Home | History | Annotate | Line # | Download | only in ata
wd.c revision 1.281
      1 /*	$NetBSD: wd.c,v 1.281 2004/07/31 21:26:43 bouyer 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 1.281 2004/07/31 21:26:43 bouyer Exp $");
     70 
     71 #ifndef WDCDEBUG
     72 #define WDCDEBUG
     73 #endif /* WDCDEBUG */
     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/uio.h>
     86 #include <sys/malloc.h>
     87 #include <sys/device.h>
     88 #include <sys/disklabel.h>
     89 #include <sys/disk.h>
     90 #include <sys/syslog.h>
     91 #include <sys/proc.h>
     92 #include <sys/vnode.h>
     93 #if NRND > 0
     94 #include <sys/rnd.h>
     95 #endif
     96 
     97 #include <machine/intr.h>
     98 #include <machine/bus.h>
     99 
    100 #include <dev/ata/atareg.h>
    101 #include <dev/ata/atavar.h>
    102 #include <dev/ata/wdvar.h>
    103 #include <dev/ic/wdcreg.h>
    104 #include <sys/ataio.h>
    105 #include "locators.h"
    106 
    107 #define	LBA48_THRESHOLD		(0xfffffff)	/* 128GB / DEV_BSIZE */
    108 
    109 #define	WDIORETRIES_SINGLE 4	/* number of retries before single-sector */
    110 #define	WDIORETRIES	5	/* number of retries before giving up */
    111 #define	RECOVERYTIME hz/2	/* time to wait before retrying a cmd */
    112 
    113 #define	WDUNIT(dev)		DISKUNIT(dev)
    114 #define	WDPART(dev)		DISKPART(dev)
    115 #define	WDMINOR(unit, part)	DISKMINOR(unit, part)
    116 #define	MAKEWDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
    117 
    118 #define	WDLABELDEV(dev)	(MAKEWDDEV(major(dev), WDUNIT(dev), RAW_PART))
    119 
    120 #define DEBUG_INTR   0x01
    121 #define DEBUG_XFERS  0x02
    122 #define DEBUG_STATUS 0x04
    123 #define DEBUG_FUNCS  0x08
    124 #define DEBUG_PROBE  0x10
    125 #ifdef WDCDEBUG
    126 int wdcdebug_wd_mask = 0x0;
    127 #define WDCDEBUG_PRINT(args, level) \
    128 	if (wdcdebug_wd_mask & (level)) \
    129 		printf args
    130 #else
    131 #define WDCDEBUG_PRINT(args, level)
    132 #endif
    133 
    134 int	wdprobe(struct device *, struct cfdata *, void *);
    135 void	wdattach(struct device *, struct device *, void *);
    136 int	wddetach(struct device *, int);
    137 int	wdactivate(struct device *, enum devact);
    138 int	wdprint(void *, char *);
    139 void	wdperror(const struct wd_softc *);
    140 
    141 CFATTACH_DECL(wd, sizeof(struct wd_softc),
    142     wdprobe, wdattach, wddetach, wdactivate);
    143 
    144 extern struct cfdriver wd_cd;
    145 
    146 dev_type_open(wdopen);
    147 dev_type_close(wdclose);
    148 dev_type_read(wdread);
    149 dev_type_write(wdwrite);
    150 dev_type_ioctl(wdioctl);
    151 dev_type_strategy(wdstrategy);
    152 dev_type_dump(wddump);
    153 dev_type_size(wdsize);
    154 
    155 const struct bdevsw wd_bdevsw = {
    156 	wdopen, wdclose, wdstrategy, wdioctl, wddump, wdsize, D_DISK
    157 };
    158 
    159 const struct cdevsw wd_cdevsw = {
    160 	wdopen, wdclose, wdread, wdwrite, wdioctl,
    161 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
    162 };
    163 
    164 /*
    165  * Glue necessary to hook WDCIOCCOMMAND into physio
    166  */
    167 
    168 struct wd_ioctl {
    169 	LIST_ENTRY(wd_ioctl) wi_list;
    170 	struct buf wi_bp;
    171 	struct uio wi_uio;
    172 	struct iovec wi_iov;
    173 	atareq_t wi_atareq;
    174 	struct wd_softc *wi_softc;
    175 };
    176 
    177 LIST_HEAD(, wd_ioctl) wi_head;
    178 
    179 struct	wd_ioctl *wi_find(struct buf *);
    180 void	wi_free(struct wd_ioctl *);
    181 struct	wd_ioctl *wi_get(void);
    182 void	wdioctlstrategy(struct buf *);
    183 
    184 void  wdgetdefaultlabel(struct wd_softc *, struct disklabel *);
    185 void  wdgetdisklabel(struct wd_softc *);
    186 void  wdstart(void *);
    187 void  __wdstart(struct wd_softc*, struct buf *);
    188 void  wdrestart(void *);
    189 void  wddone(void *);
    190 int   wd_get_params(struct wd_softc *, u_int8_t, struct ataparams *);
    191 int   wd_flushcache(struct wd_softc *, int);
    192 void  wd_shutdown(void *);
    193 
    194 int   wd_getcache(struct wd_softc *, int *);
    195 int   wd_setcache(struct wd_softc *, int);
    196 
    197 struct dkdriver wddkdriver = { wdstrategy };
    198 
    199 #ifdef HAS_BAD144_HANDLING
    200 static void bad144intern(struct wd_softc *);
    201 #endif
    202 
    203 #define	WD_QUIRK_SPLIT_MOD15_WRITE	0x0001	/* must split certain writes */
    204 
    205 /*
    206  * Quirk table for IDE drives.  Put more-specific matches first, since
    207  * a simple globbing routine is used for matching.
    208  */
    209 static const struct wd_quirk {
    210 	const char *wdq_match;		/* inquiry pattern to match */
    211 	int wdq_quirks;			/* drive quirks */
    212 } wd_quirk_table[] = {
    213 	/*
    214 	 * Some Seagate S-ATA drives have a PHY which can get confused
    215 	 * with the way data is packetized by some S-ATA controllers.
    216 	 *
    217 	 * The work-around is to split in two any write transfer whose
    218 	 * sector count % 15 == 1 (assuming 512 byte sectors).
    219 	 *
    220 	 * XXX This is an incomplete list.  There are at least a couple
    221 	 * XXX more model numbers.  If you have trouble with such transfers
    222 	 * XXX (8K is the most common) on Seagate S-ATA drives, please
    223 	 * XXX notify thorpej (at) NetBSD.org.
    224 	 */
    225 	{ "ST3120023AS",
    226 	  WD_QUIRK_SPLIT_MOD15_WRITE },
    227 	{ "ST380023AS",
    228 	  WD_QUIRK_SPLIT_MOD15_WRITE },
    229 
    230 	{ NULL,
    231 	  0 }
    232 };
    233 
    234 static const struct wd_quirk *
    235 wd_lookup_quirks(const char *name)
    236 {
    237 	const struct wd_quirk *wdq;
    238 	const char *estr;
    239 
    240 	for (wdq = wd_quirk_table; wdq->wdq_match != NULL; wdq++) {
    241 		/*
    242 		 * We only want exact matches (which include matches
    243 		 * against globbing characters).
    244 		 */
    245 		if (pmatch(name, wdq->wdq_match, &estr) == 2)
    246 			return (wdq);
    247 	}
    248 	return (NULL);
    249 }
    250 
    251 int
    252 wdprobe(struct device *parent, struct cfdata *match, void *aux)
    253 {
    254 	struct ata_device *adev = aux;
    255 
    256 	if (adev == NULL)
    257 		return 0;
    258 	if (adev->adev_bustype->bustype_type != SCSIPI_BUSTYPE_ATA)
    259 		return 0;
    260 
    261 	if (match->cf_loc[ATA_HLCF_DRIVE] != ATA_HLCF_DRIVE_DEFAULT &&
    262 	    match->cf_loc[ATA_HLCF_DRIVE] != adev->adev_drv_data->drive)
    263 		return 0;
    264 	return 1;
    265 }
    266 
    267 void
    268 wdattach(struct device *parent, struct device *self, void *aux)
    269 {
    270 	struct wd_softc *wd = (void *)self;
    271 	struct ata_device *adev= aux;
    272 	int i, blank;
    273 	char buf[41], pbuf[9], c, *p, *q;
    274 	const struct wd_quirk *wdq;
    275 	WDCDEBUG_PRINT(("wdattach\n"), DEBUG_FUNCS | DEBUG_PROBE);
    276 
    277 	lockinit(&wd->sc_lock, PRIBIO | PCATCH, "wdlock", 0, 0);
    278 
    279 	callout_init(&wd->sc_restart_ch);
    280 	bufq_alloc(&wd->sc_q, BUFQ_DISK_DEFAULT_STRAT()|BUFQ_SORT_RAWBLOCK);
    281 	SLIST_INIT(&wd->sc_bslist);
    282 
    283 	wd->atabus = adev->adev_bustype;
    284 	wd->openings = adev->adev_openings;
    285 	wd->drvp = adev->adev_drv_data;
    286 
    287 	wd->drvp->drv_done = wddone;
    288 	wd->drvp->drv_softc = &wd->sc_dev;
    289 
    290 	aprint_naive("\n");
    291 
    292 	/* read our drive info */
    293 	if (wd_get_params(wd, AT_WAIT, &wd->sc_params) != 0) {
    294 		aprint_error("\n%s: IDENTIFY failed\n", wd->sc_dev.dv_xname);
    295 		return;
    296 	}
    297 
    298 	for (blank = 0, p = wd->sc_params.atap_model, q = buf, i = 0;
    299 	    i < sizeof(wd->sc_params.atap_model); i++) {
    300 		c = *p++;
    301 		if (c == '\0')
    302 			break;
    303 		if (c != ' ') {
    304 			if (blank) {
    305 				*q++ = ' ';
    306 				blank = 0;
    307 			}
    308 			*q++ = c;
    309 		} else
    310 			blank = 1;
    311 	}
    312 	*q++ = '\0';
    313 
    314 	aprint_normal(": <%s>\n", buf);
    315 
    316 	wdq = wd_lookup_quirks(buf);
    317 	if (wdq != NULL)
    318 		wd->sc_quirks = wdq->wdq_quirks;
    319 
    320 	if ((wd->sc_params.atap_multi & 0xff) > 1) {
    321 		wd->sc_multi = wd->sc_params.atap_multi & 0xff;
    322 	} else {
    323 		wd->sc_multi = 1;
    324 	}
    325 
    326 	aprint_normal("%s: drive supports %d-sector PIO transfers,",
    327 	    wd->sc_dev.dv_xname, wd->sc_multi);
    328 
    329 	/* 48-bit LBA addressing */
    330 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0)
    331 		wd->sc_flags |= WDF_LBA48;
    332 
    333 	/* Prior to ATA-4, LBA was optional. */
    334 	if ((wd->sc_params.atap_capabilities1 & WDC_CAP_LBA) != 0)
    335 		wd->sc_flags |= WDF_LBA;
    336 #if 0
    337 	/* ATA-4 requires LBA. */
    338 	if (wd->sc_params.atap_ataversion != 0xffff &&
    339 	    wd->sc_params.atap_ataversion >= WDC_VER_ATA4)
    340 		wd->sc_flags |= WDF_LBA;
    341 #endif
    342 
    343 	if ((wd->sc_flags & WDF_LBA48) != 0) {
    344 		aprint_normal(" LBA48 addressing\n");
    345 		wd->sc_capacity =
    346 		    ((u_int64_t) wd->sc_params.__reserved6[11] << 48) |
    347 		    ((u_int64_t) wd->sc_params.__reserved6[10] << 32) |
    348 		    ((u_int64_t) wd->sc_params.__reserved6[9]  << 16) |
    349 		    ((u_int64_t) wd->sc_params.__reserved6[8]  << 0);
    350 	} else if ((wd->sc_flags & WDF_LBA) != 0) {
    351 		aprint_normal(" LBA addressing\n");
    352 		wd->sc_capacity =
    353 		    (wd->sc_params.atap_capacity[1] << 16) |
    354 		    wd->sc_params.atap_capacity[0];
    355 	} else {
    356 		aprint_normal(" chs addressing\n");
    357 		wd->sc_capacity =
    358 		    wd->sc_params.atap_cylinders *
    359 		    wd->sc_params.atap_heads *
    360 		    wd->sc_params.atap_sectors;
    361 	}
    362 	format_bytes(pbuf, sizeof(pbuf), wd->sc_capacity * DEV_BSIZE);
    363 	aprint_normal("%s: %s, %d cyl, %d head, %d sec, "
    364 	    "%d bytes/sect x %llu sectors\n",
    365 	    self->dv_xname, pbuf,
    366 	    (wd->sc_flags & WDF_LBA) ? (int)(wd->sc_capacity /
    367 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors)) :
    368 		wd->sc_params.atap_cylinders,
    369 	    wd->sc_params.atap_heads, wd->sc_params.atap_sectors,
    370 	    DEV_BSIZE, (unsigned long long)wd->sc_capacity);
    371 
    372 	WDCDEBUG_PRINT(("%s: atap_dmatiming_mimi=%d, atap_dmatiming_recom=%d\n",
    373 	    self->dv_xname, wd->sc_params.atap_dmatiming_mimi,
    374 	    wd->sc_params.atap_dmatiming_recom), DEBUG_PROBE);
    375 	/*
    376 	 * Initialize and attach the disk structure.
    377 	 */
    378 	wd->sc_dk.dk_driver = &wddkdriver;
    379 	wd->sc_dk.dk_name = wd->sc_dev.dv_xname;
    380 	disk_attach(&wd->sc_dk);
    381 	wd->sc_wdc_bio.lp = wd->sc_dk.dk_label;
    382 	wd->sc_sdhook = shutdownhook_establish(wd_shutdown, wd);
    383 	if (wd->sc_sdhook == NULL)
    384 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
    385 		    wd->sc_dev.dv_xname);
    386 #if NRND > 0
    387 	rnd_attach_source(&wd->rnd_source, wd->sc_dev.dv_xname,
    388 			  RND_TYPE_DISK, 0);
    389 #endif
    390 }
    391 
    392 int
    393 wdactivate(struct device *self, enum devact act)
    394 {
    395 	int rv = 0;
    396 
    397 	switch (act) {
    398 	case DVACT_ACTIVATE:
    399 		rv = EOPNOTSUPP;
    400 		break;
    401 
    402 	case DVACT_DEACTIVATE:
    403 		/*
    404 		 * Nothing to do; we key off the device's DVF_ACTIVATE.
    405 		 */
    406 		break;
    407 	}
    408 	return (rv);
    409 }
    410 
    411 int
    412 wddetach(struct device *self, int flags)
    413 {
    414 	struct wd_softc *sc = (struct wd_softc *)self;
    415 	struct buf *bp;
    416 	int s, bmaj, cmaj, i, mn;
    417 
    418 	lockmgr(&sc->sc_lock, LK_DRAIN, NULL);
    419 
    420 	/* Clean out the bad sector list */
    421 	while (!SLIST_EMPTY(&sc->sc_bslist)) {
    422 		void *head = SLIST_FIRST(&sc->sc_bslist);
    423 		SLIST_REMOVE_HEAD(&sc->sc_bslist, dbs_next);
    424 		free(head, M_TEMP);
    425 	}
    426 	sc->sc_bscount = 0;
    427 
    428 	/* locate the major number */
    429 	bmaj = bdevsw_lookup_major(&wd_bdevsw);
    430 	cmaj = cdevsw_lookup_major(&wd_cdevsw);
    431 
    432 	s = splbio();
    433 
    434 	/* Kill off any queued buffers. */
    435 	while ((bp = BUFQ_GET(&sc->sc_q)) != NULL) {
    436 		bp->b_error = EIO;
    437 		bp->b_flags |= B_ERROR;
    438 		bp->b_resid = bp->b_bcount;
    439 		biodone(bp);
    440 	}
    441 
    442 	bufq_free(&sc->sc_q);
    443 
    444 	splx(s);
    445 
    446 	/* Nuke the vnodes for any open instances. */
    447 	for (i = 0; i < MAXPARTITIONS; i++) {
    448 		mn = WDMINOR(self->dv_unit, i);
    449 		vdevgone(bmaj, mn, mn, VBLK);
    450 		vdevgone(cmaj, mn, mn, VCHR);
    451 	}
    452 
    453 	/* Detach disk. */
    454 	disk_detach(&sc->sc_dk);
    455 
    456 	/* Get rid of the shutdown hook. */
    457 	if (sc->sc_sdhook != NULL)
    458 		shutdownhook_disestablish(sc->sc_sdhook);
    459 
    460 #if NRND > 0
    461 	/* Unhook the entropy source. */
    462 	rnd_detach_source(&sc->rnd_source);
    463 #endif
    464 
    465 	return (0);
    466 }
    467 
    468 /*
    469  * Read/write routine for a buffer.  Validates the arguments and schedules the
    470  * transfer.  Does not wait for the transfer to complete.
    471  */
    472 void
    473 wdstrategy(struct buf *bp)
    474 {
    475 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(bp->b_dev));
    476 	struct disklabel *lp = wd->sc_dk.dk_label;
    477 	daddr_t blkno;
    478 	int s;
    479 
    480 	WDCDEBUG_PRINT(("wdstrategy (%s)\n", wd->sc_dev.dv_xname),
    481 	    DEBUG_XFERS);
    482 
    483 	/* Valid request?  */
    484 	if (bp->b_blkno < 0 ||
    485 	    (bp->b_bcount % lp->d_secsize) != 0 ||
    486 	    (bp->b_bcount / lp->d_secsize) >= (1 << NBBY)) {
    487 		bp->b_error = EINVAL;
    488 		goto bad;
    489 	}
    490 
    491 	/* If device invalidated (e.g. media change, door open), error. */
    492 	if ((wd->sc_flags & WDF_LOADED) == 0) {
    493 		bp->b_error = EIO;
    494 		goto bad;
    495 	}
    496 
    497 	/* If it's a null transfer, return immediately. */
    498 	if (bp->b_bcount == 0)
    499 		goto done;
    500 
    501 	/*
    502 	 * Do bounds checking, adjust transfer. if error, process.
    503 	 * If end of partition, just return.
    504 	 */
    505 	if (WDPART(bp->b_dev) == RAW_PART) {
    506 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
    507 		    wd->sc_capacity) <= 0)
    508 			goto done;
    509 	} else {
    510 		if (bounds_check_with_label(&wd->sc_dk, bp,
    511 		    (wd->sc_flags & (WDF_WLABEL|WDF_LABELLING)) != 0) <= 0)
    512 			goto done;
    513 	}
    514 
    515 	/*
    516 	 * Now convert the block number to absolute and put it in
    517 	 * terms of the device's logical block size.
    518 	 */
    519 	if (lp->d_secsize >= DEV_BSIZE)
    520 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    521 	else
    522 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    523 
    524 	if (WDPART(bp->b_dev) != RAW_PART)
    525 		blkno += lp->d_partitions[WDPART(bp->b_dev)].p_offset;
    526 
    527 	bp->b_rawblkno = blkno;
    528 
    529 	/*
    530 	 * If the transfer about to be attempted contains only a block that
    531 	 * is known to be bad then return an error for the transfer without
    532 	 * even attempting to start a transfer up under the premis that we
    533 	 * will just end up doing more retries for a transfer that will end
    534 	 * up failing again.
    535 	 * XXX:SMP - mutex required to protect with DIOCBSFLUSH
    536 	 */
    537 	if (__predict_false(!SLIST_EMPTY(&wd->sc_bslist))) {
    538 		struct disk_badsectors *dbs;
    539 		daddr_t maxblk = blkno + (bp->b_bcount >> DEV_BSHIFT) - 1;
    540 
    541 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next)
    542 			if ((dbs->dbs_min <= blkno && blkno <= dbs->dbs_max) ||
    543 			    (dbs->dbs_min <= maxblk && maxblk <= dbs->dbs_max)){
    544 				bp->b_error = EIO;
    545 				goto bad;
    546 			}
    547 	}
    548 
    549 	/* Queue transfer on drive, activate drive and controller if idle. */
    550 	s = splbio();
    551 	BUFQ_PUT(&wd->sc_q, bp);
    552 	wdstart(wd);
    553 	splx(s);
    554 	return;
    555 bad:
    556 	bp->b_flags |= B_ERROR;
    557 done:
    558 	/* Toss transfer; we're done early. */
    559 	bp->b_resid = bp->b_bcount;
    560 	biodone(bp);
    561 }
    562 
    563 /*
    564  * Queue a drive for I/O.
    565  */
    566 void
    567 wdstart(void *arg)
    568 {
    569 	struct wd_softc *wd = arg;
    570 	struct buf *bp = NULL;
    571 
    572 	WDCDEBUG_PRINT(("wdstart %s\n", wd->sc_dev.dv_xname),
    573 	    DEBUG_XFERS);
    574 	while (wd->openings > 0) {
    575 
    576 		/* Is there a buf for us ? */
    577 		if ((bp = BUFQ_GET(&wd->sc_q)) == NULL)
    578 			return;
    579 
    580 		/*
    581 		 * Make the command. First lock the device
    582 		 */
    583 		wd->openings--;
    584 
    585 		wd->retries = 0;
    586 		__wdstart(wd, bp);
    587 	}
    588 }
    589 
    590 static void
    591 wd_split_mod15_write(struct buf *bp)
    592 {
    593 	struct buf *obp = bp->b_private;
    594 	struct wd_softc *sc = wd_cd.cd_devs[DISKUNIT(obp->b_dev)];
    595 
    596 	if (__predict_false(bp->b_flags & B_ERROR) != 0) {
    597 		/*
    598 		 * Propagate the error.  If this was the first half of
    599 		 * the original transfer, make sure to account for that
    600 		 * in the residual.
    601 		 */
    602 		if (bp->b_data == obp->b_data)
    603 			bp->b_resid += bp->b_bcount;
    604 		goto done;
    605 	}
    606 
    607 	/*
    608 	 * If this was the second half of the transfer, we're all done!
    609 	 */
    610 	if (bp->b_data != obp->b_data)
    611 		goto done;
    612 
    613 	/*
    614 	 * Advance the pointer to the second half and issue that command
    615 	 * using the same opening.
    616 	 */
    617 	bp->b_flags = obp->b_flags | B_CALL;
    618 	bp->b_data += bp->b_bcount;
    619 	bp->b_blkno += (bp->b_bcount / 512);
    620 	bp->b_rawblkno += (bp->b_bcount / 512);
    621 	__wdstart(sc, bp);
    622 	return;
    623 
    624  done:
    625 	obp->b_flags |= (bp->b_flags & (B_EINTR|B_ERROR));
    626 	obp->b_error = bp->b_error;
    627 	obp->b_resid = bp->b_resid;
    628 	pool_put(&bufpool, bp);
    629 	biodone(obp);
    630 	sc->openings++;
    631 	/* wddone() will call wdstart() */
    632 }
    633 
    634 void
    635 __wdstart(struct wd_softc *wd, struct buf *bp)
    636 {
    637 
    638 	/*
    639 	 * Deal with the "split mod15 write" quirk.  We just divide the
    640 	 * transfer in two, doing the first half and then then second half
    641 	 * with the same command opening.
    642 	 *
    643 	 * Note we MUST do this here, because we can't let insertion
    644 	 * into the bufq cause the transfers to be re-merged.
    645 	 */
    646 	if (__predict_false((wd->sc_quirks & WD_QUIRK_SPLIT_MOD15_WRITE) != 0 &&
    647 			    (bp->b_flags & B_READ) == 0 &&
    648 			    bp->b_bcount > 512 &&
    649 			    ((bp->b_bcount / 512) % 15) == 1)) {
    650 		struct buf *nbp;
    651 
    652 		/* already at splbio */
    653 		nbp = pool_get(&bufpool, PR_NOWAIT);
    654 		if (__predict_false(nbp == NULL)) {
    655 			/* No memory -- fail the iop. */
    656 			bp->b_error = ENOMEM;
    657 			bp->b_flags |= B_ERROR;
    658 			bp->b_resid = bp->b_bcount;
    659 			biodone(bp);
    660 			wd->openings++;
    661 			return;
    662 		}
    663 
    664 		BUF_INIT(nbp);
    665 		nbp->b_error = 0;
    666 		nbp->b_proc = bp->b_proc;
    667 		nbp->b_vp = NULLVP;
    668 		nbp->b_dev = bp->b_dev;
    669 
    670 		nbp->b_bcount = bp->b_bcount / 2;
    671 		nbp->b_bufsize = bp->b_bcount / 2;
    672 		nbp->b_data = bp->b_data;
    673 
    674 		nbp->b_blkno = bp->b_blkno;
    675 		nbp->b_rawblkno = bp->b_rawblkno;
    676 
    677 		nbp->b_flags = bp->b_flags | B_CALL;
    678 		nbp->b_iodone = wd_split_mod15_write;
    679 
    680 		/* Put ptr to orig buf in b_private and use new buf */
    681 		nbp->b_private = bp;
    682 
    683 		BIO_COPYPRIO(nbp, bp);
    684 
    685 		bp = nbp;
    686 	}
    687 
    688 	wd->sc_wdc_bio.blkno = bp->b_rawblkno;
    689 	wd->sc_wdc_bio.blkdone =0;
    690 	wd->sc_bp = bp;
    691 	/*
    692 	 * If we're retrying, retry in single-sector mode. This will give us
    693 	 * the sector number of the problem, and will eventually allow the
    694 	 * transfer to succeed.
    695 	 */
    696 	if (wd->retries >= WDIORETRIES_SINGLE)
    697 		wd->sc_wdc_bio.flags = ATA_SINGLE;
    698 	else
    699 		wd->sc_wdc_bio.flags = 0;
    700 	if (wd->sc_flags & WDF_LBA48 && wd->sc_wdc_bio.blkno > LBA48_THRESHOLD)
    701 		wd->sc_wdc_bio.flags |= ATA_LBA48;
    702 	if (wd->sc_flags & WDF_LBA)
    703 		wd->sc_wdc_bio.flags |= ATA_LBA;
    704 	if (bp->b_flags & B_READ)
    705 		wd->sc_wdc_bio.flags |= ATA_READ;
    706 	wd->sc_wdc_bio.bcount = bp->b_bcount;
    707 	wd->sc_wdc_bio.databuf = bp->b_data;
    708 	/* Instrumentation. */
    709 	disk_busy(&wd->sc_dk);
    710 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
    711 	case WDC_TRY_AGAIN:
    712 		callout_reset(&wd->sc_restart_ch, hz, wdrestart, wd);
    713 		break;
    714 	case WDC_QUEUED:
    715 	case WDC_COMPLETE:
    716 		break;
    717 	default:
    718 		panic("__wdstart: bad return code from ata_bio()");
    719 	}
    720 }
    721 
    722 void
    723 wddone(void *v)
    724 {
    725 	struct wd_softc *wd = v;
    726 	struct buf *bp = wd->sc_bp;
    727 	const char *errmsg;
    728 	int do_perror = 0;
    729 	WDCDEBUG_PRINT(("wddone %s\n", wd->sc_dev.dv_xname),
    730 	    DEBUG_XFERS);
    731 
    732 	if (bp == NULL)
    733 		return;
    734 	bp->b_resid = wd->sc_wdc_bio.bcount;
    735 	switch (wd->sc_wdc_bio.error) {
    736 	case ERR_DMA:
    737 		errmsg = "DMA error";
    738 		goto retry;
    739 	case ERR_DF:
    740 		errmsg = "device fault";
    741 		goto retry;
    742 	case TIMEOUT:
    743 		errmsg = "device timeout";
    744 		goto retry;
    745 	case ERR_RESET:
    746 		errmsg = "channel reset";
    747 		goto retry2;
    748 	case ERROR:
    749 		/* Don't care about media change bits */
    750 		if (wd->sc_wdc_bio.r_error != 0 &&
    751 		    (wd->sc_wdc_bio.r_error & ~(WDCE_MC | WDCE_MCR)) == 0)
    752 			goto noerror;
    753 		errmsg = "error";
    754 		do_perror = 1;
    755 retry:		/* Just reset and retry. Can we do more ? */
    756 		wd->atabus->ata_reset_channel(wd->drvp, 0);
    757 retry2:
    758 		diskerr(bp, "wd", errmsg, LOG_PRINTF,
    759 		    wd->sc_wdc_bio.blkdone, wd->sc_dk.dk_label);
    760 		if (wd->retries < WDIORETRIES)
    761 			printf(", retrying\n");
    762 		if (do_perror)
    763 			wdperror(wd);
    764 		if (wd->retries < WDIORETRIES) {
    765 			wd->retries++;
    766 			callout_reset(&wd->sc_restart_ch, RECOVERYTIME,
    767 			    wdrestart, wd);
    768 			return;
    769 		}
    770 		printf("\n");
    771 
    772 		/*
    773 		 * Not all errors indicate a failed block but those that do,
    774 		 * put the block on the bad-block list for the device.  Only
    775 		 * do this for reads because the drive should do it for writes,
    776 		 * itself, according to Manuel.
    777 		 */
    778 		if ((bp->b_flags & B_READ) &&
    779 		    ((wd->drvp->ata_vers >= 4 && wd->sc_wdc_bio.r_error & 64) ||
    780 	     	     (wd->drvp->ata_vers < 4 && wd->sc_wdc_bio.r_error & 192))) {
    781 			struct disk_badsectors *dbs;
    782 
    783 			dbs = malloc(sizeof *dbs, M_TEMP, M_WAITOK);
    784 			dbs->dbs_min = bp->b_rawblkno;
    785 			dbs->dbs_max = dbs->dbs_min + (bp->b_bcount >> DEV_BSHIFT) - 1;
    786 			microtime(&dbs->dbs_failedat);
    787 			SLIST_INSERT_HEAD(&wd->sc_bslist, dbs, dbs_next);
    788 			wd->sc_bscount++;
    789 		}
    790 
    791 		bp->b_flags |= B_ERROR;
    792 		bp->b_error = EIO;
    793 		break;
    794 	case NOERROR:
    795 noerror:	if ((wd->sc_wdc_bio.flags & ATA_CORR) || wd->retries > 0)
    796 			printf("%s: soft error (corrected)\n",
    797 			    wd->sc_dev.dv_xname);
    798 		break;
    799 	case ERR_NODEV:
    800 		bp->b_flags |= B_ERROR;
    801 		bp->b_error = EIO;
    802 		break;
    803 	}
    804 	disk_unbusy(&wd->sc_dk, (bp->b_bcount - bp->b_resid),
    805 	    (bp->b_flags & B_READ));
    806 #if NRND > 0
    807 	rnd_add_uint32(&wd->rnd_source, bp->b_blkno);
    808 #endif
    809 	/* XXX Yuck, but we don't want to increment openings in this case */
    810 	if (__predict_false((bp->b_flags & B_CALL) != 0 &&
    811 			    bp->b_iodone == wd_split_mod15_write))
    812 		biodone(bp);
    813 	else {
    814 		biodone(bp);
    815 		wd->openings++;
    816 	}
    817 	wdstart(wd);
    818 }
    819 
    820 void
    821 wdrestart(void *v)
    822 {
    823 	struct wd_softc *wd = v;
    824 	struct buf *bp = wd->sc_bp;
    825 	int s;
    826 	WDCDEBUG_PRINT(("wdrestart %s\n", wd->sc_dev.dv_xname),
    827 	    DEBUG_XFERS);
    828 
    829 	s = splbio();
    830 	__wdstart(v, bp);
    831 	splx(s);
    832 }
    833 
    834 int
    835 wdread(dev_t dev, struct uio *uio, int flags)
    836 {
    837 
    838 	WDCDEBUG_PRINT(("wdread\n"), DEBUG_XFERS);
    839 	return (physio(wdstrategy, NULL, dev, B_READ, minphys, uio));
    840 }
    841 
    842 int
    843 wdwrite(dev_t dev, struct uio *uio, int flags)
    844 {
    845 
    846 	WDCDEBUG_PRINT(("wdwrite\n"), DEBUG_XFERS);
    847 	return (physio(wdstrategy, NULL, dev, B_WRITE, minphys, uio));
    848 }
    849 
    850 int
    851 wdopen(dev_t dev, int flag, int fmt, struct proc *p)
    852 {
    853 	struct wd_softc *wd;
    854 	int part, error;
    855 
    856 	WDCDEBUG_PRINT(("wdopen\n"), DEBUG_FUNCS);
    857 	wd = device_lookup(&wd_cd, WDUNIT(dev));
    858 	if (wd == NULL)
    859 		return (ENXIO);
    860 
    861 	/*
    862 	 * If this is the first open of this device, add a reference
    863 	 * to the adapter.
    864 	 */
    865 	if (wd->sc_dk.dk_openmask == 0 &&
    866 	    (error = wd->atabus->ata_addref(wd->drvp)) != 0)
    867 		return (error);
    868 
    869 	if ((error = lockmgr(&wd->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
    870 		goto bad4;
    871 
    872 	if (wd->sc_dk.dk_openmask != 0) {
    873 		/*
    874 		 * If any partition is open, but the disk has been invalidated,
    875 		 * disallow further opens.
    876 		 */
    877 		if ((wd->sc_flags & WDF_LOADED) == 0) {
    878 			error = EIO;
    879 			goto bad3;
    880 		}
    881 	} else {
    882 		if ((wd->sc_flags & WDF_LOADED) == 0) {
    883 			wd->sc_flags |= WDF_LOADED;
    884 
    885 			/* Load the physical device parameters. */
    886 			wd_get_params(wd, AT_WAIT, &wd->sc_params);
    887 
    888 			/* Load the partition info if not already loaded. */
    889 			wdgetdisklabel(wd);
    890 		}
    891 	}
    892 
    893 	part = WDPART(dev);
    894 
    895 	/* Check that the partition exists. */
    896 	if (part != RAW_PART &&
    897 	    (part >= wd->sc_dk.dk_label->d_npartitions ||
    898 	     wd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
    899 		error = ENXIO;
    900 		goto bad;
    901 	}
    902 
    903 	/* Insure only one open at a time. */
    904 	switch (fmt) {
    905 	case S_IFCHR:
    906 		wd->sc_dk.dk_copenmask |= (1 << part);
    907 		break;
    908 	case S_IFBLK:
    909 		wd->sc_dk.dk_bopenmask |= (1 << part);
    910 		break;
    911 	}
    912 	wd->sc_dk.dk_openmask =
    913 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
    914 
    915 	lockmgr(&wd->sc_lock, LK_RELEASE, NULL);
    916 	return 0;
    917 
    918 bad:
    919 	if (wd->sc_dk.dk_openmask == 0) {
    920 	}
    921 
    922 bad3:
    923 	lockmgr(&wd->sc_lock, LK_RELEASE, NULL);
    924 bad4:
    925 	if (wd->sc_dk.dk_openmask == 0)
    926 		wd->atabus->ata_delref(wd->drvp);
    927 	return error;
    928 }
    929 
    930 int
    931 wdclose(dev_t dev, int flag, int fmt, struct proc *p)
    932 {
    933 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
    934 	int part = WDPART(dev);
    935 	int error;
    936 
    937 	WDCDEBUG_PRINT(("wdclose\n"), DEBUG_FUNCS);
    938 	if ((error = lockmgr(&wd->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
    939 		return error;
    940 
    941 	switch (fmt) {
    942 	case S_IFCHR:
    943 		wd->sc_dk.dk_copenmask &= ~(1 << part);
    944 		break;
    945 	case S_IFBLK:
    946 		wd->sc_dk.dk_bopenmask &= ~(1 << part);
    947 		break;
    948 	}
    949 	wd->sc_dk.dk_openmask =
    950 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
    951 
    952 	if (wd->sc_dk.dk_openmask == 0) {
    953 		wd_flushcache(wd, AT_WAIT);
    954 		/* XXXX Must wait for I/O to complete! */
    955 
    956 		if (! (wd->sc_flags & WDF_KLABEL))
    957 			wd->sc_flags &= ~WDF_LOADED;
    958 
    959 		wd->atabus->ata_delref(wd->drvp);
    960 	}
    961 
    962 	lockmgr(&wd->sc_lock, LK_RELEASE, NULL);
    963 	return 0;
    964 }
    965 
    966 void
    967 wdgetdefaultlabel(struct wd_softc *wd, struct disklabel *lp)
    968 {
    969 
    970 	WDCDEBUG_PRINT(("wdgetdefaultlabel\n"), DEBUG_FUNCS);
    971 	memset(lp, 0, sizeof(struct disklabel));
    972 
    973 	lp->d_secsize = DEV_BSIZE;
    974 	lp->d_ntracks = wd->sc_params.atap_heads;
    975 	lp->d_nsectors = wd->sc_params.atap_sectors;
    976 	lp->d_ncylinders = (wd->sc_flags & WDF_LBA) ? wd->sc_capacity /
    977 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors) :
    978 		wd->sc_params.atap_cylinders;
    979 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
    980 
    981 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
    982 		lp->d_type = DTYPE_ST506;
    983 	else
    984 		lp->d_type = DTYPE_ESDI;
    985 
    986 	strncpy(lp->d_typename, wd->sc_params.atap_model, 16);
    987 	strncpy(lp->d_packname, "fictitious", 16);
    988 	if (wd->sc_capacity > UINT32_MAX)
    989 		lp->d_secperunit = UINT32_MAX;
    990 	else
    991 		lp->d_secperunit = wd->sc_capacity;
    992 	lp->d_rpm = 3600;
    993 	lp->d_interleave = 1;
    994 	lp->d_flags = 0;
    995 
    996 	lp->d_partitions[RAW_PART].p_offset = 0;
    997 	lp->d_partitions[RAW_PART].p_size =
    998 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
    999 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
   1000 	lp->d_npartitions = RAW_PART + 1;
   1001 
   1002 	lp->d_magic = DISKMAGIC;
   1003 	lp->d_magic2 = DISKMAGIC;
   1004 	lp->d_checksum = dkcksum(lp);
   1005 }
   1006 
   1007 /*
   1008  * Fabricate a default disk label, and try to read the correct one.
   1009  */
   1010 void
   1011 wdgetdisklabel(struct wd_softc *wd)
   1012 {
   1013 	struct disklabel *lp = wd->sc_dk.dk_label;
   1014 	const char *errstring;
   1015 
   1016 	WDCDEBUG_PRINT(("wdgetdisklabel\n"), DEBUG_FUNCS);
   1017 
   1018 	memset(wd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
   1019 
   1020 	wdgetdefaultlabel(wd, lp);
   1021 
   1022 	wd->sc_badsect[0] = -1;
   1023 
   1024 	if (wd->drvp->state > RESET)
   1025 		wd->drvp->drive_flags |= DRIVE_RESET;
   1026 	errstring = readdisklabel(MAKEWDDEV(0, wd->sc_dev.dv_unit, RAW_PART),
   1027 	    wdstrategy, lp, wd->sc_dk.dk_cpulabel);
   1028 	if (errstring) {
   1029 		/*
   1030 		 * This probably happened because the drive's default
   1031 		 * geometry doesn't match the DOS geometry.  We
   1032 		 * assume the DOS geometry is now in the label and try
   1033 		 * again.  XXX This is a kluge.
   1034 		 */
   1035 		if (wd->drvp->state > RESET)
   1036 			wd->drvp->drive_flags |= DRIVE_RESET;
   1037 		errstring = readdisklabel(MAKEWDDEV(0, wd->sc_dev.dv_unit,
   1038 		    RAW_PART), wdstrategy, lp, wd->sc_dk.dk_cpulabel);
   1039 	}
   1040 	if (errstring) {
   1041 		printf("%s: %s\n", wd->sc_dev.dv_xname, errstring);
   1042 		return;
   1043 	}
   1044 
   1045 	if (wd->drvp->state > RESET)
   1046 		wd->drvp->drive_flags |= DRIVE_RESET;
   1047 #ifdef HAS_BAD144_HANDLING
   1048 	if ((lp->d_flags & D_BADSECT) != 0)
   1049 		bad144intern(wd);
   1050 #endif
   1051 }
   1052 
   1053 void
   1054 wdperror(const struct wd_softc *wd)
   1055 {
   1056 	static const char *const errstr0_3[] = {"address mark not found",
   1057 	    "track 0 not found", "aborted command", "media change requested",
   1058 	    "id not found", "media changed", "uncorrectable data error",
   1059 	    "bad block detected"};
   1060 	static const char *const errstr4_5[] = {
   1061 	    "obsolete (address mark not found)",
   1062 	    "no media/write protected", "aborted command",
   1063 	    "media change requested", "id not found", "media changed",
   1064 	    "uncorrectable data error", "interface CRC error"};
   1065 	const char *const *errstr;
   1066 	int i;
   1067 	char *sep = "";
   1068 
   1069 	const char *devname = wd->sc_dev.dv_xname;
   1070 	struct ata_drive_datas *drvp = wd->drvp;
   1071 	int errno = wd->sc_wdc_bio.r_error;
   1072 
   1073 	if (drvp->ata_vers >= 4)
   1074 		errstr = errstr4_5;
   1075 	else
   1076 		errstr = errstr0_3;
   1077 
   1078 	printf("%s: (", devname);
   1079 
   1080 	if (errno == 0)
   1081 		printf("error not notified");
   1082 
   1083 	for (i = 0; i < 8; i++) {
   1084 		if (errno & (1 << i)) {
   1085 			printf("%s%s", sep, errstr[i]);
   1086 			sep = ", ";
   1087 		}
   1088 	}
   1089 	printf(")\n");
   1090 }
   1091 
   1092 int
   1093 wdioctl(dev_t dev, u_long xfer, caddr_t addr, int flag, struct proc *p)
   1094 {
   1095 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
   1096 	int error = 0;
   1097 #ifdef __HAVE_OLD_DISKLABEL
   1098 	struct disklabel *newlabel = NULL;
   1099 #endif
   1100 
   1101 	WDCDEBUG_PRINT(("wdioctl\n"), DEBUG_FUNCS);
   1102 
   1103 	if ((wd->sc_flags & WDF_LOADED) == 0)
   1104 		return EIO;
   1105 
   1106 	switch (xfer) {
   1107 #ifdef HAS_BAD144_HANDLING
   1108 	case DIOCSBAD:
   1109 		if ((flag & FWRITE) == 0)
   1110 			return EBADF;
   1111 		wd->sc_dk.dk_cpulabel->bad = *(struct dkbad *)addr;
   1112 		wd->sc_dk.dk_label->d_flags |= D_BADSECT;
   1113 		bad144intern(wd);
   1114 		return 0;
   1115 #endif
   1116 
   1117 	case DIOCBSLIST :
   1118 	{
   1119 		u_int32_t count, missing, skip;
   1120 		struct disk_badsecinfo dbsi;
   1121 		struct disk_badsectors *dbs;
   1122 		size_t available;
   1123 		caddr_t laddr;
   1124 
   1125 		dbsi = *(struct disk_badsecinfo *)addr;
   1126 		missing = wd->sc_bscount;
   1127 		count = 0;
   1128 		available = dbsi.dbsi_bufsize;
   1129 		skip = dbsi.dbsi_skip;
   1130 		laddr = dbsi.dbsi_buffer;
   1131 
   1132 		/*
   1133 		 * We start this loop with the expectation that all of the
   1134 		 * entries will be missed and decrement this counter each
   1135 		 * time we either skip over one (already copied out) or
   1136 		 * we actually copy it back to user space.  The structs
   1137 		 * holding the bad sector information are copied directly
   1138 		 * back to user space whilst the summary is returned via
   1139 		 * the struct passed in via the ioctl.
   1140 		 */
   1141 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next) {
   1142 			if (skip > 0) {
   1143 				missing--;
   1144 				skip--;
   1145 				continue;
   1146 			}
   1147 			if (available < sizeof(*dbs))
   1148 				break;
   1149 			available -= sizeof(*dbs);
   1150 			copyout(dbs, laddr, sizeof(*dbs));
   1151 			laddr += sizeof(*dbs);
   1152 			missing--;
   1153 			count++;
   1154 		}
   1155 		dbsi.dbsi_left = missing;
   1156 		dbsi.dbsi_copied = count;
   1157 		*(struct disk_badsecinfo *)addr = dbsi;
   1158 		return 0;
   1159 	}
   1160 
   1161 	case DIOCBSFLUSH :
   1162 		/* Clean out the bad sector list */
   1163 		while (!SLIST_EMPTY(&wd->sc_bslist)) {
   1164 			void *head = SLIST_FIRST(&wd->sc_bslist);
   1165 			SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
   1166 			free(head, M_TEMP);
   1167 		}
   1168 		wd->sc_bscount = 0;
   1169 		return 0;
   1170 
   1171 	case DIOCGDINFO:
   1172 		*(struct disklabel *)addr = *(wd->sc_dk.dk_label);
   1173 		return 0;
   1174 #ifdef __HAVE_OLD_DISKLABEL
   1175 	case ODIOCGDINFO:
   1176 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
   1177 		if (newlabel == NULL)
   1178 			return EIO;
   1179 		*newlabel = *(wd->sc_dk.dk_label);
   1180 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
   1181 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
   1182 		else
   1183 			error = ENOTTY;
   1184 		free(newlabel, M_TEMP);
   1185 		return error;
   1186 #endif
   1187 
   1188 	case DIOCGPART:
   1189 		((struct partinfo *)addr)->disklab = wd->sc_dk.dk_label;
   1190 		((struct partinfo *)addr)->part =
   1191 		    &wd->sc_dk.dk_label->d_partitions[WDPART(dev)];
   1192 		return 0;
   1193 
   1194 	case DIOCWDINFO:
   1195 	case DIOCSDINFO:
   1196 #ifdef __HAVE_OLD_DISKLABEL
   1197 	case ODIOCWDINFO:
   1198 	case ODIOCSDINFO:
   1199 #endif
   1200 	{
   1201 		struct disklabel *lp;
   1202 
   1203 		if ((flag & FWRITE) == 0)
   1204 			return EBADF;
   1205 
   1206 #ifdef __HAVE_OLD_DISKLABEL
   1207 		if (xfer == ODIOCSDINFO || xfer == ODIOCWDINFO) {
   1208 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
   1209 			if (newlabel == NULL)
   1210 				return EIO;
   1211 			memset(newlabel, 0, sizeof newlabel);
   1212 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
   1213 			lp = newlabel;
   1214 		} else
   1215 #endif
   1216 		lp = (struct disklabel *)addr;
   1217 
   1218 		if ((error = lockmgr(&wd->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
   1219 			goto bad;
   1220 		wd->sc_flags |= WDF_LABELLING;
   1221 
   1222 		error = setdisklabel(wd->sc_dk.dk_label,
   1223 		    lp, /*wd->sc_dk.dk_openmask : */0,
   1224 		    wd->sc_dk.dk_cpulabel);
   1225 		if (error == 0) {
   1226 			if (wd->drvp->state > RESET)
   1227 				wd->drvp->drive_flags |= DRIVE_RESET;
   1228 			if (xfer == DIOCWDINFO
   1229 #ifdef __HAVE_OLD_DISKLABEL
   1230 			    || xfer == ODIOCWDINFO
   1231 #endif
   1232 			    )
   1233 				error = writedisklabel(WDLABELDEV(dev),
   1234 				    wdstrategy, wd->sc_dk.dk_label,
   1235 				    wd->sc_dk.dk_cpulabel);
   1236 		}
   1237 
   1238 		wd->sc_flags &= ~WDF_LABELLING;
   1239 		lockmgr(&wd->sc_lock, LK_RELEASE, NULL);
   1240 bad:
   1241 #ifdef __HAVE_OLD_DISKLABEL
   1242 		if (newlabel != NULL)
   1243 			free(newlabel, M_TEMP);
   1244 #endif
   1245 		return error;
   1246 	}
   1247 
   1248 	case DIOCKLABEL:
   1249 		if (*(int *)addr)
   1250 			wd->sc_flags |= WDF_KLABEL;
   1251 		else
   1252 			wd->sc_flags &= ~WDF_KLABEL;
   1253 		return 0;
   1254 
   1255 	case DIOCWLABEL:
   1256 		if ((flag & FWRITE) == 0)
   1257 			return EBADF;
   1258 		if (*(int *)addr)
   1259 			wd->sc_flags |= WDF_WLABEL;
   1260 		else
   1261 			wd->sc_flags &= ~WDF_WLABEL;
   1262 		return 0;
   1263 
   1264 	case DIOCGDEFLABEL:
   1265 		wdgetdefaultlabel(wd, (struct disklabel *)addr);
   1266 		return 0;
   1267 #ifdef __HAVE_OLD_DISKLABEL
   1268 	case ODIOCGDEFLABEL:
   1269 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
   1270 		if (newlabel == NULL)
   1271 			return EIO;
   1272 		wdgetdefaultlabel(wd, newlabel);
   1273 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
   1274 			memcpy(addr, &newlabel, sizeof (struct olddisklabel));
   1275 		else
   1276 			error = ENOTTY;
   1277 		free(newlabel, M_TEMP);
   1278 		return error;
   1279 #endif
   1280 
   1281 #ifdef notyet
   1282 	case DIOCWFORMAT:
   1283 		if ((flag & FWRITE) == 0)
   1284 			return EBADF;
   1285 		{
   1286 		register struct format_op *fop;
   1287 		struct iovec aiov;
   1288 		struct uio auio;
   1289 
   1290 		fop = (struct format_op *)addr;
   1291 		aiov.iov_base = fop->df_buf;
   1292 		aiov.iov_len = fop->df_count;
   1293 		auio.uio_iov = &aiov;
   1294 		auio.uio_iovcnt = 1;
   1295 		auio.uio_resid = fop->df_count;
   1296 		auio.uio_segflg = 0;
   1297 		auio.uio_offset =
   1298 			fop->df_startblk * wd->sc_dk.dk_label->d_secsize;
   1299 		auio.uio_procp = p;
   1300 		error = physio(wdformat, NULL, dev, B_WRITE, minphys,
   1301 		    &auio);
   1302 		fop->df_count -= auio.uio_resid;
   1303 		fop->df_reg[0] = wdc->sc_status;
   1304 		fop->df_reg[1] = wdc->sc_error;
   1305 		return error;
   1306 		}
   1307 #endif
   1308 	case DIOCGCACHE:
   1309 		return wd_getcache(wd, (int *)addr);
   1310 
   1311 	case DIOCSCACHE:
   1312 		return wd_setcache(wd, *(int *)addr);
   1313 
   1314 	case DIOCCACHESYNC:
   1315 		return wd_flushcache(wd, AT_WAIT);
   1316 
   1317 	case ATAIOCCOMMAND:
   1318 		/*
   1319 		 * Make sure this command is (relatively) safe first
   1320 		 */
   1321 		if ((((atareq_t *) addr)->flags & ATACMD_READ) == 0 &&
   1322 		    (flag & FWRITE) == 0)
   1323 			return (EBADF);
   1324 		{
   1325 		struct wd_ioctl *wi;
   1326 		atareq_t *atareq = (atareq_t *) addr;
   1327 		int error;
   1328 
   1329 		wi = wi_get();
   1330 		wi->wi_softc = wd;
   1331 		wi->wi_atareq = *atareq;
   1332 
   1333 		if (atareq->datalen && atareq->flags &
   1334 		    (ATACMD_READ | ATACMD_WRITE)) {
   1335 			wi->wi_iov.iov_base = atareq->databuf;
   1336 			wi->wi_iov.iov_len = atareq->datalen;
   1337 			wi->wi_uio.uio_iov = &wi->wi_iov;
   1338 			wi->wi_uio.uio_iovcnt = 1;
   1339 			wi->wi_uio.uio_resid = atareq->datalen;
   1340 			wi->wi_uio.uio_offset = 0;
   1341 			wi->wi_uio.uio_segflg = UIO_USERSPACE;
   1342 			wi->wi_uio.uio_rw =
   1343 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE;
   1344 			wi->wi_uio.uio_procp = p;
   1345 			error = physio(wdioctlstrategy, &wi->wi_bp, dev,
   1346 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE,
   1347 			    minphys, &wi->wi_uio);
   1348 		} else {
   1349 			/* No need to call physio if we don't have any
   1350 			   user data */
   1351 			wi->wi_bp.b_flags = 0;
   1352 			wi->wi_bp.b_data = 0;
   1353 			wi->wi_bp.b_bcount = 0;
   1354 			wi->wi_bp.b_dev = 0;
   1355 			wi->wi_bp.b_proc = p;
   1356 			wdioctlstrategy(&wi->wi_bp);
   1357 			error = wi->wi_bp.b_error;
   1358 		}
   1359 		*atareq = wi->wi_atareq;
   1360 		wi_free(wi);
   1361 		return(error);
   1362 		}
   1363 
   1364 	default:
   1365 		return ENOTTY;
   1366 	}
   1367 
   1368 #ifdef DIAGNOSTIC
   1369 	panic("wdioctl: impossible");
   1370 #endif
   1371 }
   1372 
   1373 #ifdef B_FORMAT
   1374 int
   1375 wdformat(struct buf *bp)
   1376 {
   1377 
   1378 	bp->b_flags |= B_FORMAT;
   1379 	return wdstrategy(bp);
   1380 }
   1381 #endif
   1382 
   1383 int
   1384 wdsize(dev_t dev)
   1385 {
   1386 	struct wd_softc *wd;
   1387 	int part, omask;
   1388 	int size;
   1389 
   1390 	WDCDEBUG_PRINT(("wdsize\n"), DEBUG_FUNCS);
   1391 
   1392 	wd = device_lookup(&wd_cd, WDUNIT(dev));
   1393 	if (wd == NULL)
   1394 		return (-1);
   1395 
   1396 	part = WDPART(dev);
   1397 	omask = wd->sc_dk.dk_openmask & (1 << part);
   1398 
   1399 	if (omask == 0 && wdopen(dev, 0, S_IFBLK, NULL) != 0)
   1400 		return (-1);
   1401 	if (wd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
   1402 		size = -1;
   1403 	else
   1404 		size = wd->sc_dk.dk_label->d_partitions[part].p_size *
   1405 		    (wd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
   1406 	if (omask == 0 && wdclose(dev, 0, S_IFBLK, NULL) != 0)
   1407 		return (-1);
   1408 	return (size);
   1409 }
   1410 
   1411 /* #define WD_DUMP_NOT_TRUSTED if you just want to watch */
   1412 static int wddoingadump = 0;
   1413 static int wddumprecalibrated = 0;
   1414 static int wddumpmulti = 1;
   1415 
   1416 /*
   1417  * Dump core after a system crash.
   1418  */
   1419 int
   1420 wddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
   1421 {
   1422 	struct wd_softc *wd;	/* disk unit to do the I/O */
   1423 	struct disklabel *lp;   /* disk's disklabel */
   1424 	int part, err;
   1425 	int nblks;	/* total number of sectors left to write */
   1426 
   1427 	/* Check if recursive dump; if so, punt. */
   1428 	if (wddoingadump)
   1429 		return EFAULT;
   1430 	wddoingadump = 1;
   1431 
   1432 	wd = device_lookup(&wd_cd, WDUNIT(dev));
   1433 	if (wd == NULL)
   1434 		return (ENXIO);
   1435 
   1436 	part = WDPART(dev);
   1437 
   1438 	/* Convert to disk sectors.  Request must be a multiple of size. */
   1439 	lp = wd->sc_dk.dk_label;
   1440 	if ((size % lp->d_secsize) != 0)
   1441 		return EFAULT;
   1442 	nblks = size / lp->d_secsize;
   1443 	blkno = blkno / (lp->d_secsize / DEV_BSIZE);
   1444 
   1445 	/* Check transfer bounds against partition size. */
   1446 	if ((blkno < 0) || ((blkno + nblks) > lp->d_partitions[part].p_size))
   1447 		return EINVAL;
   1448 
   1449 	/* Offset block number to start of partition. */
   1450 	blkno += lp->d_partitions[part].p_offset;
   1451 
   1452 	/* Recalibrate, if first dump transfer. */
   1453 	if (wddumprecalibrated == 0) {
   1454 		wddumpmulti = wd->sc_multi;
   1455 		wddumprecalibrated = 1;
   1456 		wd->atabus->ata_reset_channel(wd->drvp, AT_POLL | AT_RST_EMERG);
   1457 		wd->drvp->state = RESET;
   1458 	}
   1459 
   1460 	while (nblks > 0) {
   1461 		wd->sc_bp = NULL;
   1462 		wd->sc_wdc_bio.blkno = blkno;
   1463 		wd->sc_wdc_bio.flags = ATA_POLL;
   1464 		if (wd->sc_flags & WDF_LBA48 && blkno > LBA48_THRESHOLD)
   1465 			wd->sc_wdc_bio.flags |= ATA_LBA48;
   1466 		if (wd->sc_flags & WDF_LBA)
   1467 			wd->sc_wdc_bio.flags |= ATA_LBA;
   1468 		wd->sc_wdc_bio.bcount =
   1469 			min(nblks, wddumpmulti) * lp->d_secsize;
   1470 		wd->sc_wdc_bio.databuf = va;
   1471 #ifndef WD_DUMP_NOT_TRUSTED
   1472 		switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
   1473 		case WDC_TRY_AGAIN:
   1474 			panic("wddump: try again");
   1475 			break;
   1476 		case WDC_QUEUED:
   1477 			panic("wddump: polled command has been queued");
   1478 			break;
   1479 		case WDC_COMPLETE:
   1480 			break;
   1481 		}
   1482 		switch(wd->sc_wdc_bio.error) {
   1483 		case TIMEOUT:
   1484 			printf("wddump: device timed out");
   1485 			err = EIO;
   1486 			break;
   1487 		case ERR_DF:
   1488 			printf("wddump: drive fault");
   1489 			err = EIO;
   1490 			break;
   1491 		case ERR_DMA:
   1492 			printf("wddump: DMA error");
   1493 			err = EIO;
   1494 			break;
   1495 		case ERROR:
   1496 			printf("wddump: ");
   1497 			wdperror(wd);
   1498 			err = EIO;
   1499 			break;
   1500 		case NOERROR:
   1501 			err = 0;
   1502 			break;
   1503 		default:
   1504 			panic("wddump: unknown error type");
   1505 		}
   1506 		if (err != 0) {
   1507 			printf("\n");
   1508 			return err;
   1509 		}
   1510 #else	/* WD_DUMP_NOT_TRUSTED */
   1511 		/* Let's just talk about this first... */
   1512 		printf("wd%d: dump addr 0x%x, cylin %d, head %d, sector %d\n",
   1513 		    unit, va, cylin, head, sector);
   1514 		delay(500 * 1000);	/* half a second */
   1515 #endif
   1516 
   1517 		/* update block count */
   1518 		nblks -= min(nblks, wddumpmulti);
   1519 		blkno += min(nblks, wddumpmulti);
   1520 		va += min(nblks, wddumpmulti) * lp->d_secsize;
   1521 	}
   1522 
   1523 	wddoingadump = 0;
   1524 	return 0;
   1525 }
   1526 
   1527 #ifdef HAS_BAD144_HANDLING
   1528 /*
   1529  * Internalize the bad sector table.
   1530  */
   1531 void
   1532 bad144intern(struct wd_softc *wd)
   1533 {
   1534 	struct dkbad *bt = &wd->sc_dk.dk_cpulabel->bad;
   1535 	struct disklabel *lp = wd->sc_dk.dk_label;
   1536 	int i = 0;
   1537 
   1538 	WDCDEBUG_PRINT(("bad144intern\n"), DEBUG_XFERS);
   1539 
   1540 	for (; i < NBT_BAD; i++) {
   1541 		if (bt->bt_bad[i].bt_cyl == 0xffff)
   1542 			break;
   1543 		wd->sc_badsect[i] =
   1544 		    bt->bt_bad[i].bt_cyl * lp->d_secpercyl +
   1545 		    (bt->bt_bad[i].bt_trksec >> 8) * lp->d_nsectors +
   1546 		    (bt->bt_bad[i].bt_trksec & 0xff);
   1547 	}
   1548 	for (; i < NBT_BAD+1; i++)
   1549 		wd->sc_badsect[i] = -1;
   1550 }
   1551 #endif
   1552 
   1553 int
   1554 wd_get_params(struct wd_softc *wd, u_int8_t flags, struct ataparams *params)
   1555 {
   1556 	switch (wd->atabus->ata_get_params(wd->drvp, flags, params)) {
   1557 	case CMD_AGAIN:
   1558 		return 1;
   1559 	case CMD_ERR:
   1560 		/*
   1561 		 * We `know' there's a drive here; just assume it's old.
   1562 		 * This geometry is only used to read the MBR and print a
   1563 		 * (false) attach message.
   1564 		 */
   1565 		strncpy(params->atap_model, "ST506",
   1566 		    sizeof params->atap_model);
   1567 		params->atap_config = ATA_CFG_FIXED;
   1568 		params->atap_cylinders = 1024;
   1569 		params->atap_heads = 8;
   1570 		params->atap_sectors = 17;
   1571 		params->atap_multi = 1;
   1572 		params->atap_capabilities1 = params->atap_capabilities2 = 0;
   1573 		wd->drvp->ata_vers = -1; /* Mark it as pre-ATA */
   1574 		return 0;
   1575 	case CMD_OK:
   1576 		return 0;
   1577 	default:
   1578 		panic("wd_get_params: bad return code from ata_get_params");
   1579 		/* NOTREACHED */
   1580 	}
   1581 }
   1582 
   1583 int
   1584 wd_getcache(struct wd_softc *wd, int *bitsp)
   1585 {
   1586 	struct ataparams params;
   1587 
   1588 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
   1589 		return EIO;
   1590 	if (params.atap_cmd_set1 == 0x0000 ||
   1591 	    params.atap_cmd_set1 == 0xffff ||
   1592 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0) {
   1593 		*bitsp = 0;
   1594 		return 0;
   1595 	}
   1596 	*bitsp = DKCACHE_WCHANGE | DKCACHE_READ;
   1597 	if (params.atap_cmd1_en & WDC_CMD1_CACHE)
   1598 		*bitsp |= DKCACHE_WRITE;
   1599 
   1600 	return 0;
   1601 }
   1602 
   1603 int
   1604 wd_setcache(struct wd_softc *wd, int bits)
   1605 {
   1606 	struct ataparams params;
   1607 	struct wdc_command wdc_c;
   1608 
   1609 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
   1610 		return EIO;
   1611 
   1612 	if (params.atap_cmd_set1 == 0x0000 ||
   1613 	    params.atap_cmd_set1 == 0xffff ||
   1614 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0)
   1615 		return EOPNOTSUPP;
   1616 
   1617 	if ((bits & DKCACHE_READ) == 0 ||
   1618 	    (bits & DKCACHE_SAVE) != 0)
   1619 		return EOPNOTSUPP;
   1620 
   1621 	memset(&wdc_c, 0, sizeof(struct wdc_command));
   1622 	wdc_c.r_command = SET_FEATURES;
   1623 	wdc_c.r_st_bmask = 0;
   1624 	wdc_c.r_st_pmask = 0;
   1625 	wdc_c.timeout = 30000; /* 30s timeout */
   1626 	wdc_c.flags = AT_WAIT;
   1627 	if (bits & DKCACHE_WRITE)
   1628 		wdc_c.r_features = WDSF_WRITE_CACHE_EN;
   1629 	else
   1630 		wdc_c.r_features = WDSF_WRITE_CACHE_DS;
   1631 	if (wd->atabus->ata_exec_command(wd->drvp, &wdc_c) != WDC_COMPLETE) {
   1632 		printf("%s: wd_setcache command not complete\n",
   1633 		    wd->sc_dev.dv_xname);
   1634 		return EIO;
   1635 	}
   1636 	if (wdc_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   1637 		printf("%s: wd_setcache command error 0x%x\n",
   1638 		    wd->sc_dev.dv_xname, wdc_c.flags);
   1639 		return EIO;
   1640 	}
   1641 	if (wdc_c.flags & ERR_NODEV)
   1642 		return ENODEV;
   1643 	return 0;
   1644 }
   1645 
   1646 int
   1647 wd_flushcache(struct wd_softc *wd, int flags)
   1648 {
   1649 	struct wdc_command wdc_c;
   1650 
   1651 	if (wd->drvp->ata_vers < 4) /* WDCC_FLUSHCACHE is here since ATA-4 */
   1652 		return ENODEV;
   1653 	memset(&wdc_c, 0, sizeof(struct wdc_command));
   1654 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0 &&
   1655 	    (wd->sc_params.atap_cmd2_en & ATA_CMD2_FCE) != 0)
   1656 		wdc_c.r_command = WDCC_FLUSHCACHE_EXT;
   1657 	else
   1658 		wdc_c.r_command = WDCC_FLUSHCACHE;
   1659 	wdc_c.r_st_bmask = WDCS_DRDY;
   1660 	wdc_c.r_st_pmask = WDCS_DRDY;
   1661 	wdc_c.flags = flags;
   1662 	wdc_c.timeout = 30000; /* 30s timeout */
   1663 	if (wd->atabus->ata_exec_command(wd->drvp, &wdc_c) != WDC_COMPLETE) {
   1664 		printf("%s: flush cache command didn't complete\n",
   1665 		    wd->sc_dev.dv_xname);
   1666 		return EIO;
   1667 	}
   1668 	if (wdc_c.flags & ERR_NODEV)
   1669 		return ENODEV;
   1670 	if (wdc_c.flags & AT_TIMEOU) {
   1671 		printf("%s: flush cache command timeout\n",
   1672 		    wd->sc_dev.dv_xname);
   1673 		return EIO;
   1674 	}
   1675 	if (wdc_c.flags & AT_ERROR) {
   1676 		if (wdc_c.r_error == WDCE_ABRT) /* command not supported */
   1677 			return ENODEV;
   1678 		printf("%s: flush cache command: error 0x%x\n",
   1679 		    wd->sc_dev.dv_xname, wdc_c.r_error);
   1680 		return EIO;
   1681 	}
   1682 	if (wdc_c.flags & AT_DF) {
   1683 		printf("%s: flush cache command: drive fault\n",
   1684 		    wd->sc_dev.dv_xname);
   1685 		return EIO;
   1686 	}
   1687 	return 0;
   1688 }
   1689 
   1690 void
   1691 wd_shutdown(void *arg)
   1692 {
   1693 	struct wd_softc *wd = arg;
   1694 	wd_flushcache(wd, AT_POLL);
   1695 }
   1696 
   1697 /*
   1698  * Allocate space for a ioctl queue structure.  Mostly taken from
   1699  * scsipi_ioctl.c
   1700  */
   1701 struct wd_ioctl *
   1702 wi_get(void)
   1703 {
   1704 	struct wd_ioctl *wi;
   1705 	int s;
   1706 
   1707 	wi = malloc(sizeof(struct wd_ioctl), M_TEMP, M_WAITOK|M_ZERO);
   1708 	simple_lock_init(&wi->wi_bp.b_interlock);
   1709 	s = splbio();
   1710 	LIST_INSERT_HEAD(&wi_head, wi, wi_list);
   1711 	splx(s);
   1712 	return (wi);
   1713 }
   1714 
   1715 /*
   1716  * Free an ioctl structure and remove it from our list
   1717  */
   1718 
   1719 void
   1720 wi_free(struct wd_ioctl *wi)
   1721 {
   1722 	int s;
   1723 
   1724 	s = splbio();
   1725 	LIST_REMOVE(wi, wi_list);
   1726 	splx(s);
   1727 	free(wi, M_TEMP);
   1728 }
   1729 
   1730 /*
   1731  * Find a wd_ioctl structure based on the struct buf.
   1732  */
   1733 
   1734 struct wd_ioctl *
   1735 wi_find(struct buf *bp)
   1736 {
   1737 	struct wd_ioctl *wi;
   1738 	int s;
   1739 
   1740 	s = splbio();
   1741 	for (wi = wi_head.lh_first; wi != 0; wi = wi->wi_list.le_next)
   1742 		if (bp == &wi->wi_bp)
   1743 			break;
   1744 	splx(s);
   1745 	return (wi);
   1746 }
   1747 
   1748 /*
   1749  * Ioctl pseudo strategy routine
   1750  *
   1751  * This is mostly stolen from scsipi_ioctl.c:scsistrategy().  What
   1752  * happens here is:
   1753  *
   1754  * - wdioctl() queues a wd_ioctl structure.
   1755  *
   1756  * - wdioctl() calls physio/wdioctlstrategy based on whether or not
   1757  *   user space I/O is required.  If physio() is called, physio() eventually
   1758  *   calls wdioctlstrategy().
   1759  *
   1760  * - In either case, wdioctlstrategy() calls wd->atabus->ata_exec_command()
   1761  *   to perform the actual command
   1762  *
   1763  * The reason for the use of the pseudo strategy routine is because
   1764  * when doing I/O to/from user space, physio _really_ wants to be in
   1765  * the loop.  We could put the entire buffer into the ioctl request
   1766  * structure, but that won't scale if we want to do things like download
   1767  * microcode.
   1768  */
   1769 
   1770 void
   1771 wdioctlstrategy(struct buf *bp)
   1772 {
   1773 	struct wd_ioctl *wi;
   1774 	struct wdc_command wdc_c;
   1775 	int error = 0;
   1776 
   1777 	wi = wi_find(bp);
   1778 	if (wi == NULL) {
   1779 		printf("user_strat: No ioctl\n");
   1780 		error = EINVAL;
   1781 		goto bad;
   1782 	}
   1783 
   1784 	memset(&wdc_c, 0, sizeof(wdc_c));
   1785 
   1786 	/*
   1787 	 * Abort if physio broke up the transfer
   1788 	 */
   1789 
   1790 	if (bp->b_bcount != wi->wi_atareq.datalen) {
   1791 		printf("physio split wd ioctl request... cannot proceed\n");
   1792 		error = EIO;
   1793 		goto bad;
   1794 	}
   1795 
   1796 	/*
   1797 	 * Abort if we didn't get a buffer size that was a multiple of
   1798 	 * our sector size (or was larger than NBBY)
   1799 	 */
   1800 
   1801 	if ((bp->b_bcount % wi->wi_softc->sc_dk.dk_label->d_secsize) != 0 ||
   1802 	    (bp->b_bcount / wi->wi_softc->sc_dk.dk_label->d_secsize) >=
   1803 	     (1 << NBBY)) {
   1804 		error = EINVAL;
   1805 		goto bad;
   1806 	}
   1807 
   1808 	/*
   1809 	 * Make sure a timeout was supplied in the ioctl request
   1810 	 */
   1811 
   1812 	if (wi->wi_atareq.timeout == 0) {
   1813 		error = EINVAL;
   1814 		goto bad;
   1815 	}
   1816 
   1817 	if (wi->wi_atareq.flags & ATACMD_READ)
   1818 		wdc_c.flags |= AT_READ;
   1819 	else if (wi->wi_atareq.flags & ATACMD_WRITE)
   1820 		wdc_c.flags |= AT_WRITE;
   1821 
   1822 	if (wi->wi_atareq.flags & ATACMD_READREG)
   1823 		wdc_c.flags |= AT_READREG;
   1824 
   1825 	wdc_c.flags |= AT_WAIT;
   1826 
   1827 	wdc_c.timeout = wi->wi_atareq.timeout;
   1828 	wdc_c.r_command = wi->wi_atareq.command;
   1829 	wdc_c.r_head = wi->wi_atareq.head & 0x0f;
   1830 	wdc_c.r_cyl = wi->wi_atareq.cylinder;
   1831 	wdc_c.r_sector = wi->wi_atareq.sec_num;
   1832 	wdc_c.r_count = wi->wi_atareq.sec_count;
   1833 	wdc_c.r_features = wi->wi_atareq.features;
   1834 	wdc_c.r_st_bmask = WDCS_DRDY;
   1835 	wdc_c.r_st_pmask = WDCS_DRDY;
   1836 	wdc_c.data = wi->wi_bp.b_data;
   1837 	wdc_c.bcount = wi->wi_bp.b_bcount;
   1838 
   1839 	if (wi->wi_softc->atabus->ata_exec_command(wi->wi_softc->drvp, &wdc_c)
   1840 	    != WDC_COMPLETE) {
   1841 		wi->wi_atareq.retsts = ATACMD_ERROR;
   1842 		goto bad;
   1843 	}
   1844 
   1845 	if (wdc_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   1846 		if (wdc_c.flags & AT_ERROR) {
   1847 			wi->wi_atareq.retsts = ATACMD_ERROR;
   1848 			wi->wi_atareq.error = wdc_c.r_error;
   1849 		} else if (wdc_c.flags & AT_DF)
   1850 			wi->wi_atareq.retsts = ATACMD_DF;
   1851 		else
   1852 			wi->wi_atareq.retsts = ATACMD_TIMEOUT;
   1853 	} else {
   1854 		wi->wi_atareq.retsts = ATACMD_OK;
   1855 		if (wi->wi_atareq.flags & ATACMD_READREG) {
   1856 			wi->wi_atareq.head = wdc_c.r_head ;
   1857 			wi->wi_atareq.cylinder = wdc_c.r_cyl;
   1858 			wi->wi_atareq.sec_num = wdc_c.r_sector;
   1859 			wi->wi_atareq.sec_count = wdc_c.r_count;
   1860 			wi->wi_atareq.features = wdc_c.r_features;
   1861 			wi->wi_atareq.error = wdc_c.r_error;
   1862 		}
   1863 	}
   1864 
   1865 	bp->b_error = 0;
   1866 	biodone(bp);
   1867 	return;
   1868 bad:
   1869 	bp->b_flags |= B_ERROR;
   1870 	bp->b_error = error;
   1871 	biodone(bp);
   1872 }
   1873