Home | History | Annotate | Line # | Download | only in ata
wd.c revision 1.350.2.1
      1 /*	$NetBSD: wd.c,v 1.350.2.1 2007/12/04 13:02:56 ad 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 1.350.2.1 2007/12/04 13:02:56 ad Exp $");
     70 
     71 #include "opt_ata.h"
     72 
     73 #include "rnd.h"
     74 
     75 #include <sys/param.h>
     76 #include <sys/systm.h>
     77 #include <sys/kernel.h>
     78 #include <sys/conf.h>
     79 #include <sys/file.h>
     80 #include <sys/stat.h>
     81 #include <sys/ioctl.h>
     82 #include <sys/buf.h>
     83 #include <sys/bufq.h>
     84 #include <sys/uio.h>
     85 #include <sys/malloc.h>
     86 #include <sys/device.h>
     87 #include <sys/disklabel.h>
     88 #include <sys/disk.h>
     89 #include <sys/syslog.h>
     90 #include <sys/proc.h>
     91 #include <sys/vnode.h>
     92 #if NRND > 0
     93 #include <sys/rnd.h>
     94 #endif
     95 
     96 #include <sys/intr.h>
     97 #include <sys/bus.h>
     98 
     99 #include <dev/ata/atareg.h>
    100 #include <dev/ata/atavar.h>
    101 #include <dev/ata/wdvar.h>
    102 #include <dev/ic/wdcreg.h>
    103 #include <sys/ataio.h>
    104 #include "locators.h"
    105 
    106 #include <prop/proplib.h>
    107 
    108 #define	LBA48_THRESHOLD		(0xfffffff)	/* 128GB / DEV_BSIZE */
    109 
    110 #define	WDIORETRIES_SINGLE 4	/* number of retries before single-sector */
    111 #define	WDIORETRIES	5	/* number of retries before giving up */
    112 #define	RECOVERYTIME hz/2	/* time to wait before retrying a cmd */
    113 
    114 #define	WDUNIT(dev)		DISKUNIT(dev)
    115 #define	WDPART(dev)		DISKPART(dev)
    116 #define	WDMINOR(unit, part)	DISKMINOR(unit, part)
    117 #define	MAKEWDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
    118 
    119 #define	WDLABELDEV(dev)	(MAKEWDDEV(major(dev), WDUNIT(dev), RAW_PART))
    120 
    121 #define DEBUG_INTR   0x01
    122 #define DEBUG_XFERS  0x02
    123 #define DEBUG_STATUS 0x04
    124 #define DEBUG_FUNCS  0x08
    125 #define DEBUG_PROBE  0x10
    126 #ifdef ATADEBUG
    127 int wdcdebug_wd_mask = 0x0;
    128 #define ATADEBUG_PRINT(args, level) \
    129 	if (wdcdebug_wd_mask & (level)) \
    130 		printf args
    131 #else
    132 #define ATADEBUG_PRINT(args, level)
    133 #endif
    134 
    135 int	wdprobe(struct device *, struct cfdata *, void *);
    136 void	wdattach(struct device *, struct device *, void *);
    137 int	wddetach(struct device *, int);
    138 int	wdactivate(struct device *, enum devact);
    139 int	wdprint(void *, char *);
    140 void	wdperror(const struct wd_softc *);
    141 
    142 CFATTACH_DECL(wd, sizeof(struct wd_softc),
    143     wdprobe, wdattach, wddetach, wdactivate);
    144 
    145 extern struct cfdriver wd_cd;
    146 
    147 dev_type_open(wdopen);
    148 dev_type_close(wdclose);
    149 dev_type_read(wdread);
    150 dev_type_write(wdwrite);
    151 dev_type_ioctl(wdioctl);
    152 dev_type_strategy(wdstrategy);
    153 dev_type_dump(wddump);
    154 dev_type_size(wdsize);
    155 
    156 const struct bdevsw wd_bdevsw = {
    157 	wdopen, wdclose, wdstrategy, wdioctl, wddump, wdsize, D_DISK
    158 };
    159 
    160 const struct cdevsw wd_cdevsw = {
    161 	wdopen, wdclose, wdread, wdwrite, wdioctl,
    162 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
    163 };
    164 
    165 /*
    166  * Glue necessary to hook WDCIOCCOMMAND into physio
    167  */
    168 
    169 struct wd_ioctl {
    170 	LIST_ENTRY(wd_ioctl) wi_list;
    171 	struct buf wi_bp;
    172 	struct uio wi_uio;
    173 	struct iovec wi_iov;
    174 	atareq_t wi_atareq;
    175 	struct wd_softc *wi_softc;
    176 };
    177 
    178 LIST_HEAD(, wd_ioctl) wi_head;
    179 
    180 struct	wd_ioctl *wi_find(struct buf *);
    181 void	wi_free(struct wd_ioctl *);
    182 struct	wd_ioctl *wi_get(void);
    183 void	wdioctlstrategy(struct buf *);
    184 
    185 void  wdgetdefaultlabel(struct wd_softc *, struct disklabel *);
    186 void  wdgetdisklabel(struct wd_softc *);
    187 void  wdstart(void *);
    188 void  __wdstart(struct wd_softc*, struct buf *);
    189 void  wdrestart(void *);
    190 void  wddone(void *);
    191 int   wd_get_params(struct wd_softc *, u_int8_t, struct ataparams *);
    192 int   wd_standby(struct wd_softc *, int);
    193 int   wd_flushcache(struct wd_softc *, int);
    194 void  wd_shutdown(void *);
    195 
    196 int   wd_getcache(struct wd_softc *, int *);
    197 int   wd_setcache(struct wd_softc *, int);
    198 
    199 struct dkdriver wddkdriver = { wdstrategy, minphys };
    200 
    201 #ifdef HAS_BAD144_HANDLING
    202 static void bad144intern(struct wd_softc *);
    203 #endif
    204 
    205 #define	WD_QUIRK_SPLIT_MOD15_WRITE	0x0001	/* must split certain writes */
    206 #define	WD_QUIRK_FORCE_LBA48		0x0002	/* must use LBA48 commands */
    207 
    208 #define	WD_QUIRK_FMT "\20\1SPLIT_MOD15_WRITE\2FORCE_LBA48"
    209 
    210 /*
    211  * Quirk table for IDE drives.  Put more-specific matches first, since
    212  * a simple globbing routine is used for matching.
    213  */
    214 static const struct wd_quirk {
    215 	const char *wdq_match;		/* inquiry pattern to match */
    216 	int wdq_quirks;			/* drive quirks */
    217 } wd_quirk_table[] = {
    218 	/*
    219 	 * Some Seagate S-ATA drives have a PHY which can get confused
    220 	 * with the way data is packetized by some S-ATA controllers.
    221 	 *
    222 	 * The work-around is to split in two any write transfer whose
    223 	 * sector count % 15 == 1 (assuming 512 byte sectors).
    224 	 *
    225 	 * XXX This is an incomplete list.  There are at least a couple
    226 	 * XXX more model numbers.  If you have trouble with such transfers
    227 	 * XXX (8K is the most common) on Seagate S-ATA drives, please
    228 	 * XXX notify thorpej (at) NetBSD.org.
    229 	 */
    230 	{ "ST3120023AS",
    231 	  WD_QUIRK_SPLIT_MOD15_WRITE },
    232 	{ "ST380023AS",
    233 	  WD_QUIRK_SPLIT_MOD15_WRITE },
    234 
    235 	/*
    236 	 * These seagate drives seems to have issue addressing sector 0xfffffff
    237 	 * (aka LBA48_THRESHOLD) in LBA mode. The workaround is to force
    238 	 * LBA48
    239 	 * Note that we can't just change the code to always use LBA48 for
    240 	 * sector 0xfffffff, because this would break valid and working
    241 	 * setups using LBA48 drives on non-LBA48-capable controllers
    242 	 * (and it's hard to get a list of such controllers)
    243 	 */
    244 	{ "ST3160021A*",
    245 	  WD_QUIRK_FORCE_LBA48 },
    246 	{ "ST3160811A*",
    247 	  WD_QUIRK_FORCE_LBA48 },
    248 	{ "ST3160812A*",
    249 	  WD_QUIRK_FORCE_LBA48 },
    250 	{ "ST3160023A*",
    251 	  WD_QUIRK_FORCE_LBA48 },
    252 	{ "ST3160827A*",
    253 	  WD_QUIRK_FORCE_LBA48 },
    254 	/* Attempt to catch all seagate drives larger than 200GB */
    255 	{ "ST3[2-9][0-9][0-9][0-9][0-9][0-9][A-Z]*",
    256 	  WD_QUIRK_FORCE_LBA48 },
    257 	{ "ST3[1-9][0-9][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 
    305 	ATADEBUG_PRINT(("wdattach\n"), DEBUG_FUNCS | DEBUG_PROBE);
    306 	callout_init(&wd->sc_restart_ch, 0);
    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_verbose("%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_verbose(" 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_verbose(" 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_verbose(" 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 	/* we fill in dk_info later */
    414 	disk_init(&wd->sc_dk, wd->sc_dev.dv_xname, &wddkdriver);
    415 	disk_attach(&wd->sc_dk);
    416 	wd->sc_wdc_bio.lp = wd->sc_dk.dk_label;
    417 	wd->sc_sdhook = shutdownhook_establish(wd_shutdown, wd);
    418 	if (wd->sc_sdhook == NULL)
    419 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
    420 		    wd->sc_dev.dv_xname);
    421 #if NRND > 0
    422 	rnd_attach_source(&wd->rnd_source, wd->sc_dev.dv_xname,
    423 			  RND_TYPE_DISK, 0);
    424 #endif
    425 
    426 	/* Discover wedges on this disk. */
    427 	dkwedge_discover(&wd->sc_dk);
    428 }
    429 
    430 int
    431 wdactivate(struct device *self, enum devact act)
    432 {
    433 	int rv = 0;
    434 
    435 	switch (act) {
    436 	case DVACT_ACTIVATE:
    437 		rv = EOPNOTSUPP;
    438 		break;
    439 
    440 	case DVACT_DEACTIVATE:
    441 		/*
    442 		 * Nothing to do; we key off the device's DVF_ACTIVATE.
    443 		 */
    444 		break;
    445 	}
    446 	return (rv);
    447 }
    448 
    449 int
    450 wddetach(struct device *self, int flags)
    451 {
    452 	struct wd_softc *sc = (struct wd_softc *)self;
    453 	int s, bmaj, cmaj, i, mn;
    454 
    455 	/* locate the major number */
    456 	bmaj = bdevsw_lookup_major(&wd_bdevsw);
    457 	cmaj = cdevsw_lookup_major(&wd_cdevsw);
    458 
    459 	/* Nuke the vnodes for any open instances. */
    460 	for (i = 0; i < MAXPARTITIONS; i++) {
    461 		mn = WDMINOR(device_unit(self), i);
    462 		vdevgone(bmaj, mn, mn, VBLK);
    463 		vdevgone(cmaj, mn, mn, VCHR);
    464 	}
    465 
    466 	/* Delete all of our wedges. */
    467 	dkwedge_delall(&sc->sc_dk);
    468 
    469 	s = splbio();
    470 
    471 	/* Kill off any queued buffers. */
    472 	bufq_drain(sc->sc_q);
    473 
    474 	bufq_free(sc->sc_q);
    475 	sc->atabus->ata_killpending(sc->drvp);
    476 
    477 	splx(s);
    478 
    479 	/* Detach disk. */
    480 	disk_detach(&sc->sc_dk);
    481 
    482 #ifdef WD_SOFTBADSECT
    483 	/* Clean out the bad sector list */
    484 	while (!SLIST_EMPTY(&sc->sc_bslist)) {
    485 		void *head = SLIST_FIRST(&sc->sc_bslist);
    486 		SLIST_REMOVE_HEAD(&sc->sc_bslist, dbs_next);
    487 		free(head, M_TEMP);
    488 	}
    489 	sc->sc_bscount = 0;
    490 #endif
    491 
    492 	/* Get rid of the shutdown hook. */
    493 	if (sc->sc_sdhook != NULL)
    494 		shutdownhook_disestablish(sc->sc_sdhook);
    495 
    496 #if NRND > 0
    497 	/* Unhook the entropy source. */
    498 	rnd_detach_source(&sc->rnd_source);
    499 #endif
    500 
    501 	sc->drvp->drive_flags = 0; /* no drive any more here */
    502 
    503 	return (0);
    504 }
    505 
    506 /*
    507  * Read/write routine for a buffer.  Validates the arguments and schedules the
    508  * transfer.  Does not wait for the transfer to complete.
    509  */
    510 void
    511 wdstrategy(struct buf *bp)
    512 {
    513 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(bp->b_dev));
    514 	struct disklabel *lp = wd->sc_dk.dk_label;
    515 	daddr_t blkno;
    516 	int s;
    517 
    518 	ATADEBUG_PRINT(("wdstrategy (%s)\n", wd->sc_dev.dv_xname),
    519 	    DEBUG_XFERS);
    520 
    521 	/* Valid request?  */
    522 	if (bp->b_blkno < 0 ||
    523 	    (bp->b_bcount % lp->d_secsize) != 0 ||
    524 	    (bp->b_bcount / lp->d_secsize) >= (1 << NBBY)) {
    525 		bp->b_error = EINVAL;
    526 		goto done;
    527 	}
    528 
    529 	/* If device invalidated (e.g. media change, door open), error. */
    530 	if ((wd->sc_flags & WDF_LOADED) == 0) {
    531 		bp->b_error = EIO;
    532 		goto done;
    533 	}
    534 
    535 	/* If it's a null transfer, return immediately. */
    536 	if (bp->b_bcount == 0)
    537 		goto done;
    538 
    539 	/*
    540 	 * Do bounds checking, adjust transfer. if error, process.
    541 	 * If end of partition, just return.
    542 	 */
    543 	if (WDPART(bp->b_dev) == RAW_PART) {
    544 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
    545 		    wd->sc_capacity) <= 0)
    546 			goto done;
    547 	} else {
    548 		if (bounds_check_with_label(&wd->sc_dk, bp,
    549 		    (wd->sc_flags & (WDF_WLABEL|WDF_LABELLING)) != 0) <= 0)
    550 			goto done;
    551 	}
    552 
    553 	/*
    554 	 * Now convert the block number to absolute and put it in
    555 	 * terms of the device's logical block size.
    556 	 */
    557 	if (lp->d_secsize >= DEV_BSIZE)
    558 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    559 	else
    560 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    561 
    562 	if (WDPART(bp->b_dev) != RAW_PART)
    563 		blkno += lp->d_partitions[WDPART(bp->b_dev)].p_offset;
    564 
    565 	bp->b_rawblkno = blkno;
    566 
    567 #ifdef WD_SOFTBADSECT
    568 	/*
    569 	 * If the transfer about to be attempted contains only a block that
    570 	 * is known to be bad then return an error for the transfer without
    571 	 * even attempting to start a transfer up under the premis that we
    572 	 * will just end up doing more retries for a transfer that will end
    573 	 * up failing again.
    574 	 * XXX:SMP - mutex required to protect with DIOCBSFLUSH
    575 	 */
    576 	if (__predict_false(!SLIST_EMPTY(&wd->sc_bslist))) {
    577 		struct disk_badsectors *dbs;
    578 		daddr_t maxblk = blkno + (bp->b_bcount >> DEV_BSHIFT) - 1;
    579 
    580 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next)
    581 			if ((dbs->dbs_min <= blkno && blkno <= dbs->dbs_max) ||
    582 			    (dbs->dbs_min <= maxblk && maxblk <= dbs->dbs_max)){
    583 				bp->b_error = EIO;
    584 				goto done;
    585 			}
    586 	}
    587 #endif
    588 
    589 	/* Queue transfer on drive, activate drive and controller if idle. */
    590 	s = splbio();
    591 	BUFQ_PUT(wd->sc_q, bp);
    592 	wdstart(wd);
    593 	splx(s);
    594 	return;
    595 done:
    596 	/* Toss transfer; we're done early. */
    597 	bp->b_resid = bp->b_bcount;
    598 	biodone(bp);
    599 }
    600 
    601 /*
    602  * Queue a drive for I/O.
    603  */
    604 void
    605 wdstart(void *arg)
    606 {
    607 	struct wd_softc *wd = arg;
    608 	struct buf *bp = NULL;
    609 
    610 	ATADEBUG_PRINT(("wdstart %s\n", wd->sc_dev.dv_xname),
    611 	    DEBUG_XFERS);
    612 	while (wd->openings > 0) {
    613 
    614 		/* Is there a buf for us ? */
    615 		if ((bp = BUFQ_GET(wd->sc_q)) == NULL)
    616 			return;
    617 
    618 		/*
    619 		 * Make the command. First lock the device
    620 		 */
    621 		wd->openings--;
    622 
    623 		wd->retries = 0;
    624 		__wdstart(wd, bp);
    625 	}
    626 }
    627 
    628 static void
    629 wd_split_mod15_write(struct buf *bp)
    630 {
    631 	struct buf *obp = bp->b_private;
    632 	struct wd_softc *sc = wd_cd.cd_devs[DISKUNIT(obp->b_dev)];
    633 
    634 	if (__predict_false(bp->b_error != 0)) {
    635 		/*
    636 		 * Propagate the error.  If this was the first half of
    637 		 * the original transfer, make sure to account for that
    638 		 * in the residual.
    639 		 */
    640 		if (bp->b_data == obp->b_data)
    641 			bp->b_resid += bp->b_bcount;
    642 		goto done;
    643 	}
    644 
    645 	/*
    646 	 * If this was the second half of the transfer, we're all done!
    647 	 */
    648 	if (bp->b_data != obp->b_data)
    649 		goto done;
    650 
    651 	/*
    652 	 * Advance the pointer to the second half and issue that command
    653 	 * using the same opening.
    654 	 */
    655 	bp->b_flags = obp->b_flags;
    656 	bp->b_oflags = obp->b_oflags;
    657 	bp->b_cflags = obp->b_cflags;
    658 	bp->b_data = (char *)bp->b_data + bp->b_bcount;
    659 	bp->b_blkno += (bp->b_bcount / 512);
    660 	bp->b_rawblkno += (bp->b_bcount / 512);
    661 	__wdstart(sc, bp);
    662 	return;
    663 
    664  done:
    665 	obp->b_error = bp->b_error;
    666 	obp->b_resid = bp->b_resid;
    667 	putiobuf(bp);
    668 	biodone(obp);
    669 	sc->openings++;
    670 	/* wddone() will call wdstart() */
    671 }
    672 
    673 void
    674 __wdstart(struct wd_softc *wd, struct buf *bp)
    675 {
    676 
    677 	/*
    678 	 * Deal with the "split mod15 write" quirk.  We just divide the
    679 	 * transfer in two, doing the first half and then then second half
    680 	 * with the same command opening.
    681 	 *
    682 	 * Note we MUST do this here, because we can't let insertion
    683 	 * into the bufq cause the transfers to be re-merged.
    684 	 */
    685 	if (__predict_false((wd->sc_quirks & WD_QUIRK_SPLIT_MOD15_WRITE) != 0 &&
    686 			    (bp->b_flags & B_READ) == 0 &&
    687 			    bp->b_bcount > 512 &&
    688 			    ((bp->b_bcount / 512) % 15) == 1)) {
    689 		struct buf *nbp;
    690 
    691 		/* already at splbio */
    692 		nbp = getiobuf(NULL, false);
    693 		if (__predict_false(nbp == NULL)) {
    694 			/* No memory -- fail the iop. */
    695 			bp->b_error = ENOMEM;
    696 			bp->b_resid = bp->b_bcount;
    697 			biodone(bp);
    698 			wd->openings++;
    699 			return;
    700 		}
    701 
    702 		nbp->b_error = 0;
    703 		nbp->b_proc = bp->b_proc;
    704 		nbp->b_dev = bp->b_dev;
    705 
    706 		nbp->b_bcount = bp->b_bcount / 2;
    707 		nbp->b_bufsize = bp->b_bcount / 2;
    708 		nbp->b_data = bp->b_data;
    709 
    710 		nbp->b_blkno = bp->b_blkno;
    711 		nbp->b_rawblkno = bp->b_rawblkno;
    712 
    713 		nbp->b_flags = bp->b_flags;
    714 		nbp->b_oflags = bp->b_oflags;
    715 		nbp->b_cflags = bp->b_cflags;
    716 		nbp->b_iodone = wd_split_mod15_write;
    717 
    718 		/* Put ptr to orig buf in b_private and use new buf */
    719 		nbp->b_private = bp;
    720 
    721 		BIO_COPYPRIO(nbp, bp);
    722 
    723 		bp = nbp;
    724 	}
    725 
    726 	wd->sc_wdc_bio.blkno = bp->b_rawblkno;
    727 	wd->sc_wdc_bio.blkdone =0;
    728 	wd->sc_bp = bp;
    729 	/*
    730 	 * If we're retrying, retry in single-sector mode. This will give us
    731 	 * the sector number of the problem, and will eventually allow the
    732 	 * transfer to succeed.
    733 	 */
    734 	if (wd->retries >= WDIORETRIES_SINGLE)
    735 		wd->sc_wdc_bio.flags = ATA_SINGLE;
    736 	else
    737 		wd->sc_wdc_bio.flags = 0;
    738 	if (wd->sc_flags & WDF_LBA48 &&
    739 	    (wd->sc_wdc_bio.blkno > LBA48_THRESHOLD ||
    740 	    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) != 0))
    741 		wd->sc_wdc_bio.flags |= ATA_LBA48;
    742 	if (wd->sc_flags & WDF_LBA)
    743 		wd->sc_wdc_bio.flags |= ATA_LBA;
    744 	if (bp->b_flags & B_READ)
    745 		wd->sc_wdc_bio.flags |= ATA_READ;
    746 	wd->sc_wdc_bio.bcount = bp->b_bcount;
    747 	wd->sc_wdc_bio.databuf = bp->b_data;
    748 	/* Instrumentation. */
    749 	disk_busy(&wd->sc_dk);
    750 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
    751 	case ATACMD_TRY_AGAIN:
    752 		callout_reset(&wd->sc_restart_ch, hz, wdrestart, wd);
    753 		break;
    754 	case ATACMD_QUEUED:
    755 	case ATACMD_COMPLETE:
    756 		break;
    757 	default:
    758 		panic("__wdstart: bad return code from ata_bio()");
    759 	}
    760 }
    761 
    762 void
    763 wddone(void *v)
    764 {
    765 	struct wd_softc *wd = v;
    766 	struct buf *bp = wd->sc_bp;
    767 	const char *errmsg;
    768 	int do_perror = 0;
    769 	int nblks;
    770 
    771 	ATADEBUG_PRINT(("wddone %s\n", wd->sc_dev.dv_xname),
    772 	    DEBUG_XFERS);
    773 	if (bp == NULL)
    774 		return;
    775 	bp->b_resid = wd->sc_wdc_bio.bcount;
    776 	switch (wd->sc_wdc_bio.error) {
    777 	case ERR_DMA:
    778 		errmsg = "DMA error";
    779 		goto retry;
    780 	case ERR_DF:
    781 		errmsg = "device fault";
    782 		goto retry;
    783 	case TIMEOUT:
    784 		errmsg = "device timeout";
    785 		goto retry;
    786 	case ERR_RESET:
    787 		errmsg = "channel reset";
    788 		goto retry2;
    789 	case ERROR:
    790 		/* Don't care about media change bits */
    791 		if (wd->sc_wdc_bio.r_error != 0 &&
    792 		    (wd->sc_wdc_bio.r_error & ~(WDCE_MC | WDCE_MCR)) == 0)
    793 			goto noerror;
    794 		errmsg = "error";
    795 		do_perror = 1;
    796 		if (wd->sc_wdc_bio.r_error & WDCE_IDNF &&
    797 		    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) == 0) {
    798 			nblks = wd->sc_wdc_bio.bcount /
    799 			    wd->sc_dk.dk_label->d_secsize;
    800 			/*
    801 			 * If we get a "id not found" when crossing the
    802 			 * LBA48_THRESHOLD, and the drive is larger than
    803 			 * 128GB, then we can assume the drive has the
    804 			 * LBA48 bug and we switch to LBA48.
    805 			 */
    806 			if (wd->sc_wdc_bio.blkno <= LBA48_THRESHOLD &&
    807 			    wd->sc_wdc_bio.blkno + nblks > LBA48_THRESHOLD &&
    808 			    wd->sc_capacity > LBA48_THRESHOLD + 1) {
    809 				errmsg = "LBA48 bug";
    810 				wd->sc_quirks |= WD_QUIRK_FORCE_LBA48;
    811 				do_perror = 0;
    812 				goto retry2;
    813 			}
    814 		}
    815 retry:		/* Just reset and retry. Can we do more ? */
    816 		(*wd->atabus->ata_reset_drive)(wd->drvp, AT_RST_NOCMD);
    817 retry2:
    818 		diskerr(bp, "wd", errmsg, LOG_PRINTF,
    819 		    wd->sc_wdc_bio.blkdone, wd->sc_dk.dk_label);
    820 		if (wd->retries < WDIORETRIES)
    821 			printf(", retrying");
    822 		printf("\n");
    823 		if (do_perror)
    824 			wdperror(wd);
    825 		if (wd->retries < WDIORETRIES) {
    826 			wd->retries++;
    827 			callout_reset(&wd->sc_restart_ch, RECOVERYTIME,
    828 			    wdrestart, wd);
    829 			return;
    830 		}
    831 
    832 #ifdef WD_SOFTBADSECT
    833 		/*
    834 		 * Not all errors indicate a failed block but those that do,
    835 		 * put the block on the bad-block list for the device.  Only
    836 		 * do this for reads because the drive should do it for writes,
    837 		 * itself, according to Manuel.
    838 		 */
    839 		if ((bp->b_flags & B_READ) &&
    840 		    ((wd->drvp->ata_vers >= 4 && wd->sc_wdc_bio.r_error & 64) ||
    841 	     	     (wd->drvp->ata_vers < 4 && wd->sc_wdc_bio.r_error & 192))) {
    842 			struct disk_badsectors *dbs;
    843 
    844 			dbs = malloc(sizeof *dbs, M_TEMP, M_WAITOK);
    845 			dbs->dbs_min = bp->b_rawblkno;
    846 			dbs->dbs_max = dbs->dbs_min + (bp->b_bcount >> DEV_BSHIFT) - 1;
    847 			microtime(&dbs->dbs_failedat);
    848 			SLIST_INSERT_HEAD(&wd->sc_bslist, dbs, dbs_next);
    849 			wd->sc_bscount++;
    850 		}
    851 #endif
    852 		bp->b_error = EIO;
    853 		break;
    854 	case NOERROR:
    855 noerror:	if ((wd->sc_wdc_bio.flags & ATA_CORR) || wd->retries > 0)
    856 			printf("%s: soft error (corrected)\n",
    857 			    wd->sc_dev.dv_xname);
    858 		break;
    859 	case ERR_NODEV:
    860 		bp->b_error = EIO;
    861 		break;
    862 	}
    863 	disk_unbusy(&wd->sc_dk, (bp->b_bcount - bp->b_resid),
    864 	    (bp->b_flags & B_READ));
    865 #if NRND > 0
    866 	rnd_add_uint32(&wd->rnd_source, bp->b_blkno);
    867 #endif
    868 	/* XXX Yuck, but we don't want to increment openings in this case */
    869 	if (__predict_false(bp->b_iodone == wd_split_mod15_write))
    870 		biodone(bp);
    871 	else {
    872 		biodone(bp);
    873 		wd->openings++;
    874 	}
    875 	wdstart(wd);
    876 }
    877 
    878 void
    879 wdrestart(void *v)
    880 {
    881 	struct wd_softc *wd = v;
    882 	struct buf *bp = wd->sc_bp;
    883 	int s;
    884 
    885 	ATADEBUG_PRINT(("wdrestart %s\n", wd->sc_dev.dv_xname),
    886 	    DEBUG_XFERS);
    887 	s = splbio();
    888 	__wdstart(v, bp);
    889 	splx(s);
    890 }
    891 
    892 int
    893 wdread(dev_t dev, struct uio *uio, int flags)
    894 {
    895 
    896 	ATADEBUG_PRINT(("wdread\n"), DEBUG_XFERS);
    897 	return (physio(wdstrategy, NULL, dev, B_READ, minphys, uio));
    898 }
    899 
    900 int
    901 wdwrite(dev_t dev, struct uio *uio, int flags)
    902 {
    903 
    904 	ATADEBUG_PRINT(("wdwrite\n"), DEBUG_XFERS);
    905 	return (physio(wdstrategy, NULL, dev, B_WRITE, minphys, uio));
    906 }
    907 
    908 int
    909 wdopen(dev_t dev, int flag, int fmt, struct lwp *l)
    910 {
    911 	struct wd_softc *wd;
    912 	int part, error;
    913 
    914 	ATADEBUG_PRINT(("wdopen\n"), DEBUG_FUNCS);
    915 	wd = device_lookup(&wd_cd, WDUNIT(dev));
    916 	if (wd == NULL)
    917 		return (ENXIO);
    918 
    919 	if (! device_is_active(&wd->sc_dev))
    920 		return (ENODEV);
    921 
    922 	part = WDPART(dev);
    923 
    924 	mutex_enter(&wd->sc_dk.dk_openlock);
    925 
    926 	/*
    927 	 * If there are wedges, and this is not RAW_PART, then we
    928 	 * need to fail.
    929 	 */
    930 	if (wd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
    931 		error = EBUSY;
    932 		goto bad1;
    933 	}
    934 
    935 	/*
    936 	 * If this is the first open of this device, add a reference
    937 	 * to the adapter.
    938 	 */
    939 	if (wd->sc_dk.dk_openmask == 0 &&
    940 	    (error = wd->atabus->ata_addref(wd->drvp)) != 0)
    941 		goto bad1;
    942 
    943 	if (wd->sc_dk.dk_openmask != 0) {
    944 		/*
    945 		 * If any partition is open, but the disk has been invalidated,
    946 		 * disallow further opens.
    947 		 */
    948 		if ((wd->sc_flags & WDF_LOADED) == 0) {
    949 			error = EIO;
    950 			goto bad2;
    951 		}
    952 	} else {
    953 		if ((wd->sc_flags & WDF_LOADED) == 0) {
    954 			wd->sc_flags |= WDF_LOADED;
    955 
    956 			/* Load the physical device parameters. */
    957 			wd_get_params(wd, AT_WAIT, &wd->sc_params);
    958 
    959 			/* Load the partition info if not already loaded. */
    960 			wdgetdisklabel(wd);
    961 		}
    962 	}
    963 
    964 	/* Check that the partition exists. */
    965 	if (part != RAW_PART &&
    966 	    (part >= wd->sc_dk.dk_label->d_npartitions ||
    967 	     wd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
    968 		error = ENXIO;
    969 		goto bad2;
    970 	}
    971 
    972 	/* Insure only one open at a time. */
    973 	switch (fmt) {
    974 	case S_IFCHR:
    975 		wd->sc_dk.dk_copenmask |= (1 << part);
    976 		break;
    977 	case S_IFBLK:
    978 		wd->sc_dk.dk_bopenmask |= (1 << part);
    979 		break;
    980 	}
    981 	wd->sc_dk.dk_openmask =
    982 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
    983 
    984 	mutex_exit(&wd->sc_dk.dk_openlock);
    985 	return 0;
    986 
    987  bad2:
    988 	if (wd->sc_dk.dk_openmask == 0)
    989 		wd->atabus->ata_delref(wd->drvp);
    990  bad1:
    991 	mutex_exit(&wd->sc_dk.dk_openlock);
    992 	return error;
    993 }
    994 
    995 int
    996 wdclose(dev_t dev, int flag, int fmt, struct lwp *l)
    997 {
    998 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
    999 	int part = WDPART(dev);
   1000 
   1001 	ATADEBUG_PRINT(("wdclose\n"), DEBUG_FUNCS);
   1002 
   1003 	mutex_enter(&wd->sc_dk.dk_openlock);
   1004 
   1005 	switch (fmt) {
   1006 	case S_IFCHR:
   1007 		wd->sc_dk.dk_copenmask &= ~(1 << part);
   1008 		break;
   1009 	case S_IFBLK:
   1010 		wd->sc_dk.dk_bopenmask &= ~(1 << part);
   1011 		break;
   1012 	}
   1013 	wd->sc_dk.dk_openmask =
   1014 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
   1015 
   1016 	if (wd->sc_dk.dk_openmask == 0) {
   1017 		wd_flushcache(wd, AT_WAIT);
   1018 
   1019 		if (! (wd->sc_flags & WDF_KLABEL))
   1020 			wd->sc_flags &= ~WDF_LOADED;
   1021 
   1022 		wd->atabus->ata_delref(wd->drvp);
   1023 	}
   1024 
   1025 	mutex_exit(&wd->sc_dk.dk_openlock);
   1026 	return 0;
   1027 }
   1028 
   1029 void
   1030 wdgetdefaultlabel(struct wd_softc *wd, struct disklabel *lp)
   1031 {
   1032 
   1033 	ATADEBUG_PRINT(("wdgetdefaultlabel\n"), DEBUG_FUNCS);
   1034 	memset(lp, 0, sizeof(struct disklabel));
   1035 
   1036 	lp->d_secsize = DEV_BSIZE;
   1037 	lp->d_ntracks = wd->sc_params.atap_heads;
   1038 	lp->d_nsectors = wd->sc_params.atap_sectors;
   1039 	lp->d_ncylinders = (wd->sc_flags & WDF_LBA) ? wd->sc_capacity /
   1040 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors) :
   1041 		wd->sc_params.atap_cylinders;
   1042 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
   1043 
   1044 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
   1045 		lp->d_type = DTYPE_ST506;
   1046 	else
   1047 		lp->d_type = DTYPE_ESDI;
   1048 
   1049 	strncpy(lp->d_typename, wd->sc_params.atap_model, 16);
   1050 	strncpy(lp->d_packname, "fictitious", 16);
   1051 	if (wd->sc_capacity > UINT32_MAX)
   1052 		lp->d_secperunit = UINT32_MAX;
   1053 	else
   1054 		lp->d_secperunit = wd->sc_capacity;
   1055 	lp->d_rpm = 3600;
   1056 	lp->d_interleave = 1;
   1057 	lp->d_flags = 0;
   1058 
   1059 	lp->d_partitions[RAW_PART].p_offset = 0;
   1060 	lp->d_partitions[RAW_PART].p_size =
   1061 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
   1062 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
   1063 	lp->d_npartitions = RAW_PART + 1;
   1064 
   1065 	lp->d_magic = DISKMAGIC;
   1066 	lp->d_magic2 = DISKMAGIC;
   1067 	lp->d_checksum = dkcksum(lp);
   1068 }
   1069 
   1070 /*
   1071  * Fabricate a default disk label, and try to read the correct one.
   1072  */
   1073 void
   1074 wdgetdisklabel(struct wd_softc *wd)
   1075 {
   1076 	struct disklabel *lp = wd->sc_dk.dk_label;
   1077 	const char *errstring;
   1078 	int s;
   1079 
   1080 	ATADEBUG_PRINT(("wdgetdisklabel\n"), DEBUG_FUNCS);
   1081 
   1082 	memset(wd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
   1083 
   1084 	wdgetdefaultlabel(wd, lp);
   1085 
   1086 	wd->sc_badsect[0] = -1;
   1087 
   1088 	if (wd->drvp->state > RESET) {
   1089 		s = splbio();
   1090 		wd->drvp->drive_flags |= DRIVE_RESET;
   1091 		splx(s);
   1092 	}
   1093 	errstring = readdisklabel(MAKEWDDEV(0, device_unit(&wd->sc_dev),
   1094 				  RAW_PART), wdstrategy, lp,
   1095 				  wd->sc_dk.dk_cpulabel);
   1096 	if (errstring) {
   1097 		/*
   1098 		 * This probably happened because the drive's default
   1099 		 * geometry doesn't match the DOS geometry.  We
   1100 		 * assume the DOS geometry is now in the label and try
   1101 		 * again.  XXX This is a kluge.
   1102 		 */
   1103 		if (wd->drvp->state > RESET) {
   1104 			s = splbio();
   1105 			wd->drvp->drive_flags |= DRIVE_RESET;
   1106 			splx(s);
   1107 		}
   1108 		errstring = readdisklabel(MAKEWDDEV(0, device_unit(&wd->sc_dev),
   1109 		    RAW_PART), wdstrategy, lp, wd->sc_dk.dk_cpulabel);
   1110 	}
   1111 	if (errstring) {
   1112 		printf("%s: %s\n", wd->sc_dev.dv_xname, errstring);
   1113 		return;
   1114 	}
   1115 
   1116 	if (wd->drvp->state > RESET) {
   1117 		s = splbio();
   1118 		wd->drvp->drive_flags |= DRIVE_RESET;
   1119 		splx(s);
   1120 	}
   1121 #ifdef HAS_BAD144_HANDLING
   1122 	if ((lp->d_flags & D_BADSECT) != 0)
   1123 		bad144intern(wd);
   1124 #endif
   1125 }
   1126 
   1127 void
   1128 wdperror(const struct wd_softc *wd)
   1129 {
   1130 	static const char *const errstr0_3[] = {"address mark not found",
   1131 	    "track 0 not found", "aborted command", "media change requested",
   1132 	    "id not found", "media changed", "uncorrectable data error",
   1133 	    "bad block detected"};
   1134 	static const char *const errstr4_5[] = {
   1135 	    "obsolete (address mark not found)",
   1136 	    "no media/write protected", "aborted command",
   1137 	    "media change requested", "id not found", "media changed",
   1138 	    "uncorrectable data error", "interface CRC error"};
   1139 	const char *const *errstr;
   1140 	int i;
   1141 	const char *sep = "";
   1142 
   1143 	const char *devname = wd->sc_dev.dv_xname;
   1144 	struct ata_drive_datas *drvp = wd->drvp;
   1145 	int errno = wd->sc_wdc_bio.r_error;
   1146 
   1147 	if (drvp->ata_vers >= 4)
   1148 		errstr = errstr4_5;
   1149 	else
   1150 		errstr = errstr0_3;
   1151 
   1152 	printf("%s: (", devname);
   1153 
   1154 	if (errno == 0)
   1155 		printf("error not notified");
   1156 
   1157 	for (i = 0; i < 8; i++) {
   1158 		if (errno & (1 << i)) {
   1159 			printf("%s%s", sep, errstr[i]);
   1160 			sep = ", ";
   1161 		}
   1162 	}
   1163 	printf(")\n");
   1164 }
   1165 
   1166 int
   1167 wdioctl(dev_t dev, u_long xfer, void *addr, int flag, struct lwp *l)
   1168 {
   1169 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
   1170 	int error = 0, s;
   1171 #ifdef __HAVE_OLD_DISKLABEL
   1172 	struct disklabel *newlabel = NULL;
   1173 #endif
   1174 
   1175 	ATADEBUG_PRINT(("wdioctl\n"), DEBUG_FUNCS);
   1176 
   1177 	if ((wd->sc_flags & WDF_LOADED) == 0)
   1178 		return EIO;
   1179 
   1180 	error = disk_ioctl(&wd->sc_dk, xfer, addr, flag, l);
   1181 	if (error != EPASSTHROUGH)
   1182 		return (error);
   1183 
   1184 	switch (xfer) {
   1185 #ifdef HAS_BAD144_HANDLING
   1186 	case DIOCSBAD:
   1187 		if ((flag & FWRITE) == 0)
   1188 			return EBADF;
   1189 		wd->sc_dk.dk_cpulabel->bad = *(struct dkbad *)addr;
   1190 		wd->sc_dk.dk_label->d_flags |= D_BADSECT;
   1191 		bad144intern(wd);
   1192 		return 0;
   1193 #endif
   1194 #ifdef WD_SOFTBADSECT
   1195 	case DIOCBSLIST :
   1196 	{
   1197 		u_int32_t count, missing, skip;
   1198 		struct disk_badsecinfo dbsi;
   1199 		struct disk_badsectors *dbs;
   1200 		size_t available;
   1201 		uint8_t *laddr;
   1202 
   1203 		dbsi = *(struct disk_badsecinfo *)addr;
   1204 		missing = wd->sc_bscount;
   1205 		count = 0;
   1206 		available = dbsi.dbsi_bufsize;
   1207 		skip = dbsi.dbsi_skip;
   1208 		laddr = (uint8_t *)dbsi.dbsi_buffer;
   1209 
   1210 		/*
   1211 		 * We start this loop with the expectation that all of the
   1212 		 * entries will be missed and decrement this counter each
   1213 		 * time we either skip over one (already copied out) or
   1214 		 * we actually copy it back to user space.  The structs
   1215 		 * holding the bad sector information are copied directly
   1216 		 * back to user space whilst the summary is returned via
   1217 		 * the struct passed in via the ioctl.
   1218 		 */
   1219 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next) {
   1220 			if (skip > 0) {
   1221 				missing--;
   1222 				skip--;
   1223 				continue;
   1224 			}
   1225 			if (available < sizeof(*dbs))
   1226 				break;
   1227 			available -= sizeof(*dbs);
   1228 			copyout(dbs, laddr, sizeof(*dbs));
   1229 			laddr += sizeof(*dbs);
   1230 			missing--;
   1231 			count++;
   1232 		}
   1233 		dbsi.dbsi_left = missing;
   1234 		dbsi.dbsi_copied = count;
   1235 		*(struct disk_badsecinfo *)addr = dbsi;
   1236 		return 0;
   1237 	}
   1238 
   1239 	case DIOCBSFLUSH :
   1240 		/* Clean out the bad sector list */
   1241 		while (!SLIST_EMPTY(&wd->sc_bslist)) {
   1242 			void *head = SLIST_FIRST(&wd->sc_bslist);
   1243 			SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
   1244 			free(head, M_TEMP);
   1245 		}
   1246 		wd->sc_bscount = 0;
   1247 		return 0;
   1248 #endif
   1249 	case DIOCGDINFO:
   1250 		*(struct disklabel *)addr = *(wd->sc_dk.dk_label);
   1251 		return 0;
   1252 #ifdef __HAVE_OLD_DISKLABEL
   1253 	case ODIOCGDINFO:
   1254 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
   1255 		if (newlabel == NULL)
   1256 			return EIO;
   1257 		*newlabel = *(wd->sc_dk.dk_label);
   1258 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
   1259 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
   1260 		else
   1261 			error = ENOTTY;
   1262 		free(newlabel, M_TEMP);
   1263 		return error;
   1264 #endif
   1265 
   1266 	case DIOCGPART:
   1267 		((struct partinfo *)addr)->disklab = wd->sc_dk.dk_label;
   1268 		((struct partinfo *)addr)->part =
   1269 		    &wd->sc_dk.dk_label->d_partitions[WDPART(dev)];
   1270 		return 0;
   1271 
   1272 	case DIOCWDINFO:
   1273 	case DIOCSDINFO:
   1274 #ifdef __HAVE_OLD_DISKLABEL
   1275 	case ODIOCWDINFO:
   1276 	case ODIOCSDINFO:
   1277 #endif
   1278 	{
   1279 		struct disklabel *lp;
   1280 
   1281 		if ((flag & FWRITE) == 0)
   1282 			return EBADF;
   1283 
   1284 #ifdef __HAVE_OLD_DISKLABEL
   1285 		if (xfer == ODIOCSDINFO || xfer == ODIOCWDINFO) {
   1286 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
   1287 			if (newlabel == NULL)
   1288 				return EIO;
   1289 			memset(newlabel, 0, sizeof newlabel);
   1290 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
   1291 			lp = newlabel;
   1292 		} else
   1293 #endif
   1294 		lp = (struct disklabel *)addr;
   1295 
   1296 		mutex_enter(&wd->sc_dk.dk_openlock);
   1297 		wd->sc_flags |= WDF_LABELLING;
   1298 
   1299 		error = setdisklabel(wd->sc_dk.dk_label,
   1300 		    lp, /*wd->sc_dk.dk_openmask : */0,
   1301 		    wd->sc_dk.dk_cpulabel);
   1302 		if (error == 0) {
   1303 			if (wd->drvp->state > RESET) {
   1304 				s = splbio();
   1305 				wd->drvp->drive_flags |= DRIVE_RESET;
   1306 				splx(s);
   1307 			}
   1308 			if (xfer == DIOCWDINFO
   1309 #ifdef __HAVE_OLD_DISKLABEL
   1310 			    || xfer == ODIOCWDINFO
   1311 #endif
   1312 			    )
   1313 				error = writedisklabel(WDLABELDEV(dev),
   1314 				    wdstrategy, wd->sc_dk.dk_label,
   1315 				    wd->sc_dk.dk_cpulabel);
   1316 		}
   1317 
   1318 		wd->sc_flags &= ~WDF_LABELLING;
   1319 		mutex_exit(&wd->sc_dk.dk_openlock);
   1320 #ifdef __HAVE_OLD_DISKLABEL
   1321 		if (newlabel != NULL)
   1322 			free(newlabel, M_TEMP);
   1323 #endif
   1324 		return error;
   1325 	}
   1326 
   1327 	case DIOCKLABEL:
   1328 		if (*(int *)addr)
   1329 			wd->sc_flags |= WDF_KLABEL;
   1330 		else
   1331 			wd->sc_flags &= ~WDF_KLABEL;
   1332 		return 0;
   1333 
   1334 	case DIOCWLABEL:
   1335 		if ((flag & FWRITE) == 0)
   1336 			return EBADF;
   1337 		if (*(int *)addr)
   1338 			wd->sc_flags |= WDF_WLABEL;
   1339 		else
   1340 			wd->sc_flags &= ~WDF_WLABEL;
   1341 		return 0;
   1342 
   1343 	case DIOCGDEFLABEL:
   1344 		wdgetdefaultlabel(wd, (struct disklabel *)addr);
   1345 		return 0;
   1346 #ifdef __HAVE_OLD_DISKLABEL
   1347 	case ODIOCGDEFLABEL:
   1348 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
   1349 		if (newlabel == NULL)
   1350 			return EIO;
   1351 		wdgetdefaultlabel(wd, newlabel);
   1352 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
   1353 			memcpy(addr, &newlabel, sizeof (struct olddisklabel));
   1354 		else
   1355 			error = ENOTTY;
   1356 		free(newlabel, M_TEMP);
   1357 		return error;
   1358 #endif
   1359 
   1360 #ifdef notyet
   1361 	case DIOCWFORMAT:
   1362 		if ((flag & FWRITE) == 0)
   1363 			return EBADF;
   1364 		{
   1365 		register struct format_op *fop;
   1366 		struct iovec aiov;
   1367 		struct uio auio;
   1368 
   1369 		fop = (struct format_op *)addr;
   1370 		aiov.iov_base = fop->df_buf;
   1371 		aiov.iov_len = fop->df_count;
   1372 		auio.uio_iov = &aiov;
   1373 		auio.uio_iovcnt = 1;
   1374 		auio.uio_resid = fop->df_count;
   1375 		auio.uio_offset =
   1376 			fop->df_startblk * wd->sc_dk.dk_label->d_secsize;
   1377 		auio.uio_vmspace = l->l_proc->p_vmspace;
   1378 		error = physio(wdformat, NULL, dev, B_WRITE, minphys,
   1379 		    &auio);
   1380 		fop->df_count -= auio.uio_resid;
   1381 		fop->df_reg[0] = wdc->sc_status;
   1382 		fop->df_reg[1] = wdc->sc_error;
   1383 		return error;
   1384 		}
   1385 #endif
   1386 	case DIOCGCACHE:
   1387 		return wd_getcache(wd, (int *)addr);
   1388 
   1389 	case DIOCSCACHE:
   1390 		return wd_setcache(wd, *(int *)addr);
   1391 
   1392 	case DIOCCACHESYNC:
   1393 		return wd_flushcache(wd, AT_WAIT);
   1394 
   1395 	case ATAIOCCOMMAND:
   1396 		/*
   1397 		 * Make sure this command is (relatively) safe first
   1398 		 */
   1399 		if ((((atareq_t *) addr)->flags & ATACMD_READ) == 0 &&
   1400 		    (flag & FWRITE) == 0)
   1401 			return (EBADF);
   1402 		{
   1403 		struct wd_ioctl *wi;
   1404 		atareq_t *atareq = (atareq_t *) addr;
   1405 		int error1;
   1406 
   1407 		wi = wi_get();
   1408 		wi->wi_softc = wd;
   1409 		wi->wi_atareq = *atareq;
   1410 
   1411 		if (atareq->datalen && atareq->flags &
   1412 		    (ATACMD_READ | ATACMD_WRITE)) {
   1413 			void *tbuf;
   1414 			if (atareq->datalen < DEV_BSIZE
   1415 			    && atareq->command == WDCC_IDENTIFY) {
   1416 				tbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
   1417 				wi->wi_iov.iov_base = tbuf;
   1418 				wi->wi_iov.iov_len = DEV_BSIZE;
   1419 				UIO_SETUP_SYSSPACE(&wi->wi_uio);
   1420 			} else {
   1421 				tbuf = NULL;
   1422 				wi->wi_iov.iov_base = atareq->databuf;
   1423 				wi->wi_iov.iov_len = atareq->datalen;
   1424 				wi->wi_uio.uio_vmspace = l->l_proc->p_vmspace;
   1425 			}
   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 			error1 = physio(wdioctlstrategy, &wi->wi_bp, dev,
   1433 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE,
   1434 			    minphys, &wi->wi_uio);
   1435 			if (tbuf != NULL && error1 == 0) {
   1436 				error1 = copyout(tbuf, atareq->databuf,
   1437 				    atareq->datalen);
   1438 				free(tbuf, M_TEMP);
   1439 			}
   1440 		} else {
   1441 			/* No need to call physio if we don't have any
   1442 			   user data */
   1443 			wi->wi_bp.b_flags = 0;
   1444 			wi->wi_bp.b_data = 0;
   1445 			wi->wi_bp.b_bcount = 0;
   1446 			wi->wi_bp.b_dev = 0;
   1447 			wi->wi_bp.b_proc = l->l_proc;
   1448 			wdioctlstrategy(&wi->wi_bp);
   1449 			error1 = wi->wi_bp.b_error;
   1450 		}
   1451 		*atareq = wi->wi_atareq;
   1452 		wi_free(wi);
   1453 		return(error1);
   1454 		}
   1455 
   1456 	case DIOCAWEDGE:
   1457 	    {
   1458 	    	struct dkwedge_info *dkw = (void *) addr;
   1459 
   1460 		if ((flag & FWRITE) == 0)
   1461 			return (EBADF);
   1462 
   1463 		/* If the ioctl happens here, the parent is us. */
   1464 		strcpy(dkw->dkw_parent, wd->sc_dev.dv_xname);
   1465 		return (dkwedge_add(dkw));
   1466 	    }
   1467 
   1468 	case DIOCDWEDGE:
   1469 	    {
   1470 	    	struct dkwedge_info *dkw = (void *) addr;
   1471 
   1472 		if ((flag & FWRITE) == 0)
   1473 			return (EBADF);
   1474 
   1475 		/* If the ioctl happens here, the parent is us. */
   1476 		strcpy(dkw->dkw_parent, wd->sc_dev.dv_xname);
   1477 		return (dkwedge_del(dkw));
   1478 	    }
   1479 
   1480 	case DIOCLWEDGES:
   1481 	    {
   1482 	    	struct dkwedge_list *dkwl = (void *) addr;
   1483 
   1484 		return (dkwedge_list(&wd->sc_dk, dkwl, l));
   1485 	    }
   1486 
   1487 	case DIOCGSTRATEGY:
   1488 	    {
   1489 		struct disk_strategy *dks = (void *)addr;
   1490 
   1491 		s = splbio();
   1492 		strlcpy(dks->dks_name, bufq_getstrategyname(wd->sc_q),
   1493 		    sizeof(dks->dks_name));
   1494 		splx(s);
   1495 		dks->dks_paramlen = 0;
   1496 
   1497 		return 0;
   1498 	    }
   1499 
   1500 	case DIOCSSTRATEGY:
   1501 	    {
   1502 		struct disk_strategy *dks = (void *)addr;
   1503 		struct bufq_state *new;
   1504 		struct bufq_state *old;
   1505 
   1506 		if ((flag & FWRITE) == 0) {
   1507 			return EBADF;
   1508 		}
   1509 		if (dks->dks_param != NULL) {
   1510 			return EINVAL;
   1511 		}
   1512 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
   1513 		error = bufq_alloc(&new, dks->dks_name,
   1514 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
   1515 		if (error) {
   1516 			return error;
   1517 		}
   1518 		s = splbio();
   1519 		old = wd->sc_q;
   1520 		bufq_move(new, old);
   1521 		wd->sc_q = new;
   1522 		splx(s);
   1523 		bufq_free(old);
   1524 
   1525 		return 0;
   1526 	    }
   1527 
   1528 	default:
   1529 		return ENOTTY;
   1530 	}
   1531 
   1532 #ifdef DIAGNOSTIC
   1533 	panic("wdioctl: impossible");
   1534 #endif
   1535 }
   1536 
   1537 #ifdef B_FORMAT
   1538 int
   1539 wdformat(struct buf *bp)
   1540 {
   1541 
   1542 	bp->b_flags |= B_FORMAT;
   1543 	return wdstrategy(bp);
   1544 }
   1545 #endif
   1546 
   1547 int
   1548 wdsize(dev_t dev)
   1549 {
   1550 	struct wd_softc *wd;
   1551 	int part, omask;
   1552 	int size;
   1553 
   1554 	ATADEBUG_PRINT(("wdsize\n"), DEBUG_FUNCS);
   1555 
   1556 	wd = device_lookup(&wd_cd, WDUNIT(dev));
   1557 	if (wd == NULL)
   1558 		return (-1);
   1559 
   1560 	part = WDPART(dev);
   1561 	omask = wd->sc_dk.dk_openmask & (1 << part);
   1562 
   1563 	if (omask == 0 && wdopen(dev, 0, S_IFBLK, NULL) != 0)
   1564 		return (-1);
   1565 	if (wd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
   1566 		size = -1;
   1567 	else
   1568 		size = wd->sc_dk.dk_label->d_partitions[part].p_size *
   1569 		    (wd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
   1570 	if (omask == 0 && wdclose(dev, 0, S_IFBLK, NULL) != 0)
   1571 		return (-1);
   1572 	return (size);
   1573 }
   1574 
   1575 /* #define WD_DUMP_NOT_TRUSTED if you just want to watch */
   1576 static int wddoingadump = 0;
   1577 static int wddumprecalibrated = 0;
   1578 
   1579 /*
   1580  * Dump core after a system crash.
   1581  */
   1582 int
   1583 wddump(dev_t dev, daddr_t blkno, void *va, size_t size)
   1584 {
   1585 	struct wd_softc *wd;	/* disk unit to do the I/O */
   1586 	struct disklabel *lp;   /* disk's disklabel */
   1587 	int part, err;
   1588 	int nblks;	/* total number of sectors left to write */
   1589 
   1590 	/* Check if recursive dump; if so, punt. */
   1591 	if (wddoingadump)
   1592 		return EFAULT;
   1593 	wddoingadump = 1;
   1594 
   1595 	wd = device_lookup(&wd_cd, WDUNIT(dev));
   1596 	if (wd == NULL)
   1597 		return (ENXIO);
   1598 
   1599 	part = WDPART(dev);
   1600 
   1601 	/* Convert to disk sectors.  Request must be a multiple of size. */
   1602 	lp = wd->sc_dk.dk_label;
   1603 	if ((size % lp->d_secsize) != 0)
   1604 		return EFAULT;
   1605 	nblks = size / lp->d_secsize;
   1606 	blkno = blkno / (lp->d_secsize / DEV_BSIZE);
   1607 
   1608 	/* Check transfer bounds against partition size. */
   1609 	if ((blkno < 0) || ((blkno + nblks) > lp->d_partitions[part].p_size))
   1610 		return EINVAL;
   1611 
   1612 	/* Offset block number to start of partition. */
   1613 	blkno += lp->d_partitions[part].p_offset;
   1614 
   1615 	/* Recalibrate, if first dump transfer. */
   1616 	if (wddumprecalibrated == 0) {
   1617 		wddumprecalibrated = 1;
   1618 		(*wd->atabus->ata_reset_drive)(wd->drvp,
   1619 					       AT_POLL | AT_RST_EMERG);
   1620 		wd->drvp->state = RESET;
   1621 	}
   1622 
   1623 	wd->sc_bp = NULL;
   1624 	wd->sc_wdc_bio.blkno = blkno;
   1625 	wd->sc_wdc_bio.flags = ATA_POLL;
   1626 	if (wd->sc_flags & WDF_LBA48 &&
   1627 	    (blkno > LBA48_THRESHOLD ||
   1628     	    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) != 0))
   1629 		wd->sc_wdc_bio.flags |= ATA_LBA48;
   1630 	if (wd->sc_flags & WDF_LBA)
   1631 		wd->sc_wdc_bio.flags |= ATA_LBA;
   1632 	wd->sc_wdc_bio.bcount = nblks * lp->d_secsize;
   1633 	wd->sc_wdc_bio.databuf = va;
   1634 #ifndef WD_DUMP_NOT_TRUSTED
   1635 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
   1636 	case ATACMD_TRY_AGAIN:
   1637 		panic("wddump: try again");
   1638 		break;
   1639 	case ATACMD_QUEUED:
   1640 		panic("wddump: polled command has been queued");
   1641 		break;
   1642 	case ATACMD_COMPLETE:
   1643 		break;
   1644 	}
   1645 	switch(wd->sc_wdc_bio.error) {
   1646 	case TIMEOUT:
   1647 		printf("wddump: device timed out");
   1648 		err = EIO;
   1649 		break;
   1650 	case ERR_DF:
   1651 		printf("wddump: drive fault");
   1652 		err = EIO;
   1653 		break;
   1654 	case ERR_DMA:
   1655 		printf("wddump: DMA error");
   1656 		err = EIO;
   1657 		break;
   1658 	case ERROR:
   1659 		printf("wddump: ");
   1660 		wdperror(wd);
   1661 		err = EIO;
   1662 		break;
   1663 	case NOERROR:
   1664 		err = 0;
   1665 		break;
   1666 	default:
   1667 		panic("wddump: unknown error type");
   1668 	}
   1669 	if (err != 0) {
   1670 		printf("\n");
   1671 		return err;
   1672 	}
   1673 #else	/* WD_DUMP_NOT_TRUSTED */
   1674 	/* Let's just talk about this first... */
   1675 	printf("wd%d: dump addr 0x%x, cylin %d, head %d, sector %d\n",
   1676 	    unit, va, cylin, head, sector);
   1677 	delay(500 * 1000);	/* half a second */
   1678 #endif
   1679 
   1680 	wddoingadump = 0;
   1681 	return 0;
   1682 }
   1683 
   1684 #ifdef HAS_BAD144_HANDLING
   1685 /*
   1686  * Internalize the bad sector table.
   1687  */
   1688 void
   1689 bad144intern(struct wd_softc *wd)
   1690 {
   1691 	struct dkbad *bt = &wd->sc_dk.dk_cpulabel->bad;
   1692 	struct disklabel *lp = wd->sc_dk.dk_label;
   1693 	int i = 0;
   1694 
   1695 	ATADEBUG_PRINT(("bad144intern\n"), DEBUG_XFERS);
   1696 
   1697 	for (; i < NBT_BAD; i++) {
   1698 		if (bt->bt_bad[i].bt_cyl == 0xffff)
   1699 			break;
   1700 		wd->sc_badsect[i] =
   1701 		    bt->bt_bad[i].bt_cyl * lp->d_secpercyl +
   1702 		    (bt->bt_bad[i].bt_trksec >> 8) * lp->d_nsectors +
   1703 		    (bt->bt_bad[i].bt_trksec & 0xff);
   1704 	}
   1705 	for (; i < NBT_BAD+1; i++)
   1706 		wd->sc_badsect[i] = -1;
   1707 }
   1708 #endif
   1709 
   1710 static void
   1711 wd_params_to_properties(struct wd_softc *wd, struct ataparams *params)
   1712 {
   1713 	prop_dictionary_t disk_info, odisk_info, geom;
   1714 	const char *cp;
   1715 
   1716 	disk_info = prop_dictionary_create();
   1717 
   1718 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
   1719 		cp = "ST506";
   1720 	else {
   1721 		/* XXX Should have a case for ATA here, too. */
   1722 		cp = "ESDI";
   1723 	}
   1724 	prop_dictionary_set_cstring_nocopy(disk_info, "type", cp);
   1725 
   1726 	geom = prop_dictionary_create();
   1727 
   1728 	prop_dictionary_set_uint64(geom, "sectors-per-unit", wd->sc_capacity);
   1729 
   1730 	prop_dictionary_set_uint32(geom, "sector-size",
   1731 				   DEV_BSIZE /* XXX 512? */);
   1732 
   1733 	prop_dictionary_set_uint16(geom, "sectors-per-track",
   1734 				   wd->sc_params.atap_sectors);
   1735 
   1736 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
   1737 				   wd->sc_params.atap_heads);
   1738 
   1739 	if (wd->sc_flags & WDF_LBA)
   1740 		prop_dictionary_set_uint64(geom, "cylinders-per-unit",
   1741 					   wd->sc_capacity /
   1742 					       (wd->sc_params.atap_heads *
   1743 					        wd->sc_params.atap_sectors));
   1744 	else
   1745 		prop_dictionary_set_uint16(geom, "cylinders-per-unit",
   1746 					   wd->sc_params.atap_cylinders);
   1747 
   1748 	prop_dictionary_set(disk_info, "geometry", geom);
   1749 	prop_object_release(geom);
   1750 
   1751 	prop_dictionary_set(device_properties(&wd->sc_dev),
   1752 			    "disk-info", disk_info);
   1753 
   1754 	/*
   1755 	 * Don't release disk_info here; we keep a reference to it.
   1756 	 * disk_detach() will release it when we go away.
   1757 	 */
   1758 
   1759 	odisk_info = wd->sc_dk.dk_info;
   1760 	wd->sc_dk.dk_info = disk_info;
   1761 	if (odisk_info)
   1762 		prop_object_release(odisk_info);
   1763 }
   1764 
   1765 int
   1766 wd_get_params(struct wd_softc *wd, u_int8_t flags, struct ataparams *params)
   1767 {
   1768 
   1769 	switch (wd->atabus->ata_get_params(wd->drvp, flags, params)) {
   1770 	case CMD_AGAIN:
   1771 		return 1;
   1772 	case CMD_ERR:
   1773 		/*
   1774 		 * We `know' there's a drive here; just assume it's old.
   1775 		 * This geometry is only used to read the MBR and print a
   1776 		 * (false) attach message.
   1777 		 */
   1778 		strncpy(params->atap_model, "ST506",
   1779 		    sizeof params->atap_model);
   1780 		params->atap_config = ATA_CFG_FIXED;
   1781 		params->atap_cylinders = 1024;
   1782 		params->atap_heads = 8;
   1783 		params->atap_sectors = 17;
   1784 		params->atap_multi = 1;
   1785 		params->atap_capabilities1 = params->atap_capabilities2 = 0;
   1786 		wd->drvp->ata_vers = -1; /* Mark it as pre-ATA */
   1787 		/* FALLTHROUGH */
   1788 	case CMD_OK:
   1789 		wd_params_to_properties(wd, params);
   1790 		return 0;
   1791 	default:
   1792 		panic("wd_get_params: bad return code from ata_get_params");
   1793 		/* NOTREACHED */
   1794 	}
   1795 }
   1796 
   1797 int
   1798 wd_getcache(struct wd_softc *wd, int *bitsp)
   1799 {
   1800 	struct ataparams params;
   1801 
   1802 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
   1803 		return EIO;
   1804 	if (params.atap_cmd_set1 == 0x0000 ||
   1805 	    params.atap_cmd_set1 == 0xffff ||
   1806 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0) {
   1807 		*bitsp = 0;
   1808 		return 0;
   1809 	}
   1810 	*bitsp = DKCACHE_WCHANGE | DKCACHE_READ;
   1811 	if (params.atap_cmd1_en & WDC_CMD1_CACHE)
   1812 		*bitsp |= DKCACHE_WRITE;
   1813 
   1814 	return 0;
   1815 }
   1816 
   1817 const char at_errbits[] = "\20\10ERROR\11TIMEOU\12DF";
   1818 
   1819 int
   1820 wd_setcache(struct wd_softc *wd, int bits)
   1821 {
   1822 	struct ataparams params;
   1823 	struct ata_command ata_c;
   1824 
   1825 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
   1826 		return EIO;
   1827 
   1828 	if (params.atap_cmd_set1 == 0x0000 ||
   1829 	    params.atap_cmd_set1 == 0xffff ||
   1830 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0)
   1831 		return EOPNOTSUPP;
   1832 
   1833 	if ((bits & DKCACHE_READ) == 0 ||
   1834 	    (bits & DKCACHE_SAVE) != 0)
   1835 		return EOPNOTSUPP;
   1836 
   1837 	memset(&ata_c, 0, sizeof(struct ata_command));
   1838 	ata_c.r_command = SET_FEATURES;
   1839 	ata_c.r_st_bmask = 0;
   1840 	ata_c.r_st_pmask = 0;
   1841 	ata_c.timeout = 30000; /* 30s timeout */
   1842 	ata_c.flags = AT_WAIT;
   1843 	if (bits & DKCACHE_WRITE)
   1844 		ata_c.r_features = WDSF_WRITE_CACHE_EN;
   1845 	else
   1846 		ata_c.r_features = WDSF_WRITE_CACHE_DS;
   1847 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
   1848 		printf("%s: wd_setcache command not complete\n",
   1849 		    wd->sc_dev.dv_xname);
   1850 		return EIO;
   1851 	}
   1852 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   1853 		char sbuf[sizeof(at_errbits) + 64];
   1854 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
   1855 		printf("%s: wd_setcache: status=%s\n", wd->sc_dev.dv_xname,
   1856 		    sbuf);
   1857 		return EIO;
   1858 	}
   1859 	return 0;
   1860 }
   1861 
   1862 int
   1863 wd_standby(struct wd_softc *wd, int flags)
   1864 {
   1865 	struct ata_command ata_c;
   1866 
   1867 	memset(&ata_c, 0, sizeof(struct ata_command));
   1868 	ata_c.r_command = WDCC_STANDBY_IMMED;
   1869 	ata_c.r_st_bmask = WDCS_DRDY;
   1870 	ata_c.r_st_pmask = WDCS_DRDY;
   1871 	ata_c.flags = flags;
   1872 	ata_c.timeout = 30000; /* 30s timeout */
   1873 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
   1874 		printf("%s: standby immediate command didn't complete\n",
   1875 		    wd->sc_dev.dv_xname);
   1876 		return EIO;
   1877 	}
   1878 	if (ata_c.flags & AT_ERROR) {
   1879 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
   1880 			return ENODEV;
   1881 	}
   1882 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   1883 		char sbuf[sizeof(at_errbits) + 64];
   1884 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
   1885 		printf("%s: wd_standby: status=%s\n", wd->sc_dev.dv_xname,
   1886 		    sbuf);
   1887 		return EIO;
   1888 	}
   1889 	return 0;
   1890 }
   1891 
   1892 int
   1893 wd_flushcache(struct wd_softc *wd, int flags)
   1894 {
   1895 	struct ata_command ata_c;
   1896 
   1897 	/*
   1898 	 * WDCC_FLUSHCACHE is here since ATA-4, but some drives report
   1899 	 * only ATA-2 and still support it.
   1900 	 */
   1901 	if (wd->drvp->ata_vers < 4 &&
   1902 	    ((wd->sc_params.atap_cmd_set2 & WDC_CMD2_FC) == 0 ||
   1903 	    wd->sc_params.atap_cmd_set2 == 0xffff))
   1904 		return ENODEV;
   1905 	memset(&ata_c, 0, sizeof(struct ata_command));
   1906 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0 &&
   1907 	    (wd->sc_params.atap_cmd2_en & ATA_CMD2_FCE) != 0)
   1908 		ata_c.r_command = WDCC_FLUSHCACHE_EXT;
   1909 	else
   1910 		ata_c.r_command = WDCC_FLUSHCACHE;
   1911 	ata_c.r_st_bmask = WDCS_DRDY;
   1912 	ata_c.r_st_pmask = WDCS_DRDY;
   1913 	ata_c.flags = flags;
   1914 	ata_c.timeout = 30000; /* 30s timeout */
   1915 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
   1916 		printf("%s: flush cache command didn't complete\n",
   1917 		    wd->sc_dev.dv_xname);
   1918 		return EIO;
   1919 	}
   1920 	if (ata_c.flags & AT_ERROR) {
   1921 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
   1922 			return ENODEV;
   1923 	}
   1924 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   1925 		char sbuf[sizeof(at_errbits) + 64];
   1926 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
   1927 		printf("%s: wd_flushcache: status=%s\n", wd->sc_dev.dv_xname,
   1928 		    sbuf);
   1929 		return EIO;
   1930 	}
   1931 	return 0;
   1932 }
   1933 
   1934 void
   1935 wd_shutdown(void *arg)
   1936 {
   1937 
   1938 	struct wd_softc *wd = arg;
   1939 	wd_flushcache(wd, AT_POLL);
   1940 }
   1941 
   1942 /*
   1943  * Allocate space for a ioctl queue structure.  Mostly taken from
   1944  * scsipi_ioctl.c
   1945  */
   1946 struct wd_ioctl *
   1947 wi_get(void)
   1948 {
   1949 	struct wd_ioctl *wi;
   1950 	int s;
   1951 
   1952 	wi = malloc(sizeof(struct wd_ioctl), M_TEMP, M_WAITOK|M_ZERO);
   1953 	buf_init(&wi->wi_bp);
   1954 	s = splbio();
   1955 	LIST_INSERT_HEAD(&wi_head, wi, wi_list);
   1956 	splx(s);
   1957 	return (wi);
   1958 }
   1959 
   1960 /*
   1961  * Free an ioctl structure and remove it from our list
   1962  */
   1963 
   1964 void
   1965 wi_free(struct wd_ioctl *wi)
   1966 {
   1967 	int s;
   1968 
   1969 	s = splbio();
   1970 	LIST_REMOVE(wi, wi_list);
   1971 	splx(s);
   1972 	buf_destroy(&wi->wi_bp);
   1973 	free(wi, M_TEMP);
   1974 }
   1975 
   1976 /*
   1977  * Find a wd_ioctl structure based on the struct buf.
   1978  */
   1979 
   1980 struct wd_ioctl *
   1981 wi_find(struct buf *bp)
   1982 {
   1983 	struct wd_ioctl *wi;
   1984 	int s;
   1985 
   1986 	s = splbio();
   1987 	for (wi = wi_head.lh_first; wi != 0; wi = wi->wi_list.le_next)
   1988 		if (bp == &wi->wi_bp)
   1989 			break;
   1990 	splx(s);
   1991 	return (wi);
   1992 }
   1993 
   1994 /*
   1995  * Ioctl pseudo strategy routine
   1996  *
   1997  * This is mostly stolen from scsipi_ioctl.c:scsistrategy().  What
   1998  * happens here is:
   1999  *
   2000  * - wdioctl() queues a wd_ioctl structure.
   2001  *
   2002  * - wdioctl() calls physio/wdioctlstrategy based on whether or not
   2003  *   user space I/O is required.  If physio() is called, physio() eventually
   2004  *   calls wdioctlstrategy().
   2005  *
   2006  * - In either case, wdioctlstrategy() calls wd->atabus->ata_exec_command()
   2007  *   to perform the actual command
   2008  *
   2009  * The reason for the use of the pseudo strategy routine is because
   2010  * when doing I/O to/from user space, physio _really_ wants to be in
   2011  * the loop.  We could put the entire buffer into the ioctl request
   2012  * structure, but that won't scale if we want to do things like download
   2013  * microcode.
   2014  */
   2015 
   2016 void
   2017 wdioctlstrategy(struct buf *bp)
   2018 {
   2019 	struct wd_ioctl *wi;
   2020 	struct ata_command ata_c;
   2021 	int error = 0;
   2022 
   2023 	wi = wi_find(bp);
   2024 	if (wi == NULL) {
   2025 		printf("wdioctlstrategy: "
   2026 		    "No matching ioctl request found in queue\n");
   2027 		error = EINVAL;
   2028 		goto bad;
   2029 	}
   2030 
   2031 	memset(&ata_c, 0, sizeof(ata_c));
   2032 
   2033 	/*
   2034 	 * Abort if physio broke up the transfer
   2035 	 */
   2036 
   2037 	if (bp->b_bcount != wi->wi_atareq.datalen) {
   2038 		printf("physio split wd ioctl request... cannot proceed\n");
   2039 		error = EIO;
   2040 		goto bad;
   2041 	}
   2042 
   2043 	/*
   2044 	 * Abort if we didn't get a buffer size that was a multiple of
   2045 	 * our sector size (or was larger than NBBY)
   2046 	 */
   2047 
   2048 	if ((bp->b_bcount % wi->wi_softc->sc_dk.dk_label->d_secsize) != 0 ||
   2049 	    (bp->b_bcount / wi->wi_softc->sc_dk.dk_label->d_secsize) >=
   2050 	     (1 << NBBY)) {
   2051 		error = EINVAL;
   2052 		goto bad;
   2053 	}
   2054 
   2055 	/*
   2056 	 * Make sure a timeout was supplied in the ioctl request
   2057 	 */
   2058 
   2059 	if (wi->wi_atareq.timeout == 0) {
   2060 		error = EINVAL;
   2061 		goto bad;
   2062 	}
   2063 
   2064 	if (wi->wi_atareq.flags & ATACMD_READ)
   2065 		ata_c.flags |= AT_READ;
   2066 	else if (wi->wi_atareq.flags & ATACMD_WRITE)
   2067 		ata_c.flags |= AT_WRITE;
   2068 
   2069 	if (wi->wi_atareq.flags & ATACMD_READREG)
   2070 		ata_c.flags |= AT_READREG;
   2071 
   2072 	ata_c.flags |= AT_WAIT;
   2073 
   2074 	ata_c.timeout = wi->wi_atareq.timeout;
   2075 	ata_c.r_command = wi->wi_atareq.command;
   2076 	ata_c.r_head = wi->wi_atareq.head & 0x0f;
   2077 	ata_c.r_cyl = wi->wi_atareq.cylinder;
   2078 	ata_c.r_sector = wi->wi_atareq.sec_num;
   2079 	ata_c.r_count = wi->wi_atareq.sec_count;
   2080 	ata_c.r_features = wi->wi_atareq.features;
   2081 	ata_c.r_st_bmask = WDCS_DRDY;
   2082 	ata_c.r_st_pmask = WDCS_DRDY;
   2083 	ata_c.data = wi->wi_bp.b_data;
   2084 	ata_c.bcount = wi->wi_bp.b_bcount;
   2085 
   2086 	if (wi->wi_softc->atabus->ata_exec_command(wi->wi_softc->drvp, &ata_c)
   2087 	    != ATACMD_COMPLETE) {
   2088 		wi->wi_atareq.retsts = ATACMD_ERROR;
   2089 		goto bad;
   2090 	}
   2091 
   2092 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   2093 		if (ata_c.flags & AT_ERROR) {
   2094 			wi->wi_atareq.retsts = ATACMD_ERROR;
   2095 			wi->wi_atareq.error = ata_c.r_error;
   2096 		} else if (ata_c.flags & AT_DF)
   2097 			wi->wi_atareq.retsts = ATACMD_DF;
   2098 		else
   2099 			wi->wi_atareq.retsts = ATACMD_TIMEOUT;
   2100 	} else {
   2101 		wi->wi_atareq.retsts = ATACMD_OK;
   2102 		if (wi->wi_atareq.flags & ATACMD_READREG) {
   2103 			wi->wi_atareq.head = ata_c.r_head ;
   2104 			wi->wi_atareq.cylinder = ata_c.r_cyl;
   2105 			wi->wi_atareq.sec_num = ata_c.r_sector;
   2106 			wi->wi_atareq.sec_count = ata_c.r_count;
   2107 			wi->wi_atareq.features = ata_c.r_features;
   2108 			wi->wi_atareq.error = ata_c.r_error;
   2109 		}
   2110 	}
   2111 
   2112 	bp->b_error = 0;
   2113 	biodone(bp);
   2114 	return;
   2115 bad:
   2116 	bp->b_error = error;
   2117 	biodone(bp);
   2118 }
   2119