xref: /freebsd/stand/i386/libi386/biosdisk.c (revision d3d381b2b194b4d24853e92eecef55f262688d1a)
1 /*-
2  * Copyright (c) 1998 Michael Smith <msmith@freebsd.org>
3  * Copyright (c) 2012 Andrey V. Elsukov <ae@FreeBSD.org>
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  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 /*
32  * BIOS disk device handling.
33  *
34  * Ideas and algorithms from:
35  *
36  * - NetBSD libi386/biosdisk.c
37  * - FreeBSD biosboot/disk.c
38  *
39  */
40 
41 #include <sys/disk.h>
42 #include <sys/limits.h>
43 #include <stand.h>
44 #include <machine/bootinfo.h>
45 #include <stdarg.h>
46 
47 #include <bootstrap.h>
48 #include <btxv86.h>
49 #include <edd.h>
50 #include "disk.h"
51 #include "libi386.h"
52 
53 #define	BIOS_NUMDRIVES		0x475
54 #define	BIOSDISK_SECSIZE	512
55 #define	BUFSIZE			(1 * BIOSDISK_SECSIZE)
56 
57 #define	DT_ATAPI	0x10	/* disk type for ATAPI floppies */
58 #define	WDMAJOR		0	/* major numbers for devices we frontend for */
59 #define	WFDMAJOR	1
60 #define	FDMAJOR		2
61 #define	DAMAJOR		4
62 
63 #ifdef DISK_DEBUG
64 #define	DEBUG(fmt, args...)	printf("%s: " fmt "\n", __func__, ## args)
65 #else
66 #define	DEBUG(fmt, args...)
67 #endif
68 
69 /*
70  * List of BIOS devices, translation from disk unit number to
71  * BIOS unit number.
72  */
73 static struct bdinfo
74 {
75 	int		bd_unit;	/* BIOS unit number */
76 	int		bd_cyl;		/* BIOS geometry */
77 	int		bd_hds;
78 	int		bd_sec;
79 	int		bd_flags;
80 #define	BD_MODEINT13	0x0000
81 #define	BD_MODEEDD1	0x0001
82 #define	BD_MODEEDD3	0x0002
83 #define	BD_MODEMASK	0x0003
84 #define	BD_FLOPPY	0x0004
85 	int		bd_type;	/* BIOS 'drive type' (floppy only) */
86 	uint16_t	bd_sectorsize;	/* Sector size */
87 	uint64_t	bd_sectors;	/* Disk size */
88 	int		bd_open;	/* reference counter */
89 	void		*bd_bcache;	/* buffer cache data */
90 } bdinfo [MAXBDDEV];
91 static int nbdinfo = 0;
92 
93 #define	BD(dev)		(bdinfo[(dev)->dd.d_unit])
94 #define	BD_RD		0
95 #define	BD_WR		1
96 
97 static void bd_io_workaround(struct disk_devdesc *dev);
98 
99 static int bd_io(struct disk_devdesc *, daddr_t, int, caddr_t, int);
100 static int bd_int13probe(struct bdinfo *bd);
101 
102 static int bd_init(void);
103 static int bd_strategy(void *devdata, int flag, daddr_t dblk, size_t size,
104     char *buf, size_t *rsize);
105 static int bd_realstrategy(void *devdata, int flag, daddr_t dblk, size_t size,
106     char *buf, size_t *rsize);
107 static int bd_open(struct open_file *f, ...);
108 static int bd_close(struct open_file *f);
109 static int bd_ioctl(struct open_file *f, u_long cmd, void *data);
110 static int bd_print(int verbose);
111 
112 struct devsw biosdisk = {
113 	"disk",
114 	DEVT_DISK,
115 	bd_init,
116 	bd_strategy,
117 	bd_open,
118 	bd_close,
119 	bd_ioctl,
120 	bd_print,
121 	NULL
122 };
123 
124 /*
125  * Translate between BIOS device numbers and our private unit numbers.
126  */
127 int
128 bd_bios2unit(int biosdev)
129 {
130 	int i;
131 
132 	DEBUG("looking for bios device 0x%x", biosdev);
133 	for (i = 0; i < nbdinfo; i++) {
134 		DEBUG("bd unit %d is BIOS device 0x%x", i, bdinfo[i].bd_unit);
135 		if (bdinfo[i].bd_unit == biosdev)
136 			return (i);
137 	}
138 	return (-1);
139 }
140 
141 int
142 bd_unit2bios(int unit)
143 {
144 
145 	if ((unit >= 0) && (unit < nbdinfo))
146 		return (bdinfo[unit].bd_unit);
147 	return (-1);
148 }
149 
150 /*
151  * Quiz the BIOS for disk devices, save a little info about them.
152  */
153 static int
154 bd_init(void)
155 {
156 	int base, unit, nfd = 0;
157 
158 	/* sequence 0, 0x80 */
159 	for (base = 0; base <= 0x80; base += 0x80) {
160 		for (unit = base; (nbdinfo < MAXBDDEV); unit++) {
161 #ifndef VIRTUALBOX
162 			/*
163 			 * Check the BIOS equipment list for number
164 			 * of fixed disks.
165 			 */
166 			if (base == 0x80 &&
167 			    (nfd >= *(unsigned char *)PTOV(BIOS_NUMDRIVES)))
168 				break;
169 #endif
170 			bdinfo[nbdinfo].bd_open = 0;
171 			bdinfo[nbdinfo].bd_bcache = NULL;
172 			bdinfo[nbdinfo].bd_unit = unit;
173 			bdinfo[nbdinfo].bd_flags = unit < 0x80 ? BD_FLOPPY: 0;
174 			if (!bd_int13probe(&bdinfo[nbdinfo]))
175 				break;
176 
177 			/* XXX we need "disk aliases" to make this simpler */
178 			printf("BIOS drive %c: is disk%d\n", (unit < 0x80) ?
179 			    ('A' + unit): ('C' + unit - 0x80), nbdinfo);
180 			nbdinfo++;
181 			if (base == 0x80)
182 				nfd++;
183 		}
184 	}
185 	bcache_add_dev(nbdinfo);
186 	return (0);
187 }
188 
189 /*
190  * Try to detect a device supported by the legacy int13 BIOS
191  */
192 static int
193 bd_int13probe(struct bdinfo *bd)
194 {
195 	struct edd_params params;
196 	int ret = 1;	/* assume success */
197 
198 	v86.ctl = V86_FLAGS;
199 	v86.addr = 0x13;
200 	v86.eax = 0x800;
201 	v86.edx = bd->bd_unit;
202 	v86int();
203 
204 	/* Don't error out if we get bad sector number, try EDD as well */
205 	if (V86_CY(v86.efl) ||	/* carry set */
206 	    (v86.edx & 0xff) <= (unsigned)(bd->bd_unit & 0x7f))	/* unit # bad */
207 		return (0);	/* skip device */
208 
209 	if ((v86.ecx & 0x3f) == 0)	/* absurd sector number */
210 		ret = 0;	/* set error */
211 
212 	/* Convert max cyl # -> # of cylinders */
213 	bd->bd_cyl = ((v86.ecx & 0xc0) << 2) + ((v86.ecx & 0xff00) >> 8) + 1;
214 	/* Convert max head # -> # of heads */
215 	bd->bd_hds = ((v86.edx & 0xff00) >> 8) + 1;
216 	bd->bd_sec = v86.ecx & 0x3f;
217 	bd->bd_type = v86.ebx & 0xff;
218 	bd->bd_flags |= BD_MODEINT13;
219 
220 	/* Calculate sectors count from the geometry */
221 	bd->bd_sectors = bd->bd_cyl * bd->bd_hds * bd->bd_sec;
222 	bd->bd_sectorsize = BIOSDISK_SECSIZE;
223 	DEBUG("unit 0x%x geometry %d/%d/%d", bd->bd_unit, bd->bd_cyl,
224 	    bd->bd_hds, bd->bd_sec);
225 
226 	/* Determine if we can use EDD with this device. */
227 	v86.ctl = V86_FLAGS;
228 	v86.addr = 0x13;
229 	v86.eax = 0x4100;
230 	v86.edx = bd->bd_unit;
231 	v86.ebx = 0x55aa;
232 	v86int();
233 	if (V86_CY(v86.efl) ||	/* carry set */
234 	    (v86.ebx & 0xffff) != 0xaa55 || /* signature */
235 	    (v86.ecx & EDD_INTERFACE_FIXED_DISK) == 0)
236 		return (ret);	/* return code from int13 AH=08 */
237 
238 	/* EDD supported */
239 	bd->bd_flags |= BD_MODEEDD1;
240 	if ((v86.eax & 0xff00) >= 0x3000)
241 		bd->bd_flags |= BD_MODEEDD3;
242 	/* Get disk params */
243 	params.len = sizeof(struct edd_params);
244 	v86.ctl = V86_FLAGS;
245 	v86.addr = 0x13;
246 	v86.eax = 0x4800;
247 	v86.edx = bd->bd_unit;
248 	v86.ds = VTOPSEG(&params);
249 	v86.esi = VTOPOFF(&params);
250 	v86int();
251 	if (!V86_CY(v86.efl)) {
252 		uint64_t total;
253 
254 		/*
255 		 * Sector size must be a multiple of 512 bytes.
256 		 * An alternate test would be to check power of 2,
257 		 * powerof2(params.sector_size).
258 		 */
259 		if (params.sector_size % BIOSDISK_SECSIZE)
260 			bd->bd_sectorsize = BIOSDISK_SECSIZE;
261 		else
262 			bd->bd_sectorsize = params.sector_size;
263 
264 		total = bd->bd_sectorsize * params.sectors;
265 		if (params.sectors != 0) {
266 			/* Only update if we did not overflow. */
267 			if (total > params.sectors)
268 				bd->bd_sectors = params.sectors;
269 		}
270 
271 		total = (uint64_t)params.cylinders *
272 		    params.heads * params.sectors_per_track;
273 		if (total > 0 && bd->bd_sectors > total)
274 			bd->bd_sectors = total;
275 
276 		ret = 1;
277 	}
278 	DEBUG("unit 0x%x flags %x, sectors %llu, sectorsize %u",
279 	    bd->bd_unit, bd->bd_flags, bd->bd_sectors, bd->bd_sectorsize);
280 	return (ret);
281 }
282 
283 /*
284  * Print information about disks
285  */
286 static int
287 bd_print(int verbose)
288 {
289 	static char line[80];
290 	struct disk_devdesc dev;
291 	int i, ret = 0;
292 
293 	if (nbdinfo == 0)
294 		return (0);
295 
296 	printf("%s devices:", biosdisk.dv_name);
297 	if ((ret = pager_output("\n")) != 0)
298 		return (ret);
299 
300 	for (i = 0; i < nbdinfo; i++) {
301 		snprintf(line, sizeof(line),
302 		    "    disk%d:   BIOS drive %c (%ju X %u):\n", i,
303 		    (bdinfo[i].bd_unit < 0x80) ? ('A' + bdinfo[i].bd_unit):
304 		    ('C' + bdinfo[i].bd_unit - 0x80),
305 		    (uintmax_t)bdinfo[i].bd_sectors,
306 		    bdinfo[i].bd_sectorsize);
307 		if ((ret = pager_output(line)) != 0)
308 			break;
309 
310 		dev.dd.d_dev = &biosdisk;
311 		dev.dd.d_unit = i;
312 		dev.d_slice = -1;
313 		dev.d_partition = -1;
314 		if (disk_open(&dev,
315 		    bdinfo[i].bd_sectorsize * bdinfo[i].bd_sectors,
316 		    bdinfo[i].bd_sectorsize) == 0) {
317 			snprintf(line, sizeof(line), "    disk%d", i);
318 			ret = disk_print(&dev, line, verbose);
319 			disk_close(&dev);
320 			if (ret != 0)
321 				break;
322 		}
323 	}
324 	return (ret);
325 }
326 
327 /*
328  * Attempt to open the disk described by (dev) for use by (f).
329  *
330  * Note that the philosophy here is "give them exactly what
331  * they ask for".  This is necessary because being too "smart"
332  * about what the user might want leads to complications.
333  * (eg. given no slice or partition value, with a disk that is
334  *  sliced - are they after the first BSD slice, or the DOS
335  *  slice before it?)
336  */
337 static int
338 bd_open(struct open_file *f, ...)
339 {
340 	struct disk_devdesc *dev;
341 	struct disk_devdesc disk;
342 	va_list ap;
343 	uint64_t size;
344 	int rc;
345 
346 	va_start(ap, f);
347 	dev = va_arg(ap, struct disk_devdesc *);
348 	va_end(ap);
349 
350 	if (dev->dd.d_unit < 0 || dev->dd.d_unit >= nbdinfo)
351 		return (EIO);
352 	BD(dev).bd_open++;
353 	if (BD(dev).bd_bcache == NULL)
354 	    BD(dev).bd_bcache = bcache_allocate();
355 
356 	/*
357 	 * Read disk size from partition.
358 	 * This is needed to work around buggy BIOS systems returning
359 	 * wrong (truncated) disk media size.
360 	 * During bd_probe() we tested if the mulitplication of bd_sectors
361 	 * would overflow so it should be safe to perform here.
362 	 */
363 	disk.dd.d_dev = dev->dd.d_dev;
364 	disk.dd.d_unit = dev->dd.d_unit;
365 	disk.d_slice = -1;
366 	disk.d_partition = -1;
367 	disk.d_offset = 0;
368 
369 	if (disk_open(&disk, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
370 	    BD(dev).bd_sectorsize) == 0) {
371 
372 		if (disk_ioctl(&disk, DIOCGMEDIASIZE, &size) == 0) {
373 			size /= BD(dev).bd_sectorsize;
374 			if (size > BD(dev).bd_sectors)
375 				BD(dev).bd_sectors = size;
376 		}
377 		disk_close(&disk);
378 	}
379 
380 	rc = disk_open(dev, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
381 	    BD(dev).bd_sectorsize);
382 	if (rc != 0) {
383 		BD(dev).bd_open--;
384 		if (BD(dev).bd_open == 0) {
385 			bcache_free(BD(dev).bd_bcache);
386 			BD(dev).bd_bcache = NULL;
387 		}
388 	}
389 	return (rc);
390 }
391 
392 static int
393 bd_close(struct open_file *f)
394 {
395 	struct disk_devdesc *dev;
396 
397 	dev = (struct disk_devdesc *)f->f_devdata;
398 	BD(dev).bd_open--;
399 	if (BD(dev).bd_open == 0) {
400 	    bcache_free(BD(dev).bd_bcache);
401 	    BD(dev).bd_bcache = NULL;
402 	}
403 	return (disk_close(dev));
404 }
405 
406 static int
407 bd_ioctl(struct open_file *f, u_long cmd, void *data)
408 {
409 	struct disk_devdesc *dev;
410 	int rc;
411 
412 	dev = (struct disk_devdesc *)f->f_devdata;
413 
414 	rc = disk_ioctl(dev, cmd, data);
415 	if (rc != ENOTTY)
416 		return (rc);
417 
418 	switch (cmd) {
419 	case DIOCGSECTORSIZE:
420 		*(uint32_t *)data = BD(dev).bd_sectorsize;
421 		break;
422 	case DIOCGMEDIASIZE:
423 		*(uint64_t *)data = BD(dev).bd_sectors * BD(dev).bd_sectorsize;
424 		break;
425 	default:
426 		return (ENOTTY);
427 	}
428 	return (0);
429 }
430 
431 static int
432 bd_strategy(void *devdata, int rw, daddr_t dblk, size_t size,
433     char *buf, size_t *rsize)
434 {
435 	struct bcache_devdata bcd;
436 	struct disk_devdesc *dev;
437 
438 	dev = (struct disk_devdesc *)devdata;
439 	bcd.dv_strategy = bd_realstrategy;
440 	bcd.dv_devdata = devdata;
441 	bcd.dv_cache = BD(dev).bd_bcache;
442 	return (bcache_strategy(&bcd, rw, dblk + dev->d_offset, size,
443 	    buf, rsize));
444 }
445 
446 static int
447 bd_realstrategy(void *devdata, int rw, daddr_t dblk, size_t size,
448     char *buf, size_t *rsize)
449 {
450 	struct disk_devdesc *dev = (struct disk_devdesc *)devdata;
451 	uint64_t		disk_blocks;
452 	int			blks, rc;
453 
454 	if (size % BD(dev).bd_sectorsize) {
455 		panic("bd_strategy: %d bytes I/O not multiple of block size",
456 		    size);
457 	}
458 
459 	DEBUG("open_disk %p", dev);
460 
461 	/*
462 	 * Check the value of the size argument. We do have quite small
463 	 * heap (64MB), but we do not know good upper limit, so we check against
464 	 * INT_MAX here. This will also protect us against possible overflows
465 	 * while translating block count to bytes.
466 	 */
467 	if (size > INT_MAX) {
468 		DEBUG("too large read: %zu bytes", size);
469 		return (EIO);
470 	}
471 
472 	blks = size / BD(dev).bd_sectorsize;
473 	if (dblk > dblk + blks)
474 		return (EIO);
475 
476 	if (rsize)
477 		*rsize = 0;
478 
479 	/*
480 	 * Get disk blocks, this value is either for whole disk or for
481 	 * partition.
482 	 */
483 	if (disk_ioctl(dev, DIOCGMEDIASIZE, &disk_blocks) == 0) {
484 		/* DIOCGMEDIASIZE returns bytes. */
485 		disk_blocks /= BD(dev).bd_sectorsize;
486 	} else {
487 		/* We should not get here. Just try to survive. */
488 		disk_blocks = BD(dev).bd_sectors - dev->d_offset;
489 	}
490 
491 	/* Validate source block address. */
492 	if (dblk < dev->d_offset || dblk >= dev->d_offset + disk_blocks)
493 		return (EIO);
494 
495 	/*
496 	 * Truncate if we are crossing disk or partition end.
497 	 */
498 	if (dblk + blks >= dev->d_offset + disk_blocks) {
499 		blks = dev->d_offset + disk_blocks - dblk;
500 		size = blks * BD(dev).bd_sectorsize;
501 		DEBUG("short read %d", blks);
502 	}
503 
504 	switch (rw & F_MASK) {
505 	case F_READ:
506 		DEBUG("read %d from %lld to %p", blks, dblk, buf);
507 
508 		if (blks && (rc = bd_io(dev, dblk, blks, buf, BD_RD))) {
509 			/* Filter out floppy controller errors */
510 			if (BD(dev).bd_flags != BD_FLOPPY || rc != 0x20) {
511 				printf("read %d from %lld to %p, error: 0x%x\n",
512 				    blks, dblk, buf, rc);
513 			}
514 			return (EIO);
515 		}
516 		break;
517 	case F_WRITE :
518 		DEBUG("write %d from %lld to %p", blks, dblk, buf);
519 
520 		if (blks && bd_io(dev, dblk, blks, buf, BD_WR)) {
521 			DEBUG("write error");
522 			return (EIO);
523 		}
524 		break;
525 	default:
526 		/* DO NOTHING */
527 		return (EROFS);
528 	}
529 
530 	if (rsize)
531 		*rsize = size;
532 	return (0);
533 }
534 
535 static int
536 bd_edd_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest,
537     int dowrite)
538 {
539 	static struct edd_packet packet;
540 
541 	packet.len = sizeof(struct edd_packet);
542 	packet.count = blks;
543 	packet.off = VTOPOFF(dest);
544 	packet.seg = VTOPSEG(dest);
545 	packet.lba = dblk;
546 	v86.ctl = V86_FLAGS;
547 	v86.addr = 0x13;
548 	/* Should we Write with verify ?? 0x4302 ? */
549 	if (dowrite == BD_WR)
550 		v86.eax = 0x4300;
551 	else
552 		v86.eax = 0x4200;
553 	v86.edx = BD(dev).bd_unit;
554 	v86.ds = VTOPSEG(&packet);
555 	v86.esi = VTOPOFF(&packet);
556 	v86int();
557 	if (V86_CY(v86.efl))
558 		return (v86.eax >> 8);
559 	return (0);
560 }
561 
562 static int
563 bd_chs_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest,
564     int dowrite)
565 {
566 	uint32_t x, bpc, cyl, hd, sec;
567 
568 	bpc = BD(dev).bd_sec * BD(dev).bd_hds;	/* blocks per cylinder */
569 	x = dblk;
570 	cyl = x / bpc;			/* block # / blocks per cylinder */
571 	x %= bpc;				/* block offset into cylinder */
572 	hd = x / BD(dev).bd_sec;		/* offset / blocks per track */
573 	sec = x % BD(dev).bd_sec;		/* offset into track */
574 
575 	/* correct sector number for 1-based BIOS numbering */
576 	sec++;
577 
578 	if (cyl > 1023) {
579 		/* CHS doesn't support cylinders > 1023. */
580 		return (1);
581 	}
582 
583 	v86.ctl = V86_FLAGS;
584 	v86.addr = 0x13;
585 	if (dowrite == BD_WR)
586 		v86.eax = 0x300 | blks;
587 	else
588 		v86.eax = 0x200 | blks;
589 	v86.ecx = ((cyl & 0xff) << 8) | ((cyl & 0x300) >> 2) | sec;
590 	v86.edx = (hd << 8) | BD(dev).bd_unit;
591 	v86.es = VTOPSEG(dest);
592 	v86.ebx = VTOPOFF(dest);
593 	v86int();
594 	if (V86_CY(v86.efl))
595 		return (v86.eax >> 8);
596 	return (0);
597 }
598 
599 static void
600 bd_io_workaround(struct disk_devdesc *dev)
601 {
602 	uint8_t buf[8 * 1024];
603 
604 	bd_edd_io(dev, 0xffffffff, 1, (caddr_t)buf, BD_RD);
605 }
606 
607 
608 static int
609 bd_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest,
610     int dowrite)
611 {
612 	u_int	x, sec, result, resid, retry, maxfer;
613 	caddr_t	p, xp, bbuf;
614 
615 	/* Just in case some idiot actually tries to read/write -1 blocks... */
616 	if (blks < 0)
617 		return (-1);
618 
619 	resid = blks;
620 	p = dest;
621 
622 	/*
623 	 * Workaround for a problem with some HP ProLiant BIOS failing to work
624 	 * out the boot disk after installation. hrs and kuriyama discovered
625 	 * this problem with an HP ProLiant DL320e Gen 8 with a 3TB HDD, and
626 	 * discovered that an int13h call seems to cause a buffer overrun in
627 	 * the bios. The problem is alleviated by doing an extra read before
628 	 * the buggy read. It is not immediately known whether other models
629 	 * are similarly affected.
630 	 */
631 	if (dowrite == BD_RD && dblk >= 0x100000000)
632 		bd_io_workaround(dev);
633 
634 	/* Decide whether we have to bounce */
635 	if (VTOP(dest) >> 20 != 0 || (BD(dev).bd_unit < 0x80 &&
636 	    (VTOP(dest) >> 16) !=
637 	    (VTOP(dest + blks * BD(dev).bd_sectorsize) >> 16))) {
638 
639 		/*
640 		 * There is a 64k physical boundary somewhere in the
641 		 * destination buffer, or the destination buffer is above
642 		 * first 1MB of physical memory so we have to arrange a
643 		 * suitable bounce buffer.  Allocate a buffer twice as large
644 		 * as we need to.  Use the bottom half unless there is a break
645 		 * there, in which case we use the top half.
646 		 */
647 		x = V86_IO_BUFFER_SIZE / BD(dev).bd_sectorsize;
648 		x = min(x, (unsigned)blks);
649 		bbuf = PTOV(V86_IO_BUFFER);
650 		maxfer = x;	/* limit transfers to bounce region size */
651 	} else {
652 		bbuf = NULL;
653 		maxfer = 0;
654 	}
655 
656 	while (resid > 0) {
657 		/*
658 		 * Play it safe and don't cross track boundaries.
659 		 * (XXX this is probably unnecessary)
660 		 */
661 		sec = dblk % BD(dev).bd_sec;	/* offset into track */
662 		x = min(BD(dev).bd_sec - sec, resid);
663 		if (maxfer > 0)
664 			x = min(x, maxfer);		/* fit bounce buffer */
665 
666 		/* where do we transfer to? */
667 		xp = bbuf == NULL ? p : bbuf;
668 
669 		/*
670 		 * Put your Data In, Put your Data out,
671 		 * Put your Data In, and shake it all about
672 		 */
673 		if (dowrite == BD_WR && bbuf != NULL)
674 			bcopy(p, bbuf, x * BD(dev).bd_sectorsize);
675 
676 		/*
677 		 * Loop retrying the operation a couple of times.  The BIOS
678 		 * may also retry.
679 		 */
680 		for (retry = 0; retry < 3; retry++) {
681 			/* if retrying, reset the drive */
682 			if (retry > 0) {
683 				v86.ctl = V86_FLAGS;
684 				v86.addr = 0x13;
685 				v86.eax = 0;
686 				v86.edx = BD(dev).bd_unit;
687 				v86int();
688 			}
689 
690 			if (BD(dev).bd_flags & BD_MODEEDD1)
691 				result = bd_edd_io(dev, dblk, x, xp, dowrite);
692 			else
693 				result = bd_chs_io(dev, dblk, x, xp, dowrite);
694 			if (result == 0)
695 				break;
696 		}
697 
698 		if (dowrite == BD_WR)
699 			DEBUG("Write %d sector(s) from %p (0x%x) to %lld %s", x,
700 			    p, VTOP(p), dblk, result ? "failed" : "ok");
701 		else
702 			DEBUG("Read %d sector(s) from %lld to %p (0x%x) %s", x,
703 			    dblk, p, VTOP(p), result ? "failed" : "ok");
704 		if (result) {
705 			return (result);
706 		}
707 		if (dowrite == BD_RD && bbuf != NULL)
708 			bcopy(bbuf, p, x * BD(dev).bd_sectorsize);
709 		p += (x * BD(dev).bd_sectorsize);
710 		dblk += x;
711 		resid -= x;
712 	}
713 
714 	return (0);
715 }
716 
717 /*
718  * Return the BIOS geometry of a given "fixed drive" in a format
719  * suitable for the legacy bootinfo structure.  Since the kernel is
720  * expecting raw int 0x13/0x8 values for N_BIOS_GEOM drives, we
721  * prefer to get the information directly, rather than rely on being
722  * able to put it together from information already maintained for
723  * different purposes and for a probably different number of drives.
724  *
725  * For valid drives, the geometry is expected in the format (31..0)
726  * "000000cc cccccccc hhhhhhhh 00ssssss"; and invalid drives are
727  * indicated by returning the geometry of a "1.2M" PC-format floppy
728  * disk.  And, incidentally, what is returned is not the geometry as
729  * such but the highest valid cylinder, head, and sector numbers.
730  */
731 uint32_t
732 bd_getbigeom(int bunit)
733 {
734 
735 	v86.ctl = V86_FLAGS;
736 	v86.addr = 0x13;
737 	v86.eax = 0x800;
738 	v86.edx = 0x80 + bunit;
739 	v86int();
740 	if (V86_CY(v86.efl))
741 		return (0x4f010f);
742 	return (((v86.ecx & 0xc0) << 18) | ((v86.ecx & 0xff00) << 8) |
743 	    (v86.edx & 0xff00) | (v86.ecx & 0x3f));
744 }
745 
746 /*
747  * Return a suitable dev_t value for (dev).
748  *
749  * In the case where it looks like (dev) is a SCSI disk, we allow the number of
750  * IDE disks to be specified in $num_ide_disks.  There should be a Better Way.
751  */
752 int
753 bd_getdev(struct i386_devdesc *d)
754 {
755 	struct disk_devdesc *dev;
756 	int	biosdev;
757 	int	major;
758 	int	rootdev;
759 	char	*nip, *cp;
760 	int	i, unit;
761 
762 	dev = (struct disk_devdesc *)d;
763 	biosdev = bd_unit2bios(dev->dd.d_unit);
764 	DEBUG("unit %d BIOS device %d", dev->dd.d_unit, biosdev);
765 	if (biosdev == -1)			/* not a BIOS device */
766 		return (-1);
767 	if (disk_open(dev, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
768 	    BD(dev).bd_sectorsize) != 0)	/* oops, not a viable device */
769 		return (-1);
770 	else
771 		disk_close(dev);
772 
773 	if (biosdev < 0x80) {
774 		/* floppy (or emulated floppy) or ATAPI device */
775 		if (bdinfo[dev->dd.d_unit].bd_type == DT_ATAPI) {
776 			/* is an ATAPI disk */
777 			major = WFDMAJOR;
778 		} else {
779 			/* is a floppy disk */
780 			major = FDMAJOR;
781 		}
782 	} else {
783 		/* assume an IDE disk */
784 		major = WDMAJOR;
785 	}
786 	/* default root disk unit number */
787 	unit = biosdev & 0x7f;
788 
789 	/* XXX a better kludge to set the root disk unit number */
790 	if ((nip = getenv("root_disk_unit")) != NULL) {
791 		i = strtol(nip, &cp, 0);
792 		/* check for parse error */
793 		if ((cp != nip) && (*cp == 0))
794 			unit = i;
795 	}
796 
797 	rootdev = MAKEBOOTDEV(major, dev->d_slice + 1, unit, dev->d_partition);
798 	DEBUG("dev is 0x%x\n", rootdev);
799 	return (rootdev);
800 }
801