Home | History | Annotate | Line # | Download | only in ata
wd.c revision 1.344
      1 /*	$NetBSD: wd.c,v 1.344 2007/09/05 05:36:19 jnemeth 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 1.344 2007/09/05 05:36:19 jnemeth 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 <machine/intr.h>
     97 #include <machine/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 	{ NULL,
    258 	  0 }
    259 };
    260 
    261 static const struct wd_quirk *
    262 wd_lookup_quirks(const char *name)
    263 {
    264 	const struct wd_quirk *wdq;
    265 	const char *estr;
    266 
    267 	for (wdq = wd_quirk_table; wdq->wdq_match != NULL; wdq++) {
    268 		/*
    269 		 * We only want exact matches (which include matches
    270 		 * against globbing characters).
    271 		 */
    272 		if (pmatch(name, wdq->wdq_match, &estr) == 2)
    273 			return (wdq);
    274 	}
    275 	return (NULL);
    276 }
    277 
    278 int
    279 wdprobe(struct device *parent, struct cfdata *match, void *aux)
    280 {
    281 	struct ata_device *adev = aux;
    282 
    283 	if (adev == NULL)
    284 		return 0;
    285 	if (adev->adev_bustype->bustype_type != SCSIPI_BUSTYPE_ATA)
    286 		return 0;
    287 
    288 	if (match->cf_loc[ATA_HLCF_DRIVE] != ATA_HLCF_DRIVE_DEFAULT &&
    289 	    match->cf_loc[ATA_HLCF_DRIVE] != adev->adev_drv_data->drive)
    290 		return 0;
    291 	return 1;
    292 }
    293 
    294 void
    295 wdattach(struct device *parent, struct device *self, void *aux)
    296 {
    297 	struct wd_softc *wd = (void *)self;
    298 	struct ata_device *adev= aux;
    299 	int i, blank;
    300 	char tbuf[41], pbuf[9], c, *p, *q;
    301 	const struct wd_quirk *wdq;
    302 	ATADEBUG_PRINT(("wdattach\n"), DEBUG_FUNCS | DEBUG_PROBE);
    303 
    304 	callout_init(&wd->sc_restart_ch, 0);
    305 	bufq_alloc(&wd->sc_q, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
    306 #ifdef WD_SOFTBADSECT
    307 	SLIST_INIT(&wd->sc_bslist);
    308 #endif
    309 	wd->atabus = adev->adev_bustype;
    310 	wd->openings = adev->adev_openings;
    311 	wd->drvp = adev->adev_drv_data;
    312 
    313 	wd->drvp->drv_done = wddone;
    314 	wd->drvp->drv_softc = &wd->sc_dev;
    315 
    316 	aprint_naive("\n");
    317 
    318 	/* read our drive info */
    319 	if (wd_get_params(wd, AT_WAIT, &wd->sc_params) != 0) {
    320 		aprint_error("\n%s: IDENTIFY failed\n", wd->sc_dev.dv_xname);
    321 		return;
    322 	}
    323 
    324 	for (blank = 0, p = wd->sc_params.atap_model, q = tbuf, i = 0;
    325 	    i < sizeof(wd->sc_params.atap_model); i++) {
    326 		c = *p++;
    327 		if (c == '\0')
    328 			break;
    329 		if (c != ' ') {
    330 			if (blank) {
    331 				*q++ = ' ';
    332 				blank = 0;
    333 			}
    334 			*q++ = c;
    335 		} else
    336 			blank = 1;
    337 	}
    338 	*q++ = '\0';
    339 
    340 	aprint_normal(": <%s>\n", tbuf);
    341 
    342 	wdq = wd_lookup_quirks(tbuf);
    343 	if (wdq != NULL)
    344 		wd->sc_quirks = wdq->wdq_quirks;
    345 
    346 	if (wd->sc_quirks != 0) {
    347 		char sbuf[sizeof(WD_QUIRK_FMT) + 64];
    348 		bitmask_snprintf(wd->sc_quirks, WD_QUIRK_FMT,
    349 		    sbuf, sizeof(sbuf));
    350 		aprint_normal("%s: quirks %s\n", wd->sc_dev.dv_xname, sbuf);
    351 	}
    352 
    353 	if ((wd->sc_params.atap_multi & 0xff) > 1) {
    354 		wd->sc_multi = wd->sc_params.atap_multi & 0xff;
    355 	} else {
    356 		wd->sc_multi = 1;
    357 	}
    358 
    359 	aprint_verbose("%s: drive supports %d-sector PIO transfers,",
    360 	    wd->sc_dev.dv_xname, wd->sc_multi);
    361 
    362 	/* 48-bit LBA addressing */
    363 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0)
    364 		wd->sc_flags |= WDF_LBA48;
    365 
    366 	/* Prior to ATA-4, LBA was optional. */
    367 	if ((wd->sc_params.atap_capabilities1 & WDC_CAP_LBA) != 0)
    368 		wd->sc_flags |= WDF_LBA;
    369 #if 0
    370 	/* ATA-4 requires LBA. */
    371 	if (wd->sc_params.atap_ataversion != 0xffff &&
    372 	    wd->sc_params.atap_ataversion >= WDC_VER_ATA4)
    373 		wd->sc_flags |= WDF_LBA;
    374 #endif
    375 
    376 	if ((wd->sc_flags & WDF_LBA48) != 0) {
    377 		aprint_verbose(" LBA48 addressing\n");
    378 		wd->sc_capacity =
    379 		    ((u_int64_t) wd->sc_params.__reserved6[11] << 48) |
    380 		    ((u_int64_t) wd->sc_params.__reserved6[10] << 32) |
    381 		    ((u_int64_t) wd->sc_params.__reserved6[9]  << 16) |
    382 		    ((u_int64_t) wd->sc_params.__reserved6[8]  << 0);
    383 	} else if ((wd->sc_flags & WDF_LBA) != 0) {
    384 		aprint_verbose(" LBA addressing\n");
    385 		wd->sc_capacity =
    386 		    ((u_int64_t)wd->sc_params.atap_capacity[1] << 16) |
    387 		    wd->sc_params.atap_capacity[0];
    388 	} else {
    389 		aprint_verbose(" chs addressing\n");
    390 		wd->sc_capacity =
    391 		    wd->sc_params.atap_cylinders *
    392 		    wd->sc_params.atap_heads *
    393 		    wd->sc_params.atap_sectors;
    394 	}
    395 	format_bytes(pbuf, sizeof(pbuf), wd->sc_capacity * DEV_BSIZE);
    396 	aprint_normal("%s: %s, %d cyl, %d head, %d sec, "
    397 	    "%d bytes/sect x %llu sectors\n",
    398 	    self->dv_xname, pbuf,
    399 	    (wd->sc_flags & WDF_LBA) ? (int)(wd->sc_capacity /
    400 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors)) :
    401 		wd->sc_params.atap_cylinders,
    402 	    wd->sc_params.atap_heads, wd->sc_params.atap_sectors,
    403 	    DEV_BSIZE, (unsigned long long)wd->sc_capacity);
    404 
    405 	ATADEBUG_PRINT(("%s: atap_dmatiming_mimi=%d, atap_dmatiming_recom=%d\n",
    406 	    self->dv_xname, wd->sc_params.atap_dmatiming_mimi,
    407 	    wd->sc_params.atap_dmatiming_recom), DEBUG_PROBE);
    408 	/*
    409 	 * Initialize and attach the disk structure.
    410 	 */
    411 	wd->sc_dk.dk_driver = &wddkdriver;
    412 	wd->sc_dk.dk_name = wd->sc_dev.dv_xname;
    413 	/* we fill in dk_info later */
    414 	disk_attach(&wd->sc_dk);
    415 	wd->sc_wdc_bio.lp = wd->sc_dk.dk_label;
    416 	wd->sc_sdhook = shutdownhook_establish(wd_shutdown, wd);
    417 	if (wd->sc_sdhook == NULL)
    418 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
    419 		    wd->sc_dev.dv_xname);
    420 #if NRND > 0
    421 	rnd_attach_source(&wd->rnd_source, wd->sc_dev.dv_xname,
    422 			  RND_TYPE_DISK, 0);
    423 #endif
    424 
    425 	/* Discover wedges on this disk. */
    426 	dkwedge_discover(&wd->sc_dk);
    427 }
    428 
    429 int
    430 wdactivate(struct device *self, enum devact act)
    431 {
    432 	int rv = 0;
    433 
    434 	switch (act) {
    435 	case DVACT_ACTIVATE:
    436 		rv = EOPNOTSUPP;
    437 		break;
    438 
    439 	case DVACT_DEACTIVATE:
    440 		/*
    441 		 * Nothing to do; we key off the device's DVF_ACTIVATE.
    442 		 */
    443 		break;
    444 	}
    445 	return (rv);
    446 }
    447 
    448 int
    449 wddetach(struct device *self, int flags)
    450 {
    451 	struct wd_softc *sc = (struct wd_softc *)self;
    452 	int s, bmaj, cmaj, i, mn;
    453 
    454 	/* locate the major number */
    455 	bmaj = bdevsw_lookup_major(&wd_bdevsw);
    456 	cmaj = cdevsw_lookup_major(&wd_cdevsw);
    457 
    458 	/* Nuke the vnodes for any open instances. */
    459 	for (i = 0; i < MAXPARTITIONS; i++) {
    460 		mn = WDMINOR(device_unit(self), i);
    461 		vdevgone(bmaj, mn, mn, VBLK);
    462 		vdevgone(cmaj, mn, mn, VCHR);
    463 	}
    464 
    465 	/* Delete all of our wedges. */
    466 	dkwedge_delall(&sc->sc_dk);
    467 
    468 	s = splbio();
    469 
    470 	/* Kill off any queued buffers. */
    471 	bufq_drain(sc->sc_q);
    472 
    473 	bufq_free(sc->sc_q);
    474 	sc->atabus->ata_killpending(sc->drvp);
    475 
    476 	splx(s);
    477 
    478 	/* Detach disk. */
    479 	disk_detach(&sc->sc_dk);
    480 
    481 #ifdef WD_SOFTBADSECT
    482 	/* Clean out the bad sector list */
    483 	while (!SLIST_EMPTY(&sc->sc_bslist)) {
    484 		void *head = SLIST_FIRST(&sc->sc_bslist);
    485 		SLIST_REMOVE_HEAD(&sc->sc_bslist, dbs_next);
    486 		free(head, M_TEMP);
    487 	}
    488 	sc->sc_bscount = 0;
    489 #endif
    490 
    491 	/* Get rid of the shutdown hook. */
    492 	if (sc->sc_sdhook != NULL)
    493 		shutdownhook_disestablish(sc->sc_sdhook);
    494 
    495 #if NRND > 0
    496 	/* Unhook the entropy source. */
    497 	rnd_detach_source(&sc->rnd_source);
    498 #endif
    499 
    500 	sc->drvp->drive_flags = 0; /* no drive any more here */
    501 
    502 	return (0);
    503 }
    504 
    505 /*
    506  * Read/write routine for a buffer.  Validates the arguments and schedules the
    507  * transfer.  Does not wait for the transfer to complete.
    508  */
    509 void
    510 wdstrategy(struct buf *bp)
    511 {
    512 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(bp->b_dev));
    513 	struct disklabel *lp = wd->sc_dk.dk_label;
    514 	daddr_t blkno;
    515 	int s;
    516 
    517 	ATADEBUG_PRINT(("wdstrategy (%s)\n", wd->sc_dev.dv_xname),
    518 	    DEBUG_XFERS);
    519 
    520 	/* Valid request?  */
    521 	if (bp->b_blkno < 0 ||
    522 	    (bp->b_bcount % lp->d_secsize) != 0 ||
    523 	    (bp->b_bcount / lp->d_secsize) >= (1 << NBBY)) {
    524 		bp->b_error = EINVAL;
    525 		goto done;
    526 	}
    527 
    528 	/* If device invalidated (e.g. media change, door open), error. */
    529 	if ((wd->sc_flags & WDF_LOADED) == 0) {
    530 		bp->b_error = EIO;
    531 		goto done;
    532 	}
    533 
    534 	/* If it's a null transfer, return immediately. */
    535 	if (bp->b_bcount == 0)
    536 		goto done;
    537 
    538 	/*
    539 	 * Do bounds checking, adjust transfer. if error, process.
    540 	 * If end of partition, just return.
    541 	 */
    542 	if (WDPART(bp->b_dev) == RAW_PART) {
    543 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
    544 		    wd->sc_capacity) <= 0)
    545 			goto done;
    546 	} else {
    547 		if (bounds_check_with_label(&wd->sc_dk, bp,
    548 		    (wd->sc_flags & (WDF_WLABEL|WDF_LABELLING)) != 0) <= 0)
    549 			goto done;
    550 	}
    551 
    552 	/*
    553 	 * Now convert the block number to absolute and put it in
    554 	 * terms of the device's logical block size.
    555 	 */
    556 	if (lp->d_secsize >= DEV_BSIZE)
    557 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    558 	else
    559 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    560 
    561 	if (WDPART(bp->b_dev) != RAW_PART)
    562 		blkno += lp->d_partitions[WDPART(bp->b_dev)].p_offset;
    563 
    564 	bp->b_rawblkno = blkno;
    565 
    566 #ifdef WD_SOFTBADSECT
    567 	/*
    568 	 * If the transfer about to be attempted contains only a block that
    569 	 * is known to be bad then return an error for the transfer without
    570 	 * even attempting to start a transfer up under the premis that we
    571 	 * will just end up doing more retries for a transfer that will end
    572 	 * up failing again.
    573 	 * XXX:SMP - mutex required to protect with DIOCBSFLUSH
    574 	 */
    575 	if (__predict_false(!SLIST_EMPTY(&wd->sc_bslist))) {
    576 		struct disk_badsectors *dbs;
    577 		daddr_t maxblk = blkno + (bp->b_bcount >> DEV_BSHIFT) - 1;
    578 
    579 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next)
    580 			if ((dbs->dbs_min <= blkno && blkno <= dbs->dbs_max) ||
    581 			    (dbs->dbs_min <= maxblk && maxblk <= dbs->dbs_max)){
    582 				bp->b_error = EIO;
    583 				goto done;
    584 			}
    585 	}
    586 #endif
    587 
    588 	/* Queue transfer on drive, activate drive and controller if idle. */
    589 	s = splbio();
    590 	BUFQ_PUT(wd->sc_q, bp);
    591 	wdstart(wd);
    592 	splx(s);
    593 	return;
    594 done:
    595 	/* Toss transfer; we're done early. */
    596 	bp->b_resid = bp->b_bcount;
    597 	biodone(bp);
    598 }
    599 
    600 /*
    601  * Queue a drive for I/O.
    602  */
    603 void
    604 wdstart(void *arg)
    605 {
    606 	struct wd_softc *wd = arg;
    607 	struct buf *bp = NULL;
    608 
    609 	ATADEBUG_PRINT(("wdstart %s\n", wd->sc_dev.dv_xname),
    610 	    DEBUG_XFERS);
    611 	while (wd->openings > 0) {
    612 
    613 		/* Is there a buf for us ? */
    614 		if ((bp = BUFQ_GET(wd->sc_q)) == NULL)
    615 			return;
    616 
    617 		/*
    618 		 * Make the command. First lock the device
    619 		 */
    620 		wd->openings--;
    621 
    622 		wd->retries = 0;
    623 		__wdstart(wd, bp);
    624 	}
    625 }
    626 
    627 static void
    628 wd_split_mod15_write(struct buf *bp)
    629 {
    630 	struct buf *obp = bp->b_private;
    631 	struct wd_softc *sc = wd_cd.cd_devs[DISKUNIT(obp->b_dev)];
    632 
    633 	if (__predict_false(bp->b_error != 0)) {
    634 		/*
    635 		 * Propagate the error.  If this was the first half of
    636 		 * the original transfer, make sure to account for that
    637 		 * in the residual.
    638 		 */
    639 		if (bp->b_data == obp->b_data)
    640 			bp->b_resid += bp->b_bcount;
    641 		goto done;
    642 	}
    643 
    644 	/*
    645 	 * If this was the second half of the transfer, we're all done!
    646 	 */
    647 	if (bp->b_data != obp->b_data)
    648 		goto done;
    649 
    650 	/*
    651 	 * Advance the pointer to the second half and issue that command
    652 	 * using the same opening.
    653 	 */
    654 	bp->b_flags = obp->b_flags | B_CALL;
    655 	bp->b_data = (char *)bp->b_data + bp->b_bcount;
    656 	bp->b_blkno += (bp->b_bcount / 512);
    657 	bp->b_rawblkno += (bp->b_bcount / 512);
    658 	__wdstart(sc, bp);
    659 	return;
    660 
    661  done:
    662 	obp->b_error = bp->b_error;
    663 	obp->b_resid = bp->b_resid;
    664 	putiobuf(bp);
    665 	biodone(obp);
    666 	sc->openings++;
    667 	/* wddone() will call wdstart() */
    668 }
    669 
    670 void
    671 __wdstart(struct wd_softc *wd, struct buf *bp)
    672 {
    673 
    674 	/*
    675 	 * Deal with the "split mod15 write" quirk.  We just divide the
    676 	 * transfer in two, doing the first half and then then second half
    677 	 * with the same command opening.
    678 	 *
    679 	 * Note we MUST do this here, because we can't let insertion
    680 	 * into the bufq cause the transfers to be re-merged.
    681 	 */
    682 	if (__predict_false((wd->sc_quirks & WD_QUIRK_SPLIT_MOD15_WRITE) != 0 &&
    683 			    (bp->b_flags & B_READ) == 0 &&
    684 			    bp->b_bcount > 512 &&
    685 			    ((bp->b_bcount / 512) % 15) == 1)) {
    686 		struct buf *nbp;
    687 
    688 		/* already at splbio */
    689 		nbp = getiobuf_nowait();
    690 		if (__predict_false(nbp == NULL)) {
    691 			/* No memory -- fail the iop. */
    692 			bp->b_error = ENOMEM;
    693 			bp->b_resid = bp->b_bcount;
    694 			biodone(bp);
    695 			wd->openings++;
    696 			return;
    697 		}
    698 
    699 		nbp->b_error = 0;
    700 		nbp->b_proc = bp->b_proc;
    701 		nbp->b_vp = NULLVP;
    702 		nbp->b_dev = bp->b_dev;
    703 
    704 		nbp->b_bcount = bp->b_bcount / 2;
    705 		nbp->b_bufsize = bp->b_bcount / 2;
    706 		nbp->b_data = bp->b_data;
    707 
    708 		nbp->b_blkno = bp->b_blkno;
    709 		nbp->b_rawblkno = bp->b_rawblkno;
    710 
    711 		nbp->b_flags = bp->b_flags | B_CALL;
    712 		nbp->b_iodone = wd_split_mod15_write;
    713 
    714 		/* Put ptr to orig buf in b_private and use new buf */
    715 		nbp->b_private = bp;
    716 
    717 		BIO_COPYPRIO(nbp, bp);
    718 
    719 		bp = nbp;
    720 	}
    721 
    722 	wd->sc_wdc_bio.blkno = bp->b_rawblkno;
    723 	wd->sc_wdc_bio.blkdone =0;
    724 	wd->sc_bp = bp;
    725 	/*
    726 	 * If we're retrying, retry in single-sector mode. This will give us
    727 	 * the sector number of the problem, and will eventually allow the
    728 	 * transfer to succeed.
    729 	 */
    730 	if (wd->retries >= WDIORETRIES_SINGLE)
    731 		wd->sc_wdc_bio.flags = ATA_SINGLE;
    732 	else
    733 		wd->sc_wdc_bio.flags = 0;
    734 	if (wd->sc_flags & WDF_LBA48 &&
    735 	    (wd->sc_wdc_bio.blkno > LBA48_THRESHOLD ||
    736 	    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) != 0))
    737 		wd->sc_wdc_bio.flags |= ATA_LBA48;
    738 	if (wd->sc_flags & WDF_LBA)
    739 		wd->sc_wdc_bio.flags |= ATA_LBA;
    740 	if (bp->b_flags & B_READ)
    741 		wd->sc_wdc_bio.flags |= ATA_READ;
    742 	wd->sc_wdc_bio.bcount = bp->b_bcount;
    743 	wd->sc_wdc_bio.databuf = bp->b_data;
    744 	/* Instrumentation. */
    745 	disk_busy(&wd->sc_dk);
    746 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
    747 	case ATACMD_TRY_AGAIN:
    748 		callout_reset(&wd->sc_restart_ch, hz, wdrestart, wd);
    749 		break;
    750 	case ATACMD_QUEUED:
    751 	case ATACMD_COMPLETE:
    752 		break;
    753 	default:
    754 		panic("__wdstart: bad return code from ata_bio()");
    755 	}
    756 }
    757 
    758 void
    759 wddone(void *v)
    760 {
    761 	struct wd_softc *wd = v;
    762 	struct buf *bp = wd->sc_bp;
    763 	const char *errmsg;
    764 	int do_perror = 0;
    765 	ATADEBUG_PRINT(("wddone %s\n", wd->sc_dev.dv_xname),
    766 	    DEBUG_XFERS);
    767 
    768 	if (bp == NULL)
    769 		return;
    770 	bp->b_resid = wd->sc_wdc_bio.bcount;
    771 	switch (wd->sc_wdc_bio.error) {
    772 	case ERR_DMA:
    773 		errmsg = "DMA error";
    774 		goto retry;
    775 	case ERR_DF:
    776 		errmsg = "device fault";
    777 		goto retry;
    778 	case TIMEOUT:
    779 		errmsg = "device timeout";
    780 		goto retry;
    781 	case ERR_RESET:
    782 		errmsg = "channel reset";
    783 		goto retry2;
    784 	case ERROR:
    785 		/* Don't care about media change bits */
    786 		if (wd->sc_wdc_bio.r_error != 0 &&
    787 		    (wd->sc_wdc_bio.r_error & ~(WDCE_MC | WDCE_MCR)) == 0)
    788 			goto noerror;
    789 		errmsg = "error";
    790 		do_perror = 1;
    791 retry:		/* Just reset and retry. Can we do more ? */
    792 		(*wd->atabus->ata_reset_drive)(wd->drvp, AT_RST_NOCMD);
    793 retry2:
    794 		diskerr(bp, "wd", errmsg, LOG_PRINTF,
    795 		    wd->sc_wdc_bio.blkdone, wd->sc_dk.dk_label);
    796 		if (wd->retries < WDIORETRIES)
    797 			printf(", retrying\n");
    798 		if (do_perror)
    799 			wdperror(wd);
    800 		if (wd->retries < WDIORETRIES) {
    801 			wd->retries++;
    802 			callout_reset(&wd->sc_restart_ch, RECOVERYTIME,
    803 			    wdrestart, wd);
    804 			return;
    805 		}
    806 		printf("\n");
    807 
    808 #ifdef WD_SOFTBADSECT
    809 		/*
    810 		 * Not all errors indicate a failed block but those that do,
    811 		 * put the block on the bad-block list for the device.  Only
    812 		 * do this for reads because the drive should do it for writes,
    813 		 * itself, according to Manuel.
    814 		 */
    815 		if ((bp->b_flags & B_READ) &&
    816 		    ((wd->drvp->ata_vers >= 4 && wd->sc_wdc_bio.r_error & 64) ||
    817 	     	     (wd->drvp->ata_vers < 4 && wd->sc_wdc_bio.r_error & 192))) {
    818 			struct disk_badsectors *dbs;
    819 
    820 			dbs = malloc(sizeof *dbs, M_TEMP, M_WAITOK);
    821 			dbs->dbs_min = bp->b_rawblkno;
    822 			dbs->dbs_max = dbs->dbs_min + (bp->b_bcount >> DEV_BSHIFT) - 1;
    823 			microtime(&dbs->dbs_failedat);
    824 			SLIST_INSERT_HEAD(&wd->sc_bslist, dbs, dbs_next);
    825 			wd->sc_bscount++;
    826 		}
    827 #endif
    828 		bp->b_error = EIO;
    829 		break;
    830 	case NOERROR:
    831 noerror:	if ((wd->sc_wdc_bio.flags & ATA_CORR) || wd->retries > 0)
    832 			printf("%s: soft error (corrected)\n",
    833 			    wd->sc_dev.dv_xname);
    834 		break;
    835 	case ERR_NODEV:
    836 		bp->b_error = EIO;
    837 		break;
    838 	}
    839 	disk_unbusy(&wd->sc_dk, (bp->b_bcount - bp->b_resid),
    840 	    (bp->b_flags & B_READ));
    841 #if NRND > 0
    842 	rnd_add_uint32(&wd->rnd_source, bp->b_blkno);
    843 #endif
    844 	/* XXX Yuck, but we don't want to increment openings in this case */
    845 	if (__predict_false((bp->b_flags & B_CALL) != 0 &&
    846 			    bp->b_iodone == wd_split_mod15_write))
    847 		biodone(bp);
    848 	else {
    849 		biodone(bp);
    850 		wd->openings++;
    851 	}
    852 	wdstart(wd);
    853 }
    854 
    855 void
    856 wdrestart(void *v)
    857 {
    858 	struct wd_softc *wd = v;
    859 	struct buf *bp = wd->sc_bp;
    860 	int s;
    861 	ATADEBUG_PRINT(("wdrestart %s\n", wd->sc_dev.dv_xname),
    862 	    DEBUG_XFERS);
    863 
    864 	s = splbio();
    865 	__wdstart(v, bp);
    866 	splx(s);
    867 }
    868 
    869 int
    870 wdread(dev_t dev, struct uio *uio, int flags)
    871 {
    872 
    873 	ATADEBUG_PRINT(("wdread\n"), DEBUG_XFERS);
    874 	return (physio(wdstrategy, NULL, dev, B_READ, minphys, uio));
    875 }
    876 
    877 int
    878 wdwrite(dev_t dev, struct uio *uio, int flags)
    879 {
    880 
    881 	ATADEBUG_PRINT(("wdwrite\n"), DEBUG_XFERS);
    882 	return (physio(wdstrategy, NULL, dev, B_WRITE, minphys, uio));
    883 }
    884 
    885 int
    886 wdopen(dev_t dev, int flag, int fmt, struct lwp *l)
    887 {
    888 	struct wd_softc *wd;
    889 	int part, error;
    890 
    891 	ATADEBUG_PRINT(("wdopen\n"), DEBUG_FUNCS);
    892 	wd = device_lookup(&wd_cd, WDUNIT(dev));
    893 	if (wd == NULL)
    894 		return (ENXIO);
    895 
    896 	if (! device_is_active(&wd->sc_dev))
    897 		return (ENODEV);
    898 
    899 	part = WDPART(dev);
    900 
    901 	mutex_enter(&wd->sc_dk.dk_openlock);
    902 
    903 	/*
    904 	 * If there are wedges, and this is not RAW_PART, then we
    905 	 * need to fail.
    906 	 */
    907 	if (wd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
    908 		error = EBUSY;
    909 		goto bad1;
    910 	}
    911 
    912 	/*
    913 	 * If this is the first open of this device, add a reference
    914 	 * to the adapter.
    915 	 */
    916 	if (wd->sc_dk.dk_openmask == 0 &&
    917 	    (error = wd->atabus->ata_addref(wd->drvp)) != 0)
    918 		goto bad1;
    919 
    920 	if (wd->sc_dk.dk_openmask != 0) {
    921 		/*
    922 		 * If any partition is open, but the disk has been invalidated,
    923 		 * disallow further opens.
    924 		 */
    925 		if ((wd->sc_flags & WDF_LOADED) == 0) {
    926 			error = EIO;
    927 			goto bad2;
    928 		}
    929 	} else {
    930 		if ((wd->sc_flags & WDF_LOADED) == 0) {
    931 			wd->sc_flags |= WDF_LOADED;
    932 
    933 			/* Load the physical device parameters. */
    934 			wd_get_params(wd, AT_WAIT, &wd->sc_params);
    935 
    936 			/* Load the partition info if not already loaded. */
    937 			wdgetdisklabel(wd);
    938 		}
    939 	}
    940 
    941 	/* Check that the partition exists. */
    942 	if (part != RAW_PART &&
    943 	    (part >= wd->sc_dk.dk_label->d_npartitions ||
    944 	     wd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
    945 		error = ENXIO;
    946 		goto bad2;
    947 	}
    948 
    949 	/* Insure only one open at a time. */
    950 	switch (fmt) {
    951 	case S_IFCHR:
    952 		wd->sc_dk.dk_copenmask |= (1 << part);
    953 		break;
    954 	case S_IFBLK:
    955 		wd->sc_dk.dk_bopenmask |= (1 << part);
    956 		break;
    957 	}
    958 	wd->sc_dk.dk_openmask =
    959 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
    960 
    961 	mutex_exit(&wd->sc_dk.dk_openlock);
    962 	return 0;
    963 
    964  bad2:
    965 	if (wd->sc_dk.dk_openmask == 0)
    966 		wd->atabus->ata_delref(wd->drvp);
    967  bad1:
    968 	mutex_exit(&wd->sc_dk.dk_openlock);
    969 	return error;
    970 }
    971 
    972 int
    973 wdclose(dev_t dev, int flag, int fmt, struct lwp *l)
    974 {
    975 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
    976 	int part = WDPART(dev);
    977 
    978 	ATADEBUG_PRINT(("wdclose\n"), DEBUG_FUNCS);
    979 
    980 	mutex_enter(&wd->sc_dk.dk_openlock);
    981 
    982 	switch (fmt) {
    983 	case S_IFCHR:
    984 		wd->sc_dk.dk_copenmask &= ~(1 << part);
    985 		break;
    986 	case S_IFBLK:
    987 		wd->sc_dk.dk_bopenmask &= ~(1 << part);
    988 		break;
    989 	}
    990 	wd->sc_dk.dk_openmask =
    991 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
    992 
    993 	if (wd->sc_dk.dk_openmask == 0) {
    994 		wd_flushcache(wd, AT_WAIT);
    995 
    996 		if (! (wd->sc_flags & WDF_KLABEL))
    997 			wd->sc_flags &= ~WDF_LOADED;
    998 
    999 		wd->atabus->ata_delref(wd->drvp);
   1000 	}
   1001 
   1002 	mutex_exit(&wd->sc_dk.dk_openlock);
   1003 	return 0;
   1004 }
   1005 
   1006 void
   1007 wdgetdefaultlabel(struct wd_softc *wd, struct disklabel *lp)
   1008 {
   1009 
   1010 	ATADEBUG_PRINT(("wdgetdefaultlabel\n"), DEBUG_FUNCS);
   1011 	memset(lp, 0, sizeof(struct disklabel));
   1012 
   1013 	lp->d_secsize = DEV_BSIZE;
   1014 	lp->d_ntracks = wd->sc_params.atap_heads;
   1015 	lp->d_nsectors = wd->sc_params.atap_sectors;
   1016 	lp->d_ncylinders = (wd->sc_flags & WDF_LBA) ? wd->sc_capacity /
   1017 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors) :
   1018 		wd->sc_params.atap_cylinders;
   1019 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
   1020 
   1021 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
   1022 		lp->d_type = DTYPE_ST506;
   1023 	else
   1024 		lp->d_type = DTYPE_ESDI;
   1025 
   1026 	strncpy(lp->d_typename, wd->sc_params.atap_model, 16);
   1027 	strncpy(lp->d_packname, "fictitious", 16);
   1028 	if (wd->sc_capacity > UINT32_MAX)
   1029 		lp->d_secperunit = UINT32_MAX;
   1030 	else
   1031 		lp->d_secperunit = wd->sc_capacity;
   1032 	lp->d_rpm = 3600;
   1033 	lp->d_interleave = 1;
   1034 	lp->d_flags = 0;
   1035 
   1036 	lp->d_partitions[RAW_PART].p_offset = 0;
   1037 	lp->d_partitions[RAW_PART].p_size =
   1038 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
   1039 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
   1040 	lp->d_npartitions = RAW_PART + 1;
   1041 
   1042 	lp->d_magic = DISKMAGIC;
   1043 	lp->d_magic2 = DISKMAGIC;
   1044 	lp->d_checksum = dkcksum(lp);
   1045 }
   1046 
   1047 /*
   1048  * Fabricate a default disk label, and try to read the correct one.
   1049  */
   1050 void
   1051 wdgetdisklabel(struct wd_softc *wd)
   1052 {
   1053 	struct disklabel *lp = wd->sc_dk.dk_label;
   1054 	const char *errstring;
   1055 	int s;
   1056 
   1057 	ATADEBUG_PRINT(("wdgetdisklabel\n"), DEBUG_FUNCS);
   1058 
   1059 	memset(wd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
   1060 
   1061 	wdgetdefaultlabel(wd, lp);
   1062 
   1063 	wd->sc_badsect[0] = -1;
   1064 
   1065 	if (wd->drvp->state > RESET) {
   1066 		s = splbio();
   1067 		wd->drvp->drive_flags |= DRIVE_RESET;
   1068 		splx(s);
   1069 	}
   1070 	errstring = readdisklabel(MAKEWDDEV(0, device_unit(&wd->sc_dev),
   1071 				  RAW_PART), wdstrategy, lp,
   1072 				  wd->sc_dk.dk_cpulabel);
   1073 	if (errstring) {
   1074 		/*
   1075 		 * This probably happened because the drive's default
   1076 		 * geometry doesn't match the DOS geometry.  We
   1077 		 * assume the DOS geometry is now in the label and try
   1078 		 * again.  XXX This is a kluge.
   1079 		 */
   1080 		if (wd->drvp->state > RESET) {
   1081 			s = splbio();
   1082 			wd->drvp->drive_flags |= DRIVE_RESET;
   1083 			splx(s);
   1084 		}
   1085 		errstring = readdisklabel(MAKEWDDEV(0, device_unit(&wd->sc_dev),
   1086 		    RAW_PART), wdstrategy, lp, wd->sc_dk.dk_cpulabel);
   1087 	}
   1088 	if (errstring) {
   1089 		printf("%s: %s\n", wd->sc_dev.dv_xname, errstring);
   1090 		return;
   1091 	}
   1092 
   1093 	if (wd->drvp->state > RESET) {
   1094 		s = splbio();
   1095 		wd->drvp->drive_flags |= DRIVE_RESET;
   1096 		splx(s);
   1097 	}
   1098 #ifdef HAS_BAD144_HANDLING
   1099 	if ((lp->d_flags & D_BADSECT) != 0)
   1100 		bad144intern(wd);
   1101 #endif
   1102 }
   1103 
   1104 void
   1105 wdperror(const struct wd_softc *wd)
   1106 {
   1107 	static const char *const errstr0_3[] = {"address mark not found",
   1108 	    "track 0 not found", "aborted command", "media change requested",
   1109 	    "id not found", "media changed", "uncorrectable data error",
   1110 	    "bad block detected"};
   1111 	static const char *const errstr4_5[] = {
   1112 	    "obsolete (address mark not found)",
   1113 	    "no media/write protected", "aborted command",
   1114 	    "media change requested", "id not found", "media changed",
   1115 	    "uncorrectable data error", "interface CRC error"};
   1116 	const char *const *errstr;
   1117 	int i;
   1118 	const char *sep = "";
   1119 
   1120 	const char *devname = wd->sc_dev.dv_xname;
   1121 	struct ata_drive_datas *drvp = wd->drvp;
   1122 	int errno = wd->sc_wdc_bio.r_error;
   1123 
   1124 	if (drvp->ata_vers >= 4)
   1125 		errstr = errstr4_5;
   1126 	else
   1127 		errstr = errstr0_3;
   1128 
   1129 	printf("%s: (", devname);
   1130 
   1131 	if (errno == 0)
   1132 		printf("error not notified");
   1133 
   1134 	for (i = 0; i < 8; i++) {
   1135 		if (errno & (1 << i)) {
   1136 			printf("%s%s", sep, errstr[i]);
   1137 			sep = ", ";
   1138 		}
   1139 	}
   1140 	printf(")\n");
   1141 }
   1142 
   1143 int
   1144 wdioctl(dev_t dev, u_long xfer, void *addr, int flag, struct lwp *l)
   1145 {
   1146 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
   1147 	int error = 0, s;
   1148 #ifdef __HAVE_OLD_DISKLABEL
   1149 	struct disklabel *newlabel = NULL;
   1150 #endif
   1151 
   1152 	ATADEBUG_PRINT(("wdioctl\n"), DEBUG_FUNCS);
   1153 
   1154 	if ((wd->sc_flags & WDF_LOADED) == 0)
   1155 		return EIO;
   1156 
   1157 	error = disk_ioctl(&wd->sc_dk, xfer, addr, flag, l);
   1158 	if (error != EPASSTHROUGH)
   1159 		return (error);
   1160 
   1161 	switch (xfer) {
   1162 #ifdef HAS_BAD144_HANDLING
   1163 	case DIOCSBAD:
   1164 		if ((flag & FWRITE) == 0)
   1165 			return EBADF;
   1166 		wd->sc_dk.dk_cpulabel->bad = *(struct dkbad *)addr;
   1167 		wd->sc_dk.dk_label->d_flags |= D_BADSECT;
   1168 		bad144intern(wd);
   1169 		return 0;
   1170 #endif
   1171 #ifdef WD_SOFTBADSECT
   1172 	case DIOCBSLIST :
   1173 	{
   1174 		u_int32_t count, missing, skip;
   1175 		struct disk_badsecinfo dbsi;
   1176 		struct disk_badsectors *dbs;
   1177 		size_t available;
   1178 		uint8_t *laddr;
   1179 
   1180 		dbsi = *(struct disk_badsecinfo *)addr;
   1181 		missing = wd->sc_bscount;
   1182 		count = 0;
   1183 		available = dbsi.dbsi_bufsize;
   1184 		skip = dbsi.dbsi_skip;
   1185 		laddr = (uint8_t *)dbsi.dbsi_buffer;
   1186 
   1187 		/*
   1188 		 * We start this loop with the expectation that all of the
   1189 		 * entries will be missed and decrement this counter each
   1190 		 * time we either skip over one (already copied out) or
   1191 		 * we actually copy it back to user space.  The structs
   1192 		 * holding the bad sector information are copied directly
   1193 		 * back to user space whilst the summary is returned via
   1194 		 * the struct passed in via the ioctl.
   1195 		 */
   1196 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next) {
   1197 			if (skip > 0) {
   1198 				missing--;
   1199 				skip--;
   1200 				continue;
   1201 			}
   1202 			if (available < sizeof(*dbs))
   1203 				break;
   1204 			available -= sizeof(*dbs);
   1205 			copyout(dbs, laddr, sizeof(*dbs));
   1206 			laddr += sizeof(*dbs);
   1207 			missing--;
   1208 			count++;
   1209 		}
   1210 		dbsi.dbsi_left = missing;
   1211 		dbsi.dbsi_copied = count;
   1212 		*(struct disk_badsecinfo *)addr = dbsi;
   1213 		return 0;
   1214 	}
   1215 
   1216 	case DIOCBSFLUSH :
   1217 		/* Clean out the bad sector list */
   1218 		while (!SLIST_EMPTY(&wd->sc_bslist)) {
   1219 			void *head = SLIST_FIRST(&wd->sc_bslist);
   1220 			SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
   1221 			free(head, M_TEMP);
   1222 		}
   1223 		wd->sc_bscount = 0;
   1224 		return 0;
   1225 #endif
   1226 	case DIOCGDINFO:
   1227 		*(struct disklabel *)addr = *(wd->sc_dk.dk_label);
   1228 		return 0;
   1229 #ifdef __HAVE_OLD_DISKLABEL
   1230 	case ODIOCGDINFO:
   1231 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
   1232 		if (newlabel == NULL)
   1233 			return EIO;
   1234 		*newlabel = *(wd->sc_dk.dk_label);
   1235 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
   1236 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
   1237 		else
   1238 			error = ENOTTY;
   1239 		free(newlabel, M_TEMP);
   1240 		return error;
   1241 #endif
   1242 
   1243 	case DIOCGPART:
   1244 		((struct partinfo *)addr)->disklab = wd->sc_dk.dk_label;
   1245 		((struct partinfo *)addr)->part =
   1246 		    &wd->sc_dk.dk_label->d_partitions[WDPART(dev)];
   1247 		return 0;
   1248 
   1249 	case DIOCWDINFO:
   1250 	case DIOCSDINFO:
   1251 #ifdef __HAVE_OLD_DISKLABEL
   1252 	case ODIOCWDINFO:
   1253 	case ODIOCSDINFO:
   1254 #endif
   1255 	{
   1256 		struct disklabel *lp;
   1257 
   1258 		if ((flag & FWRITE) == 0)
   1259 			return EBADF;
   1260 
   1261 #ifdef __HAVE_OLD_DISKLABEL
   1262 		if (xfer == ODIOCSDINFO || xfer == ODIOCWDINFO) {
   1263 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
   1264 			if (newlabel == NULL)
   1265 				return EIO;
   1266 			memset(newlabel, 0, sizeof newlabel);
   1267 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
   1268 			lp = newlabel;
   1269 		} else
   1270 #endif
   1271 		lp = (struct disklabel *)addr;
   1272 
   1273 		mutex_enter(&wd->sc_dk.dk_openlock);
   1274 		wd->sc_flags |= WDF_LABELLING;
   1275 
   1276 		error = setdisklabel(wd->sc_dk.dk_label,
   1277 		    lp, /*wd->sc_dk.dk_openmask : */0,
   1278 		    wd->sc_dk.dk_cpulabel);
   1279 		if (error == 0) {
   1280 			if (wd->drvp->state > RESET) {
   1281 				s = splbio();
   1282 				wd->drvp->drive_flags |= DRIVE_RESET;
   1283 				splx(s);
   1284 			}
   1285 			if (xfer == DIOCWDINFO
   1286 #ifdef __HAVE_OLD_DISKLABEL
   1287 			    || xfer == ODIOCWDINFO
   1288 #endif
   1289 			    )
   1290 				error = writedisklabel(WDLABELDEV(dev),
   1291 				    wdstrategy, wd->sc_dk.dk_label,
   1292 				    wd->sc_dk.dk_cpulabel);
   1293 		}
   1294 
   1295 		wd->sc_flags &= ~WDF_LABELLING;
   1296 		mutex_exit(&wd->sc_dk.dk_openlock);
   1297 #ifdef __HAVE_OLD_DISKLABEL
   1298 		if (newlabel != NULL)
   1299 			free(newlabel, M_TEMP);
   1300 #endif
   1301 		return error;
   1302 	}
   1303 
   1304 	case DIOCKLABEL:
   1305 		if (*(int *)addr)
   1306 			wd->sc_flags |= WDF_KLABEL;
   1307 		else
   1308 			wd->sc_flags &= ~WDF_KLABEL;
   1309 		return 0;
   1310 
   1311 	case DIOCWLABEL:
   1312 		if ((flag & FWRITE) == 0)
   1313 			return EBADF;
   1314 		if (*(int *)addr)
   1315 			wd->sc_flags |= WDF_WLABEL;
   1316 		else
   1317 			wd->sc_flags &= ~WDF_WLABEL;
   1318 		return 0;
   1319 
   1320 	case DIOCGDEFLABEL:
   1321 		wdgetdefaultlabel(wd, (struct disklabel *)addr);
   1322 		return 0;
   1323 #ifdef __HAVE_OLD_DISKLABEL
   1324 	case ODIOCGDEFLABEL:
   1325 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
   1326 		if (newlabel == NULL)
   1327 			return EIO;
   1328 		wdgetdefaultlabel(wd, newlabel);
   1329 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
   1330 			memcpy(addr, &newlabel, sizeof (struct olddisklabel));
   1331 		else
   1332 			error = ENOTTY;
   1333 		free(newlabel, M_TEMP);
   1334 		return error;
   1335 #endif
   1336 
   1337 #ifdef notyet
   1338 	case DIOCWFORMAT:
   1339 		if ((flag & FWRITE) == 0)
   1340 			return EBADF;
   1341 		{
   1342 		register struct format_op *fop;
   1343 		struct iovec aiov;
   1344 		struct uio auio;
   1345 
   1346 		fop = (struct format_op *)addr;
   1347 		aiov.iov_base = fop->df_buf;
   1348 		aiov.iov_len = fop->df_count;
   1349 		auio.uio_iov = &aiov;
   1350 		auio.uio_iovcnt = 1;
   1351 		auio.uio_resid = fop->df_count;
   1352 		auio.uio_offset =
   1353 			fop->df_startblk * wd->sc_dk.dk_label->d_secsize;
   1354 		auio.uio_vmspace = l->l_proc->p_vmspace;
   1355 		error = physio(wdformat, NULL, dev, B_WRITE, minphys,
   1356 		    &auio);
   1357 		fop->df_count -= auio.uio_resid;
   1358 		fop->df_reg[0] = wdc->sc_status;
   1359 		fop->df_reg[1] = wdc->sc_error;
   1360 		return error;
   1361 		}
   1362 #endif
   1363 	case DIOCGCACHE:
   1364 		return wd_getcache(wd, (int *)addr);
   1365 
   1366 	case DIOCSCACHE:
   1367 		return wd_setcache(wd, *(int *)addr);
   1368 
   1369 	case DIOCCACHESYNC:
   1370 		return wd_flushcache(wd, AT_WAIT);
   1371 
   1372 	case ATAIOCCOMMAND:
   1373 		/*
   1374 		 * Make sure this command is (relatively) safe first
   1375 		 */
   1376 		if ((((atareq_t *) addr)->flags & ATACMD_READ) == 0 &&
   1377 		    (flag & FWRITE) == 0)
   1378 			return (EBADF);
   1379 		{
   1380 		struct wd_ioctl *wi;
   1381 		atareq_t *atareq = (atareq_t *) addr;
   1382 		int error1;
   1383 
   1384 		wi = wi_get();
   1385 		wi->wi_softc = wd;
   1386 		wi->wi_atareq = *atareq;
   1387 
   1388 		if (atareq->datalen && atareq->flags &
   1389 		    (ATACMD_READ | ATACMD_WRITE)) {
   1390 			void *tbuf;
   1391 			if (atareq->datalen < DEV_BSIZE
   1392 			    && atareq->command == WDCC_IDENTIFY) {
   1393 				tbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
   1394 				wi->wi_iov.iov_base = tbuf;
   1395 				wi->wi_iov.iov_len = DEV_BSIZE;
   1396 				UIO_SETUP_SYSSPACE(&wi->wi_uio);
   1397 			} else {
   1398 				tbuf = NULL;
   1399 				wi->wi_iov.iov_base = atareq->databuf;
   1400 				wi->wi_iov.iov_len = atareq->datalen;
   1401 				wi->wi_uio.uio_vmspace = l->l_proc->p_vmspace;
   1402 			}
   1403 			wi->wi_uio.uio_iov = &wi->wi_iov;
   1404 			wi->wi_uio.uio_iovcnt = 1;
   1405 			wi->wi_uio.uio_resid = atareq->datalen;
   1406 			wi->wi_uio.uio_offset = 0;
   1407 			wi->wi_uio.uio_rw =
   1408 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE;
   1409 			error1 = physio(wdioctlstrategy, &wi->wi_bp, dev,
   1410 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE,
   1411 			    minphys, &wi->wi_uio);
   1412 			if (tbuf != NULL && error1 == 0) {
   1413 				error1 = copyout(tbuf, atareq->databuf,
   1414 				    atareq->datalen);
   1415 				free(tbuf, M_TEMP);
   1416 			}
   1417 		} else {
   1418 			/* No need to call physio if we don't have any
   1419 			   user data */
   1420 			wi->wi_bp.b_flags = 0;
   1421 			wi->wi_bp.b_data = 0;
   1422 			wi->wi_bp.b_bcount = 0;
   1423 			wi->wi_bp.b_dev = 0;
   1424 			wi->wi_bp.b_proc = l->l_proc;
   1425 			wdioctlstrategy(&wi->wi_bp);
   1426 			error1 = wi->wi_bp.b_error;
   1427 		}
   1428 		*atareq = wi->wi_atareq;
   1429 		wi_free(wi);
   1430 		return(error1);
   1431 		}
   1432 
   1433 	case DIOCAWEDGE:
   1434 	    {
   1435 	    	struct dkwedge_info *dkw = (void *) addr;
   1436 
   1437 		if ((flag & FWRITE) == 0)
   1438 			return (EBADF);
   1439 
   1440 		/* If the ioctl happens here, the parent is us. */
   1441 		strcpy(dkw->dkw_parent, wd->sc_dev.dv_xname);
   1442 		return (dkwedge_add(dkw));
   1443 	    }
   1444 
   1445 	case DIOCDWEDGE:
   1446 	    {
   1447 	    	struct dkwedge_info *dkw = (void *) addr;
   1448 
   1449 		if ((flag & FWRITE) == 0)
   1450 			return (EBADF);
   1451 
   1452 		/* If the ioctl happens here, the parent is us. */
   1453 		strcpy(dkw->dkw_parent, wd->sc_dev.dv_xname);
   1454 		return (dkwedge_del(dkw));
   1455 	    }
   1456 
   1457 	case DIOCLWEDGES:
   1458 	    {
   1459 	    	struct dkwedge_list *dkwl = (void *) addr;
   1460 
   1461 		return (dkwedge_list(&wd->sc_dk, dkwl, l));
   1462 	    }
   1463 
   1464 	case DIOCGSTRATEGY:
   1465 	    {
   1466 		struct disk_strategy *dks = (void *)addr;
   1467 
   1468 		s = splbio();
   1469 		strlcpy(dks->dks_name, bufq_getstrategyname(wd->sc_q),
   1470 		    sizeof(dks->dks_name));
   1471 		splx(s);
   1472 		dks->dks_paramlen = 0;
   1473 
   1474 		return 0;
   1475 	    }
   1476 
   1477 	case DIOCSSTRATEGY:
   1478 	    {
   1479 		struct disk_strategy *dks = (void *)addr;
   1480 		struct bufq_state *new;
   1481 		struct bufq_state *old;
   1482 
   1483 		if ((flag & FWRITE) == 0) {
   1484 			return EBADF;
   1485 		}
   1486 		if (dks->dks_param != NULL) {
   1487 			return EINVAL;
   1488 		}
   1489 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
   1490 		error = bufq_alloc(&new, dks->dks_name,
   1491 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
   1492 		if (error) {
   1493 			return error;
   1494 		}
   1495 		s = splbio();
   1496 		old = wd->sc_q;
   1497 		bufq_move(new, old);
   1498 		wd->sc_q = new;
   1499 		splx(s);
   1500 		bufq_free(old);
   1501 
   1502 		return 0;
   1503 	    }
   1504 
   1505 	default:
   1506 		return ENOTTY;
   1507 	}
   1508 
   1509 #ifdef DIAGNOSTIC
   1510 	panic("wdioctl: impossible");
   1511 #endif
   1512 }
   1513 
   1514 #ifdef B_FORMAT
   1515 int
   1516 wdformat(struct buf *bp)
   1517 {
   1518 
   1519 	bp->b_flags |= B_FORMAT;
   1520 	return wdstrategy(bp);
   1521 }
   1522 #endif
   1523 
   1524 int
   1525 wdsize(dev_t dev)
   1526 {
   1527 	struct wd_softc *wd;
   1528 	int part, omask;
   1529 	int size;
   1530 
   1531 	ATADEBUG_PRINT(("wdsize\n"), DEBUG_FUNCS);
   1532 
   1533 	wd = device_lookup(&wd_cd, WDUNIT(dev));
   1534 	if (wd == NULL)
   1535 		return (-1);
   1536 
   1537 	part = WDPART(dev);
   1538 	omask = wd->sc_dk.dk_openmask & (1 << part);
   1539 
   1540 	if (omask == 0 && wdopen(dev, 0, S_IFBLK, NULL) != 0)
   1541 		return (-1);
   1542 	if (wd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
   1543 		size = -1;
   1544 	else
   1545 		size = wd->sc_dk.dk_label->d_partitions[part].p_size *
   1546 		    (wd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
   1547 	if (omask == 0 && wdclose(dev, 0, S_IFBLK, NULL) != 0)
   1548 		return (-1);
   1549 	return (size);
   1550 }
   1551 
   1552 /* #define WD_DUMP_NOT_TRUSTED if you just want to watch */
   1553 static int wddoingadump = 0;
   1554 static int wddumprecalibrated = 0;
   1555 
   1556 /*
   1557  * Dump core after a system crash.
   1558  */
   1559 int
   1560 wddump(dev_t dev, daddr_t blkno, void *va, size_t size)
   1561 {
   1562 	struct wd_softc *wd;	/* disk unit to do the I/O */
   1563 	struct disklabel *lp;   /* disk's disklabel */
   1564 	int part, err;
   1565 	int nblks;	/* total number of sectors left to write */
   1566 
   1567 	/* Check if recursive dump; if so, punt. */
   1568 	if (wddoingadump)
   1569 		return EFAULT;
   1570 	wddoingadump = 1;
   1571 
   1572 	wd = device_lookup(&wd_cd, WDUNIT(dev));
   1573 	if (wd == NULL)
   1574 		return (ENXIO);
   1575 
   1576 	part = WDPART(dev);
   1577 
   1578 	/* Convert to disk sectors.  Request must be a multiple of size. */
   1579 	lp = wd->sc_dk.dk_label;
   1580 	if ((size % lp->d_secsize) != 0)
   1581 		return EFAULT;
   1582 	nblks = size / lp->d_secsize;
   1583 	blkno = blkno / (lp->d_secsize / DEV_BSIZE);
   1584 
   1585 	/* Check transfer bounds against partition size. */
   1586 	if ((blkno < 0) || ((blkno + nblks) > lp->d_partitions[part].p_size))
   1587 		return EINVAL;
   1588 
   1589 	/* Offset block number to start of partition. */
   1590 	blkno += lp->d_partitions[part].p_offset;
   1591 
   1592 	/* Recalibrate, if first dump transfer. */
   1593 	if (wddumprecalibrated == 0) {
   1594 		wddumprecalibrated = 1;
   1595 		(*wd->atabus->ata_reset_drive)(wd->drvp,
   1596 					       AT_POLL | AT_RST_EMERG);
   1597 		wd->drvp->state = RESET;
   1598 	}
   1599 
   1600 	wd->sc_bp = NULL;
   1601 	wd->sc_wdc_bio.blkno = blkno;
   1602 	wd->sc_wdc_bio.flags = ATA_POLL;
   1603 	if (wd->sc_flags & WDF_LBA48 &&
   1604 	    (blkno > LBA48_THRESHOLD ||
   1605     	    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) != 0))
   1606 		wd->sc_wdc_bio.flags |= ATA_LBA48;
   1607 	if (wd->sc_flags & WDF_LBA)
   1608 		wd->sc_wdc_bio.flags |= ATA_LBA;
   1609 	wd->sc_wdc_bio.bcount = nblks * lp->d_secsize;
   1610 	wd->sc_wdc_bio.databuf = va;
   1611 #ifndef WD_DUMP_NOT_TRUSTED
   1612 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
   1613 	case ATACMD_TRY_AGAIN:
   1614 		panic("wddump: try again");
   1615 		break;
   1616 	case ATACMD_QUEUED:
   1617 		panic("wddump: polled command has been queued");
   1618 		break;
   1619 	case ATACMD_COMPLETE:
   1620 		break;
   1621 	}
   1622 	switch(wd->sc_wdc_bio.error) {
   1623 	case TIMEOUT:
   1624 		printf("wddump: device timed out");
   1625 		err = EIO;
   1626 		break;
   1627 	case ERR_DF:
   1628 		printf("wddump: drive fault");
   1629 		err = EIO;
   1630 		break;
   1631 	case ERR_DMA:
   1632 		printf("wddump: DMA error");
   1633 		err = EIO;
   1634 		break;
   1635 	case ERROR:
   1636 		printf("wddump: ");
   1637 		wdperror(wd);
   1638 		err = EIO;
   1639 		break;
   1640 	case NOERROR:
   1641 		err = 0;
   1642 		break;
   1643 	default:
   1644 		panic("wddump: unknown error type");
   1645 	}
   1646 	if (err != 0) {
   1647 		printf("\n");
   1648 		return err;
   1649 	}
   1650 #else	/* WD_DUMP_NOT_TRUSTED */
   1651 	/* Let's just talk about this first... */
   1652 	printf("wd%d: dump addr 0x%x, cylin %d, head %d, sector %d\n",
   1653 	    unit, va, cylin, head, sector);
   1654 	delay(500 * 1000);	/* half a second */
   1655 #endif
   1656 
   1657 	wddoingadump = 0;
   1658 	return 0;
   1659 }
   1660 
   1661 #ifdef HAS_BAD144_HANDLING
   1662 /*
   1663  * Internalize the bad sector table.
   1664  */
   1665 void
   1666 bad144intern(struct wd_softc *wd)
   1667 {
   1668 	struct dkbad *bt = &wd->sc_dk.dk_cpulabel->bad;
   1669 	struct disklabel *lp = wd->sc_dk.dk_label;
   1670 	int i = 0;
   1671 
   1672 	ATADEBUG_PRINT(("bad144intern\n"), DEBUG_XFERS);
   1673 
   1674 	for (; i < NBT_BAD; i++) {
   1675 		if (bt->bt_bad[i].bt_cyl == 0xffff)
   1676 			break;
   1677 		wd->sc_badsect[i] =
   1678 		    bt->bt_bad[i].bt_cyl * lp->d_secpercyl +
   1679 		    (bt->bt_bad[i].bt_trksec >> 8) * lp->d_nsectors +
   1680 		    (bt->bt_bad[i].bt_trksec & 0xff);
   1681 	}
   1682 	for (; i < NBT_BAD+1; i++)
   1683 		wd->sc_badsect[i] = -1;
   1684 }
   1685 #endif
   1686 
   1687 static void
   1688 wd_params_to_properties(struct wd_softc *wd, struct ataparams *params)
   1689 {
   1690 	prop_dictionary_t disk_info, odisk_info, geom;
   1691 	const char *cp;
   1692 
   1693 	disk_info = prop_dictionary_create();
   1694 
   1695 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
   1696 		cp = "ST506";
   1697 	else {
   1698 		/* XXX Should have a case for ATA here, too. */
   1699 		cp = "ESDI";
   1700 	}
   1701 	prop_dictionary_set_cstring_nocopy(disk_info, "type", cp);
   1702 
   1703 	geom = prop_dictionary_create();
   1704 
   1705 	prop_dictionary_set_uint64(geom, "sectors-per-unit", wd->sc_capacity);
   1706 
   1707 	prop_dictionary_set_uint32(geom, "sector-size",
   1708 				   DEV_BSIZE /* XXX 512? */);
   1709 
   1710 	prop_dictionary_set_uint16(geom, "sectors-per-track",
   1711 				   wd->sc_params.atap_sectors);
   1712 
   1713 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
   1714 				   wd->sc_params.atap_heads);
   1715 
   1716 	if (wd->sc_flags & WDF_LBA)
   1717 		prop_dictionary_set_uint64(geom, "cylinders-per-unit",
   1718 					   wd->sc_capacity /
   1719 					       (wd->sc_params.atap_heads *
   1720 					        wd->sc_params.atap_sectors));
   1721 	else
   1722 		prop_dictionary_set_uint16(geom, "cylinders-per-unit",
   1723 					   wd->sc_params.atap_cylinders);
   1724 
   1725 	prop_dictionary_set(disk_info, "geometry", geom);
   1726 	prop_object_release(geom);
   1727 
   1728 	prop_dictionary_set(device_properties(&wd->sc_dev),
   1729 			    "disk-info", disk_info);
   1730 
   1731 	/*
   1732 	 * Don't release disk_info here; we keep a reference to it.
   1733 	 * disk_detach() will release it when we go away.
   1734 	 */
   1735 
   1736 	odisk_info = wd->sc_dk.dk_info;
   1737 	wd->sc_dk.dk_info = disk_info;
   1738 	if (odisk_info)
   1739 		prop_object_release(odisk_info);
   1740 }
   1741 
   1742 int
   1743 wd_get_params(struct wd_softc *wd, u_int8_t flags, struct ataparams *params)
   1744 {
   1745 	switch (wd->atabus->ata_get_params(wd->drvp, flags, params)) {
   1746 	case CMD_AGAIN:
   1747 		return 1;
   1748 	case CMD_ERR:
   1749 		/*
   1750 		 * We `know' there's a drive here; just assume it's old.
   1751 		 * This geometry is only used to read the MBR and print a
   1752 		 * (false) attach message.
   1753 		 */
   1754 		strncpy(params->atap_model, "ST506",
   1755 		    sizeof params->atap_model);
   1756 		params->atap_config = ATA_CFG_FIXED;
   1757 		params->atap_cylinders = 1024;
   1758 		params->atap_heads = 8;
   1759 		params->atap_sectors = 17;
   1760 		params->atap_multi = 1;
   1761 		params->atap_capabilities1 = params->atap_capabilities2 = 0;
   1762 		wd->drvp->ata_vers = -1; /* Mark it as pre-ATA */
   1763 		/* FALLTHROUGH */
   1764 	case CMD_OK:
   1765 		wd_params_to_properties(wd, params);
   1766 		return 0;
   1767 	default:
   1768 		panic("wd_get_params: bad return code from ata_get_params");
   1769 		/* NOTREACHED */
   1770 	}
   1771 }
   1772 
   1773 int
   1774 wd_getcache(struct wd_softc *wd, int *bitsp)
   1775 {
   1776 	struct ataparams params;
   1777 
   1778 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
   1779 		return EIO;
   1780 	if (params.atap_cmd_set1 == 0x0000 ||
   1781 	    params.atap_cmd_set1 == 0xffff ||
   1782 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0) {
   1783 		*bitsp = 0;
   1784 		return 0;
   1785 	}
   1786 	*bitsp = DKCACHE_WCHANGE | DKCACHE_READ;
   1787 	if (params.atap_cmd1_en & WDC_CMD1_CACHE)
   1788 		*bitsp |= DKCACHE_WRITE;
   1789 
   1790 	return 0;
   1791 }
   1792 
   1793 const char at_errbits[] = "\20\10ERROR\11TIMEOU\12DF";
   1794 
   1795 int
   1796 wd_setcache(struct wd_softc *wd, int bits)
   1797 {
   1798 	struct ataparams params;
   1799 	struct ata_command ata_c;
   1800 
   1801 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
   1802 		return EIO;
   1803 
   1804 	if (params.atap_cmd_set1 == 0x0000 ||
   1805 	    params.atap_cmd_set1 == 0xffff ||
   1806 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0)
   1807 		return EOPNOTSUPP;
   1808 
   1809 	if ((bits & DKCACHE_READ) == 0 ||
   1810 	    (bits & DKCACHE_SAVE) != 0)
   1811 		return EOPNOTSUPP;
   1812 
   1813 	memset(&ata_c, 0, sizeof(struct ata_command));
   1814 	ata_c.r_command = SET_FEATURES;
   1815 	ata_c.r_st_bmask = 0;
   1816 	ata_c.r_st_pmask = 0;
   1817 	ata_c.timeout = 30000; /* 30s timeout */
   1818 	ata_c.flags = AT_WAIT;
   1819 	if (bits & DKCACHE_WRITE)
   1820 		ata_c.r_features = WDSF_WRITE_CACHE_EN;
   1821 	else
   1822 		ata_c.r_features = WDSF_WRITE_CACHE_DS;
   1823 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
   1824 		printf("%s: wd_setcache command not complete\n",
   1825 		    wd->sc_dev.dv_xname);
   1826 		return EIO;
   1827 	}
   1828 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   1829 		char sbuf[sizeof(at_errbits) + 64];
   1830 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
   1831 		printf("%s: wd_setcache: status=%s\n", wd->sc_dev.dv_xname,
   1832 		    sbuf);
   1833 		return EIO;
   1834 	}
   1835 	return 0;
   1836 }
   1837 
   1838 int
   1839 wd_standby(struct wd_softc *wd, int flags)
   1840 {
   1841 	struct ata_command ata_c;
   1842 
   1843 	memset(&ata_c, 0, sizeof(struct ata_command));
   1844 	ata_c.r_command = WDCC_STANDBY_IMMED;
   1845 	ata_c.r_st_bmask = WDCS_DRDY;
   1846 	ata_c.r_st_pmask = WDCS_DRDY;
   1847 	ata_c.flags = flags;
   1848 	ata_c.timeout = 30000; /* 30s timeout */
   1849 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
   1850 		printf("%s: standby immediate command didn't complete\n",
   1851 		    wd->sc_dev.dv_xname);
   1852 		return EIO;
   1853 	}
   1854 	if (ata_c.flags & AT_ERROR) {
   1855 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
   1856 			return ENODEV;
   1857 	}
   1858 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   1859 		char sbuf[sizeof(at_errbits) + 64];
   1860 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
   1861 		printf("%s: wd_standby: status=%s\n", wd->sc_dev.dv_xname,
   1862 		    sbuf);
   1863 		return EIO;
   1864 	}
   1865 	return 0;
   1866 }
   1867 
   1868 int
   1869 wd_flushcache(struct wd_softc *wd, int flags)
   1870 {
   1871 	struct ata_command ata_c;
   1872 
   1873 	/*
   1874 	 * WDCC_FLUSHCACHE is here since ATA-4, but some drives report
   1875 	 * only ATA-2 and still support it.
   1876 	 */
   1877 	if (wd->drvp->ata_vers < 4 &&
   1878 	    ((wd->sc_params.atap_cmd_set2 & WDC_CMD2_FC) == 0 ||
   1879 	    wd->sc_params.atap_cmd_set2 == 0xffff))
   1880 		return ENODEV;
   1881 	memset(&ata_c, 0, sizeof(struct ata_command));
   1882 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0 &&
   1883 	    (wd->sc_params.atap_cmd2_en & ATA_CMD2_FCE) != 0)
   1884 		ata_c.r_command = WDCC_FLUSHCACHE_EXT;
   1885 	else
   1886 		ata_c.r_command = WDCC_FLUSHCACHE;
   1887 	ata_c.r_st_bmask = WDCS_DRDY;
   1888 	ata_c.r_st_pmask = WDCS_DRDY;
   1889 	ata_c.flags = flags;
   1890 	ata_c.timeout = 30000; /* 30s timeout */
   1891 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
   1892 		printf("%s: flush cache command didn't complete\n",
   1893 		    wd->sc_dev.dv_xname);
   1894 		return EIO;
   1895 	}
   1896 	if (ata_c.flags & AT_ERROR) {
   1897 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
   1898 			return ENODEV;
   1899 	}
   1900 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   1901 		char sbuf[sizeof(at_errbits) + 64];
   1902 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
   1903 		printf("%s: wd_flushcache: status=%s\n", wd->sc_dev.dv_xname,
   1904 		    sbuf);
   1905 		return EIO;
   1906 	}
   1907 	return 0;
   1908 }
   1909 
   1910 void
   1911 wd_shutdown(void *arg)
   1912 {
   1913 	struct wd_softc *wd = arg;
   1914 	wd_flushcache(wd, AT_POLL);
   1915 }
   1916 
   1917 /*
   1918  * Allocate space for a ioctl queue structure.  Mostly taken from
   1919  * scsipi_ioctl.c
   1920  */
   1921 struct wd_ioctl *
   1922 wi_get(void)
   1923 {
   1924 	struct wd_ioctl *wi;
   1925 	int s;
   1926 
   1927 	wi = malloc(sizeof(struct wd_ioctl), M_TEMP, M_WAITOK|M_ZERO);
   1928 	simple_lock_init(&wi->wi_bp.b_interlock);
   1929 	s = splbio();
   1930 	LIST_INSERT_HEAD(&wi_head, wi, wi_list);
   1931 	splx(s);
   1932 	return (wi);
   1933 }
   1934 
   1935 /*
   1936  * Free an ioctl structure and remove it from our list
   1937  */
   1938 
   1939 void
   1940 wi_free(struct wd_ioctl *wi)
   1941 {
   1942 	int s;
   1943 
   1944 	s = splbio();
   1945 	LIST_REMOVE(wi, wi_list);
   1946 	splx(s);
   1947 	free(wi, M_TEMP);
   1948 }
   1949 
   1950 /*
   1951  * Find a wd_ioctl structure based on the struct buf.
   1952  */
   1953 
   1954 struct wd_ioctl *
   1955 wi_find(struct buf *bp)
   1956 {
   1957 	struct wd_ioctl *wi;
   1958 	int s;
   1959 
   1960 	s = splbio();
   1961 	for (wi = wi_head.lh_first; wi != 0; wi = wi->wi_list.le_next)
   1962 		if (bp == &wi->wi_bp)
   1963 			break;
   1964 	splx(s);
   1965 	return (wi);
   1966 }
   1967 
   1968 /*
   1969  * Ioctl pseudo strategy routine
   1970  *
   1971  * This is mostly stolen from scsipi_ioctl.c:scsistrategy().  What
   1972  * happens here is:
   1973  *
   1974  * - wdioctl() queues a wd_ioctl structure.
   1975  *
   1976  * - wdioctl() calls physio/wdioctlstrategy based on whether or not
   1977  *   user space I/O is required.  If physio() is called, physio() eventually
   1978  *   calls wdioctlstrategy().
   1979  *
   1980  * - In either case, wdioctlstrategy() calls wd->atabus->ata_exec_command()
   1981  *   to perform the actual command
   1982  *
   1983  * The reason for the use of the pseudo strategy routine is because
   1984  * when doing I/O to/from user space, physio _really_ wants to be in
   1985  * the loop.  We could put the entire buffer into the ioctl request
   1986  * structure, but that won't scale if we want to do things like download
   1987  * microcode.
   1988  */
   1989 
   1990 void
   1991 wdioctlstrategy(struct buf *bp)
   1992 {
   1993 	struct wd_ioctl *wi;
   1994 	struct ata_command ata_c;
   1995 	int error = 0;
   1996 
   1997 	wi = wi_find(bp);
   1998 	if (wi == NULL) {
   1999 		printf("wdioctlstrategy: "
   2000 		    "No matching ioctl request found in queue\n");
   2001 		error = EINVAL;
   2002 		goto done;
   2003 	}
   2004 
   2005 	memset(&ata_c, 0, sizeof(ata_c));
   2006 
   2007 	/*
   2008 	 * Abort if physio broke up the transfer
   2009 	 */
   2010 
   2011 	if (bp->b_bcount != wi->wi_atareq.datalen) {
   2012 		printf("physio split wd ioctl request... cannot proceed\n");
   2013 		error = EIO;
   2014 		goto done;
   2015 	}
   2016 
   2017 	/*
   2018 	 * Abort if we didn't get a buffer size that was a multiple of
   2019 	 * our sector size (or was larger than NBBY)
   2020 	 */
   2021 
   2022 	if ((bp->b_bcount % wi->wi_softc->sc_dk.dk_label->d_secsize) != 0 ||
   2023 	    (bp->b_bcount / wi->wi_softc->sc_dk.dk_label->d_secsize) >=
   2024 	     (1 << NBBY)) {
   2025 		error = EINVAL;
   2026 		goto done;
   2027 	}
   2028 
   2029 	/*
   2030 	 * Make sure a timeout was supplied in the ioctl request
   2031 	 */
   2032 
   2033 	if (wi->wi_atareq.timeout == 0) {
   2034 		error = EINVAL;
   2035 		goto done;
   2036 	}
   2037 
   2038 	if (wi->wi_atareq.flags & ATACMD_READ)
   2039 		ata_c.flags |= AT_READ;
   2040 	else if (wi->wi_atareq.flags & ATACMD_WRITE)
   2041 		ata_c.flags |= AT_WRITE;
   2042 
   2043 	if (wi->wi_atareq.flags & ATACMD_READREG)
   2044 		ata_c.flags |= AT_READREG;
   2045 
   2046 	ata_c.flags |= AT_WAIT;
   2047 
   2048 	ata_c.timeout = wi->wi_atareq.timeout;
   2049 	ata_c.r_command = wi->wi_atareq.command;
   2050 	ata_c.r_head = wi->wi_atareq.head & 0x0f;
   2051 	ata_c.r_cyl = wi->wi_atareq.cylinder;
   2052 	ata_c.r_sector = wi->wi_atareq.sec_num;
   2053 	ata_c.r_count = wi->wi_atareq.sec_count;
   2054 	ata_c.r_features = wi->wi_atareq.features;
   2055 	ata_c.r_st_bmask = WDCS_DRDY;
   2056 	ata_c.r_st_pmask = WDCS_DRDY;
   2057 	ata_c.data = wi->wi_bp.b_data;
   2058 	ata_c.bcount = wi->wi_bp.b_bcount;
   2059 
   2060 	if (wi->wi_softc->atabus->ata_exec_command(wi->wi_softc->drvp, &ata_c)
   2061 	    != ATACMD_COMPLETE) {
   2062 		wi->wi_atareq.retsts = ATACMD_ERROR;
   2063 		error = EIO;
   2064 		goto done;
   2065 	}
   2066 
   2067 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   2068 		if (ata_c.flags & AT_ERROR) {
   2069 			wi->wi_atareq.retsts = ATACMD_ERROR;
   2070 			wi->wi_atareq.error = ata_c.r_error;
   2071 		} else if (ata_c.flags & AT_DF)
   2072 			wi->wi_atareq.retsts = ATACMD_DF;
   2073 		else
   2074 			wi->wi_atareq.retsts = ATACMD_TIMEOUT;
   2075 	} else {
   2076 		wi->wi_atareq.retsts = ATACMD_OK;
   2077 		if (wi->wi_atareq.flags & ATACMD_READREG) {
   2078 			wi->wi_atareq.head = ata_c.r_head ;
   2079 			wi->wi_atareq.cylinder = ata_c.r_cyl;
   2080 			wi->wi_atareq.sec_num = ata_c.r_sector;
   2081 			wi->wi_atareq.sec_count = ata_c.r_count;
   2082 			wi->wi_atareq.features = ata_c.r_features;
   2083 			wi->wi_atareq.error = ata_c.r_error;
   2084 		}
   2085 	}
   2086 
   2087 done:
   2088 	bp->b_error = error;
   2089 	biodone(bp);
   2090 }
   2091