2 * Copyright (c) 1998 Michael Smith <msmith@freebsd.org>
3 * Copyright (c) 2012 Andrey V. Elsukov <ae@FreeBSD.org>
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.
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
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
32 * BIOS disk device handling.
34 * Ideas and algorithms from:
36 * - NetBSD libi386/biosdisk.c
37 * - FreeBSD biosboot/disk.c
42 #include <sys/limits.h>
44 #include <machine/bootinfo.h>
47 #include <bootstrap.h>
53 #define BIOS_NUMDRIVES 0x475
54 #define BIOSDISK_SECSIZE 512
55 #define BUFSIZE (1 * BIOSDISK_SECSIZE)
57 #define DT_ATAPI 0x10 /* disk type for ATAPI floppies */
58 #define WDMAJOR 0 /* major numbers for devices we frontend for */
64 # define DEBUG(fmt, args...) printf("%s: " fmt "\n" , __func__ , ## args)
66 # define DEBUG(fmt, args...)
70 * List of BIOS devices, translation from disk unit number to
75 int bd_unit; /* BIOS unit number */
76 int bd_cyl; /* BIOS geometry */
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 */
91 static int nbdinfo = 0;
93 #define BD(dev) (bdinfo[(dev)->dd.d_unit])
95 static void bd_io_workaround(struct disk_devdesc *dev);
97 static int bd_read(struct disk_devdesc *dev, daddr_t dblk, int blks,
99 static int bd_write(struct disk_devdesc *dev, daddr_t dblk, int blks,
101 static int bd_int13probe(struct bdinfo *bd);
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);
113 struct devsw biosdisk = {
126 * Translate between BIOS device numbers and our private unit numbers.
129 bd_bios2unit(int biosdev)
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)
143 bd_unit2bios(int unit)
146 if ((unit >= 0) && (unit < nbdinfo))
147 return (bdinfo[unit].bd_unit);
152 * Quiz the BIOS for disk devices, save a little info about them.
157 int base, unit, nfd = 0;
159 /* sequence 0, 0x80 */
160 for (base = 0; base <= 0x80; base += 0x80) {
161 for (unit = base; (nbdinfo < MAXBDDEV); unit++) {
164 * Check the BIOS equipment list for number
168 (nfd >= *(unsigned char *)PTOV(BIOS_NUMDRIVES)))
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]))
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);
186 bcache_add_dev(nbdinfo);
191 * Try to detect a device supported by the legacy int13 BIOS
194 bd_int13probe(struct bdinfo *bd)
196 struct edd_params params;
197 int ret = 1; /* assume success */
202 v86.edx = bd->bd_unit;
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 */
210 if ((v86.ecx & 0x3f) == 0) /* absurd sector number */
211 ret = 0; /* set error */
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;
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);
227 /* Determine if we can use EDD with this device. */
231 v86.edx = bd->bd_unit;
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 */
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);
248 v86.edx = bd->bd_unit;
249 v86.ds = VTOPSEG(¶ms);
250 v86.esi = VTOPOFF(¶ms);
252 if (!V86_CY(v86.efl)) {
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).
260 if (params.sector_size % BIOSDISK_SECSIZE)
261 bd->bd_sectorsize = BIOSDISK_SECSIZE;
263 bd->bd_sectorsize = params.sector_size;
265 total = bd->bd_sectorsize * params.sectors;
266 if (params.sectors != 0) {
267 /* Only update if we did not overflow. */
268 if (total > params.sectors)
269 bd->bd_sectors = params.sectors;
272 total = (uint64_t)params.cylinders *
273 params.heads * params.sectors_per_track;
274 if (bd->bd_sectors < total)
275 bd->bd_sectors = total;
279 DEBUG("unit 0x%x flags %x, sectors %llu, sectorsize %u",
280 bd->bd_unit, bd->bd_flags, bd->bd_sectors, bd->bd_sectorsize);
285 * Print information about disks
288 bd_print(int verbose)
290 static char line[80];
291 struct disk_devdesc dev;
297 printf("%s devices:", biosdisk.dv_name);
298 if ((ret = pager_output("\n")) != 0)
301 for (i = 0; i < nbdinfo; i++) {
302 snprintf(line, sizeof(line),
303 " disk%d: BIOS drive %c (%ju X %u):\n", i,
304 (bdinfo[i].bd_unit < 0x80) ? ('A' + bdinfo[i].bd_unit):
305 ('C' + bdinfo[i].bd_unit - 0x80),
306 (uintmax_t)bdinfo[i].bd_sectors,
307 bdinfo[i].bd_sectorsize);
308 if ((ret = pager_output(line)) != 0)
310 dev.dd.d_dev = &biosdisk;
313 dev.d_partition = -1;
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);
328 * Attempt to open the disk described by (dev) for use by (f).
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
338 bd_open(struct open_file *f, ...)
340 struct disk_devdesc *dev, rdev;
341 struct disk_devdesc disk;
347 dev = va_arg(ap, struct disk_devdesc *);
350 if (dev->dd.d_unit < 0 || dev->dd.d_unit >= nbdinfo)
353 if (BD(dev).bd_bcache == NULL)
354 BD(dev).bd_bcache = bcache_allocate();
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.
363 disk.dd.d_dev = dev->dd.d_dev;
364 disk.dd.d_unit = dev->dd.d_unit;
366 disk.d_partition = -1;
368 if (disk_open(&disk, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
369 BD(dev).bd_sectorsize) == 0) {
371 if (disk_ioctl(&disk, DIOCGMEDIASIZE, &size) == 0) {
372 size /= BD(dev).bd_sectorsize;
373 if (size > BD(dev).bd_sectors)
374 BD(dev).bd_sectors = size;
379 err = disk_open(dev, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
380 BD(dev).bd_sectorsize);
386 bd_close(struct open_file *f)
388 struct disk_devdesc *dev;
390 dev = (struct disk_devdesc *)f->f_devdata;
392 if (BD(dev).bd_open == 0) {
393 bcache_free(BD(dev).bd_bcache);
394 BD(dev).bd_bcache = NULL;
396 return (disk_close(dev));
400 bd_ioctl(struct open_file *f, u_long cmd, void *data)
402 struct disk_devdesc *dev;
405 dev = (struct disk_devdesc *)f->f_devdata;
407 rc = disk_ioctl(dev, cmd, data);
412 case DIOCGSECTORSIZE:
413 *(u_int *)data = BD(dev).bd_sectorsize;
416 *(uint64_t *)data = BD(dev).bd_sectors * BD(dev).bd_sectorsize;
425 bd_strategy(void *devdata, int rw, daddr_t dblk, size_t size,
426 char *buf, size_t *rsize)
428 struct bcache_devdata bcd;
429 struct disk_devdesc *dev;
431 dev = (struct disk_devdesc *)devdata;
432 bcd.dv_strategy = bd_realstrategy;
433 bcd.dv_devdata = devdata;
434 bcd.dv_cache = BD(dev).bd_bcache;
435 return (bcache_strategy(&bcd, rw, dblk + dev->d_offset,
440 bd_realstrategy(void *devdata, int rw, daddr_t dblk, size_t size,
441 char *buf, size_t *rsize)
443 struct disk_devdesc *dev = (struct disk_devdesc *)devdata;
444 uint64_t disk_blocks;
446 #ifdef BD_SUPPORT_FRAGS /* XXX: sector size */
447 char fragbuf[BIOSDISK_SECSIZE];
450 fragsize = size % BIOSDISK_SECSIZE;
452 if (size % BD(dev).bd_sectorsize)
453 panic("bd_strategy: %d bytes I/O not multiple of block size", size);
456 DEBUG("open_disk %p", dev);
459 * Check the value of the size argument. We do have quite small
460 * heap (64MB), but we do not know good upper limit, so we check against
461 * INT_MAX here. This will also protect us against possible overflows
462 * while translating block count to bytes.
464 if (size > INT_MAX) {
465 DEBUG("too large read: %zu bytes", size);
469 blks = size / BD(dev).bd_sectorsize;
470 if (dblk > dblk + blks)
476 /* Get disk blocks, this value is either for whole disk or for partition */
477 if (disk_ioctl(dev, DIOCGMEDIASIZE, &disk_blocks) == 0) {
478 /* DIOCGMEDIASIZE returns bytes. */
479 disk_blocks /= BD(dev).bd_sectorsize;
481 /* We should not get here. Just try to survive. */
482 disk_blocks = BD(dev).bd_sectors - dev->d_offset;
485 /* Validate source block address. */
486 if (dblk < dev->d_offset || dblk >= dev->d_offset + disk_blocks)
490 * Truncate if we are crossing disk or partition end.
492 if (dblk + blks >= dev->d_offset + disk_blocks) {
493 blks = dev->d_offset + disk_blocks - dblk;
494 size = blks * BD(dev).bd_sectorsize;
495 DEBUG("short read %d", blks);
498 switch (rw & F_MASK) {
500 DEBUG("read %d from %lld to %p", blks, dblk, buf);
502 if (blks && (rc = bd_read(dev, dblk, blks, buf))) {
503 /* Filter out floppy controller errors */
504 if (BD(dev).bd_flags != BD_FLOPPY || rc != 0x20) {
505 printf("read %d from %lld to %p, error: 0x%x\n", blks, dblk,
510 #ifdef BD_SUPPORT_FRAGS /* XXX: sector size */
511 DEBUG("bd_strategy: frag read %d from %d+%d to %p",
512 fragsize, dblk, blks, buf + (blks * BIOSDISK_SECSIZE));
513 if (fragsize && bd_read(od, dblk + blks, 1, fragsize)) {
514 DEBUG("frag read error");
517 bcopy(fragbuf, buf + (blks * BIOSDISK_SECSIZE), fragsize);
521 DEBUG("write %d from %lld to %p", blks, dblk, buf);
523 if (blks && bd_write(dev, dblk, blks, buf)) {
524 DEBUG("write error");
527 #ifdef BD_SUPPORT_FRAGS
529 DEBUG("Attempted to write a frag");
545 bd_edd_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest,
548 static struct edd_packet packet;
550 packet.len = sizeof(struct edd_packet);
552 packet.off = VTOPOFF(dest);
553 packet.seg = VTOPSEG(dest);
558 /* Should we Write with verify ?? 0x4302 ? */
562 v86.edx = BD(dev).bd_unit;
563 v86.ds = VTOPSEG(&packet);
564 v86.esi = VTOPOFF(&packet);
567 return (v86.eax >> 8);
572 bd_chs_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest,
575 u_int x, bpc, cyl, hd, sec;
577 bpc = BD(dev).bd_sec * BD(dev).bd_hds; /* blocks per cylinder */
579 cyl = x / bpc; /* block # / blocks per cylinder */
580 x %= bpc; /* block offset into cylinder */
581 hd = x / BD(dev).bd_sec; /* offset / blocks per track */
582 sec = x % BD(dev).bd_sec; /* offset into track */
584 /* correct sector number for 1-based BIOS numbering */
588 /* CHS doesn't support cylinders > 1023. */
594 v86.eax = 0x300 | blks;
596 v86.eax = 0x200 | blks;
597 v86.ecx = ((cyl & 0xff) << 8) | ((cyl & 0x300) >> 2) | sec;
598 v86.edx = (hd << 8) | BD(dev).bd_unit;
599 v86.es = VTOPSEG(dest);
600 v86.ebx = VTOPOFF(dest);
603 return (v86.eax >> 8);
608 bd_io_workaround(struct disk_devdesc *dev)
610 uint8_t buf[8 * 1024];
612 bd_edd_io(dev, 0xffffffff, 1, (caddr_t)buf, 0);
617 bd_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest, int write)
619 u_int x, sec, result, resid, retry, maxfer;
622 /* Just in case some idiot actually tries to read/write -1 blocks... */
630 * Workaround for a problem with some HP ProLiant BIOS failing to work out
631 * the boot disk after installation. hrs and kuriyama discovered this
632 * problem with an HP ProLiant DL320e Gen 8 with a 3TB HDD, and discovered
633 * that an int13h call seems to cause a buffer overrun in the bios. The
634 * problem is alleviated by doing an extra read before the buggy read. It
635 * is not immediately known whether other models are similarly affected.
637 if (dblk >= 0x100000000)
638 bd_io_workaround(dev);
640 /* Decide whether we have to bounce */
641 if (VTOP(dest) >> 20 != 0 || (BD(dev).bd_unit < 0x80 &&
642 (VTOP(dest) >> 16) != (VTOP(dest +
643 blks * BD(dev).bd_sectorsize) >> 16))) {
646 * There is a 64k physical boundary somewhere in the
647 * destination buffer, or the destination buffer is above
648 * first 1MB of physical memory so we have to arrange a
649 * suitable bounce buffer. Allocate a buffer twice as large
650 * as we need to. Use the bottom half unless there is a break
651 * there, in which case we use the top half.
653 x = V86_IO_BUFFER_SIZE / BD(dev).bd_sectorsize;
654 x = min(x, (unsigned)blks);
655 bbuf = PTOV(V86_IO_BUFFER);
656 maxfer = x; /* limit transfers to bounce region size */
664 * Play it safe and don't cross track boundaries.
665 * (XXX this is probably unnecessary)
667 sec = dblk % BD(dev).bd_sec; /* offset into track */
668 x = min(BD(dev).bd_sec - sec, resid);
670 x = min(x, maxfer); /* fit bounce buffer */
672 /* where do we transfer to? */
673 xp = bbuf == NULL ? p : bbuf;
676 * Put your Data In, Put your Data out,
677 * Put your Data In, and shake it all about
679 if (write && bbuf != NULL)
680 bcopy(p, bbuf, x * BD(dev).bd_sectorsize);
683 * Loop retrying the operation a couple of times. The BIOS
686 for (retry = 0; retry < 3; retry++) {
687 /* if retrying, reset the drive */
692 v86.edx = BD(dev).bd_unit;
696 if (BD(dev).bd_flags & BD_MODEEDD1)
697 result = bd_edd_io(dev, dblk, x, xp, write);
699 result = bd_chs_io(dev, dblk, x, xp, write);
705 DEBUG("Write %d sector(s) from %p (0x%x) to %lld %s", x,
706 p, VTOP(p), dblk, result ? "failed" : "ok");
708 DEBUG("Read %d sector(s) from %lld to %p (0x%x) %s", x,
709 dblk, p, VTOP(p), result ? "failed" : "ok");
713 if (!write && bbuf != NULL)
714 bcopy(bbuf, p, x * BD(dev).bd_sectorsize);
715 p += (x * BD(dev).bd_sectorsize);
720 /* hexdump(dest, (blks * BD(dev).bd_sectorsize)); */
725 bd_read(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest)
728 return (bd_io(dev, dblk, blks, dest, 0));
732 bd_write(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest)
735 return (bd_io(dev, dblk, blks, dest, 1));
739 * Return the BIOS geometry of a given "fixed drive" in a format
740 * suitable for the legacy bootinfo structure. Since the kernel is
741 * expecting raw int 0x13/0x8 values for N_BIOS_GEOM drives, we
742 * prefer to get the information directly, rather than rely on being
743 * able to put it together from information already maintained for
744 * different purposes and for a probably different number of drives.
746 * For valid drives, the geometry is expected in the format (31..0)
747 * "000000cc cccccccc hhhhhhhh 00ssssss"; and invalid drives are
748 * indicated by returning the geometry of a "1.2M" PC-format floppy
749 * disk. And, incidentally, what is returned is not the geometry as
750 * such but the highest valid cylinder, head, and sector numbers.
753 bd_getbigeom(int bunit)
759 v86.edx = 0x80 + bunit;
763 return ((v86.ecx & 0xc0) << 18) | ((v86.ecx & 0xff00) << 8) |
764 (v86.edx & 0xff00) | (v86.ecx & 0x3f);
768 * Return a suitable dev_t value for (dev).
770 * In the case where it looks like (dev) is a SCSI disk, we allow the number of
771 * IDE disks to be specified in $num_ide_disks. There should be a Better Way.
774 bd_getdev(struct i386_devdesc *d)
776 struct disk_devdesc *dev;
783 dev = (struct disk_devdesc *)d;
784 biosdev = bd_unit2bios(dev->dd.d_unit);
785 DEBUG("unit %d BIOS device %d", dev->dd.d_unit, biosdev);
786 if (biosdev == -1) /* not a BIOS device */
788 if (disk_open(dev, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
789 BD(dev).bd_sectorsize) != 0) /* oops, not a viable device */
794 if (biosdev < 0x80) {
795 /* floppy (or emulated floppy) or ATAPI device */
796 if (bdinfo[dev->dd.d_unit].bd_type == DT_ATAPI) {
797 /* is an ATAPI disk */
800 /* is a floppy disk */
804 /* assume an IDE disk */
807 /* default root disk unit number */
808 unit = biosdev & 0x7f;
810 /* XXX a better kludge to set the root disk unit number */
811 if ((nip = getenv("root_disk_unit")) != NULL) {
812 i = strtol(nip, &cp, 0);
813 /* check for parse error */
814 if ((cp != nip) && (*cp == 0))
818 rootdev = MAKEBOOTDEV(major, dev->d_slice + 1, unit, dev->d_partition);
819 DEBUG("dev is 0x%x\n", rootdev);