Home | History | Annotate | Line # | Download | only in ata
wd.c revision 1.433
      1 /*	$NetBSD: wd.c,v 1.433 2017/10/19 20:45:07 jdolecek 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  *
     15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     25  */
     26 
     27 /*-
     28  * Copyright (c) 1998, 2003, 2004 The NetBSD Foundation, Inc.
     29  * All rights reserved.
     30  *
     31  * This code is derived from software contributed to The NetBSD Foundation
     32  * by Charles M. Hannum and by Onno van der Linden.
     33  *
     34  * Redistribution and use in source and binary forms, with or without
     35  * modification, are permitted provided that the following conditions
     36  * are met:
     37  * 1. Redistributions of source code must retain the above copyright
     38  *    notice, this list of conditions and the following disclaimer.
     39  * 2. Redistributions in binary form must reproduce the above copyright
     40  *    notice, this list of conditions and the following disclaimer in the
     41  *    documentation and/or other materials provided with the distribution.
     42  *
     43  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     44  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     45  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     46  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     47  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     48  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     49  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     50  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     51  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     52  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     53  * POSSIBILITY OF SUCH DAMAGE.
     54  */
     55 
     56 #include <sys/cdefs.h>
     57 __KERNEL_RCSID(0, "$NetBSD: wd.c,v 1.433 2017/10/19 20:45:07 jdolecek Exp $");
     58 
     59 #include "opt_ata.h"
     60 #include "opt_wd.h"
     61 
     62 #include <sys/param.h>
     63 #include <sys/systm.h>
     64 #include <sys/kernel.h>
     65 #include <sys/conf.h>
     66 #include <sys/file.h>
     67 #include <sys/stat.h>
     68 #include <sys/ioctl.h>
     69 #include <sys/buf.h>
     70 #include <sys/bufq.h>
     71 #include <sys/uio.h>
     72 #include <sys/malloc.h>
     73 #include <sys/device.h>
     74 #include <sys/disklabel.h>
     75 #include <sys/disk.h>
     76 #include <sys/syslog.h>
     77 #include <sys/proc.h>
     78 #include <sys/reboot.h>
     79 #include <sys/vnode.h>
     80 #include <sys/rndsource.h>
     81 
     82 #include <sys/intr.h>
     83 #include <sys/bus.h>
     84 
     85 #include <dev/ata/atareg.h>
     86 #include <dev/ata/atavar.h>
     87 #include <dev/ata/wdvar.h>
     88 #include <dev/ic/wdcreg.h>
     89 #include <sys/ataio.h>
     90 #include "locators.h"
     91 
     92 #include <prop/proplib.h>
     93 
     94 #define	WDIORETRIES_SINGLE 4	/* number of retries for single-sector */
     95 #define	WDIORETRIES	5	/* number of retries before giving up */
     96 #define	RECOVERYTIME hz/2	/* time to wait before retrying a cmd */
     97 
     98 #define	WDUNIT(dev)		DISKUNIT(dev)
     99 #define	WDPART(dev)		DISKPART(dev)
    100 #define	WDMINOR(unit, part)	DISKMINOR(unit, part)
    101 #define	MAKEWDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
    102 
    103 #define	WDLABELDEV(dev)	(MAKEWDDEV(major(dev), WDUNIT(dev), RAW_PART))
    104 
    105 #define DEBUG_FUNCS  0x08
    106 #define DEBUG_PROBE  0x10
    107 #define DEBUG_DETACH 0x20
    108 #define	DEBUG_XFERS  0x40
    109 #ifdef ATADEBUG
    110 int wdcdebug_wd_mask = 0x0;
    111 #define ATADEBUG_PRINT(args, level) \
    112 	if (wdcdebug_wd_mask & (level)) \
    113 		printf args
    114 #else
    115 #define ATADEBUG_PRINT(args, level)
    116 #endif
    117 
    118 int	wdprobe(device_t, cfdata_t, void *);
    119 void	wdattach(device_t, device_t, void *);
    120 int	wddetach(device_t, int);
    121 int	wdprint(void *, char *);
    122 void	wdperror(const struct wd_softc *, struct ata_xfer *);
    123 
    124 static void	wdminphys(struct buf *);
    125 
    126 static int	wdlastclose(device_t);
    127 static bool	wd_suspend(device_t, const pmf_qual_t *);
    128 static int	wd_standby(struct wd_softc *, int);
    129 
    130 CFATTACH_DECL3_NEW(wd, sizeof(struct wd_softc),
    131     wdprobe, wdattach, wddetach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
    132 
    133 extern struct cfdriver wd_cd;
    134 
    135 dev_type_open(wdopen);
    136 dev_type_close(wdclose);
    137 dev_type_read(wdread);
    138 dev_type_write(wdwrite);
    139 dev_type_ioctl(wdioctl);
    140 dev_type_strategy(wdstrategy);
    141 dev_type_dump(wddump);
    142 dev_type_size(wdsize);
    143 static dev_type_discard(wddiscard);
    144 
    145 const struct bdevsw wd_bdevsw = {
    146 	.d_open = wdopen,
    147 	.d_close = wdclose,
    148 	.d_strategy = wdstrategy,
    149 	.d_ioctl = wdioctl,
    150 	.d_dump = wddump,
    151 	.d_psize = wdsize,
    152 	.d_discard = wddiscard,
    153 	.d_flag = D_DISK
    154 };
    155 
    156 const struct cdevsw wd_cdevsw = {
    157 	.d_open = wdopen,
    158 	.d_close = wdclose,
    159 	.d_read = wdread,
    160 	.d_write = wdwrite,
    161 	.d_ioctl = wdioctl,
    162 	.d_stop = nostop,
    163 	.d_tty = notty,
    164 	.d_poll = nopoll,
    165 	.d_mmap = nommap,
    166 	.d_kqfilter = nokqfilter,
    167 	.d_discard = wddiscard,
    168 	.d_flag = D_DISK
    169 };
    170 
    171 /* #define WD_DUMP_NOT_TRUSTED if you just want to watch */
    172 static int wddoingadump = 0;
    173 static int wddumprecalibrated = 0;
    174 
    175 /*
    176  * Glue necessary to hook WDCIOCCOMMAND into physio
    177  */
    178 
    179 struct wd_ioctl {
    180 	LIST_ENTRY(wd_ioctl) wi_list;
    181 	struct buf wi_bp;
    182 	struct uio wi_uio;
    183 	struct iovec wi_iov;
    184 	atareq_t wi_atareq;
    185 	struct wd_softc *wi_softc;
    186 };
    187 
    188 struct	wd_ioctl *wi_find(struct buf *);
    189 void	wi_free(struct wd_ioctl *);
    190 struct	wd_ioctl *wi_get(struct wd_softc *);
    191 void	wdioctlstrategy(struct buf *);
    192 
    193 void  wdgetdefaultlabel(struct wd_softc *, struct disklabel *);
    194 void  wdgetdisklabel(struct wd_softc *);
    195 void  wdstart(device_t);
    196 void  wdstart1(struct wd_softc *, struct buf *, struct ata_xfer *);
    197 static void  wdbiorestart(void *);
    198 void  wddone(device_t, struct ata_xfer *);
    199 static void wd_params_to_properties(struct wd_softc *);
    200 int   wd_get_params(struct wd_softc *, uint8_t, struct ataparams *);
    201 int   wd_flushcache(struct wd_softc *, int, bool);
    202 int   wd_trim(struct wd_softc *, int, daddr_t, long);
    203 bool  wd_shutdown(device_t, int);
    204 
    205 int   wd_getcache(struct wd_softc *, int *);
    206 int   wd_setcache(struct wd_softc *, int);
    207 
    208 static void wd_sysctl_attach(struct wd_softc *);
    209 static void wd_sysctl_detach(struct wd_softc *);
    210 
    211 struct dkdriver wddkdriver = {
    212 	.d_strategy = wdstrategy,
    213 	.d_minphys = wdminphys
    214 };
    215 
    216 #ifdef HAS_BAD144_HANDLING
    217 static void bad144intern(struct wd_softc *);
    218 #endif
    219 
    220 #define	WD_QUIRK_SPLIT_MOD15_WRITE	0x0001	/* must split certain writes */
    221 
    222 #define	WD_QUIRK_FMT "\20\1SPLIT_MOD15_WRITE\2FORCE_LBA48"
    223 
    224 /*
    225  * Quirk table for IDE drives.  Put more-specific matches first, since
    226  * a simple globing routine is used for matching.
    227  */
    228 static const struct wd_quirk {
    229 	const char *wdq_match;		/* inquiry pattern to match */
    230 	int wdq_quirks;			/* drive quirks */
    231 } wd_quirk_table[] = {
    232 	/*
    233 	 * Some Seagate S-ATA drives have a PHY which can get confused
    234 	 * with the way data is packetized by some S-ATA controllers.
    235 	 *
    236 	 * The work-around is to split in two any write transfer whose
    237 	 * sector count % 15 == 1 (assuming 512 byte sectors).
    238 	 *
    239 	 * XXX This is an incomplete list.  There are at least a couple
    240 	 * XXX more model numbers.  If you have trouble with such transfers
    241 	 * XXX (8K is the most common) on Seagate S-ATA drives, please
    242 	 * XXX notify thorpej (at) NetBSD.org.
    243 	 *
    244 	 * The ST360015AS has not yet been confirmed to have this
    245 	 * issue, however, it is the only other drive in the
    246 	 * Seagate Barracuda Serial ATA V family.
    247 	 *
    248 	 */
    249 	{ "ST3120023AS",
    250 	  WD_QUIRK_SPLIT_MOD15_WRITE },
    251 	{ "ST380023AS",
    252 	  WD_QUIRK_SPLIT_MOD15_WRITE },
    253 	{ "ST360015AS",
    254 	  WD_QUIRK_SPLIT_MOD15_WRITE },
    255 	{ NULL,
    256 	  0 }
    257 };
    258 
    259 static const struct wd_quirk *
    260 wd_lookup_quirks(const char *name)
    261 {
    262 	const struct wd_quirk *wdq;
    263 	const char *estr;
    264 
    265 	for (wdq = wd_quirk_table; wdq->wdq_match != NULL; wdq++) {
    266 		/*
    267 		 * We only want exact matches (which include matches
    268 		 * against globbing characters).
    269 		 */
    270 		if (pmatch(name, wdq->wdq_match, &estr) == 2)
    271 			return (wdq);
    272 	}
    273 	return (NULL);
    274 }
    275 
    276 int
    277 wdprobe(device_t parent, cfdata_t match, void *aux)
    278 {
    279 	struct ata_device *adev = aux;
    280 
    281 	if (adev == NULL)
    282 		return 0;
    283 	if (adev->adev_bustype->bustype_type != SCSIPI_BUSTYPE_ATA)
    284 		return 0;
    285 
    286 	if (match->cf_loc[ATA_HLCF_DRIVE] != ATA_HLCF_DRIVE_DEFAULT &&
    287 	    match->cf_loc[ATA_HLCF_DRIVE] != adev->adev_drv_data->drive)
    288 		return 0;
    289 	return 1;
    290 }
    291 
    292 void
    293 wdattach(device_t parent, device_t self, void *aux)
    294 {
    295 	struct wd_softc *wd = device_private(self);
    296 	struct ata_device *adev= aux;
    297 	int i, blank;
    298 	char tbuf[41], pbuf[9], c, *p, *q;
    299 	const struct wd_quirk *wdq;
    300 
    301 	wd->sc_dev = self;
    302 
    303 	ATADEBUG_PRINT(("wdattach\n"), DEBUG_FUNCS | DEBUG_PROBE);
    304 	mutex_init(&wd->sc_lock, MUTEX_DEFAULT, IPL_BIO);
    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->drvp = adev->adev_drv_data;
    311 
    312 	wd->drvp->drv_openings = 1;
    313 	wd->drvp->drv_start = wdstart;
    314 	wd->drvp->drv_done = wddone;
    315 	wd->drvp->drv_softc = wd->sc_dev; /* done in atabusconfig_thread()
    316 					     but too late */
    317 
    318 	aprint_naive("\n");
    319 	aprint_normal("\n");
    320 
    321 	/* read our drive info */
    322 	if (wd_get_params(wd, AT_WAIT, &wd->sc_params) != 0) {
    323 		aprint_error_dev(self, "IDENTIFY failed\n");
    324 		goto out;
    325 	}
    326 
    327 	for (blank = 0, p = wd->sc_params.atap_model, q = tbuf, i = 0;
    328 	    i < sizeof(wd->sc_params.atap_model); i++) {
    329 		c = *p++;
    330 		if (c == '\0')
    331 			break;
    332 		if (c != ' ') {
    333 			if (blank) {
    334 				*q++ = ' ';
    335 				blank = 0;
    336 			}
    337 			*q++ = c;
    338 		} else
    339 			blank = 1;
    340 	}
    341 	*q++ = '\0';
    342 
    343 	aprint_normal_dev(self, "<%s>\n", tbuf);
    344 
    345 	wdq = wd_lookup_quirks(tbuf);
    346 	if (wdq != NULL)
    347 		wd->sc_quirks = wdq->wdq_quirks;
    348 
    349 	if (wd->sc_quirks != 0) {
    350 		char sbuf[sizeof(WD_QUIRK_FMT) + 64];
    351 		snprintb(sbuf, sizeof(sbuf), WD_QUIRK_FMT, wd->sc_quirks);
    352 		aprint_normal_dev(self, "quirks %s\n", sbuf);
    353 
    354 		if (wd->sc_quirks & WD_QUIRK_SPLIT_MOD15_WRITE) {
    355 			aprint_error_dev(self, "drive corrupts write transfers with certain controllers, consider replacing\n");
    356 		}
    357 	}
    358 
    359 	if ((wd->sc_params.atap_multi & 0xff) > 1) {
    360 		wd->drvp->multi = wd->sc_params.atap_multi & 0xff;
    361 	} else {
    362 		wd->drvp->multi = 1;
    363 	}
    364 
    365 	aprint_verbose_dev(self, "drive supports %d-sector PIO transfers,",
    366 	    wd->drvp->multi);
    367 
    368 	/* 48-bit LBA addressing */
    369 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0)
    370 		wd->sc_flags |= WDF_LBA48;
    371 
    372 	/* Prior to ATA-4, LBA was optional. */
    373 	if ((wd->sc_params.atap_capabilities1 & WDC_CAP_LBA) != 0)
    374 		wd->sc_flags |= WDF_LBA;
    375 #if 0
    376 	/* ATA-4 requires LBA. */
    377 	if (wd->sc_params.atap_ataversion != 0xffff &&
    378 	    wd->sc_params.atap_ataversion >= WDC_VER_ATA4)
    379 		wd->sc_flags |= WDF_LBA;
    380 #endif
    381 
    382 	if ((wd->sc_flags & WDF_LBA48) != 0) {
    383 		aprint_verbose(" LBA48 addressing\n");
    384 		wd->sc_capacity =
    385 		    ((uint64_t) wd->sc_params.atap_max_lba[3] << 48) |
    386 		    ((uint64_t) wd->sc_params.atap_max_lba[2] << 32) |
    387 		    ((uint64_t) wd->sc_params.atap_max_lba[1] << 16) |
    388 		    ((uint64_t) wd->sc_params.atap_max_lba[0] <<  0);
    389 		wd->sc_capacity28 =
    390 		    (wd->sc_params.atap_capacity[1] << 16) |
    391 		    wd->sc_params.atap_capacity[0];
    392 	} else if ((wd->sc_flags & WDF_LBA) != 0) {
    393 		aprint_verbose(" LBA addressing\n");
    394 		wd->sc_capacity28 = wd->sc_capacity =
    395 		    (wd->sc_params.atap_capacity[1] << 16) |
    396 		    wd->sc_params.atap_capacity[0];
    397 	} else {
    398 		aprint_verbose(" chs addressing\n");
    399 		wd->sc_capacity28 = wd->sc_capacity =
    400 		    wd->sc_params.atap_cylinders *
    401 		    wd->sc_params.atap_heads *
    402 		    wd->sc_params.atap_sectors;
    403 	}
    404 	if ((wd->sc_params.atap_secsz & ATA_SECSZ_VALID_MASK) == ATA_SECSZ_VALID
    405 	    && ((wd->sc_params.atap_secsz & ATA_SECSZ_LLS) != 0)) {
    406 		wd->sc_blksize = 2ULL *
    407 		    ((uint32_t)((wd->sc_params.atap_lls_secsz[1] << 16) |
    408 		    wd->sc_params.atap_lls_secsz[0]));
    409 	} else {
    410 		wd->sc_blksize = 512;
    411 	}
    412 	wd->sc_capacity512 = (wd->sc_capacity * wd->sc_blksize) / DEV_BSIZE;
    413 	format_bytes(pbuf, sizeof(pbuf), wd->sc_capacity * wd->sc_blksize);
    414 	aprint_normal_dev(self, "%s, %d cyl, %d head, %d sec, "
    415 	    "%d bytes/sect x %llu sectors\n",
    416 	    pbuf,
    417 	    (wd->sc_flags & WDF_LBA) ? (int)(wd->sc_capacity /
    418 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors)) :
    419 		wd->sc_params.atap_cylinders,
    420 	    wd->sc_params.atap_heads, wd->sc_params.atap_sectors,
    421 	    wd->sc_blksize, (unsigned long long)wd->sc_capacity);
    422 
    423 	ATADEBUG_PRINT(("%s: atap_dmatiming_mimi=%d, atap_dmatiming_recom=%d\n",
    424 	    device_xname(self), wd->sc_params.atap_dmatiming_mimi,
    425 	    wd->sc_params.atap_dmatiming_recom), DEBUG_PROBE);
    426 
    427 	if (wd->sc_blksize <= 0 || !powerof2(wd->sc_blksize) ||
    428 	    wd->sc_blksize < DEV_BSIZE || wd->sc_blksize > MAXPHYS) {
    429 		aprint_normal_dev(self, "WARNING: block size %u "
    430 		    "might not actually work\n", wd->sc_blksize);
    431 	}
    432 
    433 out:
    434 	/*
    435 	 * Initialize and attach the disk structure.
    436 	 */
    437 	/* we fill in dk_info later */
    438 	disk_init(&wd->sc_dk, device_xname(wd->sc_dev), &wddkdriver);
    439 	disk_attach(&wd->sc_dk);
    440 	wd->drvp->lp = wd->sc_dk.dk_label;
    441 	wd_params_to_properties(wd);
    442 	rnd_attach_source(&wd->rnd_source, device_xname(wd->sc_dev),
    443 			  RND_TYPE_DISK, RND_FLAG_DEFAULT);
    444 
    445 	/* Discover wedges on this disk. */
    446 	dkwedge_discover(&wd->sc_dk);
    447 
    448 	if (!pmf_device_register1(self, wd_suspend, NULL, wd_shutdown))
    449 		aprint_error_dev(self, "couldn't establish power handler\n");
    450 
    451 	wd_sysctl_attach(wd);
    452 }
    453 
    454 static bool
    455 wd_suspend(device_t dv, const pmf_qual_t *qual)
    456 {
    457 	struct wd_softc *sc = device_private(dv);
    458 
    459 	/* the adapter needs to be enabled */
    460 	if (sc->atabus->ata_addref(sc->drvp))
    461 		return true; /* no need to complain */
    462 
    463 	wd_flushcache(sc, AT_WAIT, false);
    464 	wd_standby(sc, AT_WAIT);
    465 
    466 	sc->atabus->ata_delref(sc->drvp);
    467 	return true;
    468 }
    469 
    470 int
    471 wddetach(device_t self, int flags)
    472 {
    473 	struct wd_softc *sc = device_private(self);
    474 	int bmaj, cmaj, i, mn, rc;
    475 
    476 	if ((rc = disk_begindetach(&sc->sc_dk, wdlastclose, self, flags)) != 0)
    477 		return rc;
    478 
    479 	/* locate the major number */
    480 	bmaj = bdevsw_lookup_major(&wd_bdevsw);
    481 	cmaj = cdevsw_lookup_major(&wd_cdevsw);
    482 
    483 	/* Nuke the vnodes for any open instances. */
    484 	for (i = 0; i < MAXPARTITIONS; i++) {
    485 		mn = WDMINOR(device_unit(self), i);
    486 		vdevgone(bmaj, mn, mn, VBLK);
    487 		vdevgone(cmaj, mn, mn, VCHR);
    488 	}
    489 
    490 	/* Delete all of our wedges. */
    491 	dkwedge_delall(&sc->sc_dk);
    492 
    493 	mutex_enter(&sc->sc_lock);
    494 
    495 	/* Kill off any queued buffers. */
    496 	bufq_drain(sc->sc_q);
    497 
    498 	sc->atabus->ata_killpending(sc->drvp);
    499 	mutex_exit(&sc->sc_lock);
    500 
    501 	if (flags & DETACH_POWEROFF)
    502 		wd_standby(sc, AT_POLL);
    503 
    504 	bufq_free(sc->sc_q);
    505 
    506 	/* Detach disk. */
    507 	disk_detach(&sc->sc_dk);
    508 	disk_destroy(&sc->sc_dk);
    509 
    510 #ifdef WD_SOFTBADSECT
    511 	/* Clean out the bad sector list */
    512 	while (!SLIST_EMPTY(&sc->sc_bslist)) {
    513 		void *head = SLIST_FIRST(&sc->sc_bslist);
    514 		SLIST_REMOVE_HEAD(&sc->sc_bslist, dbs_next);
    515 		free(head, M_TEMP);
    516 	}
    517 	sc->sc_bscount = 0;
    518 #endif
    519 
    520 	pmf_device_deregister(self);
    521 
    522 	wd_sysctl_detach(sc);
    523 
    524 	/* Unhook the entropy source. */
    525 	rnd_detach_source(&sc->rnd_source);
    526 
    527 	mutex_destroy(&sc->sc_lock);
    528 
    529 	sc->drvp->drive_type = ATA_DRIVET_NONE; /* no drive any more here */
    530 	sc->drvp->drive_flags = 0;
    531 
    532 	return (0);
    533 }
    534 
    535 /*
    536  * Read/write routine for a buffer.  Validates the arguments and schedules the
    537  * transfer.  Does not wait for the transfer to complete.
    538  */
    539 void
    540 wdstrategy(struct buf *bp)
    541 {
    542 	struct wd_softc *wd =
    543 	    device_lookup_private(&wd_cd, WDUNIT(bp->b_dev));
    544 	struct disklabel *lp = wd->sc_dk.dk_label;
    545 	daddr_t blkno;
    546 
    547 	ATADEBUG_PRINT(("wdstrategy (%s)\n", device_xname(wd->sc_dev)),
    548 	    DEBUG_XFERS);
    549 
    550 	/* Valid request?  */
    551 	if (bp->b_blkno < 0 ||
    552 	    (bp->b_bcount % lp->d_secsize) != 0 ||
    553 	    (bp->b_bcount / lp->d_secsize) >= (1 << NBBY)) {
    554 		bp->b_error = EINVAL;
    555 		goto done;
    556 	}
    557 
    558 	/* If device invalidated (e.g. media change, door open,
    559 	 * device detachment), then error.
    560 	 */
    561 	if ((wd->sc_flags & WDF_LOADED) == 0 ||
    562 	    !device_is_enabled(wd->sc_dev)) {
    563 		bp->b_error = EIO;
    564 		goto done;
    565 	}
    566 
    567 	/* If it's a null transfer, return immediately. */
    568 	if (bp->b_bcount == 0)
    569 		goto done;
    570 
    571 	/*
    572 	 * Do bounds checking, adjust transfer. if error, process.
    573 	 * If end of partition, just return.
    574 	 */
    575 	if (WDPART(bp->b_dev) == RAW_PART) {
    576 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
    577 		    wd->sc_capacity512) <= 0)
    578 			goto done;
    579 	} else {
    580 		if (bounds_check_with_label(&wd->sc_dk, bp,
    581 		    (wd->sc_flags & (WDF_WLABEL|WDF_LABELLING)) != 0) <= 0)
    582 			goto done;
    583 	}
    584 
    585 	/*
    586 	 * Now convert the block number to absolute and put it in
    587 	 * terms of the device's logical block size.
    588 	 */
    589 	if (lp->d_secsize >= DEV_BSIZE)
    590 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    591 	else
    592 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
    593 
    594 	if (WDPART(bp->b_dev) != RAW_PART)
    595 		blkno += lp->d_partitions[WDPART(bp->b_dev)].p_offset;
    596 
    597 	bp->b_rawblkno = blkno;
    598 
    599 #ifdef WD_SOFTBADSECT
    600 	/*
    601 	 * If the transfer about to be attempted contains only a block that
    602 	 * is known to be bad then return an error for the transfer without
    603 	 * even attempting to start a transfer up under the premis that we
    604 	 * will just end up doing more retries for a transfer that will end
    605 	 * up failing again.
    606 	 */
    607 	if (__predict_false(!SLIST_EMPTY(&wd->sc_bslist))) {
    608 		struct disk_badsectors *dbs;
    609 		daddr_t maxblk = blkno + (bp->b_bcount / wd->sc_blksize) - 1;
    610 
    611 		mutex_enter(&wd->sc_lock);
    612 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next)
    613 			if ((dbs->dbs_min <= blkno && blkno <= dbs->dbs_max) ||
    614 			    (dbs->dbs_min <= maxblk && maxblk <= dbs->dbs_max)){
    615 				bp->b_error = EIO;
    616 				mutex_exit(&wd->sc_lock);
    617 				goto done;
    618 			}
    619 		mutex_exit(&wd->sc_lock);
    620 	}
    621 #endif
    622 
    623 	/* Queue transfer on drive, activate drive and controller if idle. */
    624 	mutex_enter(&wd->sc_lock);
    625 	disk_wait(&wd->sc_dk);
    626 	bufq_put(wd->sc_q, bp);
    627 	mutex_exit(&wd->sc_lock);
    628 
    629 	/* Try to queue on the current drive only */
    630 	wdstart(wd->sc_dev);
    631 	return;
    632 done:
    633 	/* Toss transfer; we're done early. */
    634 	bp->b_resid = bp->b_bcount;
    635 	biodone(bp);
    636 }
    637 
    638 /*
    639  * Queue a drive for I/O.
    640  */
    641 void
    642 wdstart(device_t self)
    643 {
    644 	struct wd_softc *wd = device_private(self);
    645 	struct buf *bp;
    646 	struct ata_xfer *xfer;
    647 
    648 	ATADEBUG_PRINT(("wdstart %s\n", device_xname(wd->sc_dev)),
    649 	    DEBUG_XFERS);
    650 
    651 	if (!device_is_active(wd->sc_dev))
    652 		return;
    653 
    654 	mutex_enter(&wd->sc_lock);
    655 
    656 	/*
    657 	 * Do not queue any transfers until flush is finished, so that
    658 	 * once flush is pending, it will get handled as soon as xfer
    659 	 * is available.
    660 	 */
    661 	if (ISSET(wd->sc_flags, WDF_FLUSH_PEND)) {
    662 		ATADEBUG_PRINT(("wdstart %s flush pend\n",
    663 		    device_xname(wd->sc_dev)), DEBUG_XFERS);
    664 		goto out;
    665 	}
    666 
    667 	while (bufq_peek(wd->sc_q) != NULL) {
    668 		/* First try to get xfer. Limit to drive openings iff NCQ. */
    669 		xfer = ata_get_xfer_ext(wd->drvp->chnl_softc, 0,
    670 		    WD_USE_NCQ(wd) ? WD_MAX_OPENINGS(wd) : 0);
    671 		if (xfer == NULL) {
    672 			ATADEBUG_PRINT(("wdstart %s no xfer\n",
    673 			    device_xname(wd->sc_dev)), DEBUG_XFERS);
    674 			break;
    675 		}
    676 
    677 		/* There is got to be a buf for us */
    678 		bp = bufq_get(wd->sc_q);
    679 		KASSERT(bp != NULL);
    680 
    681 		xfer->c_retries = 0;
    682 		wdstart1(wd, bp, xfer);
    683 	}
    684 
    685 out:
    686 	mutex_exit(&wd->sc_lock);
    687 }
    688 
    689 void
    690 wdstart1(struct wd_softc *wd, struct buf *bp, struct ata_xfer *xfer)
    691 {
    692 	/* must be locked on entry */
    693 	KASSERT(mutex_owned(&wd->sc_lock));
    694 
    695 	KASSERT(bp == xfer->c_bio.bp || xfer->c_bio.bp == NULL);
    696 	KASSERT((xfer->c_flags & (C_WAITACT|C_FREE)) == 0);
    697 
    698 	/* Reset state, so that retries don't use stale info */
    699 	if (__predict_false(xfer->c_retries > 0)) {
    700 		xfer->c_flags = 0;
    701 		memset(&xfer->c_bio, 0, sizeof(xfer->c_bio));
    702 	}
    703 
    704 	xfer->c_bio.blkno = bp->b_rawblkno;
    705 	xfer->c_bio.bcount = bp->b_bcount;
    706 	xfer->c_bio.databuf = bp->b_data;
    707 	xfer->c_bio.blkdone = 0;
    708 	xfer->c_bio.bp = bp;
    709 
    710 #ifdef WD_CHAOS_MONKEY
    711 	/*
    712 	 * Override blkno to be over device capacity to trigger error,
    713 	 * but only if it's read, to avoid trashing disk contents should
    714 	 * the command be clipped, or otherwise misinterpreted, by the
    715 	 * driver or controller.
    716 	 */
    717 	if (BUF_ISREAD(bp) && xfer->c_retries == 0 && wd->drv_chaos_freq > 0 &&
    718 	    (++wd->drv_chaos_cnt % wd->drv_chaos_freq) == 0) {
    719 		aprint_normal_dev(wd->sc_dev, "%s: chaos xfer %d\n",
    720 		    __func__, xfer->c_slot);
    721 		xfer->c_bio.blkno = 7777777 + wd->sc_capacity;
    722 		xfer->c_flags |= C_CHAOS;
    723 	}
    724 #endif
    725 
    726 	/*
    727 	 * If we're retrying, retry in single-sector mode. This will give us
    728 	 * the sector number of the problem, and will eventually allow the
    729 	 * transfer to succeed. If FUA is requested, we can't actually
    730 	 * do this, as ATA_SINGLE is usually executed as PIO transfer by drivers
    731 	 * which support it, and that isn't compatible with NCQ/FUA.
    732 	 */
    733 	if (xfer->c_retries >= WDIORETRIES_SINGLE &&
    734 	    (bp->b_flags & B_MEDIA_FUA) == 0)
    735 		xfer->c_bio.flags = ATA_SINGLE;
    736 	else
    737 		xfer->c_bio.flags = 0;
    738 	if (wd->sc_flags & WDF_LBA48 &&
    739 	    (((xfer->c_bio.blkno +
    740 	     xfer->c_bio.bcount / wd->sc_dk.dk_label->d_secsize) >
    741 	    wd->sc_capacity28) ||
    742 	    ((xfer->c_bio.bcount / wd->sc_dk.dk_label->d_secsize) > 128)))
    743 		xfer->c_bio.flags |= ATA_LBA48;
    744 
    745 	/*
    746 	 * If NCQ was negotiated, always use it for the first several attempts.
    747 	 * Since device cancels all outstanding requests on error, downgrade
    748 	 * to non-NCQ on retry, so that the retried transfer would not cause
    749 	 * cascade failure for the other transfers if it fails again.
    750 	 * If FUA was requested, we can't downgrade, as that would violate
    751 	 * the semantics - FUA would not be honored. In that case, continue
    752 	 * retrying with NCQ.
    753 	 */
    754 	if (WD_USE_NCQ(wd) && (xfer->c_retries < WDIORETRIES_SINGLE ||
    755 	    (bp->b_flags & B_MEDIA_FUA) != 0)) {
    756 		xfer->c_bio.flags |= ATA_LBA48;
    757 		xfer->c_flags |= C_NCQ;
    758 
    759 		if (WD_USE_NCQ_PRIO(wd) &&
    760 		    BIO_GETPRIO(bp) == BPRIO_TIMECRITICAL)
    761 			xfer->c_bio.flags |= ATA_PRIO_HIGH;
    762 	}
    763 
    764 	if (wd->sc_flags & WDF_LBA)
    765 		xfer->c_bio.flags |= ATA_LBA;
    766 	if (bp->b_flags & B_READ)
    767 		xfer->c_bio.flags |= ATA_READ;
    768 	if (bp->b_flags & B_MEDIA_FUA) {
    769 		/* If not using NCQ, the command WRITE DMA FUA EXT is LBA48 */
    770 		KASSERT((wd->sc_flags & WDF_LBA48) != 0);
    771 		if ((xfer->c_flags & C_NCQ) == 0)
    772 			xfer->c_bio.flags |= ATA_LBA48;
    773 
    774 		xfer->c_bio.flags |= ATA_FUA;
    775 	}
    776 
    777 	/* Instrumentation. */
    778 	if (xfer->c_retries == 0)
    779 		disk_busy(&wd->sc_dk);
    780 	switch (wd->atabus->ata_bio(wd->drvp, xfer)) {
    781 	case ATACMD_TRY_AGAIN:
    782 		panic("wdstart1: try again");
    783 		break;
    784 	case ATACMD_QUEUED:
    785 	case ATACMD_COMPLETE:
    786 		break;
    787 	default:
    788 		panic("wdstart1: bad return code from ata_bio()");
    789 	}
    790 }
    791 
    792 void
    793 wddone(device_t self, struct ata_xfer *xfer)
    794 {
    795 	struct wd_softc *wd = device_private(self);
    796 	const char *errmsg;
    797 	int do_perror = 0;
    798 	struct buf *bp;
    799 
    800 	ATADEBUG_PRINT(("wddone %s\n", device_xname(wd->sc_dev)),
    801 	    DEBUG_XFERS);
    802 
    803 	if (__predict_false(wddoingadump)) {
    804 		/* just drop it to the floor */
    805 		ata_free_xfer(wd->drvp->chnl_softc, xfer);
    806 		return;
    807 	}
    808 
    809 	bp = xfer->c_bio.bp;
    810 	KASSERT(bp != NULL);
    811 
    812 	bp->b_resid = xfer->c_bio.bcount;
    813 	switch (xfer->c_bio.error) {
    814 	case ERR_DMA:
    815 		errmsg = "DMA error";
    816 		goto retry;
    817 	case ERR_DF:
    818 		errmsg = "device fault";
    819 		goto retry;
    820 	case TIMEOUT:
    821 		errmsg = "device timeout";
    822 		goto retry;
    823 	case REQUEUE:
    824 		errmsg = "requeue";
    825 		goto retry2;
    826 	case ERR_RESET:
    827 		errmsg = "channel reset";
    828 		goto retry2;
    829 	case ERROR:
    830 		/* Don't care about media change bits */
    831 		if (xfer->c_bio.r_error != 0 &&
    832 		    (xfer->c_bio.r_error & ~(WDCE_MC | WDCE_MCR)) == 0)
    833 			goto noerror;
    834 		errmsg = "error";
    835 		do_perror = 1;
    836 retry:		/* Just reset and retry. Can we do more ? */
    837 		if ((xfer->c_flags & C_RECOVERED) == 0) {
    838 			int wflags = (xfer->c_flags & C_POLL) ? AT_POLL : 0;
    839 			(*wd->atabus->ata_reset_drive)(wd->drvp, wflags, NULL);
    840 		}
    841 retry2:
    842 		mutex_enter(&wd->sc_lock);
    843 
    844 		diskerr(bp, "wd", errmsg, LOG_PRINTF,
    845 		    xfer->c_bio.blkdone, wd->sc_dk.dk_label);
    846 		if (xfer->c_retries < WDIORETRIES)
    847 			printf(", slot %d, retry %d", xfer->c_slot,
    848 			    xfer->c_retries + 1);
    849 		printf("\n");
    850 		if (do_perror)
    851 			wdperror(wd, xfer);
    852 
    853 		if (xfer->c_retries < WDIORETRIES) {
    854 			xfer->c_retries++;
    855 
    856 			/* Rerun ASAP if just requeued */
    857 			callout_reset(&xfer->c_retry_callout,
    858 			    (xfer->c_bio.error == REQUEUE) ? 1 : RECOVERYTIME,
    859 			    wdbiorestart, xfer);
    860 
    861 			mutex_exit(&wd->sc_lock);
    862 			return;
    863 		}
    864 
    865 		mutex_exit(&wd->sc_lock);
    866 
    867 #ifdef WD_SOFTBADSECT
    868 		/*
    869 		 * Not all errors indicate a failed block but those that do,
    870 		 * put the block on the bad-block list for the device.  Only
    871 		 * do this for reads because the drive should do it for writes,
    872 		 * itself, according to Manuel.
    873 		 */
    874 		if ((bp->b_flags & B_READ) &&
    875 		    ((wd->drvp->ata_vers >= 4 && xfer->c_bio.r_error & 64) ||
    876 		     (wd->drvp->ata_vers < 4 && xfer->c_bio.r_error & 192))) {
    877 			struct disk_badsectors *dbs;
    878 
    879 			dbs = malloc(sizeof *dbs, M_TEMP, M_NOWAIT);
    880 			if (dbs == NULL) {
    881 				aprint_error_dev(wd->sc_dev,
    882 				    "failed to add bad block to list\n");
    883 				goto out;
    884 			}
    885 
    886 			dbs->dbs_min = bp->b_rawblkno;
    887 			dbs->dbs_max = dbs->dbs_min +
    888 			    (bp->b_bcount /wd->sc_blksize) - 1;
    889 			microtime(&dbs->dbs_failedat);
    890 
    891 			mutex_enter(&wd->sc_lock);
    892 			SLIST_INSERT_HEAD(&wd->sc_bslist, dbs, dbs_next);
    893 			wd->sc_bscount++;
    894 			mutex_exit(&wd->sc_lock);
    895 		}
    896 out:
    897 #endif
    898 		bp->b_error = EIO;
    899 		break;
    900 	case NOERROR:
    901 noerror:	if ((xfer->c_bio.flags & ATA_CORR) || xfer->c_retries > 0)
    902 			aprint_error_dev(wd->sc_dev,
    903 			    "soft error (corrected) slot %d\n", xfer->c_slot);
    904 #ifdef WD_CHAOS_MONKEY
    905 		KASSERT((xfer->c_flags & C_CHAOS) == 0);
    906 #endif
    907 		break;
    908 	case ERR_NODEV:
    909 		bp->b_error = EIO;
    910 		break;
    911 	}
    912 	if (__predict_false(bp->b_error != 0) && bp->b_resid == 0) {
    913 		/*
    914 		 * the disk or controller sometimes report a complete
    915 		 * xfer, when there has been an error. This is wrong,
    916 		 * assume nothing got transfered in this case
    917 		 */
    918 		bp->b_resid = bp->b_bcount;
    919 	}
    920 	disk_unbusy(&wd->sc_dk, (bp->b_bcount - bp->b_resid),
    921 	    (bp->b_flags & B_READ));
    922 	rnd_add_uint32(&wd->rnd_source, bp->b_blkno);
    923 	ata_free_xfer(wd->drvp->chnl_softc, xfer);
    924 	biodone(bp);
    925 	ata_channel_start(wd->drvp->chnl_softc, wd->drvp->drive);
    926 }
    927 
    928 static void
    929 wdbiorestart(void *v)
    930 {
    931 	struct ata_xfer *xfer = v;
    932 	struct buf *bp = xfer->c_bio.bp;
    933 	struct wd_softc *wd = device_lookup_private(&wd_cd, WDUNIT(bp->b_dev));
    934 
    935 	ATADEBUG_PRINT(("wdrestart %s\n", device_xname(wd->sc_dev)),
    936 	    DEBUG_XFERS);
    937 
    938 	mutex_enter(&wd->sc_lock);
    939 	wdstart1(wd, bp, xfer);
    940 	mutex_exit(&wd->sc_lock);
    941 }
    942 
    943 static void
    944 wdminphys(struct buf *bp)
    945 {
    946 	const struct wd_softc * const wd =
    947 	    device_lookup_private(&wd_cd, WDUNIT(bp->b_dev));
    948 	uint32_t maxsectors;
    949 
    950 	/*
    951 	 * The limit is actually 65536 for LBA48 and 256 for non-LBA48,
    952 	 * but that requires to set the count for the ATA command
    953 	 * to 0, which is somewhat error prone, so better stay safe.
    954 	 */
    955 	if (wd->sc_flags & WDF_LBA48)
    956 		maxsectors = 65535;
    957 	else
    958 		maxsectors = 128;
    959 
    960 	if (bp->b_bcount > (wd->sc_blksize * maxsectors))
    961 		bp->b_bcount = (wd->sc_blksize * maxsectors);
    962 
    963 	minphys(bp);
    964 }
    965 
    966 int
    967 wdread(dev_t dev, struct uio *uio, int flags)
    968 {
    969 
    970 	ATADEBUG_PRINT(("wdread\n"), DEBUG_XFERS);
    971 	return (physio(wdstrategy, NULL, dev, B_READ, wdminphys, uio));
    972 }
    973 
    974 int
    975 wdwrite(dev_t dev, struct uio *uio, int flags)
    976 {
    977 
    978 	ATADEBUG_PRINT(("wdwrite\n"), DEBUG_XFERS);
    979 	return (physio(wdstrategy, NULL, dev, B_WRITE, wdminphys, uio));
    980 }
    981 
    982 int
    983 wdopen(dev_t dev, int flag, int fmt, struct lwp *l)
    984 {
    985 	struct wd_softc *wd;
    986 	int part, error;
    987 
    988 	ATADEBUG_PRINT(("wdopen\n"), DEBUG_FUNCS);
    989 	wd = device_lookup_private(&wd_cd, WDUNIT(dev));
    990 	if (wd == NULL)
    991 		return (ENXIO);
    992 
    993 	if (! device_is_active(wd->sc_dev))
    994 		return (ENODEV);
    995 
    996 	if (wd->sc_capacity == 0)
    997 		return (ENODEV);
    998 
    999 	part = WDPART(dev);
   1000 
   1001 	mutex_enter(&wd->sc_dk.dk_openlock);
   1002 
   1003 	/*
   1004 	 * If there are wedges, and this is not RAW_PART, then we
   1005 	 * need to fail.
   1006 	 */
   1007 	if (wd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
   1008 		error = EBUSY;
   1009 		goto bad1;
   1010 	}
   1011 
   1012 	/*
   1013 	 * If this is the first open of this device, add a reference
   1014 	 * to the adapter.
   1015 	 */
   1016 	if (wd->sc_dk.dk_openmask == 0 &&
   1017 	    (error = wd->atabus->ata_addref(wd->drvp)) != 0)
   1018 		goto bad1;
   1019 
   1020 	if (wd->sc_dk.dk_openmask != 0) {
   1021 		/*
   1022 		 * If any partition is open, but the disk has been invalidated,
   1023 		 * disallow further opens.
   1024 		 */
   1025 		if ((wd->sc_flags & WDF_LOADED) == 0) {
   1026 			error = EIO;
   1027 			goto bad2;
   1028 		}
   1029 	} else {
   1030 		if ((wd->sc_flags & WDF_LOADED) == 0) {
   1031 
   1032 			/* Load the physical device parameters. */
   1033 			if (wd_get_params(wd, AT_WAIT, &wd->sc_params) != 0) {
   1034 				aprint_error_dev(wd->sc_dev,
   1035 				"IDENTIFY failed\n");
   1036 				error = EIO;
   1037 				goto bad2;
   1038 			}
   1039 			wd->sc_flags |= WDF_LOADED;
   1040 			/* Load the partition info if not already loaded. */
   1041 			wdgetdisklabel(wd);
   1042 		}
   1043 	}
   1044 
   1045 	/* Check that the partition exists. */
   1046 	if (part != RAW_PART &&
   1047 	    (part >= wd->sc_dk.dk_label->d_npartitions ||
   1048 	     wd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
   1049 		error = ENXIO;
   1050 		goto bad2;
   1051 	}
   1052 
   1053 	/* Insure only one open at a time. */
   1054 	switch (fmt) {
   1055 	case S_IFCHR:
   1056 		wd->sc_dk.dk_copenmask |= (1 << part);
   1057 		break;
   1058 	case S_IFBLK:
   1059 		wd->sc_dk.dk_bopenmask |= (1 << part);
   1060 		break;
   1061 	}
   1062 	wd->sc_dk.dk_openmask =
   1063 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
   1064 
   1065 	mutex_exit(&wd->sc_dk.dk_openlock);
   1066 	return 0;
   1067 
   1068  bad2:
   1069 	if (wd->sc_dk.dk_openmask == 0)
   1070 		wd->atabus->ata_delref(wd->drvp);
   1071  bad1:
   1072 	mutex_exit(&wd->sc_dk.dk_openlock);
   1073 	return error;
   1074 }
   1075 
   1076 /*
   1077  * Caller must hold wd->sc_dk.dk_openlock.
   1078  */
   1079 static int
   1080 wdlastclose(device_t self)
   1081 {
   1082 	struct wd_softc *wd = device_private(self);
   1083 
   1084 	wd_flushcache(wd, AT_WAIT, false);
   1085 
   1086 	if (! (wd->sc_flags & WDF_KLABEL))
   1087 		wd->sc_flags &= ~WDF_LOADED;
   1088 
   1089 	wd->atabus->ata_delref(wd->drvp);
   1090 
   1091 	return 0;
   1092 }
   1093 
   1094 int
   1095 wdclose(dev_t dev, int flag, int fmt, struct lwp *l)
   1096 {
   1097 	struct wd_softc *wd =
   1098 	    device_lookup_private(&wd_cd, WDUNIT(dev));
   1099 	int part = WDPART(dev);
   1100 
   1101 	ATADEBUG_PRINT(("wdclose\n"), DEBUG_FUNCS);
   1102 
   1103 	mutex_enter(&wd->sc_dk.dk_openlock);
   1104 
   1105 	switch (fmt) {
   1106 	case S_IFCHR:
   1107 		wd->sc_dk.dk_copenmask &= ~(1 << part);
   1108 		break;
   1109 	case S_IFBLK:
   1110 		wd->sc_dk.dk_bopenmask &= ~(1 << part);
   1111 		break;
   1112 	}
   1113 	wd->sc_dk.dk_openmask =
   1114 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
   1115 
   1116 	if (wd->sc_dk.dk_openmask == 0)
   1117 		wdlastclose(wd->sc_dev);
   1118 
   1119 	mutex_exit(&wd->sc_dk.dk_openlock);
   1120 	return 0;
   1121 }
   1122 
   1123 void
   1124 wdgetdefaultlabel(struct wd_softc *wd, struct disklabel *lp)
   1125 {
   1126 
   1127 	ATADEBUG_PRINT(("wdgetdefaultlabel\n"), DEBUG_FUNCS);
   1128 	memset(lp, 0, sizeof(struct disklabel));
   1129 
   1130 	lp->d_secsize = wd->sc_blksize;
   1131 	lp->d_ntracks = wd->sc_params.atap_heads;
   1132 	lp->d_nsectors = wd->sc_params.atap_sectors;
   1133 	lp->d_ncylinders = (wd->sc_flags & WDF_LBA) ? wd->sc_capacity /
   1134 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors) :
   1135 		wd->sc_params.atap_cylinders;
   1136 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
   1137 
   1138 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
   1139 		lp->d_type = DKTYPE_ST506;
   1140 	else
   1141 		lp->d_type = DKTYPE_ESDI;
   1142 
   1143 	strncpy(lp->d_typename, wd->sc_params.atap_model, 16);
   1144 	strncpy(lp->d_packname, "fictitious", 16);
   1145 	if (wd->sc_capacity > UINT32_MAX)
   1146 		lp->d_secperunit = UINT32_MAX;
   1147 	else
   1148 		lp->d_secperunit = wd->sc_capacity;
   1149 	lp->d_rpm = 3600;
   1150 	lp->d_interleave = 1;
   1151 	lp->d_flags = 0;
   1152 
   1153 	lp->d_partitions[RAW_PART].p_offset = 0;
   1154 	lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
   1155 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
   1156 	lp->d_npartitions = RAW_PART + 1;
   1157 
   1158 	lp->d_magic = DISKMAGIC;
   1159 	lp->d_magic2 = DISKMAGIC;
   1160 	lp->d_checksum = dkcksum(lp);
   1161 }
   1162 
   1163 /*
   1164  * Fabricate a default disk label, and try to read the correct one.
   1165  */
   1166 void
   1167 wdgetdisklabel(struct wd_softc *wd)
   1168 {
   1169 	struct disklabel *lp = wd->sc_dk.dk_label;
   1170 	const char *errstring;
   1171 
   1172 	ATADEBUG_PRINT(("wdgetdisklabel\n"), DEBUG_FUNCS);
   1173 
   1174 	memset(wd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
   1175 
   1176 	wdgetdefaultlabel(wd, lp);
   1177 
   1178 	wd->drvp->badsect[0] = -1;
   1179 
   1180 	if (wd->drvp->state > RESET) {
   1181 		mutex_enter(&wd->sc_lock);
   1182 		wd->drvp->drive_flags |= ATA_DRIVE_RESET;
   1183 		mutex_exit(&wd->sc_lock);
   1184 	}
   1185 	errstring = readdisklabel(MAKEWDDEV(0, device_unit(wd->sc_dev),
   1186 				  RAW_PART), wdstrategy, lp,
   1187 				  wd->sc_dk.dk_cpulabel);
   1188 	if (errstring) {
   1189 		/*
   1190 		 * This probably happened because the drive's default
   1191 		 * geometry doesn't match the DOS geometry.  We
   1192 		 * assume the DOS geometry is now in the label and try
   1193 		 * again.  XXX This is a kluge.
   1194 		 */
   1195 		if (wd->drvp->state > RESET) {
   1196 			mutex_enter(&wd->sc_lock);
   1197 			wd->drvp->drive_flags |= ATA_DRIVE_RESET;
   1198 			mutex_exit(&wd->sc_lock);
   1199 		}
   1200 		errstring = readdisklabel(MAKEWDDEV(0, device_unit(wd->sc_dev),
   1201 		    RAW_PART), wdstrategy, lp, wd->sc_dk.dk_cpulabel);
   1202 	}
   1203 	if (errstring) {
   1204 		aprint_error_dev(wd->sc_dev, "%s\n", errstring);
   1205 		return;
   1206 	}
   1207 
   1208 	if (wd->drvp->state > RESET) {
   1209 		mutex_enter(&wd->sc_lock);
   1210 		wd->drvp->drive_flags |= ATA_DRIVE_RESET;
   1211 		mutex_exit(&wd->sc_lock);
   1212 	}
   1213 #ifdef HAS_BAD144_HANDLING
   1214 	if ((lp->d_flags & D_BADSECT) != 0)
   1215 		bad144intern(wd);
   1216 #endif
   1217 }
   1218 
   1219 void
   1220 wdperror(const struct wd_softc *wd, struct ata_xfer *xfer)
   1221 {
   1222 	static const char *const errstr0_3[] = {"address mark not found",
   1223 	    "track 0 not found", "aborted command", "media change requested",
   1224 	    "id not found", "media changed", "uncorrectable data error",
   1225 	    "bad block detected"};
   1226 	static const char *const errstr4_5[] = {
   1227 	    "obsolete (address mark not found)",
   1228 	    "no media/write protected", "aborted command",
   1229 	    "media change requested", "id not found", "media changed",
   1230 	    "uncorrectable data error", "interface CRC error"};
   1231 	const char *const *errstr;
   1232 	int i;
   1233 	const char *sep = "";
   1234 
   1235 	const char *devname = device_xname(wd->sc_dev);
   1236 	struct ata_drive_datas *drvp = wd->drvp;
   1237 	int errno = xfer->c_bio.r_error;
   1238 
   1239 	if (drvp->ata_vers >= 4)
   1240 		errstr = errstr4_5;
   1241 	else
   1242 		errstr = errstr0_3;
   1243 
   1244 	printf("%s: (", devname);
   1245 
   1246 	if (errno == 0)
   1247 		printf("error not notified");
   1248 
   1249 	for (i = 0; i < 8; i++) {
   1250 		if (errno & (1 << i)) {
   1251 			printf("%s%s", sep, errstr[i]);
   1252 			sep = ", ";
   1253 		}
   1254 	}
   1255 	printf(")\n");
   1256 }
   1257 
   1258 int
   1259 wdioctl(dev_t dev, u_long xfer, void *addr, int flag, struct lwp *l)
   1260 {
   1261 	struct wd_softc *wd =
   1262 	    device_lookup_private(&wd_cd, WDUNIT(dev));
   1263 	int error;
   1264 #ifdef __HAVE_OLD_DISKLABEL
   1265 	struct disklabel *newlabel = NULL;
   1266 #endif
   1267 
   1268 	ATADEBUG_PRINT(("wdioctl\n"), DEBUG_FUNCS);
   1269 
   1270 	if ((wd->sc_flags & WDF_LOADED) == 0)
   1271 		return EIO;
   1272 
   1273 	error = disk_ioctl(&wd->sc_dk, dev, xfer, addr, flag, l);
   1274 	if (error != EPASSTHROUGH)
   1275 		return error;
   1276 
   1277 	error = 0;
   1278 	switch (xfer) {
   1279 #ifdef HAS_BAD144_HANDLING
   1280 	case DIOCSBAD:
   1281 		if ((flag & FWRITE) == 0)
   1282 			return EBADF;
   1283 		wd->sc_dk.dk_cpulabel->bad = *(struct dkbad *)addr;
   1284 		wd->sc_dk.dk_label->d_flags |= D_BADSECT;
   1285 		bad144intern(wd);
   1286 		return 0;
   1287 #endif
   1288 #ifdef WD_SOFTBADSECT
   1289 	case DIOCBSLIST :
   1290 	{
   1291 		uint32_t count, missing, skip;
   1292 		struct disk_badsecinfo dbsi;
   1293 		struct disk_badsectors *dbs;
   1294 		size_t available;
   1295 		uint8_t *laddr;
   1296 
   1297 		dbsi = *(struct disk_badsecinfo *)addr;
   1298 		missing = wd->sc_bscount;
   1299 		count = 0;
   1300 		available = dbsi.dbsi_bufsize;
   1301 		skip = dbsi.dbsi_skip;
   1302 		laddr = (uint8_t *)dbsi.dbsi_buffer;
   1303 
   1304 		/*
   1305 		 * We start this loop with the expectation that all of the
   1306 		 * entries will be missed and decrement this counter each
   1307 		 * time we either skip over one (already copied out) or
   1308 		 * we actually copy it back to user space.  The structs
   1309 		 * holding the bad sector information are copied directly
   1310 		 * back to user space whilst the summary is returned via
   1311 		 * the struct passed in via the ioctl.
   1312 		 */
   1313 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next) {
   1314 			if (skip > 0) {
   1315 				missing--;
   1316 				skip--;
   1317 				continue;
   1318 			}
   1319 			if (available < sizeof(*dbs))
   1320 				break;
   1321 			available -= sizeof(*dbs);
   1322 			copyout(dbs, laddr, sizeof(*dbs));
   1323 			laddr += sizeof(*dbs);
   1324 			missing--;
   1325 			count++;
   1326 		}
   1327 		dbsi.dbsi_left = missing;
   1328 		dbsi.dbsi_copied = count;
   1329 		*(struct disk_badsecinfo *)addr = dbsi;
   1330 		return 0;
   1331 	}
   1332 
   1333 	case DIOCBSFLUSH :
   1334 		/* Clean out the bad sector list */
   1335 		while (!SLIST_EMPTY(&wd->sc_bslist)) {
   1336 			void *head = SLIST_FIRST(&wd->sc_bslist);
   1337 			SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
   1338 			free(head, M_TEMP);
   1339 		}
   1340 		wd->sc_bscount = 0;
   1341 		return 0;
   1342 #endif
   1343 
   1344 	case DIOCWDINFO:
   1345 	case DIOCSDINFO:
   1346 #ifdef __HAVE_OLD_DISKLABEL
   1347 	case ODIOCWDINFO:
   1348 	case ODIOCSDINFO:
   1349 #endif
   1350 	{
   1351 		struct disklabel *lp;
   1352 
   1353 		if ((flag & FWRITE) == 0)
   1354 			return EBADF;
   1355 
   1356 #ifdef __HAVE_OLD_DISKLABEL
   1357 		if (xfer == ODIOCSDINFO || xfer == ODIOCWDINFO) {
   1358 			newlabel = malloc(sizeof *newlabel, M_TEMP,
   1359 			    M_WAITOK | M_ZERO);
   1360 			if (newlabel == NULL)
   1361 				return EIO;
   1362 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
   1363 			lp = newlabel;
   1364 		} else
   1365 #endif
   1366 		lp = (struct disklabel *)addr;
   1367 
   1368 		mutex_enter(&wd->sc_dk.dk_openlock);
   1369 		wd->sc_flags |= WDF_LABELLING;
   1370 
   1371 		error = setdisklabel(wd->sc_dk.dk_label,
   1372 		    lp, /*wd->sc_dk.dk_openmask : */0,
   1373 		    wd->sc_dk.dk_cpulabel);
   1374 		if (error == 0) {
   1375 			if (wd->drvp->state > RESET) {
   1376 				mutex_enter(&wd->sc_lock);
   1377 				wd->drvp->drive_flags |= ATA_DRIVE_RESET;
   1378 				mutex_exit(&wd->sc_lock);
   1379 			}
   1380 			if (xfer == DIOCWDINFO
   1381 #ifdef __HAVE_OLD_DISKLABEL
   1382 			    || xfer == ODIOCWDINFO
   1383 #endif
   1384 			    )
   1385 				error = writedisklabel(WDLABELDEV(dev),
   1386 				    wdstrategy, wd->sc_dk.dk_label,
   1387 				    wd->sc_dk.dk_cpulabel);
   1388 		}
   1389 
   1390 		wd->sc_flags &= ~WDF_LABELLING;
   1391 		mutex_exit(&wd->sc_dk.dk_openlock);
   1392 #ifdef __HAVE_OLD_DISKLABEL
   1393 		if (newlabel != NULL)
   1394 			free(newlabel, M_TEMP);
   1395 #endif
   1396 		return error;
   1397 	}
   1398 
   1399 	case DIOCKLABEL:
   1400 		if (*(int *)addr)
   1401 			wd->sc_flags |= WDF_KLABEL;
   1402 		else
   1403 			wd->sc_flags &= ~WDF_KLABEL;
   1404 		return 0;
   1405 
   1406 	case DIOCWLABEL:
   1407 		if ((flag & FWRITE) == 0)
   1408 			return EBADF;
   1409 		if (*(int *)addr)
   1410 			wd->sc_flags |= WDF_WLABEL;
   1411 		else
   1412 			wd->sc_flags &= ~WDF_WLABEL;
   1413 		return 0;
   1414 
   1415 	case DIOCGDEFLABEL:
   1416 		wdgetdefaultlabel(wd, (struct disklabel *)addr);
   1417 		return 0;
   1418 #ifdef __HAVE_OLD_DISKLABEL
   1419 	case ODIOCGDEFLABEL:
   1420 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
   1421 		if (newlabel == NULL)
   1422 			return EIO;
   1423 		wdgetdefaultlabel(wd, newlabel);
   1424 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
   1425 			memcpy(addr, &newlabel, sizeof (struct olddisklabel));
   1426 		else
   1427 			error = ENOTTY;
   1428 		free(newlabel, M_TEMP);
   1429 		return error;
   1430 #endif
   1431 
   1432 #ifdef notyet
   1433 	case DIOCWFORMAT:
   1434 		if ((flag & FWRITE) == 0)
   1435 			return EBADF;
   1436 		{
   1437 		register struct format_op *fop;
   1438 		struct iovec aiov;
   1439 		struct uio auio;
   1440 
   1441 		fop = (struct format_op *)addr;
   1442 		aiov.iov_base = fop->df_buf;
   1443 		aiov.iov_len = fop->df_count;
   1444 		auio.uio_iov = &aiov;
   1445 		auio.uio_iovcnt = 1;
   1446 		auio.uio_resid = fop->df_count;
   1447 		auio.uio_offset =
   1448 			fop->df_startblk * wd->sc_dk.dk_label->d_secsize;
   1449 		auio.uio_vmspace = l->l_proc->p_vmspace;
   1450 		error = physio(wdformat, NULL, dev, B_WRITE, wdminphys,
   1451 		    &auio);
   1452 		fop->df_count -= auio.uio_resid;
   1453 		fop->df_reg[0] = wdc->sc_status;
   1454 		fop->df_reg[1] = wdc->sc_error;
   1455 		return error;
   1456 		}
   1457 #endif
   1458 	case DIOCGCACHE:
   1459 		return wd_getcache(wd, (int *)addr);
   1460 
   1461 	case DIOCSCACHE:
   1462 		return wd_setcache(wd, *(int *)addr);
   1463 
   1464 	case DIOCCACHESYNC:
   1465 		return wd_flushcache(wd, AT_WAIT, true);
   1466 
   1467 	case ATAIOCCOMMAND:
   1468 		/*
   1469 		 * Make sure this command is (relatively) safe first
   1470 		 */
   1471 		if ((((atareq_t *) addr)->flags & ATACMD_READ) == 0 &&
   1472 		    (flag & FWRITE) == 0)
   1473 			return (EBADF);
   1474 		{
   1475 		struct wd_ioctl *wi;
   1476 		atareq_t *atareq = (atareq_t *) addr;
   1477 		int error1;
   1478 
   1479 		wi = wi_get(wd);
   1480 		wi->wi_atareq = *atareq;
   1481 
   1482 		if (atareq->datalen && atareq->flags &
   1483 		    (ATACMD_READ | ATACMD_WRITE)) {
   1484 			void *tbuf;
   1485 			if (atareq->datalen < DEV_BSIZE
   1486 			    && atareq->command == WDCC_IDENTIFY) {
   1487 				tbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
   1488 				wi->wi_iov.iov_base = tbuf;
   1489 				wi->wi_iov.iov_len = DEV_BSIZE;
   1490 				UIO_SETUP_SYSSPACE(&wi->wi_uio);
   1491 			} else {
   1492 				tbuf = NULL;
   1493 				wi->wi_iov.iov_base = atareq->databuf;
   1494 				wi->wi_iov.iov_len = atareq->datalen;
   1495 				wi->wi_uio.uio_vmspace = l->l_proc->p_vmspace;
   1496 			}
   1497 			wi->wi_uio.uio_iov = &wi->wi_iov;
   1498 			wi->wi_uio.uio_iovcnt = 1;
   1499 			wi->wi_uio.uio_resid = atareq->datalen;
   1500 			wi->wi_uio.uio_offset = 0;
   1501 			wi->wi_uio.uio_rw =
   1502 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE;
   1503 			error1 = physio(wdioctlstrategy, &wi->wi_bp, dev,
   1504 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE,
   1505 			    wdminphys, &wi->wi_uio);
   1506 			if (tbuf != NULL && error1 == 0) {
   1507 				error1 = copyout(tbuf, atareq->databuf,
   1508 				    atareq->datalen);
   1509 				free(tbuf, M_TEMP);
   1510 			}
   1511 		} else {
   1512 			/* No need to call physio if we don't have any
   1513 			   user data */
   1514 			wi->wi_bp.b_flags = 0;
   1515 			wi->wi_bp.b_data = 0;
   1516 			wi->wi_bp.b_bcount = 0;
   1517 			wi->wi_bp.b_dev = dev;
   1518 			wi->wi_bp.b_proc = l->l_proc;
   1519 			wdioctlstrategy(&wi->wi_bp);
   1520 			error1 = wi->wi_bp.b_error;
   1521 		}
   1522 		*atareq = wi->wi_atareq;
   1523 		wi_free(wi);
   1524 		return(error1);
   1525 		}
   1526 
   1527 	case DIOCGSTRATEGY:
   1528 	    {
   1529 		struct disk_strategy *dks = (void *)addr;
   1530 
   1531 		mutex_enter(&wd->sc_lock);
   1532 		strlcpy(dks->dks_name, bufq_getstrategyname(wd->sc_q),
   1533 		    sizeof(dks->dks_name));
   1534 		mutex_exit(&wd->sc_lock);
   1535 		dks->dks_paramlen = 0;
   1536 
   1537 		return 0;
   1538 	    }
   1539 
   1540 	case DIOCSSTRATEGY:
   1541 	    {
   1542 		struct disk_strategy *dks = (void *)addr;
   1543 		struct bufq_state *new;
   1544 		struct bufq_state *old;
   1545 
   1546 		if ((flag & FWRITE) == 0) {
   1547 			return EBADF;
   1548 		}
   1549 		if (dks->dks_param != NULL) {
   1550 			return EINVAL;
   1551 		}
   1552 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
   1553 		error = bufq_alloc(&new, dks->dks_name,
   1554 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
   1555 		if (error) {
   1556 			return error;
   1557 		}
   1558 		mutex_enter(&wd->sc_lock);
   1559 		old = wd->sc_q;
   1560 		bufq_move(new, old);
   1561 		wd->sc_q = new;
   1562 		mutex_exit(&wd->sc_lock);
   1563 		bufq_free(old);
   1564 
   1565 		return 0;
   1566 	    }
   1567 
   1568 	default:
   1569 		return ENOTTY;
   1570 	}
   1571 
   1572 #ifdef DIAGNOSTIC
   1573 	panic("wdioctl: impossible");
   1574 #endif
   1575 }
   1576 
   1577 static int
   1578 wddiscard(dev_t dev, off_t pos, off_t len)
   1579 {
   1580 	struct wd_softc *wd = device_lookup_private(&wd_cd, WDUNIT(dev));
   1581 	daddr_t bno;
   1582 	long size, done;
   1583 	long maxatonce, amount;
   1584 	int result;
   1585 
   1586 	if (!(wd->sc_params.atap_ata_major & WDC_VER_ATA7)
   1587 	    || !(wd->sc_params.support_dsm & ATA_SUPPORT_DSM_TRIM)) {
   1588 		/* not supported; ignore request */
   1589 		ATADEBUG_PRINT(("wddiscard (unsupported)\n"), DEBUG_FUNCS);
   1590 		return 0;
   1591 	}
   1592 	maxatonce = 0xffff; /*wd->sc_params.max_dsm_blocks*/
   1593 
   1594 	ATADEBUG_PRINT(("wddiscard\n"), DEBUG_FUNCS);
   1595 
   1596 	if ((wd->sc_flags & WDF_LOADED) == 0)
   1597 		return EIO;
   1598 
   1599 	/* round the start up and the end down */
   1600 	bno = (pos + wd->sc_blksize - 1) / wd->sc_blksize;
   1601 	size = ((pos + len) / wd->sc_blksize) - bno;
   1602 
   1603 	done = 0;
   1604 	while (done < size) {
   1605 	     amount = size - done;
   1606 	     if (amount > maxatonce) {
   1607 		     amount = maxatonce;
   1608 	     }
   1609 	     result = wd_trim(wd, WDPART(dev), bno + done, amount);
   1610 	     if (result) {
   1611 		     return result;
   1612 	     }
   1613 	     done += amount;
   1614 	}
   1615 	return 0;
   1616 }
   1617 
   1618 #ifdef B_FORMAT
   1619 int
   1620 wdformat(struct buf *bp)
   1621 {
   1622 
   1623 	bp->b_flags |= B_FORMAT;
   1624 	return wdstrategy(bp);
   1625 }
   1626 #endif
   1627 
   1628 int
   1629 wdsize(dev_t dev)
   1630 {
   1631 	struct wd_softc *wd;
   1632 	int part, omask;
   1633 	int size;
   1634 
   1635 	ATADEBUG_PRINT(("wdsize\n"), DEBUG_FUNCS);
   1636 
   1637 	wd = device_lookup_private(&wd_cd, WDUNIT(dev));
   1638 	if (wd == NULL)
   1639 		return (-1);
   1640 
   1641 	part = WDPART(dev);
   1642 	omask = wd->sc_dk.dk_openmask & (1 << part);
   1643 
   1644 	if (omask == 0 && wdopen(dev, 0, S_IFBLK, NULL) != 0)
   1645 		return (-1);
   1646 	if (wd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
   1647 		size = -1;
   1648 	else
   1649 		size = wd->sc_dk.dk_label->d_partitions[part].p_size *
   1650 		    (wd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
   1651 	if (omask == 0 && wdclose(dev, 0, S_IFBLK, NULL) != 0)
   1652 		return (-1);
   1653 	return (size);
   1654 }
   1655 
   1656 /*
   1657  * Dump core after a system crash.
   1658  */
   1659 int
   1660 wddump(dev_t dev, daddr_t blkno, void *va, size_t size)
   1661 {
   1662 	struct wd_softc *wd;	/* disk unit to do the I/O */
   1663 	struct disklabel *lp;   /* disk's disklabel */
   1664 	int part, err;
   1665 	int nblks;	/* total number of sectors left to write */
   1666 	struct ata_xfer *xfer;
   1667 
   1668 	/* Check if recursive dump; if so, punt. */
   1669 	if (wddoingadump)
   1670 		return EFAULT;
   1671 	wddoingadump = 1;
   1672 
   1673 	wd = device_lookup_private(&wd_cd, WDUNIT(dev));
   1674 	if (wd == NULL)
   1675 		return (ENXIO);
   1676 
   1677 	part = WDPART(dev);
   1678 
   1679 	/* Convert to disk sectors.  Request must be a multiple of size. */
   1680 	lp = wd->sc_dk.dk_label;
   1681 	if ((size % lp->d_secsize) != 0)
   1682 		return EFAULT;
   1683 	nblks = size / lp->d_secsize;
   1684 	blkno = blkno / (lp->d_secsize / DEV_BSIZE);
   1685 
   1686 	/* Check transfer bounds against partition size. */
   1687 	if ((blkno < 0) || ((blkno + nblks) > lp->d_partitions[part].p_size))
   1688 		return EINVAL;
   1689 
   1690 	/* Offset block number to start of partition. */
   1691 	blkno += lp->d_partitions[part].p_offset;
   1692 
   1693 	/* Recalibrate, if first dump transfer. */
   1694 	if (wddumprecalibrated == 0) {
   1695 		wddumprecalibrated = 1;
   1696 		(*wd->atabus->ata_reset_drive)(wd->drvp,
   1697 					       AT_POLL | AT_RST_EMERG, NULL);
   1698 		wd->drvp->state = RESET;
   1699 	}
   1700 
   1701 	xfer = ata_get_xfer_ext(wd->drvp->chnl_softc, 0, 0);
   1702 	if (xfer == NULL) {
   1703 		printf("%s: no xfer\n", __func__);
   1704 		return EAGAIN;
   1705 	}
   1706 
   1707 	xfer->c_bio.blkno = blkno;
   1708 	xfer->c_bio.flags = ATA_POLL;
   1709 	if (wd->sc_flags & WDF_LBA48 &&
   1710 	    (xfer->c_bio.blkno + nblks) > wd->sc_capacity28)
   1711 		xfer->c_bio.flags |= ATA_LBA48;
   1712 	if (wd->sc_flags & WDF_LBA)
   1713 		xfer->c_bio.flags |= ATA_LBA;
   1714 	xfer->c_bio.bcount = nblks * lp->d_secsize;
   1715 	xfer->c_bio.databuf = va;
   1716 #ifndef WD_DUMP_NOT_TRUSTED
   1717 	switch (err = wd->atabus->ata_bio(wd->drvp, xfer)) {
   1718 	case ATACMD_TRY_AGAIN:
   1719 		panic("wddump: try again");
   1720 		break;
   1721 	case ATACMD_QUEUED:
   1722 		panic("wddump: polled command has been queued");
   1723 		break;
   1724 	case ATACMD_COMPLETE:
   1725 		break;
   1726 	default:
   1727 		panic("wddump: unknown atacmd code %d", err);
   1728 	}
   1729 	switch(err = xfer->c_bio.error) {
   1730 	case TIMEOUT:
   1731 		printf("wddump: device timed out");
   1732 		err = EIO;
   1733 		break;
   1734 	case ERR_DF:
   1735 		printf("wddump: drive fault");
   1736 		err = EIO;
   1737 		break;
   1738 	case ERR_DMA:
   1739 		printf("wddump: DMA error");
   1740 		err = EIO;
   1741 		break;
   1742 	case ERROR:
   1743 		printf("wddump: ");
   1744 		wdperror(wd, xfer);
   1745 		err = EIO;
   1746 		break;
   1747 	case NOERROR:
   1748 		err = 0;
   1749 		break;
   1750 	default:
   1751 		panic("wddump: unknown error type %d", err);
   1752 	}
   1753 
   1754 	if (err != 0) {
   1755 		printf("\n");
   1756 		return err;
   1757 	}
   1758 #else	/* WD_DUMP_NOT_TRUSTED */
   1759 	/* Let's just talk about this first... */
   1760 	printf("wd%d: dump addr 0x%x, cylin %d, head %d, sector %d\n",
   1761 	    unit, va, cylin, head, sector);
   1762 	delay(500 * 1000);	/* half a second */
   1763 #endif
   1764 
   1765 	wddoingadump = 0;
   1766 	return 0;
   1767 }
   1768 
   1769 #ifdef HAS_BAD144_HANDLING
   1770 /*
   1771  * Internalize the bad sector table.
   1772  */
   1773 void
   1774 bad144intern(struct wd_softc *wd)
   1775 {
   1776 	struct dkbad *bt = &wd->sc_dk.dk_cpulabel->bad;
   1777 	struct disklabel *lp = wd->sc_dk.dk_label;
   1778 	int i = 0;
   1779 
   1780 	ATADEBUG_PRINT(("bad144intern\n"), DEBUG_XFERS);
   1781 
   1782 	for (; i < NBT_BAD; i++) {
   1783 		if (bt->bt_bad[i].bt_cyl == 0xffff)
   1784 			break;
   1785 		wd->drvp->badsect[i] =
   1786 		    bt->bt_bad[i].bt_cyl * lp->d_secpercyl +
   1787 		    (bt->bt_bad[i].bt_trksec >> 8) * lp->d_nsectors +
   1788 		    (bt->bt_bad[i].bt_trksec & 0xff);
   1789 	}
   1790 	for (; i < NBT_BAD+1; i++)
   1791 		wd->drvp->badsect[i] = -1;
   1792 }
   1793 #endif
   1794 
   1795 static void
   1796 wd_params_to_properties(struct wd_softc *wd)
   1797 {
   1798 	struct disk_geom *dg = &wd->sc_dk.dk_geom;
   1799 
   1800 	memset(dg, 0, sizeof(*dg));
   1801 
   1802 	dg->dg_secperunit = wd->sc_capacity;
   1803 	dg->dg_secsize = wd->sc_blksize;
   1804 	dg->dg_nsectors = wd->sc_params.atap_sectors;
   1805 	dg->dg_ntracks = wd->sc_params.atap_heads;
   1806 	if ((wd->sc_flags & WDF_LBA) == 0)
   1807 		dg->dg_ncylinders = wd->sc_params.atap_cylinders;
   1808 
   1809 	/* XXX Should have a case for ATA here, too. */
   1810 	const char *cp = strcmp(wd->sc_params.atap_model, "ST506") ?
   1811 	    "ST506" : "ESDI";
   1812 
   1813 	disk_set_info(wd->sc_dev, &wd->sc_dk, cp);
   1814 }
   1815 
   1816 int
   1817 wd_get_params(struct wd_softc *wd, uint8_t flags, struct ataparams *params)
   1818 {
   1819 
   1820 	switch (wd->atabus->ata_get_params(wd->drvp, flags, params)) {
   1821 	case CMD_AGAIN:
   1822 		return 1;
   1823 	case CMD_ERR:
   1824 		if (wd->drvp->drive_type != ATA_DRIVET_OLD)
   1825 			return 1;
   1826 		/*
   1827 		 * We `know' there's a drive here; just assume it's old.
   1828 		 * This geometry is only used to read the MBR and print a
   1829 		 * (false) attach message.
   1830 		 */
   1831 		strncpy(params->atap_model, "ST506",
   1832 		    sizeof params->atap_model);
   1833 		params->atap_config = ATA_CFG_FIXED;
   1834 		params->atap_cylinders = 1024;
   1835 		params->atap_heads = 8;
   1836 		params->atap_sectors = 17;
   1837 		params->atap_multi = 1;
   1838 		params->atap_capabilities1 = params->atap_capabilities2 = 0;
   1839 		wd->drvp->ata_vers = -1; /* Mark it as pre-ATA */
   1840 		/* FALLTHROUGH */
   1841 	case CMD_OK:
   1842 		return 0;
   1843 	default:
   1844 		panic("wd_get_params: bad return code from ata_get_params");
   1845 		/* NOTREACHED */
   1846 	}
   1847 }
   1848 
   1849 int
   1850 wd_getcache(struct wd_softc *wd, int *bitsp)
   1851 {
   1852 	struct ataparams params;
   1853 
   1854 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
   1855 		return EIO;
   1856 	if (params.atap_cmd_set1 == 0x0000 ||
   1857 	    params.atap_cmd_set1 == 0xffff ||
   1858 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0) {
   1859 		*bitsp = 0;
   1860 		return 0;
   1861 	}
   1862 	*bitsp = DKCACHE_WCHANGE | DKCACHE_READ;
   1863 	if (params.atap_cmd1_en & WDC_CMD1_CACHE)
   1864 		*bitsp |= DKCACHE_WRITE;
   1865 
   1866 	if (WD_USE_NCQ(wd) || (wd->drvp->drive_flags & ATA_DRIVE_WFUA))
   1867 		*bitsp |= DKCACHE_FUA;
   1868 
   1869 	return 0;
   1870 }
   1871 
   1872 const char at_errbits[] = "\20\10ERROR\11TIMEOU\12DF";
   1873 
   1874 int
   1875 wd_setcache(struct wd_softc *wd, int bits)
   1876 {
   1877 	struct ataparams params;
   1878 	struct ata_xfer *xfer;
   1879 	int error;
   1880 
   1881 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
   1882 		return EIO;
   1883 
   1884 	if (params.atap_cmd_set1 == 0x0000 ||
   1885 	    params.atap_cmd_set1 == 0xffff ||
   1886 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0)
   1887 		return EOPNOTSUPP;
   1888 
   1889 	if ((bits & DKCACHE_READ) == 0 ||
   1890 	    (bits & DKCACHE_SAVE) != 0)
   1891 		return EOPNOTSUPP;
   1892 
   1893 	xfer = ata_get_xfer(wd->drvp->chnl_softc);
   1894 	if (xfer == NULL)
   1895 		return EINTR;
   1896 
   1897 	xfer->c_ata_c.r_command = SET_FEATURES;
   1898 	xfer->c_ata_c.r_st_bmask = 0;
   1899 	xfer->c_ata_c.r_st_pmask = 0;
   1900 	xfer->c_ata_c.timeout = 30000; /* 30s timeout */
   1901 	xfer->c_ata_c.flags = AT_WAIT;
   1902 	if (bits & DKCACHE_WRITE)
   1903 		xfer->c_ata_c.r_features = WDSF_WRITE_CACHE_EN;
   1904 	else
   1905 		xfer->c_ata_c.r_features = WDSF_WRITE_CACHE_DS;
   1906 	if (wd->atabus->ata_exec_command(wd->drvp, xfer) != ATACMD_COMPLETE) {
   1907 		aprint_error_dev(wd->sc_dev,
   1908 		    "wd_setcache command not complete\n");
   1909 		error = EIO;
   1910 		goto out;
   1911 	}
   1912 
   1913 	if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   1914 		char sbuf[sizeof(at_errbits) + 64];
   1915 		snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
   1916 		aprint_error_dev(wd->sc_dev, "wd_setcache: status=%s\n", sbuf);
   1917 		error = EIO;
   1918 		goto out;
   1919 	}
   1920 
   1921 	error = 0;
   1922 
   1923 out:
   1924 	ata_free_xfer(wd->drvp->chnl_softc, xfer);
   1925 	ata_channel_start(wd->drvp->chnl_softc, wd->drvp->drive);
   1926 	return error;
   1927 }
   1928 
   1929 static int
   1930 wd_standby(struct wd_softc *wd, int flags)
   1931 {
   1932 	struct ata_xfer *xfer;
   1933 	int error;
   1934 
   1935 	xfer = ata_get_xfer(wd->drvp->chnl_softc);
   1936 	if (xfer == NULL)
   1937 		return EINTR;
   1938 
   1939 	xfer->c_ata_c.r_command = WDCC_STANDBY_IMMED;
   1940 	xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
   1941 	xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
   1942 	xfer->c_ata_c.flags = flags;
   1943 	xfer->c_ata_c.timeout = 30000; /* 30s timeout */
   1944 	if (wd->atabus->ata_exec_command(wd->drvp, xfer) != ATACMD_COMPLETE) {
   1945 		aprint_error_dev(wd->sc_dev,
   1946 		    "standby immediate command didn't complete\n");
   1947 		error = EIO;
   1948 		goto out;
   1949 	}
   1950 	if (xfer->c_ata_c.flags & AT_ERROR) {
   1951 		if (xfer->c_ata_c.r_error == WDCE_ABRT) {
   1952 			/* command not supported */
   1953 			error = ENODEV;
   1954 			goto out;
   1955 		}
   1956 	}
   1957 	if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   1958 		char sbuf[sizeof(at_errbits) + 64];
   1959 		snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
   1960 		aprint_error_dev(wd->sc_dev, "wd_standby: status=%s\n", sbuf);
   1961 		error = EIO;
   1962 		goto out;
   1963 	}
   1964 	error = 0;
   1965 
   1966 out:
   1967 	ata_free_xfer(wd->drvp->chnl_softc, xfer);
   1968 	/* drive is supposed to go idle, do not call ata_channel_start() */
   1969 	return error;
   1970 }
   1971 
   1972 int
   1973 wd_flushcache(struct wd_softc *wd, int flags, bool start)
   1974 {
   1975 	struct ata_xfer *xfer;
   1976 	int error;
   1977 
   1978 	/*
   1979 	 * WDCC_FLUSHCACHE is here since ATA-4, but some drives report
   1980 	 * only ATA-2 and still support it.
   1981 	 */
   1982 	if (wd->drvp->ata_vers < 4 &&
   1983 	    ((wd->sc_params.atap_cmd_set2 & WDC_CMD2_FC) == 0 ||
   1984 	    wd->sc_params.atap_cmd_set2 == 0xffff))
   1985 		return ENODEV;
   1986 
   1987 	mutex_enter(&wd->sc_lock);
   1988 	SET(wd->sc_flags, WDF_FLUSH_PEND);
   1989 	mutex_exit(&wd->sc_lock);
   1990 
   1991 	xfer = ata_get_xfer(wd->drvp->chnl_softc);
   1992 
   1993 	mutex_enter(&wd->sc_lock);
   1994 	CLR(wd->sc_flags, WDF_FLUSH_PEND);
   1995 	mutex_exit(&wd->sc_lock);
   1996 
   1997 	if (xfer == NULL) {
   1998 		error = EINTR;
   1999 		goto out;
   2000 	}
   2001 
   2002 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0 &&
   2003 	    (wd->sc_params.atap_cmd2_en & ATA_CMD2_FCE) != 0) {
   2004 		xfer->c_ata_c.r_command = WDCC_FLUSHCACHE_EXT;
   2005 		flags |= AT_LBA48;
   2006 	} else
   2007 		xfer->c_ata_c.r_command = WDCC_FLUSHCACHE;
   2008 	xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
   2009 	xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
   2010 	xfer->c_ata_c.flags = flags | AT_READREG;
   2011 	xfer->c_ata_c.timeout = 300000; /* 5m timeout */
   2012 	if (wd->atabus->ata_exec_command(wd->drvp, xfer) != ATACMD_COMPLETE) {
   2013 		aprint_error_dev(wd->sc_dev,
   2014 		    "flush cache command didn't complete\n");
   2015 		error = EIO;
   2016 		goto out_xfer;
   2017 	}
   2018 	if (xfer->c_ata_c.flags & AT_ERROR) {
   2019 		if (xfer->c_ata_c.r_error == WDCE_ABRT) {
   2020 			/* command not supported */
   2021 			error = ENODEV;
   2022 			goto out_xfer;
   2023 		}
   2024 	}
   2025 	if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   2026 		char sbuf[sizeof(at_errbits) + 64];
   2027 		snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
   2028 		aprint_error_dev(wd->sc_dev, "wd_flushcache: status=%s\n",
   2029 		    sbuf);
   2030 		error = EIO;
   2031 		goto out_xfer;
   2032 	}
   2033 	error = 0;
   2034 
   2035 out_xfer:
   2036 	ata_free_xfer(wd->drvp->chnl_softc, xfer);
   2037 
   2038 out:
   2039 	/* kick queue processing blocked while waiting for flush xfer */
   2040 	if (start)
   2041 		ata_channel_start(wd->drvp->chnl_softc, wd->drvp->drive);
   2042 
   2043 	return error;
   2044 }
   2045 
   2046 int
   2047 wd_trim(struct wd_softc *wd, int part, daddr_t bno, long size)
   2048 {
   2049 	struct ata_xfer *xfer;
   2050 	int error;
   2051 	unsigned char *req;
   2052 
   2053 	if (part != RAW_PART)
   2054 		bno += wd->sc_dk.dk_label->d_partitions[part].p_offset;;
   2055 
   2056 	xfer = ata_get_xfer(wd->drvp->chnl_softc);
   2057 	if (xfer == NULL)
   2058 		return EINTR;
   2059 
   2060 	req = kmem_zalloc(512, KM_SLEEP);
   2061 	req[0] = bno & 0xff;
   2062 	req[1] = (bno >> 8) & 0xff;
   2063 	req[2] = (bno >> 16) & 0xff;
   2064 	req[3] = (bno >> 24) & 0xff;
   2065 	req[4] = (bno >> 32) & 0xff;
   2066 	req[5] = (bno >> 40) & 0xff;
   2067 	req[6] = size & 0xff;
   2068 	req[7] = (size >> 8) & 0xff;
   2069 
   2070 	xfer->c_ata_c.r_command = ATA_DATA_SET_MANAGEMENT;
   2071 	xfer->c_ata_c.r_count = 1;
   2072 	xfer->c_ata_c.r_features = ATA_SUPPORT_DSM_TRIM;
   2073 	xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
   2074 	xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
   2075 	xfer->c_ata_c.timeout = 30000; /* 30s timeout */
   2076 	xfer->c_ata_c.data = req;
   2077 	xfer->c_ata_c.bcount = 512;
   2078 	xfer->c_ata_c.flags |= AT_WRITE | AT_WAIT;
   2079 	if (wd->atabus->ata_exec_command(wd->drvp, xfer) != ATACMD_COMPLETE) {
   2080 		aprint_error_dev(wd->sc_dev,
   2081 		    "trim command didn't complete\n");
   2082 		kmem_free(req, 512);
   2083 		error = EIO;
   2084 		goto out;
   2085 	}
   2086 	kmem_free(req, 512);
   2087 	if (xfer->c_ata_c.flags & AT_ERROR) {
   2088 		if (xfer->c_ata_c.r_error == WDCE_ABRT) {
   2089 			/* command not supported */
   2090 			error = ENODEV;
   2091 			goto out;
   2092 		}
   2093 	}
   2094 	if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   2095 		char sbuf[sizeof(at_errbits) + 64];
   2096 		snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
   2097 		aprint_error_dev(wd->sc_dev, "wd_trim: status=%s\n",
   2098 		    sbuf);
   2099 		error = EIO;
   2100 		goto out;
   2101 	}
   2102 	error = 0;
   2103 
   2104 out:
   2105 	ata_free_xfer(wd->drvp->chnl_softc, xfer);
   2106 	ata_channel_start(wd->drvp->chnl_softc, wd->drvp->drive);
   2107 	return error;
   2108 }
   2109 
   2110 bool
   2111 wd_shutdown(device_t dev, int how)
   2112 {
   2113 	struct wd_softc *wd = device_private(dev);
   2114 
   2115 	/* the adapter needs to be enabled */
   2116 	if (wd->atabus->ata_addref(wd->drvp))
   2117 		return true; /* no need to complain */
   2118 
   2119 	wd_flushcache(wd, AT_POLL, false);
   2120 	if ((how & RB_POWERDOWN) == RB_POWERDOWN)
   2121 		wd_standby(wd, AT_POLL);
   2122 	return true;
   2123 }
   2124 
   2125 /*
   2126  * Allocate space for a ioctl queue structure.  Mostly taken from
   2127  * scsipi_ioctl.c
   2128  */
   2129 struct wd_ioctl *
   2130 wi_get(struct wd_softc *wd)
   2131 {
   2132 	struct wd_ioctl *wi;
   2133 
   2134 	wi = malloc(sizeof(struct wd_ioctl), M_TEMP, M_WAITOK|M_ZERO);
   2135 	wi->wi_softc = wd;
   2136 	buf_init(&wi->wi_bp);
   2137 
   2138 	return (wi);
   2139 }
   2140 
   2141 /*
   2142  * Free an ioctl structure and remove it from our list
   2143  */
   2144 
   2145 void
   2146 wi_free(struct wd_ioctl *wi)
   2147 {
   2148 	buf_destroy(&wi->wi_bp);
   2149 	free(wi, M_TEMP);
   2150 }
   2151 
   2152 /*
   2153  * Find a wd_ioctl structure based on the struct buf.
   2154  */
   2155 
   2156 struct wd_ioctl *
   2157 wi_find(struct buf *bp)
   2158 {
   2159 	return container_of(bp, struct wd_ioctl, wi_bp);
   2160 }
   2161 
   2162 static uint
   2163 wi_sector_size(const struct wd_ioctl * const wi)
   2164 {
   2165 	switch (wi->wi_atareq.command) {
   2166 	case WDCC_READ:
   2167 	case WDCC_WRITE:
   2168 	case WDCC_READMULTI:
   2169 	case WDCC_WRITEMULTI:
   2170 	case WDCC_READDMA:
   2171 	case WDCC_WRITEDMA:
   2172 	case WDCC_READ_EXT:
   2173 	case WDCC_WRITE_EXT:
   2174 	case WDCC_READMULTI_EXT:
   2175 	case WDCC_WRITEMULTI_EXT:
   2176 	case WDCC_READDMA_EXT:
   2177 	case WDCC_WRITEDMA_EXT:
   2178 	case WDCC_READ_FPDMA_QUEUED:
   2179 	case WDCC_WRITE_FPDMA_QUEUED:
   2180 		return wi->wi_softc->sc_blksize;
   2181 	default:
   2182 		return 512;
   2183 	}
   2184 }
   2185 
   2186 /*
   2187  * Ioctl pseudo strategy routine
   2188  *
   2189  * This is mostly stolen from scsipi_ioctl.c:scsistrategy().  What
   2190  * happens here is:
   2191  *
   2192  * - wdioctl() queues a wd_ioctl structure.
   2193  *
   2194  * - wdioctl() calls physio/wdioctlstrategy based on whether or not
   2195  *   user space I/O is required.  If physio() is called, physio() eventually
   2196  *   calls wdioctlstrategy().
   2197  *
   2198  * - In either case, wdioctlstrategy() calls wd->atabus->ata_exec_command()
   2199  *   to perform the actual command
   2200  *
   2201  * The reason for the use of the pseudo strategy routine is because
   2202  * when doing I/O to/from user space, physio _really_ wants to be in
   2203  * the loop.  We could put the entire buffer into the ioctl request
   2204  * structure, but that won't scale if we want to do things like download
   2205  * microcode.
   2206  */
   2207 
   2208 void
   2209 wdioctlstrategy(struct buf *bp)
   2210 {
   2211 	struct wd_ioctl *wi;
   2212 	struct ata_xfer *xfer;
   2213 	int error = 0;
   2214 
   2215 	wi = wi_find(bp);
   2216 	if (wi == NULL) {
   2217 		printf("wdioctlstrategy: "
   2218 		    "No matching ioctl request found in queue\n");
   2219 		error = EINVAL;
   2220 		goto out2;
   2221 	}
   2222 
   2223 	xfer = ata_get_xfer(wi->wi_softc->drvp->chnl_softc);
   2224 	if (xfer == NULL) {
   2225 		error = EINTR;
   2226 		goto out2;
   2227 	}
   2228 
   2229 	/*
   2230 	 * Abort if physio broke up the transfer
   2231 	 */
   2232 
   2233 	if (bp->b_bcount != wi->wi_atareq.datalen) {
   2234 		printf("physio split wd ioctl request... cannot proceed\n");
   2235 		error = EIO;
   2236 		goto out;
   2237 	}
   2238 
   2239 	/*
   2240 	 * Abort if we didn't get a buffer size that was a multiple of
   2241 	 * our sector size (or overflows CHS/LBA28 sector count)
   2242 	 */
   2243 
   2244 	if ((bp->b_bcount % wi_sector_size(wi)) != 0 ||
   2245 	    (bp->b_bcount / wi_sector_size(wi)) >=
   2246 	     (1 << NBBY)) {
   2247 		error = EINVAL;
   2248 		goto out;
   2249 	}
   2250 
   2251 	/*
   2252 	 * Make sure a timeout was supplied in the ioctl request
   2253 	 */
   2254 
   2255 	if (wi->wi_atareq.timeout == 0) {
   2256 		error = EINVAL;
   2257 		goto out;
   2258 	}
   2259 
   2260 	if (wi->wi_atareq.flags & ATACMD_READ)
   2261 		xfer->c_ata_c.flags |= AT_READ;
   2262 	else if (wi->wi_atareq.flags & ATACMD_WRITE)
   2263 		xfer->c_ata_c.flags |= AT_WRITE;
   2264 
   2265 	if (wi->wi_atareq.flags & ATACMD_READREG)
   2266 		xfer->c_ata_c.flags |= AT_READREG;
   2267 
   2268 	if ((wi->wi_atareq.flags & ATACMD_LBA) != 0)
   2269 		xfer->c_ata_c.flags |= AT_LBA;
   2270 
   2271 	xfer->c_ata_c.flags |= AT_WAIT;
   2272 
   2273 	xfer->c_ata_c.timeout = wi->wi_atareq.timeout;
   2274 	xfer->c_ata_c.r_command = wi->wi_atareq.command;
   2275 	xfer->c_ata_c.r_lba = ((wi->wi_atareq.head & 0x0f) << 24) |
   2276 	    (wi->wi_atareq.cylinder << 8) |
   2277 	    wi->wi_atareq.sec_num;
   2278 	xfer->c_ata_c.r_count = wi->wi_atareq.sec_count;
   2279 	xfer->c_ata_c.r_features = wi->wi_atareq.features;
   2280 	xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
   2281 	xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
   2282 	xfer->c_ata_c.data = wi->wi_bp.b_data;
   2283 	xfer->c_ata_c.bcount = wi->wi_bp.b_bcount;
   2284 
   2285 	if (wi->wi_softc->atabus->ata_exec_command(wi->wi_softc->drvp, xfer)
   2286 	    != ATACMD_COMPLETE) {
   2287 		wi->wi_atareq.retsts = ATACMD_ERROR;
   2288 		error = EIO;
   2289 		goto out;
   2290 	}
   2291 
   2292 	if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
   2293 		if (xfer->c_ata_c.flags & AT_ERROR) {
   2294 			wi->wi_atareq.retsts = ATACMD_ERROR;
   2295 			wi->wi_atareq.error = xfer->c_ata_c.r_error;
   2296 		} else if (xfer->c_ata_c.flags & AT_DF)
   2297 			wi->wi_atareq.retsts = ATACMD_DF;
   2298 		else
   2299 			wi->wi_atareq.retsts = ATACMD_TIMEOUT;
   2300 	} else {
   2301 		wi->wi_atareq.retsts = ATACMD_OK;
   2302 		if (wi->wi_atareq.flags & ATACMD_READREG) {
   2303 			wi->wi_atareq.command = xfer->c_ata_c.r_status;
   2304 			wi->wi_atareq.features = xfer->c_ata_c.r_error;
   2305 			wi->wi_atareq.sec_count = xfer->c_ata_c.r_count;
   2306 			wi->wi_atareq.sec_num = xfer->c_ata_c.r_lba & 0xff;
   2307 			wi->wi_atareq.head = (xfer->c_ata_c.r_device & 0xf0) |
   2308 			    ((xfer->c_ata_c.r_lba >> 24) & 0x0f);
   2309 			wi->wi_atareq.cylinder =
   2310 			    (xfer->c_ata_c.r_lba >> 8) & 0xffff;
   2311 			wi->wi_atareq.error = xfer->c_ata_c.r_error;
   2312 		}
   2313 	}
   2314 
   2315 out:
   2316 	ata_free_xfer(wi->wi_softc->drvp->chnl_softc, xfer);
   2317 	ata_channel_start(wi->wi_softc->drvp->chnl_softc,
   2318 	    wi->wi_softc->drvp->drive);
   2319 out2:
   2320 	bp->b_error = error;
   2321 	if (error)
   2322 		bp->b_resid = bp->b_bcount;
   2323 	biodone(bp);
   2324 }
   2325 
   2326 static void
   2327 wd_sysctl_attach(struct wd_softc *wd)
   2328 {
   2329 	const struct sysctlnode *node;
   2330 	int error;
   2331 
   2332 	/* sysctl set-up */
   2333 	if (sysctl_createv(&wd->nodelog, 0, NULL, &node,
   2334 				0, CTLTYPE_NODE, device_xname(wd->sc_dev),
   2335 				SYSCTL_DESCR("wd driver settings"),
   2336 				NULL, 0, NULL, 0,
   2337 				CTL_HW, CTL_CREATE, CTL_EOL) != 0) {
   2338 		aprint_error_dev(wd->sc_dev,
   2339 		    "could not create %s.%s sysctl node\n",
   2340 		    "hw", device_xname(wd->sc_dev));
   2341 		return;
   2342 	}
   2343 
   2344 	wd->drv_max_tags = ATA_MAX_OPENINGS;
   2345 	if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
   2346 				CTLFLAG_READWRITE, CTLTYPE_INT, "max_tags",
   2347 				SYSCTL_DESCR("max number of NCQ tags to use"),
   2348 				NULL, 0, &wd->drv_max_tags, 0,
   2349 				CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
   2350 				!= 0) {
   2351 		aprint_error_dev(wd->sc_dev,
   2352 		    "could not create %s.%s.max_tags sysctl - error %d\n",
   2353 		    "hw", device_xname(wd->sc_dev), error);
   2354 		return;
   2355 	}
   2356 
   2357 	wd->drv_ncq = true;
   2358 	if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
   2359 				CTLFLAG_READWRITE, CTLTYPE_BOOL, "use_ncq",
   2360 				SYSCTL_DESCR("use NCQ if supported"),
   2361 				NULL, 0, &wd->drv_ncq, 0,
   2362 				CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
   2363 				!= 0) {
   2364 		aprint_error_dev(wd->sc_dev,
   2365 		    "could not create %s.%s.use_ncq sysctl - error %d\n",
   2366 		    "hw", device_xname(wd->sc_dev), error);
   2367 		return;
   2368 	}
   2369 
   2370 	wd->drv_ncq_prio = false;
   2371 	if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
   2372 				CTLFLAG_READWRITE, CTLTYPE_BOOL, "use_ncq_prio",
   2373 				SYSCTL_DESCR("use NCQ PRIORITY if supported"),
   2374 				NULL, 0, &wd->drv_ncq_prio, 0,
   2375 				CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
   2376 				!= 0) {
   2377 		aprint_error_dev(wd->sc_dev,
   2378 		    "could not create %s.%s.use_ncq_prio sysctl - error %d\n",
   2379 		    "hw", device_xname(wd->sc_dev), error);
   2380 		return;
   2381 	}
   2382 
   2383 #ifdef WD_CHAOS_MONKEY
   2384 	wd->drv_chaos_freq = 0;
   2385 	if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
   2386 				CTLFLAG_READWRITE, CTLTYPE_INT, "chaos_freq",
   2387 				SYSCTL_DESCR("simulated bio read error rate"),
   2388 				NULL, 0, &wd->drv_chaos_freq, 0,
   2389 				CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
   2390 				!= 0) {
   2391 		aprint_error_dev(wd->sc_dev,
   2392 		    "could not create %s.%s.chaos_freq sysctl - error %d\n",
   2393 		    "hw", device_xname(wd->sc_dev), error);
   2394 		return;
   2395 	}
   2396 
   2397 	wd->drv_chaos_cnt = 0;
   2398 	if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
   2399 				CTLFLAG_READONLY, CTLTYPE_INT, "chaos_cnt",
   2400 				SYSCTL_DESCR("number of processed bio reads"),
   2401 				NULL, 0, &wd->drv_chaos_cnt, 0,
   2402 				CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
   2403 				!= 0) {
   2404 		aprint_error_dev(wd->sc_dev,
   2405 		    "could not create %s.%s.chaos_cnt sysctl - error %d\n",
   2406 		    "hw", device_xname(wd->sc_dev), error);
   2407 		return;
   2408 	}
   2409 #endif
   2410 
   2411 }
   2412 
   2413 static void
   2414 wd_sysctl_detach(struct wd_softc *wd)
   2415 {
   2416 	sysctl_teardown(&wd->nodelog);
   2417 }
   2418 
   2419