Home | History | Annotate | Line # | Download | only in ata
wd.c revision 1.335.2.2
      1 /*	$NetBSD: wd.c,v 1.335.2.2 2008/03/30 14:45:03 jdc Exp $ */
      2 
      3 /*
      4  * Copyright (c) 1998, 2001 Manuel Bouyer.  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  * 3. All advertising materials mentioning features or use of this software
     15  *	must display the following acknowledgement:
     16  *  This product includes software developed by Manuel Bouyer.
     17  * 4. The name of the author may not be used to endorse or promote products
     18  *	derived from this software without specific prior written permission.
     19  *
     20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     21  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     22  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     23  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     25  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     26  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     27  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     28  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     29  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     30  */
     31 
     32 /*-
     33  * Copyright (c) 1998, 2003, 2004 The NetBSD Foundation, Inc.
     34  * All rights reserved.
     35  *
     36  * This code is derived from software contributed to The NetBSD Foundation
     37  * by Charles M. Hannum and by Onno van der Linden.
     38  *
     39  * Redistribution and use in source and binary forms, with or without
     40  * modification, are permitted provided that the following conditions
     41  * are met:
     42  * 1. Redistributions of source code must retain the above copyright
     43  *    notice, this list of conditions and the following disclaimer.
     44  * 2. Redistributions in binary form must reproduce the above copyright
     45  *    notice, this list of conditions and the following disclaimer in the
     46  *    documentation and/or other materials provided with the distribution.
     47  * 3. All advertising materials mentioning features or use of this software
     48  *    must display the following acknowledgement:
     49  *        This product includes software developed by the NetBSD
     50  *        Foundation, Inc. and its contributors.
     51  * 4. Neither the name of The NetBSD Foundation nor the names of its
     52  *    contributors may be used to endorse or promote products derived
     53  *    from this software without specific prior written permission.
     54  *
     55  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     56  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     57  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     58  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     59  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     60  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     61  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     62  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     63  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     64  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     65  * POSSIBILITY OF SUCH DAMAGE.
     66  */
     67 
     68 #include <sys/cdefs.h>
     69 __KERNEL_RCSID(0, "$NetBSD: wd.c,v 1.335.2.2 2008/03/30 14:45:03 jdc Exp $");
     70 
     71 #ifndef ATADEBUG
     72 #define ATADEBUG
     73 #endif /* ATADEBUG */
     74 
     75 #include "rnd.h"
     76 
     77 #include <sys/param.h>
     78 #include <sys/systm.h>
     79 #include <sys/kernel.h>
     80 #include <sys/conf.h>
     81 #include <sys/file.h>
     82 #include <sys/stat.h>
     83 #include <sys/ioctl.h>
     84 #include <sys/buf.h>
     85 #include <sys/bufq.h>
     86 #include <sys/uio.h>
     87 #include <sys/malloc.h>
     88 #include <sys/device.h>
     89 #include <sys/disklabel.h>
     90 #include <sys/disk.h>
     91 #include <sys/syslog.h>
     92 #include <sys/proc.h>
     93 #include <sys/vnode.h>
     94 #if NRND > 0
     95 #include <sys/rnd.h>
     96 #endif
     97 
     98 #include <machine/intr.h>
     99 #include <machine/bus.h>
    100 
    101 #include <dev/ata/atareg.h>
    102 #include <dev/ata/atavar.h>
    103 #include <dev/ata/wdvar.h>
    104 #include <dev/ic/wdcreg.h>
    105 #include <sys/ataio.h>
    106 #include "locators.h"
    107 
    108 #include <prop/proplib.h>
    109 
    110 #define	LBA48_THRESHOLD		(0xfffffff)	/* 128GB / DEV_BSIZE */
    111 
    112 #define	WDIORETRIES_SINGLE 4	/* number of retries before single-sector */
    113 #define	WDIORETRIES	5	/* number of retries before giving up */
    114 #define	RECOVERYTIME hz/2	/* time to wait before retrying a cmd */
    115 
    116 #define	WDUNIT(dev)		DISKUNIT(dev)
    117 #define	WDPART(dev)		DISKPART(dev)
    118 #define	WDMINOR(unit, part)	DISKMINOR(unit, part)
    119 #define	MAKEWDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
    120 
    121 #define	WDLABELDEV(dev)	(MAKEWDDEV(major(dev), WDUNIT(dev), RAW_PART))
    122 
    123 #define DEBUG_INTR   0x01
    124 #define DEBUG_XFERS  0x02
    125 #define DEBUG_STATUS 0x04
    126 #define DEBUG_FUNCS  0x08
    127 #define DEBUG_PROBE  0x10
    128 #ifdef ATADEBUG
    129 int wdcdebug_wd_mask = 0x0;
    130 #define ATADEBUG_PRINT(args, level) \
    131 	if (wdcdebug_wd_mask & (level)) \
    132 		printf args
    133 #else
    134 #define ATADEBUG_PRINT(args, level)
    135 #endif
    136 
    137 int	wdprobe(struct device *, struct cfdata *, void *);
    138 void	wdattach(struct device *, struct device *, void *);
    139 int	wddetach(struct device *, int);
    140 int	wdactivate(struct device *, enum devact);
    141 int	wdprint(void *, char *);
    142 void	wdperror(const struct wd_softc *);
    143 
    144 CFATTACH_DECL(wd, sizeof(struct wd_softc),
    145     wdprobe, wdattach, wddetach, wdactivate);
    146 
    147 extern struct cfdriver wd_cd;
    148 
    149 dev_type_open(wdopen);
    150 dev_type_close(wdclose);
    151 dev_type_read(wdread);
    152 dev_type_write(wdwrite);
    153 dev_type_ioctl(wdioctl);
    154 dev_type_strategy(wdstrategy);
    155 dev_type_dump(wddump);
    156 dev_type_size(wdsize);
    157 
    158 const struct bdevsw wd_bdevsw = {
    159 	wdopen, wdclose, wdstrategy, wdioctl, wddump, wdsize, D_DISK
    160 };
    161 
    162 const struct cdevsw wd_cdevsw = {
    163 	wdopen, wdclose, wdread, wdwrite, wdioctl,
    164 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
    165 };
    166 
    167 /*
    168  * Glue necessary to hook WDCIOCCOMMAND into physio
    169  */
    170 
    171 struct wd_ioctl {
    172 	LIST_ENTRY(wd_ioctl) wi_list;
    173 	struct buf wi_bp;
    174 	struct uio wi_uio;
    175 	struct iovec wi_iov;
    176 	atareq_t wi_atareq;
    177 	struct wd_softc *wi_softc;
    178 };
    179 
    180 LIST_HEAD(, wd_ioctl) wi_head;
    181 
    182 struct	wd_ioctl *wi_find(struct buf *);
    183 void	wi_free(struct wd_ioctl *);
    184 struct	wd_ioctl *wi_get(void);
    185 void	wdioctlstrategy(struct buf *);
    186 
    187 void  wdgetdefaultlabel(struct wd_softc *, struct disklabel *);
    188 void  wdgetdisklabel(struct wd_softc *);
    189 void  wdstart(void *);
    190 void  __wdstart(struct wd_softc*, struct buf *);
    191 void  wdrestart(void *);
    192 void  wddone(void *);
    193 int   wd_get_params(struct wd_softc *, u_int8_t, struct ataparams *);
    194 int   wd_standby(struct wd_softc *, int);
    195 int   wd_flushcache(struct wd_softc *, int);
    196 void  wd_shutdown(void *);
    197 
    198 int   wd_getcache(struct wd_softc *, int *);
    199 int   wd_setcache(struct wd_softc *, int);
    200 
    201 struct dkdriver wddkdriver = { wdstrategy, minphys };
    202 
    203 #ifdef HAS_BAD144_HANDLING
    204 static void bad144intern(struct wd_softc *);
    205 #endif
    206 
    207 #define	WD_QUIRK_SPLIT_MOD15_WRITE	0x0001	/* must split certain writes */
    208 #define	WD_QUIRK_FORCE_LBA48		0x0002	/* must use LBA48 commands */
    209 
    210 #define	WD_QUIRK_FMT "\20\1SPLIT_MOD15_WRITE\2FORCE_LBA48"
    211 
    212 /*
    213  * Quirk table for IDE drives.  Put more-specific matches first, since
    214  * a simple globbing routine is used for matching.
    215  */
    216 static const struct wd_quirk {
    217 	const char *wdq_match;		/* inquiry pattern to match */
    218 	int wdq_quirks;			/* drive quirks */
    219 } wd_quirk_table[] = {
    220 	/*
    221 	 * Some Seagate S-ATA drives have a PHY which can get confused
    222 	 * with the way data is packetized by some S-ATA controllers.
    223 	 *
    224 	 * The work-around is to split in two any write transfer whose
    225 	 * sector count % 15 == 1 (assuming 512 byte sectors).
    226 	 *
    227 	 * XXX This is an incomplete list.  There are at least a couple
    228 	 * XXX more model numbers.  If you have trouble with such transfers
    229 	 * XXX (8K is the most common) on Seagate S-ATA drives, please
    230 	 * XXX notify thorpej (at) NetBSD.org.
    231 	 */
    232 	{ "ST3120023AS",
    233 	  WD_QUIRK_SPLIT_MOD15_WRITE },
    234 	{ "ST380023AS",
    235 	  WD_QUIRK_SPLIT_MOD15_WRITE },
    236 
    237 	/*
    238 	 * These seagate drives seems to have issue addressing sector 0xfffffff
    239 	 * (aka LBA48_THRESHOLD) in LBA mode. The workaround is to force
    240 	 * LBA48
    241 	 * Note that we can't just change the code to always use LBA48 for
    242 	 * sector 0xfffffff, because this would break valid and working
    243 	 * setups using LBA48 drives on non-LBA48-capable controllers
    244 	 * (and it's hard to get a list of such controllers)
    245 	 */
    246 	{ "ST3160021A*",
    247 	  WD_QUIRK_FORCE_LBA48 },
    248 	{ "ST3160811A*",
    249 	  WD_QUIRK_FORCE_LBA48 },
    250 	{ "ST3160812A*",
    251 	  WD_QUIRK_FORCE_LBA48 },
    252 	{ "ST3160023A*",
    253 	  WD_QUIRK_FORCE_LBA48 },
    254 	{ "ST3160827A*",
    255 	  WD_QUIRK_FORCE_LBA48 },
    256 	/* Attempt to catch all seagate drives larger than 200GB */
    257 	{ "ST3[2-9][0-9][0-9][0-9][0-9][0-9][A-Z]*",
    258 	  WD_QUIRK_FORCE_LBA48 },
    259 	{ NULL,
    260 	  0 }
    261 };
    262 
    263 static const struct wd_quirk *
    264 wd_lookup_quirks(const char *name)
    265 {
    266 	const struct wd_quirk *wdq;
    267 	const char *estr;
    268 
    269 	for (wdq = wd_quirk_table; wdq->wdq_match != NULL; wdq++) {
    270 		/*
    271 		 * We only want exact matches (which include matches
    272 		 * against globbing characters).
    273 		 */
    274 		if (pmatch(name, wdq->wdq_match, &estr) == 2)
    275 			return (wdq);
    276 	}
    277 	return (NULL);
    278 }
    279 
    280 int
    281 wdprobe(struct device *parent, struct cfdata *match, void *aux)
    282 {
    283 	struct ata_device *adev = aux;
    284 
    285 	if (adev == NULL)
    286 		return 0;
    287 	if (adev->adev_bustype->bustype_type != SCSIPI_BUSTYPE_ATA)
    288 		return 0;
    289 
    290 	if (match->cf_loc[ATA_HLCF_DRIVE] != ATA_HLCF_DRIVE_DEFAULT &&
    291 	    match->cf_loc[ATA_HLCF_DRIVE] != adev->adev_drv_data->drive)
    292 		return 0;
    293 	return 1;
    294 }
    295 
    296 void
    297 wdattach(struct device *parent, struct device *self, void *aux)
    298 {
    299 	struct wd_softc *wd = (void *)self;
    300 	struct ata_device *adev= aux;
    301 	int i, blank;
    302 	char tbuf[41], pbuf[9], c, *p, *q;
    303 	const struct wd_quirk *wdq;
    304 	ATADEBUG_PRINT(("wdattach\n"), DEBUG_FUNCS | DEBUG_PROBE);
    305 
    306 	callout_init(&wd->sc_restart_ch);
    307 	bufq_alloc(&wd->sc_q, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
    308 #ifdef WD_SOFTBADSECT
    309 	SLIST_INIT(&wd->sc_bslist);
    310 #endif
    311 	wd->atabus = adev->adev_bustype;
    312 	wd->openings = adev->adev_openings;
    313 	wd->drvp = adev->adev_drv_data;
    314 
    315 	wd->drvp->drv_done = wddone;
    316 	wd->drvp->drv_softc = &wd->sc_dev;
    317 
    318 	aprint_naive("\n");
    319 
    320 	/* read our drive info */
    321 	if (wd_get_params(wd, AT_WAIT, &wd->sc_params) != 0) {
    322 		aprint_error("\n%s: IDENTIFY failed\n", wd->sc_dev.dv_xname);
    323 		return;
    324 	}
    325 
    326 	for (blank = 0, p = wd->sc_params.atap_model, q = tbuf, i = 0;
    327 	    i < sizeof(wd->sc_params.atap_model); i++) {
    328 		c = *p++;
    329 		if (c == '\0')
    330 			break;
    331 		if (c != ' ') {
    332 			if (blank) {
    333 				*q++ = ' ';
    334 				blank = 0;
    335 			}
    336 			*q++ = c;
    337 		} else
    338 			blank = 1;
    339 	}
    340 	*q++ = '\0';
    341 
    342 	aprint_normal(": <%s>\n", tbuf);
    343 
    344 	wdq = wd_lookup_quirks(tbuf);
    345 	if (wdq != NULL)
    346 		wd->sc_quirks = wdq->wdq_quirks;
    347 
    348 	if (wd->sc_quirks != 0) {
    349 		char sbuf[sizeof(WD_QUIRK_FMT) + 64];
    350 		bitmask_snprintf(wd->sc_quirks, WD_QUIRK_FMT,
    351 		    sbuf, sizeof(sbuf));
    352 		aprint_normal("%s: quirks %s\n", wd->sc_dev.dv_xname, sbuf);
    353 	}
    354 
    355 	if ((wd->sc_params.atap_multi & 0xff) > 1) {
    356 		wd->sc_multi = wd->sc_params.atap_multi & 0xff;
    357 	} else {
    358 		wd->sc_multi = 1;
    359 	}
    360 
    361 	aprint_normal("%s: drive supports %d-sector PIO transfers,",
    362 	    wd->sc_dev.dv_xname, wd->sc_multi);
    363 
    364 	/* 48-bit LBA addressing */
    365 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0)
    366 		wd->sc_flags |= WDF_LBA48;
    367 
    368 	/* Prior to ATA-4, LBA was optional. */
    369 	if ((wd->sc_params.atap_capabilities1 & WDC_CAP_LBA) != 0)
    370 		wd->sc_flags |= WDF_LBA;
    371 #if 0
    372 	/* ATA-4 requires LBA. */
    373 	if (wd->sc_params.atap_ataversion != 0xffff &&
    374 	    wd->sc_params.atap_ataversion >= WDC_VER_ATA4)
    375 		wd->sc_flags |= WDF_LBA;
    376 #endif
    377 
    378 	if ((wd->sc_flags & WDF_LBA48) != 0) {
    379 		aprint_normal(" LBA48 addressing\n");
    380 		wd->sc_capacity =
    381 		    ((u_int64_t) wd->sc_params.__reserved6[11] << 48) |
    382 		    ((u_int64_t) wd->sc_params.__reserved6[10] << 32) |
    383 		    ((u_int64_t) wd->sc_params.__reserved6[9]  << 16) |
    384 		    ((u_int64_t) wd->sc_params.__reserved6[8]  << 0);
    385 	} else if ((wd->sc_flags & WDF_LBA) != 0) {
    386 		aprint_normal(" LBA addressing\n");
    387 		wd->sc_capacity =
    388 		    ((u_int64_t)wd->sc_params.atap_capacity[1] << 16) |
    389 		    wd->sc_params.atap_capacity[0];
    390 	} else {
    391 		aprint_normal(" chs addressing\n");
    392 		wd->sc_capacity =
    393 		    wd->sc_params.atap_cylinders *
    394 		    wd->sc_params.atap_heads *
    395 		    wd->sc_params.atap_sectors;
    396 	}
    397 	format_bytes(pbuf, sizeof(pbuf), wd->sc_capacity * DEV_BSIZE);
    398 	aprint_normal("%s: %s, %d cyl, %d head, %d sec, "
    399 	    "%d bytes/sect x %llu sectors\n",
    400 	    self->dv_xname, pbuf,
    401 	    (wd->sc_flags & WDF_LBA) ? (int)(wd->sc_capacity /
    402 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors)) :
    403 		wd->sc_params.atap_cylinders,
    404 	    wd->sc_params.atap_heads, wd->sc_params.atap_sectors,
    405 	    DEV_BSIZE, (unsigned long long)wd->sc_capacity);
    406 
    407 	ATADEBUG_PRINT(("%s: atap_dmatiming_mimi=%d, atap_dmatiming_recom=%d\n",
    408 	    self->dv_xname, wd->sc_params.atap_dmatiming_mimi,
    409 	    wd->sc_params.atap_dmatiming_recom), DEBUG_PROBE);
    410 	/*
    411 	 * Initialize and attach the disk structure.
    412 	 */
    413 	wd->sc_dk.dk_driver = &wddkdriver;
    414 	wd->sc_dk.dk_name = wd->sc_dev.dv_xname;
    415 	/* we fill in dk_info later */
    416 	disk_attach(&wd->sc_dk);
    417 	wd->sc_wdc_bio.lp = wd->sc_dk.dk_label;
    418 	wd->sc_sdhook = shutdownhook_establish(wd_shutdown, wd);
    419 	if (wd->sc_sdhook == NULL)
    420 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
    421 		    wd->sc_dev.dv_xname);
    422 #if NRND > 0
    423 	rnd_attach_source(&wd->rnd_source, wd->sc_dev.dv_xname,
    424 			  RND_TYPE_DISK, 0);
    425 #endif
    426 
    427 	/* Discover wedges on this disk. */
    428 	dkwedge_discover(&wd->sc_dk);
    429 }
    430 
    431 int
    432 wdactivate(struct device *self, enum devact act)
    433 {
    434 	int rv = 0;
    435 
    436 	switch (act) {
    437 	case DVACT_ACTIVATE:
    438 		rv = EOPNOTSUPP;
    439 		break;
    440 
    441 	case DVACT_DEACTIVATE:
    442 		/*
    443 		 * Nothing to do; we key off the device's DVF_ACTIVATE.
    444 		 */
    445 		break;
    446 	}
    447 	return (rv);
    448 }
    449 
    450 int
    451 wddetach(struct device *self, int flags)
    452 {
    453 	struct wd_softc *sc = (struct wd_softc *)self;
    454 	int s, bmaj, cmaj, i, mn;
    455 
    456 	/* locate the major number */
    457 	bmaj = bdevsw_lookup_major(&wd_bdevsw);
    458 	cmaj = cdevsw_lookup_major(&wd_cdevsw);
    459 
    460 	/* Nuke the vnodes for any open instances. */
    461 	for (i = 0; i < MAXPARTITIONS; i++) {
    462 		mn = WDMINOR(device_unit(self), i);
    463 		vdevgone(bmaj, mn, mn, VBLK);
    464 		vdevgone(cmaj, mn, mn, VCHR);
    465 	}
    466 
    467 	/* Delete all of our wedges. */
    468 	dkwedge_delall(&sc->sc_dk);
    469 
    470 	s = splbio();
    471 
    472 	/* Kill off any queued buffers. */
    473 	bufq_drain(sc->sc_q);
    474 
    475 	bufq_free(sc->sc_q);
    476 	sc->atabus->ata_killpending(sc->drvp);
    477 
    478 	splx(s);
    479 
    480 	/* Detach disk. */
    481 	disk_detach(&sc->sc_dk);
    482 
    483 #ifdef WD_SOFTBADSECT
    484 	/* Clean out the bad sector list */
    485 	while (!SLIST_EMPTY(&sc->sc_bslist)) {
    486 		void *head = SLIST_FIRST(&sc->sc_bslist);
    487 		SLIST_REMOVE_HEAD(&sc->sc_bslist, dbs_next);
    488 		free(head, M_TEMP);
    489 	}
    490 	sc->sc_bscount = 0;
    491 #endif
    492 
    493 	/* Get rid of the shutdown hook. */
    494 	if (sc->sc_sdhook != NULL)
    495 		shutdownhook_disestablish(sc->sc_sdhook);
    496 
    497 #if NRND > 0
    498 	/* Unhook the entropy source. */
    499 	rnd_detach_source(&sc->rnd_source);
    500 #endif
    501 
    502 	sc->drvp->drive_flags = 0; /* no drive any more here */
    503 
    504 	return (0);
    505 }
    506 
    507 /*
    508  * Read/write routine for a buffer.  Validates the arguments and schedules the
    509  * transfer.  Does not wait for the transfer to complete.
    510  */
    511 void
    512 wdstrategy(struct buf *bp)
    513 {
    514 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(bp->b_dev));
    515 	struct disklabel *lp = wd->sc_dk.dk_label;
    516 	daddr_t blkno;
    517 	int s;
    518 
    519 	ATADEBUG_PRINT(("wdstrategy (%s)\n", wd->sc_dev.dv_xname),
    520 	    DEBUG_XFERS);
    521 
    522 	/* Valid request?  */
    523 	if (bp->b_blkno < 0 ||
    524 	    (bp->b_bcount % lp->d_secsize) != 0 ||
    525 	    (bp->b_bcount / lp->d_secsize) >= (1 << NBBY)) {
    526 		bp->b_error = EINVAL;
    527 		goto bad;
    528 	}
    529 
    530 	/* If device invalidated (e.g. media change, door open), error. */
    531 	if ((wd->sc_flags & WDF_LOADED) == 0) {
    532 		bp->b_error = EIO;
    533 		goto bad;
    534 	}
    535 
    536 	/* If it's a null transfer, return immediately. */
    537 	if (bp->b_bcount == 0)
    538 		goto done;
    539 
    540 	/*
    541 	 * Do bounds checking, adjust transfer. if error, process.
    542 	 * If end of partition, just return.
    543 	 */
    544 	if (WDPART(bp->b_dev) == RAW_PART) {
    545 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
    546 		    wd->sc_capacity) <= 0)
    547 			goto done;
    548 	} else {
    549 		if (bounds_check_with_label(&wd->sc_dk, bp,
    550 		    (wd->sc_flags & (WDF_WLABEL|WDF_LABELLING)) != 0) <= 0)
    551 			goto done;
    552 	}
    553 
    554 	/*
    555 	 * Now convert the block number to absolute and put it in
    556 	 * terms of the device's logical block size.
    557 	 */
    558 	if (lp->d_secsize >= DEV_BSIZE)
    559 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    560 	else
    561 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    562 
    563 	if (WDPART(bp->b_dev) != RAW_PART)
    564 		blkno += lp->d_partitions[WDPART(bp->b_dev)].p_offset;
    565 
    566 	bp->b_rawblkno = blkno;
    567 
    568 #ifdef WD_SOFTBADSECT
    569 	/*
    570 	 * If the transfer about to be attempted contains only a block that
    571 	 * is known to be bad then return an error for the transfer without
    572 	 * even attempting to start a transfer up under the premis that we
    573 	 * will just end up doing more retries for a transfer that will end
    574 	 * up failing again.
    575 	 * XXX:SMP - mutex required to protect with DIOCBSFLUSH
    576 	 */
    577 	if (__predict_false(!SLIST_EMPTY(&wd->sc_bslist))) {
    578 		struct disk_badsectors *dbs;
    579 		daddr_t maxblk = blkno + (bp->b_bcount >> DEV_BSHIFT) - 1;
    580 
    581 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next)
    582 			if ((dbs->dbs_min <= blkno && blkno <= dbs->dbs_max) ||
    583 			    (dbs->dbs_min <= maxblk && maxblk <= dbs->dbs_max)){
    584 				bp->b_error = EIO;
    585 				goto bad;
    586 			}
    587 	}
    588 #endif
    589 
    590 	/* Queue transfer on drive, activate drive and controller if idle. */
    591 	s = splbio();
    592 	BUFQ_PUT(wd->sc_q, bp);
    593 	wdstart(wd);
    594 	splx(s);
    595 	return;
    596 bad:
    597 	bp->b_flags |= B_ERROR;
    598 done:
    599 	/* Toss transfer; we're done early. */
    600 	bp->b_resid = bp->b_bcount;
    601 	biodone(bp);
    602 }
    603 
    604 /*
    605  * Queue a drive for I/O.
    606  */
    607 void
    608 wdstart(void *arg)
    609 {
    610 	struct wd_softc *wd = arg;
    611 	struct buf *bp = NULL;
    612 
    613 	ATADEBUG_PRINT(("wdstart %s\n", wd->sc_dev.dv_xname),
    614 	    DEBUG_XFERS);
    615 	while (wd->openings > 0) {
    616 
    617 		/* Is there a buf for us ? */
    618 		if ((bp = BUFQ_GET(wd->sc_q)) == NULL)
    619 			return;
    620 
    621 		/*
    622 		 * Make the command. First lock the device
    623 		 */
    624 		wd->openings--;
    625 
    626 		wd->retries = 0;
    627 		__wdstart(wd, bp);
    628 	}
    629 }
    630 
    631 static void
    632 wd_split_mod15_write(struct buf *bp)
    633 {
    634 	struct buf *obp = bp->b_private;
    635 	struct wd_softc *sc = wd_cd.cd_devs[DISKUNIT(obp->b_dev)];
    636 
    637 	if (__predict_false(bp->b_flags & B_ERROR) != 0) {
    638 		/*
    639 		 * Propagate the error.  If this was the first half of
    640 		 * the original transfer, make sure to account for that
    641 		 * in the residual.
    642 		 */
    643 		if (bp->b_data == obp->b_data)
    644 			bp->b_resid += bp->b_bcount;
    645 		goto done;
    646 	}
    647 
    648 	/*
    649 	 * If this was the second half of the transfer, we're all done!
    650 	 */
    651 	if (bp->b_data != obp->b_data)
    652 		goto done;
    653 
    654 	/*
    655 	 * Advance the pointer to the second half and issue that command
    656 	 * using the same opening.
    657 	 */
    658 	bp->b_flags = obp->b_flags | B_CALL;
    659 	bp->b_data += bp->b_bcount;
    660 	bp->b_blkno += (bp->b_bcount / 512);
    661 	bp->b_rawblkno += (bp->b_bcount / 512);
    662 	__wdstart(sc, bp);
    663 	return;
    664 
    665  done:
    666 	obp->b_flags |= bp->b_flags & B_ERROR;
    667 	obp->b_error = bp->b_error;
    668 	obp->b_resid = bp->b_resid;
    669 	putiobuf(bp);
    670 	biodone(obp);
    671 	sc->openings++;
    672 	/* wddone() will call wdstart() */
    673 }
    674 
    675 void
    676 __wdstart(struct wd_softc *wd, struct buf *bp)
    677 {
    678 
    679 	/*
    680 	 * Deal with the "split mod15 write" quirk.  We just divide the
    681 	 * transfer in two, doing the first half and then then second half
    682 	 * with the same command opening.
    683 	 *
    684 	 * Note we MUST do this here, because we can't let insertion
    685 	 * into the bufq cause the transfers to be re-merged.
    686 	 */
    687 	if (__predict_false((wd->sc_quirks & WD_QUIRK_SPLIT_MOD15_WRITE) != 0 &&
    688 			    (bp->b_flags & B_READ) == 0 &&
    689 			    bp->b_bcount > 512 &&
    690 			    ((bp->b_bcount / 512) % 15) == 1)) {
    691 		struct buf *nbp;
    692 
    693 		/* already at splbio */
    694 		nbp = getiobuf_nowait();
    695 		if (__predict_false(nbp == NULL)) {
    696 			/* No memory -- fail the iop. */
    697 			bp->b_error = ENOMEM;
    698 			bp->b_flags |= B_ERROR;
    699 			bp->b_resid = bp->b_bcount;
    700 			biodone(bp);
    701 			wd->openings++;
    702 			return;
    703 		}
    704 
    705 		nbp->b_error = 0;
    706 		nbp->b_proc = bp->b_proc;
    707 		nbp->b_vp = NULLVP;
    708 		nbp->b_dev = bp->b_dev;
    709 
    710 		nbp->b_bcount = bp->b_bcount / 2;
    711 		nbp->b_bufsize = bp->b_bcount / 2;
    712 		nbp->b_data = bp->b_data;
    713 
    714 		nbp->b_blkno = bp->b_blkno;
    715 		nbp->b_rawblkno = bp->b_rawblkno;
    716 
    717 		nbp->b_flags = bp->b_flags | B_CALL;
    718 		nbp->b_iodone = wd_split_mod15_write;
    719 
    720 		/* Put ptr to orig buf in b_private and use new buf */
    721 		nbp->b_private = bp;
    722 
    723 		BIO_COPYPRIO(nbp, bp);
    724 
    725 		bp = nbp;
    726 	}
    727 
    728 	wd->sc_wdc_bio.blkno = bp->b_rawblkno;
    729 	wd->sc_wdc_bio.blkdone =0;
    730 	wd->sc_bp = bp;
    731 	/*
    732 	 * If we're retrying, retry in single-sector mode. This will give us
    733 	 * the sector number of the problem, and will eventually allow the
    734 	 * transfer to succeed.
    735 	 */
    736 	if (wd->retries >= WDIORETRIES_SINGLE)
    737 		wd->sc_wdc_bio.flags = ATA_SINGLE;
    738 	else
    739 		wd->sc_wdc_bio.flags = 0;
    740 	if (wd->sc_flags & WDF_LBA48 &&
    741 	    (wd->sc_wdc_bio.blkno > LBA48_THRESHOLD ||
    742 	    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) != 0))
    743 		wd->sc_wdc_bio.flags |= ATA_LBA48;
    744 	if (wd->sc_flags & WDF_LBA)
    745 		wd->sc_wdc_bio.flags |= ATA_LBA;
    746 	if (bp->b_flags & B_READ)
    747 		wd->sc_wdc_bio.flags |= ATA_READ;
    748 	wd->sc_wdc_bio.bcount = bp->b_bcount;
    749 	wd->sc_wdc_bio.databuf = bp->b_data;
    750 	/* Instrumentation. */
    751 	disk_busy(&wd->sc_dk);
    752 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
    753 	case ATACMD_TRY_AGAIN:
    754 		callout_reset(&wd->sc_restart_ch, hz, wdrestart, wd);
    755 		break;
    756 	case ATACMD_QUEUED:
    757 	case ATACMD_COMPLETE:
    758 		break;
    759 	default:
    760 		panic("__wdstart: bad return code from ata_bio()");
    761 	}
    762 }
    763 
    764 void
    765 wddone(void *v)
    766 {
    767 	struct wd_softc *wd = v;
    768 	struct buf *bp = wd->sc_bp;
    769 	const char *errmsg;
    770 	int do_perror = 0;
    771 	ATADEBUG_PRINT(("wddone %s\n", wd->sc_dev.dv_xname),
    772 	    DEBUG_XFERS);
    773 	int nblks;
    774 
    775 	if (bp == NULL)
    776 		return;
    777 	bp->b_resid = wd->sc_wdc_bio.bcount;
    778 	switch (wd->sc_wdc_bio.error) {
    779 	case ERR_DMA:
    780 		errmsg = "DMA error";
    781 		goto retry;
    782 	case ERR_DF:
    783 		errmsg = "device fault";
    784 		goto retry;
    785 	case TIMEOUT:
    786 		errmsg = "device timeout";
    787 		goto retry;
    788 	case ERR_RESET:
    789 		errmsg = "channel reset";
    790 		goto retry2;
    791 	case ERROR:
    792 		/* Don't care about media change bits */
    793 		if (wd->sc_wdc_bio.r_error != 0 &&
    794 		    (wd->sc_wdc_bio.r_error & ~(WDCE_MC | WDCE_MCR)) == 0)
    795 			goto noerror;
    796 		errmsg = "error";
    797 		do_perror = 1;
    798 		if ((wd->sc_wdc_bio.r_error & (WDCE_IDNF | WDCE_ABRT)) &&
    799 		    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) == 0) {
    800 			nblks = wd->sc_wdc_bio.bcount /
    801 			    wd->sc_dk.dk_label->d_secsize;
    802 			/*
    803 			 * If we get a "id not found" when crossing the
    804 			 * LBA48_THRESHOLD, and the drive is larger than
    805 			 * 128GB, then we can assume the drive has the
    806 			 * LBA48 bug and we switch to LBA48.
    807 			 */
    808 			if (wd->sc_wdc_bio.blkno <= LBA48_THRESHOLD &&
    809 			    wd->sc_wdc_bio.blkno + nblks > LBA48_THRESHOLD &&
    810 			    wd->sc_capacity > LBA48_THRESHOLD + 1) {
    811 				errmsg = "LBA48 bug";
    812 				wd->sc_quirks |= WD_QUIRK_FORCE_LBA48;
    813 				do_perror = 0;
    814 				goto retry2;
    815 			}
    816 		}
    817 retry:		/* Just reset and retry. Can we do more ? */
    818 		(*wd->atabus->ata_reset_drive)(wd->drvp, AT_RST_NOCMD);
    819 retry2:
    820 		diskerr(bp, "wd", errmsg, LOG_PRINTF,
    821 		    wd->sc_wdc_bio.blkdone, wd->sc_dk.dk_label);
    822 		if (wd->retries < WDIORETRIES)
    823 			printf(", retrying");
    824 		printf("\n");
    825 		if (do_perror)
    826 			wdperror(wd);
    827 		if (wd->retries < WDIORETRIES) {
    828 			wd->retries++;
    829 			callout_reset(&wd->sc_restart_ch, RECOVERYTIME,
    830 			    wdrestart, wd);
    831 			return;
    832 		}
    833 
    834 #ifdef WD_SOFTBADSECT
    835 		/*
    836 		 * Not all errors indicate a failed block but those that do,
    837 		 * put the block on the bad-block list for the device.  Only
    838 		 * do this for reads because the drive should do it for writes,
    839 		 * itself, according to Manuel.
    840 		 */
    841 		if ((bp->b_flags & B_READ) &&
    842 		    ((wd->drvp->ata_vers >= 4 && wd->sc_wdc_bio.r_error & 64) ||
    843 	     	     (wd->drvp->ata_vers < 4 && wd->sc_wdc_bio.r_error & 192))) {
    844 			struct disk_badsectors *dbs;
    845 
    846 			dbs = malloc(sizeof *dbs, M_TEMP, M_WAITOK);
    847 			dbs->dbs_min = bp->b_rawblkno;
    848 			dbs->dbs_max = dbs->dbs_min + (bp->b_bcount >> DEV_BSHIFT) - 1;
    849 			microtime(&dbs->dbs_failedat);
    850 			SLIST_INSERT_HEAD(&wd->sc_bslist, dbs, dbs_next);
    851 			wd->sc_bscount++;
    852 		}
    853 #endif
    854 		bp->b_flags |= B_ERROR;
    855 		bp->b_error = EIO;
    856 		break;
    857 	case NOERROR:
    858 noerror:	if ((wd->sc_wdc_bio.flags & ATA_CORR) || wd->retries > 0)
    859 			printf("%s: soft error (corrected)\n",
    860 			    wd->sc_dev.dv_xname);
    861 		break;
    862 	case ERR_NODEV:
    863 		bp->b_flags |= B_ERROR;
    864 		bp->b_error = EIO;
    865 		break;
    866 	}
    867 	disk_unbusy(&wd->sc_dk, (bp->b_bcount - bp->b_resid),
    868 	    (bp->b_flags & B_READ));
    869 #if NRND > 0
    870 	rnd_add_uint32(&wd->rnd_source, bp->b_blkno);
    871 #endif
    872 	/* XXX Yuck, but we don't want to increment openings in this case */
    873 	if (__predict_false((bp->b_flags & B_CALL) != 0 &&
    874 			    bp->b_iodone == wd_split_mod15_write))
    875 		biodone(bp);
    876 	else {
    877 		biodone(bp);
    878 		wd->openings++;
    879 	}
    880 	wdstart(wd);
    881 }
    882 
    883 void
    884 wdrestart(void *v)
    885 {
    886 	struct wd_softc *wd = v;
    887 	struct buf *bp = wd->sc_bp;
    888 	int s;
    889 	ATADEBUG_PRINT(("wdrestart %s\n", wd->sc_dev.dv_xname),
    890 	    DEBUG_XFERS);
    891 
    892 	s = splbio();
    893 	__wdstart(v, bp);
    894 	splx(s);
    895 }
    896 
    897 int
    898 wdread(dev_t dev, struct uio *uio, int flags)
    899 {
    900 
    901 	ATADEBUG_PRINT(("wdread\n"), DEBUG_XFERS);
    902 	return (physio(wdstrategy, NULL, dev, B_READ, minphys, uio));
    903 }
    904 
    905 int
    906 wdwrite(dev_t dev, struct uio *uio, int flags)
    907 {
    908 
    909 	ATADEBUG_PRINT(("wdwrite\n"), DEBUG_XFERS);
    910 	return (physio(wdstrategy, NULL, dev, B_WRITE, minphys, uio));
    911 }
    912 
    913 int
    914 wdopen(dev_t dev, int flag, int fmt, struct lwp *l)
    915 {
    916 	struct wd_softc *wd;
    917 	int part, error;
    918 
    919 	ATADEBUG_PRINT(("wdopen\n"), DEBUG_FUNCS);
    920 	wd = device_lookup(&wd_cd, WDUNIT(dev));
    921 	if (wd == NULL)
    922 		return (ENXIO);
    923 
    924 	if (! device_is_active(&wd->sc_dev))
    925 		return (ENODEV);
    926 
    927 	part = WDPART(dev);
    928 
    929 	if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
    930 		return (error);
    931 
    932 	/*
    933 	 * If there are wedges, and this is not RAW_PART, then we
    934 	 * need to fail.
    935 	 */
    936 	if (wd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
    937 		error = EBUSY;
    938 		goto bad1;
    939 	}
    940 
    941 	/*
    942 	 * If this is the first open of this device, add a reference
    943 	 * to the adapter.
    944 	 */
    945 	if (wd->sc_dk.dk_openmask == 0 &&
    946 	    (error = wd->atabus->ata_addref(wd->drvp)) != 0)
    947 		goto bad1;
    948 
    949 	if (wd->sc_dk.dk_openmask != 0) {
    950 		/*
    951 		 * If any partition is open, but the disk has been invalidated,
    952 		 * disallow further opens.
    953 		 */
    954 		if ((wd->sc_flags & WDF_LOADED) == 0) {
    955 			error = EIO;
    956 			goto bad2;
    957 		}
    958 	} else {
    959 		if ((wd->sc_flags & WDF_LOADED) == 0) {
    960 			wd->sc_flags |= WDF_LOADED;
    961 
    962 			/* Load the physical device parameters. */
    963 			wd_get_params(wd, AT_WAIT, &wd->sc_params);
    964 
    965 			/* Load the partition info if not already loaded. */
    966 			wdgetdisklabel(wd);
    967 		}
    968 	}
    969 
    970 	/* Check that the partition exists. */
    971 	if (part != RAW_PART &&
    972 	    (part >= wd->sc_dk.dk_label->d_npartitions ||
    973 	     wd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
    974 		error = ENXIO;
    975 		goto bad2;
    976 	}
    977 
    978 	/* Insure only one open at a time. */
    979 	switch (fmt) {
    980 	case S_IFCHR:
    981 		wd->sc_dk.dk_copenmask |= (1 << part);
    982 		break;
    983 	case S_IFBLK:
    984 		wd->sc_dk.dk_bopenmask |= (1 << part);
    985 		break;
    986 	}
    987 	wd->sc_dk.dk_openmask =
    988 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
    989 
    990 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
    991 	return 0;
    992 
    993  bad2:
    994 	if (wd->sc_dk.dk_openmask == 0)
    995 		wd->atabus->ata_delref(wd->drvp);
    996  bad1:
    997 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
    998 	return error;
    999 }
   1000 
   1001 int
   1002 wdclose(dev_t dev, int flag, int fmt, struct lwp *l)
   1003 {
   1004 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
   1005 	int part = WDPART(dev);
   1006 	int error;
   1007 
   1008 	ATADEBUG_PRINT(("wdclose\n"), DEBUG_FUNCS);
   1009 
   1010 	if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
   1011 		return error;
   1012 
   1013 	switch (fmt) {
   1014 	case S_IFCHR:
   1015 		wd->sc_dk.dk_copenmask &= ~(1 << part);
   1016 		break;
   1017 	case S_IFBLK:
   1018 		wd->sc_dk.dk_bopenmask &= ~(1 << part);
   1019 		break;
   1020 	}
   1021 	wd->sc_dk.dk_openmask =
   1022 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
   1023 
   1024 	if (wd->sc_dk.dk_openmask == 0) {
   1025 		wd_flushcache(wd, AT_WAIT);
   1026 
   1027 		if (! (wd->sc_flags & WDF_KLABEL))
   1028 			wd->sc_flags &= ~WDF_LOADED;
   1029 
   1030 		wd->atabus->ata_delref(wd->drvp);
   1031 	}
   1032 
   1033 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
   1034 	return 0;
   1035 }
   1036 
   1037 void
   1038 wdgetdefaultlabel(struct wd_softc *wd, struct disklabel *lp)
   1039 {
   1040 
   1041 	ATADEBUG_PRINT(("wdgetdefaultlabel\n"), DEBUG_FUNCS);
   1042 	memset(lp, 0, sizeof(struct disklabel));
   1043 
   1044 	lp->d_secsize = DEV_BSIZE;
   1045 	lp->d_ntracks = wd->sc_params.atap_heads;
   1046 	lp->d_nsectors = wd->sc_params.atap_sectors;
   1047 	lp->d_ncylinders = (wd->sc_flags & WDF_LBA) ? wd->sc_capacity /
   1048 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors) :
   1049 		wd->sc_params.atap_cylinders;
   1050 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
   1051 
   1052 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
   1053 		lp->d_type = DTYPE_ST506;
   1054 	else
   1055 		lp->d_type = DTYPE_ESDI;
   1056 
   1057 	strncpy(lp->d_typename, wd->sc_params.atap_model, 16);
   1058 	strncpy(lp->d_packname, "fictitious", 16);
   1059 	if (wd->sc_capacity > UINT32_MAX)
   1060 		lp->d_secperunit = UINT32_MAX;
   1061 	else
   1062 		lp->d_secperunit = wd->sc_capacity;
   1063 	lp->d_rpm = 3600;
   1064 	lp->d_interleave = 1;
   1065 	lp->d_flags = 0;
   1066 
   1067 	lp->d_partitions[RAW_PART].p_offset = 0;
   1068 	lp->d_partitions[RAW_PART].p_size =
   1069 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
   1070 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
   1071 	lp->d_npartitions = RAW_PART + 1;
   1072 
   1073 	lp->d_magic = DISKMAGIC;
   1074 	lp->d_magic2 = DISKMAGIC;
   1075 	lp->d_checksum = dkcksum(lp);
   1076 }
   1077 
   1078 /*
   1079  * Fabricate a default disk label, and try to read the correct one.
   1080  */
   1081 void
   1082 wdgetdisklabel(struct wd_softc *wd)
   1083 {
   1084 	struct disklabel *lp = wd->sc_dk.dk_label;
   1085 	const char *errstring;
   1086 	int s;
   1087 
   1088 	ATADEBUG_PRINT(("wdgetdisklabel\n"), DEBUG_FUNCS);
   1089 
   1090 	memset(wd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
   1091 
   1092 	wdgetdefaultlabel(wd, lp);
   1093 
   1094 	wd->sc_badsect[0] = -1;
   1095 
   1096 	if (wd->drvp->state > RESET) {
   1097 		s = splbio();
   1098 		wd->drvp->drive_flags |= DRIVE_RESET;
   1099 		splx(s);
   1100 	}
   1101 	errstring = readdisklabel(MAKEWDDEV(0, device_unit(&wd->sc_dev),
   1102 				  RAW_PART), wdstrategy, lp,
   1103 				  wd->sc_dk.dk_cpulabel);
   1104 	if (errstring) {
   1105 		/*
   1106 		 * This probably happened because the drive's default
   1107 		 * geometry doesn't match the DOS geometry.  We
   1108 		 * assume the DOS geometry is now in the label and try
   1109 		 * again.  XXX This is a kluge.
   1110 		 */
   1111 		if (wd->drvp->state > RESET) {
   1112 			s = splbio();
   1113 			wd->drvp->drive_flags |= DRIVE_RESET;
   1114 			splx(s);
   1115 		}
   1116 		errstring = readdisklabel(MAKEWDDEV(0, device_unit(&wd->sc_dev),
   1117 		    RAW_PART), wdstrategy, lp, wd->sc_dk.dk_cpulabel);
   1118 	}
   1119 	if (errstring) {
   1120 		printf("%s: %s\n", wd->sc_dev.dv_xname, errstring);
   1121 		return;
   1122 	}
   1123 
   1124 	if (wd->drvp->state > RESET) {
   1125 		s = splbio();
   1126 		wd->drvp->drive_flags |= DRIVE_RESET;
   1127 		splx(s);
   1128 	}
   1129 #ifdef HAS_BAD144_HANDLING
   1130 	if ((lp->d_flags & D_BADSECT) != 0)
   1131 		bad144intern(wd);
   1132 #endif
   1133 }
   1134 
   1135 void
   1136 wdperror(const struct wd_softc *wd)
   1137 {
   1138 	static const char *const errstr0_3[] = {"address mark not found",
   1139 	    "track 0 not found", "aborted command", "media change requested",
   1140 	    "id not found", "media changed", "uncorrectable data error",
   1141 	    "bad block detected"};
   1142 	static const char *const errstr4_5[] = {
   1143 	    "obsolete (address mark not found)",
   1144 	    "no media/write protected", "aborted command",
   1145 	    "media change requested", "id not found", "media changed",
   1146 	    "uncorrectable data error", "interface CRC error"};
   1147 	const char *const *errstr;
   1148 	int i;
   1149 	const char *sep = "";
   1150 
   1151 	const char *devname = wd->sc_dev.dv_xname;
   1152 	struct ata_drive_datas *drvp = wd->drvp;
   1153 	int errno = wd->sc_wdc_bio.r_error;
   1154 
   1155 	if (drvp->ata_vers >= 4)
   1156 		errstr = errstr4_5;
   1157 	else
   1158 		errstr = errstr0_3;
   1159 
   1160 	printf("%s: (", devname);
   1161 
   1162 	if (errno == 0)
   1163 		printf("error not notified");
   1164 
   1165 	for (i = 0; i < 8; i++) {
   1166 		if (errno & (1 << i)) {
   1167 			printf("%s%s", sep, errstr[i]);
   1168 			sep = ", ";
   1169 		}
   1170 	}
   1171 	printf(")\n");
   1172 }
   1173 
   1174 int
   1175 wdioctl(dev_t dev, u_long xfer, caddr_t addr, int flag, struct lwp *l)
   1176 {
   1177 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
   1178 	int error = 0, s;
   1179 #ifdef __HAVE_OLD_DISKLABEL
   1180 	struct disklabel *newlabel = NULL;
   1181 #endif
   1182 
   1183 	ATADEBUG_PRINT(("wdioctl\n"), DEBUG_FUNCS);
   1184 
   1185 	if ((wd->sc_flags & WDF_LOADED) == 0)
   1186 		return EIO;
   1187 
   1188 	error = disk_ioctl(&wd->sc_dk, xfer, addr, flag, l);
   1189 	if (error != EPASSTHROUGH)
   1190 		return (error);
   1191 
   1192 	switch (xfer) {
   1193 #ifdef HAS_BAD144_HANDLING
   1194 	case DIOCSBAD:
   1195 		if ((flag & FWRITE) == 0)
   1196 			return EBADF;
   1197 		wd->sc_dk.dk_cpulabel->bad = *(struct dkbad *)addr;
   1198 		wd->sc_dk.dk_label->d_flags |= D_BADSECT;
   1199 		bad144intern(wd);
   1200 		return 0;
   1201 #endif
   1202 #ifdef WD_SOFTBADSECT
   1203 	case DIOCBSLIST :
   1204 	{
   1205 		u_int32_t count, missing, skip;
   1206 		struct disk_badsecinfo dbsi;
   1207 		struct disk_badsectors *dbs;
   1208 		size_t available;
   1209 		caddr_t laddr;
   1210 
   1211 		dbsi = *(struct disk_badsecinfo *)addr;
   1212 		missing = wd->sc_bscount;
   1213 		count = 0;
   1214 		available = dbsi.dbsi_bufsize;
   1215 		skip = dbsi.dbsi_skip;
   1216 		laddr = dbsi.dbsi_buffer;
   1217 
   1218 		/*
   1219 		 * We start this loop with the expectation that all of the
   1220 		 * entries will be missed and decrement this counter each
   1221 		 * time we either skip over one (already copied out) or
   1222 		 * we actually copy it back to user space.  The structs
   1223 		 * holding the bad sector information are copied directly
   1224 		 * back to user space whilst the summary is returned via
   1225 		 * the struct passed in via the ioctl.
   1226 		 */
   1227 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next) {
   1228 			if (skip > 0) {
   1229 				missing--;
   1230 				skip--;
   1231 				continue;
   1232 			}
   1233 			if (available < sizeof(*dbs))
   1234 				break;
   1235 			available -= sizeof(*dbs);
   1236 			copyout(dbs, laddr, sizeof(*dbs));
   1237 			laddr += sizeof(*dbs);
   1238 			missing--;
   1239 			count++;
   1240 		}
   1241 		dbsi.dbsi_left = missing;
   1242 		dbsi.dbsi_copied = count;
   1243 		*(struct disk_badsecinfo *)addr = dbsi;
   1244 		return 0;
   1245 	}
   1246 
   1247 	case DIOCBSFLUSH :
   1248 		/* Clean out the bad sector list */
   1249 		while (!SLIST_EMPTY(&wd->sc_bslist)) {
   1250 			void *head = SLIST_FIRST(&wd->sc_bslist);
   1251 			SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
   1252 			free(head, M_TEMP);
   1253 		}
   1254 		wd->sc_bscount = 0;
   1255 		return 0;
   1256 #endif
   1257 	case DIOCGDINFO:
   1258 		*(struct disklabel *)addr = *(wd->sc_dk.dk_label);
   1259 		return 0;
   1260 #ifdef __HAVE_OLD_DISKLABEL
   1261 	case ODIOCGDINFO:
   1262 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
   1263 		if (newlabel == NULL)
   1264 			return EIO;
   1265 		*newlabel = *(wd->sc_dk.dk_label);
   1266 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
   1267 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
   1268 		else
   1269 			error = ENOTTY;
   1270 		free(newlabel, M_TEMP);
   1271 		return error;
   1272 #endif
   1273 
   1274 	case DIOCGPART:
   1275 		((struct partinfo *)addr)->disklab = wd->sc_dk.dk_label;
   1276 		((struct partinfo *)addr)->part =
   1277 		    &wd->sc_dk.dk_label->d_partitions[WDPART(dev)];
   1278 		return 0;
   1279 
   1280 	case DIOCWDINFO:
   1281 	case DIOCSDINFO:
   1282 #ifdef __HAVE_OLD_DISKLABEL
   1283 	case ODIOCWDINFO:
   1284 	case ODIOCSDINFO:
   1285 #endif
   1286 	{
   1287 		struct disklabel *lp;
   1288 
   1289 		if ((flag & FWRITE) == 0)
   1290 			return EBADF;
   1291 
   1292 #ifdef __HAVE_OLD_DISKLABEL
   1293 		if (xfer == ODIOCSDINFO || xfer == ODIOCWDINFO) {
   1294 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
   1295 			if (newlabel == NULL)
   1296 				return EIO;
   1297 			memset(newlabel, 0, sizeof newlabel);
   1298 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
   1299 			lp = newlabel;
   1300 		} else
   1301 #endif
   1302 		lp = (struct disklabel *)addr;
   1303 
   1304 		if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE,
   1305 				     NULL)) != 0)
   1306 			goto bad;
   1307 		wd->sc_flags |= WDF_LABELLING;
   1308 
   1309 		error = setdisklabel(wd->sc_dk.dk_label,
   1310 		    lp, /*wd->sc_dk.dk_openmask : */0,
   1311 		    wd->sc_dk.dk_cpulabel);
   1312 		if (error == 0) {
   1313 			if (wd->drvp->state > RESET) {
   1314 				s = splbio();
   1315 				wd->drvp->drive_flags |= DRIVE_RESET;
   1316 				splx(s);
   1317 			}
   1318 			if (xfer == DIOCWDINFO
   1319 #ifdef __HAVE_OLD_DISKLABEL
   1320 			    || xfer == ODIOCWDINFO
   1321 #endif
   1322 			    )
   1323 				error = writedisklabel(WDLABELDEV(dev),
   1324 				    wdstrategy, wd->sc_dk.dk_label,
   1325 				    wd->sc_dk.dk_cpulabel);
   1326 		}
   1327 
   1328 		wd->sc_flags &= ~WDF_LABELLING;
   1329 		(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
   1330 bad:
   1331 #ifdef __HAVE_OLD_DISKLABEL
   1332 		if (newlabel != NULL)
   1333 			free(newlabel, M_TEMP);
   1334 #endif
   1335 		return error;
   1336 	}
   1337 
   1338 	case DIOCKLABEL:
   1339 		if (*(int *)addr)
   1340 			wd->sc_flags |= WDF_KLABEL;
   1341 		else
   1342 			wd->sc_flags &= ~WDF_KLABEL;
   1343 		return 0;
   1344 
   1345 	case DIOCWLABEL:
   1346 		if ((flag & FWRITE) == 0)
   1347 			return EBADF;
   1348 		if (*(int *)addr)
   1349 			wd->sc_flags |= WDF_WLABEL;
   1350 		else
   1351 			wd->sc_flags &= ~WDF_WLABEL;
   1352 		return 0;
   1353 
   1354 	case DIOCGDEFLABEL:
   1355 		wdgetdefaultlabel(wd, (struct disklabel *)addr);
   1356 		return 0;
   1357 #ifdef __HAVE_OLD_DISKLABEL
   1358 	case ODIOCGDEFLABEL:
   1359 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
   1360 		if (newlabel == NULL)
   1361 			return EIO;
   1362 		wdgetdefaultlabel(wd, newlabel);
   1363 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
   1364 			memcpy(addr, &newlabel, sizeof (struct olddisklabel));
   1365 		else
   1366 			error = ENOTTY;
   1367 		free(newlabel, M_TEMP);
   1368 		return error;
   1369 #endif
   1370 
   1371 #ifdef notyet
   1372 	case DIOCWFORMAT:
   1373 		if ((flag & FWRITE) == 0)
   1374 			return EBADF;
   1375 		{
   1376 		register struct format_op *fop;
   1377 		struct iovec aiov;
   1378 		struct uio auio;
   1379 
   1380 		fop = (struct format_op *)addr;
   1381 		aiov.iov_base = fop->df_buf;
   1382 		aiov.iov_len = fop->df_count;
   1383 		auio.uio_iov = &aiov;
   1384 		auio.uio_iovcnt = 1;
   1385 		auio.uio_resid = fop->df_count;
   1386 		auio.uio_offset =
   1387 			fop->df_startblk * wd->sc_dk.dk_label->d_secsize;
   1388 		auio.uio_vmspace = l->l_proc->p_vmspace;
   1389 		error = physio(wdformat, NULL, dev, B_WRITE, minphys,
   1390 		    &auio);
   1391 		fop->df_count -= auio.uio_resid;
   1392 		fop->df_reg[0] = wdc->sc_status;
   1393 		fop->df_reg[1] = wdc->sc_error;
   1394 		return error;
   1395 		}
   1396 #endif
   1397 	case DIOCGCACHE:
   1398 		return wd_getcache(wd, (int *)addr);
   1399 
   1400 	case DIOCSCACHE:
   1401 		return wd_setcache(wd, *(int *)addr);
   1402 
   1403 	case DIOCCACHESYNC:
   1404 		return wd_flushcache(wd, AT_WAIT);
   1405 
   1406 	case ATAIOCCOMMAND:
   1407 		/*
   1408 		 * Make sure this command is (relatively) safe first
   1409 		 */
   1410 		if ((((atareq_t *) addr)->flags & ATACMD_READ) == 0 &&
   1411 		    (flag & FWRITE) == 0)
   1412 			return (EBADF);
   1413 		{
   1414 		struct wd_ioctl *wi;
   1415 		atareq_t *atareq = (atareq_t *) addr;
   1416 		int error1;
   1417 
   1418 		wi = wi_get();
   1419 		wi->wi_softc = wd;
   1420 		wi->wi_atareq = *atareq;
   1421 
   1422 		if (atareq->datalen && atareq->flags &
   1423 		    (ATACMD_READ | ATACMD_WRITE)) {
   1424 			wi->wi_iov.iov_base = atareq->databuf;
   1425 			wi->wi_iov.iov_len = atareq->datalen;
   1426 			wi->wi_uio.uio_iov = &wi->wi_iov;
   1427 			wi->wi_uio.uio_iovcnt = 1;
   1428 			wi->wi_uio.uio_resid = atareq->datalen;
   1429 			wi->wi_uio.uio_offset = 0;
   1430 			wi->wi_uio.uio_rw =
   1431 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE;
   1432 			wi->wi_uio.uio_vmspace = l->l_proc->p_vmspace;
   1433 			error1 = physio(wdioctlstrategy, &wi->wi_bp, dev,
   1434 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE,
   1435 			    minphys, &wi->wi_uio);
   1436 		} else {
   1437 			/* No need to call physio if we don't have any
   1438 			   user data */
   1439 			wi->wi_bp.b_flags = 0;
   1440 			wi->wi_bp.b_data = 0;
   1441 			wi->wi_bp.b_bcount = 0;
   1442 			wi->wi_bp.b_dev = 0;
   1443 			wi->wi_bp.b_proc = l->l_proc;
   1444 			wdioctlstrategy(&wi->wi_bp);
   1445 			error1 = wi->wi_bp.b_error;
   1446 		}
   1447 		*atareq = wi->wi_atareq;
   1448 		wi_free(wi);
   1449 		return(error1);
   1450 		}
   1451 
   1452 	case DIOCAWEDGE:
   1453 	    {
   1454 	    	struct dkwedge_info *dkw = (void *) addr;
   1455 
   1456 		if ((flag & FWRITE) == 0)
   1457 			return (EBADF);
   1458 
   1459 		/* If the ioctl happens here, the parent is us. */
   1460 		strcpy(dkw->dkw_parent, wd->sc_dev.dv_xname);
   1461 		return (dkwedge_add(dkw));
   1462 	    }
   1463 
   1464 	case DIOCDWEDGE:
   1465 	    {
   1466 	    	struct dkwedge_info *dkw = (void *) addr;
   1467 
   1468 		if ((flag & FWRITE) == 0)
   1469 			return (EBADF);
   1470 
   1471 		/* If the ioctl happens here, the parent is us. */
   1472 		strcpy(dkw->dkw_parent, wd->sc_dev.dv_xname);
   1473 		return (dkwedge_del(dkw));
   1474 	    }
   1475 
   1476 	case DIOCLWEDGES:
   1477 	    {
   1478 	    	struct dkwedge_list *dkwl = (void *) addr;
   1479 
   1480 		return (dkwedge_list(&wd->sc_dk, dkwl, l));
   1481 	    }
   1482 
   1483 	case DIOCGSTRATEGY:
   1484 	    {
   1485 		struct disk_strategy *dks = (void *)addr;
   1486 
   1487 		s = splbio();
   1488 		strlcpy(dks->dks_name, bufq_getstrategyname(wd->sc_q),
   1489 		    sizeof(dks->dks_name));
   1490 		splx(s);
   1491 		dks->dks_paramlen = 0;
   1492 
   1493 		return 0;
   1494 	    }
   1495 
   1496 	case DIOCSSTRATEGY:
   1497 	    {
   1498 		struct disk_strategy *dks = (void *)addr;
   1499 		struct bufq_state *new;
   1500 		struct bufq_state *old;
   1501 
   1502 		if ((flag & FWRITE) == 0) {
   1503 			return EBADF;
   1504 		}
   1505 		if (dks->dks_param != NULL) {
   1506 			return EINVAL;
   1507 		}
   1508 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
   1509 		error = bufq_alloc(&new, dks->dks_name,
   1510 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
   1511 		if (error) {
   1512 			return error;
   1513 		}
   1514 		s = splbio();
   1515 		old = wd->sc_q;
   1516 		bufq_move(new, old);
   1517 		wd->sc_q = new;
   1518 		splx(s);
   1519 		bufq_free(old);
   1520 
   1521 		return 0;
   1522 	    }
   1523 
   1524 	default:
   1525 		return ENOTTY;
   1526 	}
   1527 
   1528 #ifdef DIAGNOSTIC
   1529 	panic("wdioctl: impossible");
   1530 #endif
   1531 }
   1532 
   1533 #ifdef B_FORMAT
   1534 int
   1535 wdformat(struct buf *bp)
   1536 {
   1537 
   1538 	bp->b_flags |= B_FORMAT;
   1539 	return wdstrategy(bp);
   1540 }
   1541 #endif
   1542 
   1543 int
   1544 wdsize(dev_t dev)
   1545 {
   1546 	struct wd_softc *wd;
   1547 	int part, omask;
   1548 	int size;
   1549 
   1550 	ATADEBUG_PRINT(("wdsize\n"), DEBUG_FUNCS);
   1551 
   1552 	wd = device_lookup(&wd_cd, WDUNIT(dev));
   1553 	if (wd == NULL)
   1554 		return (-1);
   1555 
   1556 	part = WDPART(dev);
   1557 	omask = wd->sc_dk.dk_openmask & (1 << part);
   1558 
   1559 	if (omask == 0 && wdopen(dev, 0, S_IFBLK, NULL) != 0)
   1560 		return (-1);
   1561 	if (wd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
   1562 		size = -1;
   1563 	else
   1564 		size = wd->sc_dk.dk_label->d_partitions[part].p_size *
   1565 		    (wd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
   1566 	if (omask == 0 && wdclose(dev, 0, S_IFBLK, NULL) != 0)
   1567 		return (-1);
   1568 	return (size);
   1569 }
   1570 
   1571 /* #define WD_DUMP_NOT_TRUSTED if you just want to watch */
   1572 static int wddoingadump = 0;
   1573 static int wddumprecalibrated = 0;
   1574 
   1575 /*
   1576  * Dump core after a system crash.
   1577  */
   1578 int
   1579 wddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
   1580 {
   1581 	struct wd_softc *wd;	/* disk unit to do the I/O */
   1582 	struct disklabel *lp;   /* disk's disklabel */
   1583 	int part, err;
   1584 	int nblks;	/* total number of sectors left to write */
   1585 
   1586 	/* Check if recursive dump; if so, punt. */
   1587 	if (wddoingadump)
   1588 		return EFAULT;
   1589 	wddoingadump = 1;
   1590 
   1591 	wd = device_lookup(&wd_cd, WDUNIT(dev));
   1592 	if (wd == NULL)
   1593 		return (ENXIO);
   1594 
   1595 	part = WDPART(dev);
   1596 
   1597 	/* Convert to disk sectors.  Request must be a multiple of size. */
   1598 	lp = wd->sc_dk.dk_label;
   1599 	if ((size % lp->d_secsize) != 0)
   1600 		return EFAULT;
   1601 	nblks = size / lp->d_secsize;
   1602 	blkno = blkno / (lp->d_secsize / DEV_BSIZE);
   1603 
   1604 	/* Check transfer bounds against partition size. */
   1605 	if ((blkno < 0) || ((blkno + nblks) > lp->d_partitions[part].p_size))
   1606 		return EINVAL;
   1607 
   1608 	/* Offset block number to start of partition. */
   1609 	blkno += lp->d_partitions[part].p_offset;
   1610 
   1611 	/* Recalibrate, if first dump transfer. */
   1612 	if (wddumprecalibrated == 0) {
   1613 		wddumprecalibrated = 1;
   1614 		(*wd->atabus->ata_reset_drive)(wd->drvp,
   1615 					       AT_POLL | AT_RST_EMERG);
   1616 		wd->drvp->state = RESET;
   1617 	}
   1618 
   1619 	wd->sc_bp = NULL;
   1620 	wd->sc_wdc_bio.blkno = blkno;
   1621 	wd->sc_wdc_bio.flags = ATA_POLL;
   1622 	if (wd->sc_flags & WDF_LBA48 &&
   1623 	    (blkno > LBA48_THRESHOLD ||
   1624     	    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) != 0))
   1625 		wd->sc_wdc_bio.flags |= ATA_LBA48;
   1626 	if (wd->sc_flags & WDF_LBA)
   1627 		wd->sc_wdc_bio.flags |= ATA_LBA;
   1628 	wd->sc_wdc_bio.bcount = nblks * lp->d_secsize;
   1629 	wd->sc_wdc_bio.databuf = va;
   1630 #ifndef WD_DUMP_NOT_TRUSTED
   1631 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
   1632 	case ATACMD_TRY_AGAIN:
   1633 		panic("wddump: try again");
   1634 		break;
   1635 	case ATACMD_QUEUED:
   1636 		panic("wddump: polled command has been queued");
   1637 		break;
   1638 	case ATACMD_COMPLETE:
   1639 		break;
   1640 	}
   1641 	switch(wd->sc_wdc_bio.error) {
   1642 	case TIMEOUT:
   1643 		printf("wddump: device timed out");
   1644 		err = EIO;
   1645 		break;
   1646 	case ERR_DF:
   1647 		printf("wddump: drive fault");
   1648 		err = EIO;
   1649 		break;
   1650 	case ERR_DMA:
   1651 		printf("wddump: DMA error");
   1652 		err = EIO;
   1653 		break;
   1654 	case ERROR:
   1655 		printf("wddump: ");
   1656 		wdperror(wd);
   1657 		err = EIO;
   1658 		break;
   1659 	case NOERROR:
   1660 		err = 0;
   1661 		break;
   1662 	default:
   1663 		panic("wddump: unknown error type");
   1664 	}
   1665 	if (err != 0) {
   1666 		printf("\n");
   1667 		return err;
   1668 	}
   1669 #else	/* WD_DUMP_NOT_TRUSTED */
   1670 	/* Let's just talk about this first... */
   1671 	printf("wd%d: dump addr 0x%x, cylin %d, head %d, sector %d\n",
   1672 	    unit, va, cylin, head, sector);
   1673 	delay(500 * 1000);	/* half a second */
   1674 #endif
   1675 
   1676 	wddoingadump = 0;
   1677 	return 0;
   1678 }
   1679 
   1680 #ifdef HAS_BAD144_HANDLING
   1681 /*
   1682  * Internalize the bad sector table.
   1683  */
   1684 void
   1685 bad144intern(struct wd_softc *wd)
   1686 {
   1687 	struct dkbad *bt = &wd->sc_dk.dk_cpulabel->bad;
   1688 	struct disklabel *lp = wd->sc_dk.dk_label;
   1689 	int i = 0;
   1690 
   1691 	ATADEBUG_PRINT(("bad144intern\n"), DEBUG_XFERS);
   1692 
   1693 	for (; i < NBT_BAD; i++) {
   1694 		if (bt->bt_bad[i].bt_cyl == 0xffff)
   1695 			break;
   1696 		wd->sc_badsect[i] =
   1697 		    bt->bt_bad[i].bt_cyl * lp->d_secpercyl +
   1698 		    (bt->bt_bad[i].bt_trksec >> 8) * lp->d_nsectors +
   1699 		    (bt->bt_bad[i].bt_trksec & 0xff);
   1700 	}
   1701 	for (; i < NBT_BAD+1; i++)
   1702 		wd->sc_badsect[i] = -1;
   1703 }
   1704 #endif
   1705 
   1706 static void
   1707 wd_params_to_properties(struct wd_softc *wd, struct ataparams *params)
   1708 {
   1709 	prop_dictionary_t disk_info, odisk_info, geom;
   1710 	const char *cp;
   1711 
   1712 	disk_info = prop_dictionary_create();
   1713 
   1714 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
   1715 		cp = "ST506";
   1716 	else {
   1717 		/* XXX Should have a case for ATA here, too. */
   1718 		cp = "ESDI";
   1719 	}
   1720 	prop_dictionary_set_cstring_nocopy(disk_info, "type", cp);
   1721 
   1722 	geom = prop_dictionary_create();
   1723 
   1724 	prop_dictionary_set_uint64(geom, "sectors-per-unit", wd->sc_capacity);
   1725 
   1726 	prop_dictionary_set_uint32(geom, "sector-size",
   1727 				   DEV_BSIZE /* XXX 512? */);
   1728 
   1729 	prop_dictionary_set_uint16(geom, "sectors-per-track",
   1730 				   wd->sc_params.atap_sectors);
   1731 
   1732 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
   1733 				   wd->sc_params.atap_heads);
   1734 
   1735 	if (wd->sc_flags & WDF_LBA)
   1736 		prop_dictionary_set_uint64(geom, "cylinders-per-unit",
   1737 					   wd->sc_capacity /
   1738 					       (wd->sc_params.atap_heads *
   1739 					        wd->sc_params.atap_sectors));
   1740 	else
   1741 		prop_dictionary_set_uint16(geom, "cylinders-per-unit",
   1742 					   wd->sc_params.atap_cylinders);
   1743 
   1744 	prop_dictionary_set(disk_info, "geometry", geom);
   1745 	prop_object_release(geom);
   1746 
   1747 	prop_dictionary_set(device_properties(&wd->sc_dev),
   1748 			    "disk-info", disk_info);
   1749 
   1750 	/*
   1751 	 * Don't release disk_info here; we keep a reference to it.
   1752 	 * disk_detach() will release it when we go away.
   1753 	 */
   1754 
   1755 	odisk_info = wd->sc_dk.dk_info;
   1756 	wd->sc_dk.dk_info = disk_info;
   1757 	if (odisk_info)
   1758 		prop_object_release(odisk_info);
   1759 }
   1760 
   1761 int
   1762 wd_get_params(struct wd_softc *wd, u_int8_t flags, struct ataparams *params)
   1763 {
   1764 	switch (wd->atabus->ata_get_params(wd->drvp, flags, params)) {
   1765 	case CMD_AGAIN:
   1766 		return 1;
   1767 	case CMD_ERR:
   1768 		/*
   1769 		 * We `know' there's a drive here; just assume it's old.
   1770 		 * This geometry is only used to read the MBR and print a
   1771 		 * (false) attach message.
   1772 		 */
   1773 		strncpy(params->atap_model, "ST506",
   1774 		    sizeof params->atap_model);
   1775 		params->atap_config = ATA_CFG_FIXED;
   1776 		params->atap_cylinders = 1024;
   1777 		params->atap_heads = 8;
   1778 		params->atap_sectors = 17;
   1779 		params->atap_multi = 1;
   1780 		params->atap_capabilities1 = params->atap_capabilities2 = 0;
   1781 		wd->drvp->ata_vers = -1; /* Mark it as pre-ATA */
   1782 		/* FALLTHROUGH */
   1783 	case CMD_OK:
   1784 		wd_params_to_properties(wd, params);
   1785 		return 0;
   1786 	default:
   1787 		panic("wd_get_params: bad return code from ata_get_params");
   1788 		/* NOTREACHED */
   1789 	}
   1790 }
   1791 
   1792 int
   1793 wd_getcache(struct wd_softc *wd, int *bitsp)
   1794 {
   1795 	struct ataparams params;
   1796 
   1797 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
   1798 		return EIO;
   1799 	if (params.atap_cmd_set1 == 0x0000 ||
   1800 	    params.atap_cmd_set1 == 0xffff ||
   1801 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0) {
   1802 		*bitsp = 0;
   1803 		return 0;
   1804 	}
   1805 	*bitsp = DKCACHE_WCHANGE | DKCACHE_READ;
   1806 	if (params.atap_cmd1_en & WDC_CMD1_CACHE)
   1807 		*bitsp |= DKCACHE_WRITE;
   1808 
   1809 	return 0;
   1810 }
   1811 
   1812 const char at_errbits[] = "\20\10ERROR\11TIMEOU\12DF";
   1813 
   1814 int
   1815 wd_setcache(struct wd_softc *wd, int bits)
   1816 {
   1817 	struct ataparams params;
   1818 	struct ata_command ata_c;
   1819 
   1820 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
   1821 		return EIO;
   1822 
   1823 	if (params.atap_cmd_set1 == 0x0000 ||
   1824 	    params.atap_cmd_set1 == 0xffff ||
   1825 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0)
   1826 		return EOPNOTSUPP;
   1827 
   1828 	if ((bits & DKCACHE_READ) == 0 ||
   1829 	    (bits & DKCACHE_SAVE) != 0)
   1830 		return EOPNOTSUPP;
   1831 
   1832 	memset(&ata_c, 0, sizeof(struct ata_command));
   1833 	ata_c.r_command = SET_FEATURES;
   1834 	ata_c.r_st_bmask = 0;
   1835 	ata_c.r_st_pmask = 0;
   1836 	ata_c.timeout = 30000; /* 30s timeout */
   1837 	ata_c.flags = AT_WAIT;
   1838 	if (bits & DKCACHE_WRITE)
   1839 		ata_c.r_features = WDSF_WRITE_CACHE_EN;
   1840 	else
   1841 		ata_c.r_features = WDSF_WRITE_CACHE_DS;
   1842 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
   1843 		printf("%s: wd_setcache command not complete\n",
   1844 		    wd->sc_dev.dv_xname);
   1845 		return EIO;
   1846 	}
   1847 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   1848 		char sbuf[sizeof(at_errbits) + 64];
   1849 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
   1850 		printf("%s: wd_setcache: status=%s\n", wd->sc_dev.dv_xname,
   1851 		    sbuf);
   1852 		return EIO;
   1853 	}
   1854 	return 0;
   1855 }
   1856 
   1857 int
   1858 wd_standby(struct wd_softc *wd, int flags)
   1859 {
   1860 	struct ata_command ata_c;
   1861 
   1862 	memset(&ata_c, 0, sizeof(struct ata_command));
   1863 	ata_c.r_command = WDCC_STANDBY_IMMED;
   1864 	ata_c.r_st_bmask = WDCS_DRDY;
   1865 	ata_c.r_st_pmask = WDCS_DRDY;
   1866 	ata_c.flags = flags;
   1867 	ata_c.timeout = 30000; /* 30s timeout */
   1868 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
   1869 		printf("%s: standby immediate command didn't complete\n",
   1870 		    wd->sc_dev.dv_xname);
   1871 		return EIO;
   1872 	}
   1873 	if (ata_c.flags & AT_ERROR) {
   1874 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
   1875 			return ENODEV;
   1876 	}
   1877 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   1878 		char sbuf[sizeof(at_errbits) + 64];
   1879 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
   1880 		printf("%s: wd_standby: status=%s\n", wd->sc_dev.dv_xname,
   1881 		    sbuf);
   1882 		return EIO;
   1883 	}
   1884 	return 0;
   1885 }
   1886 
   1887 int
   1888 wd_flushcache(struct wd_softc *wd, int flags)
   1889 {
   1890 	struct ata_command ata_c;
   1891 
   1892 	/*
   1893 	 * WDCC_FLUSHCACHE is here since ATA-4, but some drives report
   1894 	 * only ATA-2 and still support it.
   1895 	 */
   1896 	if (wd->drvp->ata_vers < 4 &&
   1897 	    ((wd->sc_params.atap_cmd_set2 & WDC_CMD2_FC) == 0 ||
   1898 	    wd->sc_params.atap_cmd_set2 == 0xffff))
   1899 		return ENODEV;
   1900 	memset(&ata_c, 0, sizeof(struct ata_command));
   1901 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0 &&
   1902 	    (wd->sc_params.atap_cmd2_en & ATA_CMD2_FCE) != 0)
   1903 		ata_c.r_command = WDCC_FLUSHCACHE_EXT;
   1904 	else
   1905 		ata_c.r_command = WDCC_FLUSHCACHE;
   1906 	ata_c.r_st_bmask = WDCS_DRDY;
   1907 	ata_c.r_st_pmask = WDCS_DRDY;
   1908 	ata_c.flags = flags;
   1909 	ata_c.timeout = 30000; /* 30s timeout */
   1910 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
   1911 		printf("%s: flush cache command didn't complete\n",
   1912 		    wd->sc_dev.dv_xname);
   1913 		return EIO;
   1914 	}
   1915 	if (ata_c.flags & AT_ERROR) {
   1916 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
   1917 			return ENODEV;
   1918 	}
   1919 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   1920 		char sbuf[sizeof(at_errbits) + 64];
   1921 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
   1922 		printf("%s: wd_flushcache: status=%s\n", wd->sc_dev.dv_xname,
   1923 		    sbuf);
   1924 		return EIO;
   1925 	}
   1926 	return 0;
   1927 }
   1928 
   1929 void
   1930 wd_shutdown(void *arg)
   1931 {
   1932 	struct wd_softc *wd = arg;
   1933 	wd_flushcache(wd, AT_POLL);
   1934 }
   1935 
   1936 /*
   1937  * Allocate space for a ioctl queue structure.  Mostly taken from
   1938  * scsipi_ioctl.c
   1939  */
   1940 struct wd_ioctl *
   1941 wi_get(void)
   1942 {
   1943 	struct wd_ioctl *wi;
   1944 	int s;
   1945 
   1946 	wi = malloc(sizeof(struct wd_ioctl), M_TEMP, M_WAITOK|M_ZERO);
   1947 	simple_lock_init(&wi->wi_bp.b_interlock);
   1948 	s = splbio();
   1949 	LIST_INSERT_HEAD(&wi_head, wi, wi_list);
   1950 	splx(s);
   1951 	return (wi);
   1952 }
   1953 
   1954 /*
   1955  * Free an ioctl structure and remove it from our list
   1956  */
   1957 
   1958 void
   1959 wi_free(struct wd_ioctl *wi)
   1960 {
   1961 	int s;
   1962 
   1963 	s = splbio();
   1964 	LIST_REMOVE(wi, wi_list);
   1965 	splx(s);
   1966 	free(wi, M_TEMP);
   1967 }
   1968 
   1969 /*
   1970  * Find a wd_ioctl structure based on the struct buf.
   1971  */
   1972 
   1973 struct wd_ioctl *
   1974 wi_find(struct buf *bp)
   1975 {
   1976 	struct wd_ioctl *wi;
   1977 	int s;
   1978 
   1979 	s = splbio();
   1980 	for (wi = wi_head.lh_first; wi != 0; wi = wi->wi_list.le_next)
   1981 		if (bp == &wi->wi_bp)
   1982 			break;
   1983 	splx(s);
   1984 	return (wi);
   1985 }
   1986 
   1987 /*
   1988  * Ioctl pseudo strategy routine
   1989  *
   1990  * This is mostly stolen from scsipi_ioctl.c:scsistrategy().  What
   1991  * happens here is:
   1992  *
   1993  * - wdioctl() queues a wd_ioctl structure.
   1994  *
   1995  * - wdioctl() calls physio/wdioctlstrategy based on whether or not
   1996  *   user space I/O is required.  If physio() is called, physio() eventually
   1997  *   calls wdioctlstrategy().
   1998  *
   1999  * - In either case, wdioctlstrategy() calls wd->atabus->ata_exec_command()
   2000  *   to perform the actual command
   2001  *
   2002  * The reason for the use of the pseudo strategy routine is because
   2003  * when doing I/O to/from user space, physio _really_ wants to be in
   2004  * the loop.  We could put the entire buffer into the ioctl request
   2005  * structure, but that won't scale if we want to do things like download
   2006  * microcode.
   2007  */
   2008 
   2009 void
   2010 wdioctlstrategy(struct buf *bp)
   2011 {
   2012 	struct wd_ioctl *wi;
   2013 	struct ata_command ata_c;
   2014 	int error = 0;
   2015 
   2016 	wi = wi_find(bp);
   2017 	if (wi == NULL) {
   2018 		printf("wdioctlstrategy: "
   2019 		    "No matching ioctl request found in queue\n");
   2020 		error = EINVAL;
   2021 		goto bad;
   2022 	}
   2023 
   2024 	memset(&ata_c, 0, sizeof(ata_c));
   2025 
   2026 	/*
   2027 	 * Abort if physio broke up the transfer
   2028 	 */
   2029 
   2030 	if (bp->b_bcount != wi->wi_atareq.datalen) {
   2031 		printf("physio split wd ioctl request... cannot proceed\n");
   2032 		error = EIO;
   2033 		goto bad;
   2034 	}
   2035 
   2036 	/*
   2037 	 * Abort if we didn't get a buffer size that was a multiple of
   2038 	 * our sector size (or was larger than NBBY)
   2039 	 */
   2040 
   2041 	if ((bp->b_bcount % wi->wi_softc->sc_dk.dk_label->d_secsize) != 0 ||
   2042 	    (bp->b_bcount / wi->wi_softc->sc_dk.dk_label->d_secsize) >=
   2043 	     (1 << NBBY)) {
   2044 		error = EINVAL;
   2045 		goto bad;
   2046 	}
   2047 
   2048 	/*
   2049 	 * Make sure a timeout was supplied in the ioctl request
   2050 	 */
   2051 
   2052 	if (wi->wi_atareq.timeout == 0) {
   2053 		error = EINVAL;
   2054 		goto bad;
   2055 	}
   2056 
   2057 	if (wi->wi_atareq.flags & ATACMD_READ)
   2058 		ata_c.flags |= AT_READ;
   2059 	else if (wi->wi_atareq.flags & ATACMD_WRITE)
   2060 		ata_c.flags |= AT_WRITE;
   2061 
   2062 	if (wi->wi_atareq.flags & ATACMD_READREG)
   2063 		ata_c.flags |= AT_READREG;
   2064 
   2065 	ata_c.flags |= AT_WAIT;
   2066 
   2067 	ata_c.timeout = wi->wi_atareq.timeout;
   2068 	ata_c.r_command = wi->wi_atareq.command;
   2069 	ata_c.r_head = wi->wi_atareq.head & 0x0f;
   2070 	ata_c.r_cyl = wi->wi_atareq.cylinder;
   2071 	ata_c.r_sector = wi->wi_atareq.sec_num;
   2072 	ata_c.r_count = wi->wi_atareq.sec_count;
   2073 	ata_c.r_features = wi->wi_atareq.features;
   2074 	ata_c.r_st_bmask = WDCS_DRDY;
   2075 	ata_c.r_st_pmask = WDCS_DRDY;
   2076 	ata_c.data = wi->wi_bp.b_data;
   2077 	ata_c.bcount = wi->wi_bp.b_bcount;
   2078 
   2079 	if (wi->wi_softc->atabus->ata_exec_command(wi->wi_softc->drvp, &ata_c)
   2080 	    != ATACMD_COMPLETE) {
   2081 		wi->wi_atareq.retsts = ATACMD_ERROR;
   2082 		goto bad;
   2083 	}
   2084 
   2085 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   2086 		if (ata_c.flags & AT_ERROR) {
   2087 			wi->wi_atareq.retsts = ATACMD_ERROR;
   2088 			wi->wi_atareq.error = ata_c.r_error;
   2089 		} else if (ata_c.flags & AT_DF)
   2090 			wi->wi_atareq.retsts = ATACMD_DF;
   2091 		else
   2092 			wi->wi_atareq.retsts = ATACMD_TIMEOUT;
   2093 	} else {
   2094 		wi->wi_atareq.retsts = ATACMD_OK;
   2095 		if (wi->wi_atareq.flags & ATACMD_READREG) {
   2096 			wi->wi_atareq.head = ata_c.r_head ;
   2097 			wi->wi_atareq.cylinder = ata_c.r_cyl;
   2098 			wi->wi_atareq.sec_num = ata_c.r_sector;
   2099 			wi->wi_atareq.sec_count = ata_c.r_count;
   2100 			wi->wi_atareq.features = ata_c.r_features;
   2101 			wi->wi_atareq.error = ata_c.r_error;
   2102 		}
   2103 	}
   2104 
   2105 	bp->b_error = 0;
   2106 	biodone(bp);
   2107 	return;
   2108 bad:
   2109 	bp->b_flags |= B_ERROR;
   2110 	bp->b_error = error;
   2111 	biodone(bp);
   2112 }
   2113