xref: /freebsd/stand/i386/libi386/biosdisk.c (revision f024dd3bf0c99ff38239ef8e92c12a9fc6bce14c)
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 #include <stdbool.h>
47 
48 #include <bootstrap.h>
49 #include <btxv86.h>
50 #include <edd.h>
51 #include "disk.h"
52 #include "libi386.h"
53 
54 #define	BIOS_NUMDRIVES		0x475
55 #define	BIOSDISK_SECSIZE	512
56 #define	BUFSIZE			(1 * BIOSDISK_SECSIZE)
57 
58 #define	DT_ATAPI	0x10	/* disk type for ATAPI floppies */
59 #define	WDMAJOR		0	/* major numbers for devices we frontend for */
60 #define	WFDMAJOR	1
61 #define	FDMAJOR		2
62 #define	DAMAJOR		4
63 
64 #ifdef DISK_DEBUG
65 #define	DEBUG(fmt, args...)	printf("%s: " fmt "\n", __func__, ## args)
66 #else
67 #define	DEBUG(fmt, args...)
68 #endif
69 
70 /*
71  * List of BIOS devices, translation from disk unit number to
72  * BIOS unit number.
73  */
74 static struct bdinfo
75 {
76 	int		bd_unit;	/* BIOS unit number */
77 	int		bd_cyl;		/* BIOS geometry */
78 	int		bd_hds;
79 	int		bd_sec;
80 	int		bd_flags;
81 #define	BD_MODEINT13	0x0000
82 #define	BD_MODEEDD1	0x0001
83 #define	BD_MODEEDD3	0x0002
84 #define	BD_MODEMASK	0x0003
85 #define	BD_FLOPPY	0x0004
86 	int		bd_type;	/* BIOS 'drive type' (floppy only) */
87 	uint16_t	bd_sectorsize;	/* Sector size */
88 	uint64_t	bd_sectors;	/* Disk size */
89 	int		bd_open;	/* reference counter */
90 	void		*bd_bcache;	/* buffer cache data */
91 } bdinfo [MAXBDDEV];
92 static int nbdinfo = 0;
93 
94 #define	BD(dev)		(bdinfo[(dev)->dd.d_unit])
95 #define	BD_RD		0
96 #define	BD_WR		1
97 
98 static void bd_io_workaround(struct disk_devdesc *dev);
99 
100 static int bd_io(struct disk_devdesc *, daddr_t, int, caddr_t, int);
101 static int bd_int13probe(struct bdinfo *bd);
102 
103 static int bd_init(void);
104 static int bd_strategy(void *devdata, int flag, daddr_t dblk, size_t size,
105     char *buf, size_t *rsize);
106 static int bd_realstrategy(void *devdata, int flag, daddr_t dblk, size_t size,
107     char *buf, size_t *rsize);
108 static int bd_open(struct open_file *f, ...);
109 static int bd_close(struct open_file *f);
110 static int bd_ioctl(struct open_file *f, u_long cmd, void *data);
111 static int bd_print(int verbose);
112 
113 struct devsw biosdisk = {
114 	"disk",
115 	DEVT_DISK,
116 	bd_init,
117 	bd_strategy,
118 	bd_open,
119 	bd_close,
120 	bd_ioctl,
121 	bd_print,
122 	NULL
123 };
124 
125 /*
126  * Translate between BIOS device numbers and our private unit numbers.
127  */
128 int
129 bd_bios2unit(int biosdev)
130 {
131 	int i;
132 
133 	DEBUG("looking for bios device 0x%x", biosdev);
134 	for (i = 0; i < nbdinfo; i++) {
135 		DEBUG("bd unit %d is BIOS device 0x%x", i, bdinfo[i].bd_unit);
136 		if (bdinfo[i].bd_unit == biosdev)
137 			return (i);
138 	}
139 	return (-1);
140 }
141 
142 int
143 bd_unit2bios(int unit)
144 {
145 
146 	if ((unit >= 0) && (unit < nbdinfo))
147 		return (bdinfo[unit].bd_unit);
148 	return (-1);
149 }
150 
151 /*
152  * Quiz the BIOS for disk devices, save a little info about them.
153  */
154 static int
155 bd_init(void)
156 {
157 	int base, unit, nfd = 0;
158 
159 	/* sequence 0, 0x80 */
160 	for (base = 0; base <= 0x80; base += 0x80) {
161 		for (unit = base; (nbdinfo < MAXBDDEV); unit++) {
162 #ifndef VIRTUALBOX
163 			/*
164 			 * Check the BIOS equipment list for number
165 			 * of fixed disks.
166 			 */
167 			if (base == 0x80 &&
168 			    (nfd >= *(unsigned char *)PTOV(BIOS_NUMDRIVES)))
169 				break;
170 #endif
171 			bdinfo[nbdinfo].bd_open = 0;
172 			bdinfo[nbdinfo].bd_bcache = NULL;
173 			bdinfo[nbdinfo].bd_unit = unit;
174 			bdinfo[nbdinfo].bd_flags = unit < 0x80 ? BD_FLOPPY: 0;
175 			if (!bd_int13probe(&bdinfo[nbdinfo]))
176 				break;
177 
178 			/* XXX we need "disk aliases" to make this simpler */
179 			printf("BIOS drive %c: is disk%d\n", (unit < 0x80) ?
180 			    ('A' + unit): ('C' + unit - 0x80), nbdinfo);
181 			nbdinfo++;
182 			if (base == 0x80)
183 				nfd++;
184 		}
185 	}
186 	bcache_add_dev(nbdinfo);
187 	return (0);
188 }
189 
190 /*
191  * Try to detect a device supported by the legacy int13 BIOS
192  */
193 static int
194 bd_int13probe(struct bdinfo *bd)
195 {
196 	struct edd_params params;
197 	int ret = 1;	/* assume success */
198 
199 	v86.ctl = V86_FLAGS;
200 	v86.addr = 0x13;
201 	v86.eax = 0x800;
202 	v86.edx = bd->bd_unit;
203 	v86int();
204 
205 	/* Don't error out if we get bad sector number, try EDD as well */
206 	if (V86_CY(v86.efl) ||	/* carry set */
207 	    (v86.edx & 0xff) <= (unsigned)(bd->bd_unit & 0x7f))	/* unit # bad */
208 		return (0);	/* skip device */
209 
210 	if ((v86.ecx & 0x3f) == 0)	/* absurd sector number */
211 		ret = 0;	/* set error */
212 
213 	/* Convert max cyl # -> # of cylinders */
214 	bd->bd_cyl = ((v86.ecx & 0xc0) << 2) + ((v86.ecx & 0xff00) >> 8) + 1;
215 	/* Convert max head # -> # of heads */
216 	bd->bd_hds = ((v86.edx & 0xff00) >> 8) + 1;
217 	bd->bd_sec = v86.ecx & 0x3f;
218 	bd->bd_type = v86.ebx & 0xff;
219 	bd->bd_flags |= BD_MODEINT13;
220 
221 	/* Calculate sectors count from the geometry */
222 	bd->bd_sectors = bd->bd_cyl * bd->bd_hds * bd->bd_sec;
223 	bd->bd_sectorsize = BIOSDISK_SECSIZE;
224 	DEBUG("unit 0x%x geometry %d/%d/%d", bd->bd_unit, bd->bd_cyl,
225 	    bd->bd_hds, bd->bd_sec);
226 
227 	/* Determine if we can use EDD with this device. */
228 	v86.ctl = V86_FLAGS;
229 	v86.addr = 0x13;
230 	v86.eax = 0x4100;
231 	v86.edx = bd->bd_unit;
232 	v86.ebx = 0x55aa;
233 	v86int();
234 	if (V86_CY(v86.efl) ||	/* carry set */
235 	    (v86.ebx & 0xffff) != 0xaa55 || /* signature */
236 	    (v86.ecx & EDD_INTERFACE_FIXED_DISK) == 0)
237 		return (ret);	/* return code from int13 AH=08 */
238 
239 	/* EDD supported */
240 	bd->bd_flags |= BD_MODEEDD1;
241 	if ((v86.eax & 0xff00) >= 0x3000)
242 		bd->bd_flags |= BD_MODEEDD3;
243 	/* Get disk params */
244 	params.len = sizeof(struct edd_params);
245 	v86.ctl = V86_FLAGS;
246 	v86.addr = 0x13;
247 	v86.eax = 0x4800;
248 	v86.edx = bd->bd_unit;
249 	v86.ds = VTOPSEG(&params);
250 	v86.esi = VTOPOFF(&params);
251 	v86int();
252 	if (!V86_CY(v86.efl)) {
253 		uint64_t total;
254 
255 		/*
256 		 * Sector size must be a multiple of 512 bytes.
257 		 * An alternate test would be to check power of 2,
258 		 * powerof2(params.sector_size).
259 		 * 4K is largest read buffer we can use at this time.
260 		 */
261 		if (params.sector_size >= 512 &&
262 		    params.sector_size <= 4096 &&
263 		    (params.sector_size % BIOSDISK_SECSIZE) == 0)
264 			bd->bd_sectorsize = params.sector_size;
265 
266 		total = bd->bd_sectorsize * params.sectors;
267 		if (params.sectors != 0) {
268 			/* Only update if we did not overflow. */
269 			if (total > params.sectors)
270 				bd->bd_sectors = params.sectors;
271 		}
272 
273 		total = (uint64_t)params.cylinders *
274 		    params.heads * params.sectors_per_track;
275 		if (total > 0 && bd->bd_sectors > total)
276 			bd->bd_sectors = total;
277 
278 		ret = 1;
279 	}
280 	DEBUG("unit 0x%x flags %x, sectors %llu, sectorsize %u",
281 	    bd->bd_unit, bd->bd_flags, bd->bd_sectors, bd->bd_sectorsize);
282 	return (ret);
283 }
284 
285 /*
286  * Print information about disks
287  */
288 static int
289 bd_print(int verbose)
290 {
291 	static char line[80];
292 	struct disk_devdesc dev;
293 	int i, ret = 0;
294 
295 	if (nbdinfo == 0)
296 		return (0);
297 
298 	printf("%s devices:", biosdisk.dv_name);
299 	if ((ret = pager_output("\n")) != 0)
300 		return (ret);
301 
302 	for (i = 0; i < nbdinfo; i++) {
303 		snprintf(line, sizeof(line),
304 		    "    disk%d:   BIOS drive %c (%ju X %u):\n", i,
305 		    (bdinfo[i].bd_unit < 0x80) ? ('A' + bdinfo[i].bd_unit):
306 		    ('C' + bdinfo[i].bd_unit - 0x80),
307 		    (uintmax_t)bdinfo[i].bd_sectors,
308 		    bdinfo[i].bd_sectorsize);
309 		if ((ret = pager_output(line)) != 0)
310 			break;
311 
312 		dev.dd.d_dev = &biosdisk;
313 		dev.dd.d_unit = i;
314 		dev.d_slice = -1;
315 		dev.d_partition = -1;
316 		if (disk_open(&dev,
317 		    bdinfo[i].bd_sectorsize * bdinfo[i].bd_sectors,
318 		    bdinfo[i].bd_sectorsize) == 0) {
319 			snprintf(line, sizeof(line), "    disk%d", i);
320 			ret = disk_print(&dev, line, verbose);
321 			disk_close(&dev);
322 			if (ret != 0)
323 				break;
324 		}
325 	}
326 	return (ret);
327 }
328 
329 /*
330  * Attempt to open the disk described by (dev) for use by (f).
331  *
332  * Note that the philosophy here is "give them exactly what
333  * they ask for".  This is necessary because being too "smart"
334  * about what the user might want leads to complications.
335  * (eg. given no slice or partition value, with a disk that is
336  *  sliced - are they after the first BSD slice, or the DOS
337  *  slice before it?)
338  */
339 static int
340 bd_open(struct open_file *f, ...)
341 {
342 	struct disk_devdesc *dev;
343 	struct disk_devdesc disk;
344 	va_list ap;
345 	uint64_t size;
346 	int rc;
347 
348 	va_start(ap, f);
349 	dev = va_arg(ap, struct disk_devdesc *);
350 	va_end(ap);
351 
352 	if (dev->dd.d_unit < 0 || dev->dd.d_unit >= nbdinfo)
353 		return (EIO);
354 	BD(dev).bd_open++;
355 	if (BD(dev).bd_bcache == NULL)
356 	    BD(dev).bd_bcache = bcache_allocate();
357 
358 	/*
359 	 * Read disk size from partition.
360 	 * This is needed to work around buggy BIOS systems returning
361 	 * wrong (truncated) disk media size.
362 	 * During bd_probe() we tested if the mulitplication of bd_sectors
363 	 * would overflow so it should be safe to perform here.
364 	 */
365 	disk.dd.d_dev = dev->dd.d_dev;
366 	disk.dd.d_unit = dev->dd.d_unit;
367 	disk.d_slice = -1;
368 	disk.d_partition = -1;
369 	disk.d_offset = 0;
370 
371 	if (disk_open(&disk, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
372 	    BD(dev).bd_sectorsize) == 0) {
373 
374 		if (disk_ioctl(&disk, DIOCGMEDIASIZE, &size) == 0) {
375 			size /= BD(dev).bd_sectorsize;
376 			if (size > BD(dev).bd_sectors)
377 				BD(dev).bd_sectors = size;
378 		}
379 		disk_close(&disk);
380 	}
381 
382 	rc = disk_open(dev, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
383 	    BD(dev).bd_sectorsize);
384 	if (rc != 0) {
385 		BD(dev).bd_open--;
386 		if (BD(dev).bd_open == 0) {
387 			bcache_free(BD(dev).bd_bcache);
388 			BD(dev).bd_bcache = NULL;
389 		}
390 	}
391 	return (rc);
392 }
393 
394 static int
395 bd_close(struct open_file *f)
396 {
397 	struct disk_devdesc *dev;
398 
399 	dev = (struct disk_devdesc *)f->f_devdata;
400 	BD(dev).bd_open--;
401 	if (BD(dev).bd_open == 0) {
402 	    bcache_free(BD(dev).bd_bcache);
403 	    BD(dev).bd_bcache = NULL;
404 	}
405 	return (disk_close(dev));
406 }
407 
408 static int
409 bd_ioctl(struct open_file *f, u_long cmd, void *data)
410 {
411 	struct disk_devdesc *dev;
412 	int rc;
413 
414 	dev = (struct disk_devdesc *)f->f_devdata;
415 
416 	rc = disk_ioctl(dev, cmd, data);
417 	if (rc != ENOTTY)
418 		return (rc);
419 
420 	switch (cmd) {
421 	case DIOCGSECTORSIZE:
422 		*(uint32_t *)data = BD(dev).bd_sectorsize;
423 		break;
424 	case DIOCGMEDIASIZE:
425 		*(uint64_t *)data = BD(dev).bd_sectors * BD(dev).bd_sectorsize;
426 		break;
427 	default:
428 		return (ENOTTY);
429 	}
430 	return (0);
431 }
432 
433 static int
434 bd_strategy(void *devdata, int rw, daddr_t dblk, size_t size,
435     char *buf, size_t *rsize)
436 {
437 	struct bcache_devdata bcd;
438 	struct disk_devdesc *dev;
439 
440 	dev = (struct disk_devdesc *)devdata;
441 	bcd.dv_strategy = bd_realstrategy;
442 	bcd.dv_devdata = devdata;
443 	bcd.dv_cache = BD(dev).bd_bcache;
444 	return (bcache_strategy(&bcd, rw, dblk + dev->d_offset, size,
445 	    buf, rsize));
446 }
447 
448 static int
449 bd_realstrategy(void *devdata, int rw, daddr_t dblk, size_t size,
450     char *buf, size_t *rsize)
451 {
452 	struct disk_devdesc *dev = (struct disk_devdesc *)devdata;
453 	uint64_t disk_blocks, offset;
454 	size_t blks, blkoff, bsize, rest;
455 	caddr_t bbuf;
456 	int rc;
457 
458 	/*
459 	 * First make sure the IO size is a multiple of 512 bytes. While we do
460 	 * process partial reads below, the strategy mechanism is built
461 	 * assuming IO is a multiple of 512B blocks. If the request is not
462 	 * a multiple of 512B blocks, it has to be some sort of bug.
463 	 */
464 	if (size == 0 || (size % BIOSDISK_SECSIZE) != 0) {
465 		printf("bd_strategy: %d bytes I/O not multiple of %d\n",
466 		    size, BIOSDISK_SECSIZE);
467 		return (EIO);
468 	}
469 
470 	DEBUG("open_disk %p", dev);
471 
472 	offset = dblk * BIOSDISK_SECSIZE;
473 	dblk = offset / BD(dev).bd_sectorsize;
474 	blkoff = offset % BD(dev).bd_sectorsize;
475 
476 	/*
477 	 * Check the value of the size argument. We do have quite small
478 	 * heap (64MB), but we do not know good upper limit, so we check against
479 	 * INT_MAX here. This will also protect us against possible overflows
480 	 * while translating block count to bytes.
481 	 */
482 	if (size > INT_MAX) {
483 		DEBUG("too large I/O: %zu bytes", size);
484 		return (EIO);
485 	}
486 
487 	blks = size / BD(dev).bd_sectorsize;
488 	if (blks == 0 || (size % BD(dev).bd_sectorsize) != 0)
489 		blks++;
490 
491 	if (dblk > dblk + blks)
492 		return (EIO);
493 
494 	if (rsize)
495 		*rsize = 0;
496 
497 	/*
498 	 * Get disk blocks, this value is either for whole disk or for
499 	 * partition.
500 	 */
501 	if (disk_ioctl(dev, DIOCGMEDIASIZE, &disk_blocks) == 0) {
502 		/* DIOCGMEDIASIZE returns bytes. */
503 		disk_blocks /= BD(dev).bd_sectorsize;
504 	} else {
505 		/* We should not get here. Just try to survive. */
506 		disk_blocks = BD(dev).bd_sectors - dev->d_offset;
507 	}
508 
509 	/* Validate source block address. */
510 	if (dblk < dev->d_offset || dblk >= dev->d_offset + disk_blocks)
511 		return (EIO);
512 
513 	/*
514 	 * Truncate if we are crossing disk or partition end.
515 	 */
516 	if (dblk + blks >= dev->d_offset + disk_blocks) {
517 		blks = dev->d_offset + disk_blocks - dblk;
518 		size = blks * BD(dev).bd_sectorsize;
519 		DEBUG("short I/O %d", blks);
520 	}
521 
522 	if (V86_IO_BUFFER_SIZE / BD(dev).bd_sectorsize == 0)
523 		panic("BUG: Real mode buffer is too small\n");
524 
525 	bbuf = PTOV(V86_IO_BUFFER);
526 	rest = size;
527 
528 	while (blks > 0) {
529 		int x = min(blks, V86_IO_BUFFER_SIZE / BD(dev).bd_sectorsize);
530 
531 		switch (rw & F_MASK) {
532 		case F_READ:
533 			DEBUG("read %d from %lld to %p", x, dblk, buf);
534 			bsize = BD(dev).bd_sectorsize * x - blkoff;
535 			if (rest < bsize)
536 				bsize = rest;
537 
538 			if ((rc = bd_io(dev, dblk, x, bbuf, BD_RD)) != 0)
539 				return (EIO);
540 
541 			bcopy(bbuf + blkoff, buf, bsize);
542 			break;
543 		case F_WRITE :
544 			DEBUG("write %d from %lld to %p", x, dblk, buf);
545 			if (blkoff != 0) {
546 				/*
547 				 * We got offset to sector, read 1 sector to
548 				 * bbuf.
549 				 */
550 				x = 1;
551 				bsize = BD(dev).bd_sectorsize - blkoff;
552 				bsize = min(bsize, rest);
553 				rc = bd_io(dev, dblk, x, bbuf, BD_RD);
554 			} else if (rest < BD(dev).bd_sectorsize) {
555 				/*
556 				 * The remaining block is not full
557 				 * sector. Read 1 sector to bbuf.
558 				 */
559 				x = 1;
560 				bsize = rest;
561 				rc = bd_io(dev, dblk, x, bbuf, BD_RD);
562 			} else {
563 				/* We can write full sector(s). */
564 				bsize = BD(dev).bd_sectorsize * x;
565 			}
566 			/*
567 			 * Put your Data In, Put your Data out,
568 			 * Put your Data In, and shake it all about
569 			 */
570 			bcopy(buf, bbuf + blkoff, bsize);
571 			if ((rc = bd_io(dev, dblk, x, bbuf, BD_WR)) != 0)
572 				return (EIO);
573 
574 			break;
575 		default:
576 			/* DO NOTHING */
577 			return (EROFS);
578 		}
579 
580 		blkoff = 0;
581 		buf += bsize;
582 		rest -= bsize;
583 		blks -= x;
584 		dblk += x;
585 	}
586 
587 	if (rsize != NULL)
588 		*rsize = size;
589 	return (0);
590 }
591 
592 static int
593 bd_edd_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest,
594     int dowrite)
595 {
596 	static struct edd_packet packet;
597 
598 	packet.len = sizeof(struct edd_packet);
599 	packet.count = blks;
600 	packet.off = VTOPOFF(dest);
601 	packet.seg = VTOPSEG(dest);
602 	packet.lba = dblk;
603 	v86.ctl = V86_FLAGS;
604 	v86.addr = 0x13;
605 	/* Should we Write with verify ?? 0x4302 ? */
606 	if (dowrite == BD_WR)
607 		v86.eax = 0x4300;
608 	else
609 		v86.eax = 0x4200;
610 	v86.edx = BD(dev).bd_unit;
611 	v86.ds = VTOPSEG(&packet);
612 	v86.esi = VTOPOFF(&packet);
613 	v86int();
614 	if (V86_CY(v86.efl))
615 		return (v86.eax >> 8);
616 	return (0);
617 }
618 
619 static int
620 bd_chs_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest,
621     int dowrite)
622 {
623 	uint32_t x, bpc, cyl, hd, sec;
624 
625 	bpc = BD(dev).bd_sec * BD(dev).bd_hds;	/* blocks per cylinder */
626 	x = dblk;
627 	cyl = x / bpc;			/* block # / blocks per cylinder */
628 	x %= bpc;				/* block offset into cylinder */
629 	hd = x / BD(dev).bd_sec;		/* offset / blocks per track */
630 	sec = x % BD(dev).bd_sec;		/* offset into track */
631 
632 	/* correct sector number for 1-based BIOS numbering */
633 	sec++;
634 
635 	if (cyl > 1023) {
636 		/* CHS doesn't support cylinders > 1023. */
637 		return (1);
638 	}
639 
640 	v86.ctl = V86_FLAGS;
641 	v86.addr = 0x13;
642 	if (dowrite == BD_WR)
643 		v86.eax = 0x300 | blks;
644 	else
645 		v86.eax = 0x200 | blks;
646 	v86.ecx = ((cyl & 0xff) << 8) | ((cyl & 0x300) >> 2) | sec;
647 	v86.edx = (hd << 8) | BD(dev).bd_unit;
648 	v86.es = VTOPSEG(dest);
649 	v86.ebx = VTOPOFF(dest);
650 	v86int();
651 	if (V86_CY(v86.efl))
652 		return (v86.eax >> 8);
653 	return (0);
654 }
655 
656 static void
657 bd_io_workaround(struct disk_devdesc *dev)
658 {
659 	uint8_t buf[8 * 1024];
660 
661 	bd_edd_io(dev, 0xffffffff, 1, (caddr_t)buf, BD_RD);
662 }
663 
664 static int
665 bd_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest,
666     int dowrite)
667 {
668 	int result, retry;
669 
670 	/* Just in case some idiot actually tries to read/write -1 blocks... */
671 	if (blks < 0)
672 		return (-1);
673 
674 	/*
675 	 * Workaround for a problem with some HP ProLiant BIOS failing to work
676 	 * out the boot disk after installation. hrs and kuriyama discovered
677 	 * this problem with an HP ProLiant DL320e Gen 8 with a 3TB HDD, and
678 	 * discovered that an int13h call seems to cause a buffer overrun in
679 	 * the bios. The problem is alleviated by doing an extra read before
680 	 * the buggy read. It is not immediately known whether other models
681 	 * are similarly affected.
682 	 * Loop retrying the operation a couple of times.  The BIOS
683 	 * may also retry.
684 	 */
685 	if (dowrite == BD_RD && dblk >= 0x100000000)
686 		bd_io_workaround(dev);
687 	for (retry = 0; retry < 3; retry++) {
688 		/* if retrying, reset the drive */
689 		if (retry > 0) {
690 			v86.ctl = V86_FLAGS;
691 			v86.addr = 0x13;
692 			v86.eax = 0;
693 			v86.edx = BD(dev).bd_unit;
694 			v86int();
695 		}
696 
697 		if (BD(dev).bd_flags & BD_MODEEDD1)
698 			result = bd_edd_io(dev, dblk, blks, dest, dowrite);
699 		else
700 			result = bd_chs_io(dev, dblk, blks, dest, dowrite);
701 
702 		if (result == 0)
703 			break;
704 	}
705 
706 	/*
707 	 * 0x20 - Controller failure. This is common error when the
708 	 * media is not present.
709 	 */
710 	if (result != 0 && result != 0x20) {
711 		if (dowrite == BD_WR) {
712 			printf("%s%d: Write %d sector(s) from %p (0x%x) "
713 			    "to %lld: 0x%x\n", dev->dd.d_dev->dv_name,
714 			    dev->dd.d_unit, blks, dest, VTOP(dest), dblk,
715 			    result);
716 		} else {
717 			printf("%s%d: Read %d sector(s) from %lld to %p "
718 			    "(0x%x): 0x%x\n", dev->dd.d_dev->dv_name,
719 			    dev->dd.d_unit, blks, dblk, dest, VTOP(dest),
720 			    result);
721 		}
722 	}
723 
724 	return (result);
725 }
726 
727 /*
728  * Return the BIOS geometry of a given "fixed drive" in a format
729  * suitable for the legacy bootinfo structure.  Since the kernel is
730  * expecting raw int 0x13/0x8 values for N_BIOS_GEOM drives, we
731  * prefer to get the information directly, rather than rely on being
732  * able to put it together from information already maintained for
733  * different purposes and for a probably different number of drives.
734  *
735  * For valid drives, the geometry is expected in the format (31..0)
736  * "000000cc cccccccc hhhhhhhh 00ssssss"; and invalid drives are
737  * indicated by returning the geometry of a "1.2M" PC-format floppy
738  * disk.  And, incidentally, what is returned is not the geometry as
739  * such but the highest valid cylinder, head, and sector numbers.
740  */
741 uint32_t
742 bd_getbigeom(int bunit)
743 {
744 
745 	v86.ctl = V86_FLAGS;
746 	v86.addr = 0x13;
747 	v86.eax = 0x800;
748 	v86.edx = 0x80 + bunit;
749 	v86int();
750 	if (V86_CY(v86.efl))
751 		return (0x4f010f);
752 	return (((v86.ecx & 0xc0) << 18) | ((v86.ecx & 0xff00) << 8) |
753 	    (v86.edx & 0xff00) | (v86.ecx & 0x3f));
754 }
755 
756 /*
757  * Return a suitable dev_t value for (dev).
758  *
759  * In the case where it looks like (dev) is a SCSI disk, we allow the number of
760  * IDE disks to be specified in $num_ide_disks.  There should be a Better Way.
761  */
762 int
763 bd_getdev(struct i386_devdesc *d)
764 {
765 	struct disk_devdesc *dev;
766 	int	biosdev;
767 	int	major;
768 	int	rootdev;
769 	char	*nip, *cp;
770 	int	i, unit;
771 
772 	dev = (struct disk_devdesc *)d;
773 	biosdev = bd_unit2bios(dev->dd.d_unit);
774 	DEBUG("unit %d BIOS device %d", dev->dd.d_unit, biosdev);
775 	if (biosdev == -1)			/* not a BIOS device */
776 		return (-1);
777 	if (disk_open(dev, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
778 	    BD(dev).bd_sectorsize) != 0)	/* oops, not a viable device */
779 		return (-1);
780 	else
781 		disk_close(dev);
782 
783 	if (biosdev < 0x80) {
784 		/* floppy (or emulated floppy) or ATAPI device */
785 		if (bdinfo[dev->dd.d_unit].bd_type == DT_ATAPI) {
786 			/* is an ATAPI disk */
787 			major = WFDMAJOR;
788 		} else {
789 			/* is a floppy disk */
790 			major = FDMAJOR;
791 		}
792 	} else {
793 		/* assume an IDE disk */
794 		major = WDMAJOR;
795 	}
796 	/* default root disk unit number */
797 	unit = biosdev & 0x7f;
798 
799 	/* XXX a better kludge to set the root disk unit number */
800 	if ((nip = getenv("root_disk_unit")) != NULL) {
801 		i = strtol(nip, &cp, 0);
802 		/* check for parse error */
803 		if ((cp != nip) && (*cp == 0))
804 			unit = i;
805 	}
806 
807 	rootdev = MAKEBOOTDEV(major, dev->d_slice + 1, unit, dev->d_partition);
808 	DEBUG("dev is 0x%x\n", rootdev);
809 	return (rootdev);
810 }
811