc p1 to p2.
556 	 * We do this as cheaply as possible, copying only the active
557 	 * part of the stack.  The stack and pcb need to agree;
558 	 * this is tricky, as the final pcb is constructed by savectx,
559 	 * but its frame isn't yet on the stack when the stack is copied.
560 	 * swtch compensates for this when the child eventually runs.
561 	 * This should be done differently, with a single call
562 	 * that copies and updates the pcb+stack,
563 	 * replacing the bcopy and savectx.
564 	 */
565 	p2->p_addr->u_pcb = p1->p_addr->u_pcb;
566 	offset = mvesp() - (int)kstack;
567 	bcopy((caddr_t)kstack + offset, (caddr_t)p2->p_addr + offset,
568 	    (unsigned) ctob(UPAGES) - offset);
569 	p2->p_md.md_regs = p1->p_md.md_regs;
570 
571 	pmap_activate(&p2->p_vmspace->vm_pmap, &up->u_pcb);
572 
573 	/*
574 	 *
575 	 * Arrange for a non-local goto when the new process
576 	 * is started, to resume here, returning nonzero from setjmp.
577 	 */
578 	if (savectx(&up->u_pcb, 1)) {
579 		/*
580 		 * Return 1 in child.
581 		 */
582 		return (1);
583 	}
584 	return (0);
585 }
586 
587 void
588 cpu_exit(p)
589 	register struct proc *p;
590 {
591 
592 #if NNPX > 0
593 	npxexit(p);
594 #endif	/* NNPX */
595 	cnt.v_swtch++;
596 	cpu_switch(p);
597 	panic("cpu_exit");
598 }
599 
600 void
601 cpu_wait(p) struct proc *p; {
602 /*	extern vm_map_t upages_map; */
603 
604 	/* drop per-process resources */
605  	pmap_remove(vm_map_pmap(u_map), (vm_offset_t) p->p_addr,
606 		((vm_offset_t) p->p_addr) + ctob(UPAGES));
607 	kmem_free(u_map, (vm_offset_t)p->p_addr, ctob(UPAGES));
608 	vmspace_free(p->p_vmspace);
609 }
610 
611 /*
612  * Dump the machine specific header information at the start of a core dump.
613  */
614 int
615 cpu_coredump(p, vp, cred)
616 	struct proc *p;
617 	struct vnode *vp;
618 	struct ucred *cred;
619 {
620 
621 	return (vn_rdwr(UIO_WRITE, vp, (caddr_t) p->p_addr, ctob(UPAGES),
622 	    (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, cred, (int *)NULL,
623 	    p));
624 }
625 
626 /*
627  * Set a red zone in the kernel stack after the u. area.
628  */
629 void
630 setredzone(pte, vaddr)
631 	u_short *pte;
632 	caddr_t vaddr;
633 {
634 /* eventually do this by setting up an expand-down stack segment
635    for ss0: selector, allowing stack access down to top of u.
636    this means though that protection violations need to be handled
637    thru a double fault exception that must do an integral task
638    switch to a known good context, within which a dump can be
639    taken. a sensible scheme might be to save the initial context
640    used by sched (that has physical memory mapped 1:1 at bottom)
641    and take the dump while still in mapped mode */
642 }
643 
644 /*
645  * Move pages from one kernel virtual address to another.
646  * Both addresses are assumed to reside in the Sysmap,
647  * and size must be a multiple of CLSIZE.
648  */
649 
650 void
651 pagemove(from, to, size)
652 	register caddr_t from, to;
653 	int size;
654 {
655 	register vm_offset_t pa;
656 
657 	if (size & CLOFSET)
658 		panic("pagemove");
659 	while (size > 0) {
660 		pa = pmap_kextract((vm_offset_t)from);
661 		if (pa == 0)
662 			panic("pagemove 2");
663 		if (pmap_kextract((vm_offset_t)to) != 0)
664 			panic("pagemove 3");
665 		pmap_kremove((vm_offset_t)from);
666 		pmap_kenter((vm_offset_t)to, pa);
667 		from += PAGE_SIZE;
668 		to += PAGE_SIZE;
669 		size -= PAGE_SIZE;
670 	}
671 }
672 
673 /*
674  * Convert kernel VA to physical address
675  */
676 u_long
677 kvtop(void *addr)
678 {
679 	vm_offset_t va;
680 
681 	va = pmap_kextract((vm_offset_t)addr);
682 	if (va == 0)
683 		panic("kvtop: zero page frame");
684 	return((int)va);
685 }
686 
687 /*
688  * Map an IO request into kernel virtual address space.
689  *
690  * All requests are (re)mapped into kernel VA space.
691  * Notice that we use b_bufsize for the size of the buffer
692  * to be mapped.  b_bcount might be modified by the driver.
693  */
694 void
695 vmapbuf(bp)
696 	register struct buf *bp;
697 {
698 	register int npf;
699 	register caddr_t addr;
700 	int off;
701 	vm_offset_t kva;
702 	vm_offset_t pa, v;
703 
704 	if ((bp->b_flags & B_PHYS) == 0)
705 		panic("vmapbuf");
706 
707 	/*
708 	 * this is the kva that is to be used for
709 	 * the temporary kernel mapping
710 	 */
711 	kva = (vm_offset_t) bp->b_saveaddr;
712 
713 	for (addr = (caddr_t)trunc_page(bp->b_data);
714 		addr < bp->b_data + bp->b_bufsize;
715 		addr += PAGE_SIZE) {
716 
717 /*
718  * do the vm_fault if needed, do the copy-on-write thing when
719  * reading stuff off device into memory.
720  */
721 		vm_fault_quick(addr,
722 			(bp->b_flags&B_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ);
723 		pa = pmap_kextract((vm_offset_t) addr);
724 		if (pa == 0)
725 			panic("vmapbuf: page not present");
726 /*
727  * hold the data page
728  */
729 #ifdef DIAGNOSTIC
730 		if( VM_PAGE_TO_PHYS(PHYS_TO_VM_PAGE(pa)) != pa)
731 			panic("vmapbuf: confused PHYS_TO_VM_PAGE mapping");
732 #endif
733 		vm_page_hold(PHYS_TO_VM_PAGE(pa));
734 	}
735 
736 	addr = bp->b_saveaddr = bp->b_data;
737 	off = (int)addr & PGOFSET;
738 	npf = btoc(round_page(bp->b_bufsize + off));
739 	bp->b_data = (caddr_t) (kva + off);
740 	while (npf--) {
741 		pa = pmap_kextract((vm_offset_t)addr);
742 		if (pa == 0)
743 			panic("vmapbuf: null page frame");
744 		pmap_kenter(kva, trunc_page(pa));
745 		addr += PAGE_SIZE;
746 		kva += PAGE_SIZE;
747 	}
748 }
749 
750 /*
751  * Free the io map PTEs associated with this IO operation.
752  * We also invalidate the TLB entries and restore the original b_addr.
753  */
754 void
755 vunmapbuf(bp)
756 	register struct buf *bp;
757 {
758 	register caddr_t addr;
759 	vm_offset_t v,pa;
760 
761 	if ((bp->b_flags & B_PHYS) == 0)
762 		panic("vunmapbuf");
763 
764 	for (addr = (caddr_t)trunc_page((vm_offset_t) bp->b_data);
765 		addr < bp->b_data + bp->b_bufsize;
766 		addr += NBPG)
767 		pmap_kremove((vm_offset_t) addr);
768 
769 	bp->b_data = bp->b_saveaddr;
770 	bp->b_saveaddr = NULL;
771 
772 /*
773  * unhold the pde, and data pages
774  */
775 	for (addr = (caddr_t)trunc_page((vm_offset_t) bp->b_data);
776 		addr < bp->b_data + bp->b_bufsize;
777 		addr += NBPG) {
778 	/*
779 	 * release the data page
780 	 */
781 		pa = pmap_kextract((vm_offset_t) addr);
782 		vm_page_unhold(PHYS_TO_VM_PAGE(pa));
783 	}
784 }
785 
786 /*
787  * Force reset the processor by invalidating the entire address space!
788  */
789 void
790 cpu_reset() {
791 
792 	/*
793 	 * Attempt to do a CPU reset via the keyboard controller,
794 	 * do not turn of the GateA20, as any machine that fails
795 	 * to do the reset here would then end up in no man's land.
796 	 */
797 
798 #ifndef BROKEN_KEYBOARD_RESET
799 	outb(IO_KBD + 4, 0xFE);
800 	DELAY(500000);	/* wait 0.5 sec to see if that did it */
801 	printf("Keyboard reset did not work, attempting CPU shutdown\n");
802 	DELAY(1000000);	/* wait 1 sec for printf to complete */
803 #endif
804 
805 	/* force a shutdown by unmapping entire address space ! */
806 	bzero((caddr_t) PTD, NBPG);
807 
808 	/* "good night, sweet prince .... <THUNK!>" */
809 	pmap_update();
810 	/* NOTREACHED */
811 	while(1);
812 }
813 
814 /*
815  * Grow the user stack to allow for 'sp'. This version grows the stack in
816  *	chunks of SGROWSIZ.
817  */
818 int
819 grow(p, sp)
820 	struct proc *p;
821 	u_int sp;
822 {
823 	unsigned int nss;
824 	caddr_t v;
825 	struct vmspace *vm = p->p_vmspace;
826 
827 	if ((caddr_t)sp <= vm->vm_maxsaddr || (unsigned)sp >= (unsigned)USRSTACK)
828 	    return (1);
829 
830 	nss = roundup(USRSTACK - (unsigned)sp, PAGE_SIZE);
831 
832 	if (nss > p->p_rlimit[RLIMIT_STACK].rlim_cur)
833 		return (0);
834 
835 	if (vm->vm_ssize && roundup(vm->vm_ssize << PAGE_SHIFT,
836 	    SGROWSIZ) < nss) {
837 		int grow_amount;
838 		/*
839 		 * If necessary, grow the VM that the stack occupies
840 		 * to allow for the rlimit. This allows us to not have
841 		 * to allocate all of the VM up-front in execve (which
842 		 * is expensive).
843 		 * Grow the VM by the amount requested rounded up to
844 		 * the nearest SGROWSIZ to provide for some hysteresis.
845 		 */
846 		grow_amount = roundup((nss - (vm->vm_ssize << PAGE_SHIFT)), SGROWSIZ);
847 		v = (char *)USRSTACK - roundup(vm->vm_ssize << PAGE_SHIFT,
848 		    SGROWSIZ) - grow_amount;
849 		/*
850 		 * If there isn't enough room to extend by SGROWSIZ, then
851 		 * just extend to the maximum size
852 		 */
853 		if (v < vm->vm_maxsaddr) {
854 			v = vm->vm_maxsaddr;
855 			grow_amount = MAXSSIZ - (vm->vm_ssize << PAGE_SHIFT);
856 		}
857 		if ((grow_amount == 0) || (vm_map_find(&vm->vm_map, NULL, 0, (vm_offset_t *)&v,
858 		    grow_amount, FALSE) != KERN_SUCCESS)) {
859 			return (0);
860 		}
861 		vm->vm_ssize += grow_amount >> PAGE_SHIFT;
862 	}
863 
864 	return (1);
865 }
866