xref: /netbsd-src/sys/dev/ata/wd.c (revision 8a5e2a50be13e77dd4df5daf258ddceeeeb47ce6)
1 /*	$NetBSD: wd.c,v 1.305 2005/07/02 04:29:01 dsainty 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 1.305 2005/07/02 04:29:01 dsainty Exp $");
70 
71 #ifndef ATADEBUG
72 #define ATADEBUG
73 #endif /* ATADEBUG */
74 
75 #include "rnd.h"
76 
77 #include <sys/param.h>
78 #include <sys/systm.h>
79 #include <sys/kernel.h>
80 #include <sys/conf.h>
81 #include <sys/file.h>
82 #include <sys/stat.h>
83 #include <sys/ioctl.h>
84 #include <sys/buf.h>
85 #include <sys/bufq.h>
86 #include <sys/uio.h>
87 #include <sys/malloc.h>
88 #include <sys/device.h>
89 #include <sys/disklabel.h>
90 #include <sys/disk.h>
91 #include <sys/syslog.h>
92 #include <sys/proc.h>
93 #include <sys/vnode.h>
94 #if NRND > 0
95 #include <sys/rnd.h>
96 #endif
97 
98 #include <machine/intr.h>
99 #include <machine/bus.h>
100 
101 #include <dev/ata/atareg.h>
102 #include <dev/ata/atavar.h>
103 #include <dev/ata/wdvar.h>
104 #include <dev/ic/wdcreg.h>
105 #include <sys/ataio.h>
106 #include "locators.h"
107 
108 #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 /*
209  * Quirk table for IDE drives.  Put more-specific matches first, since
210  * a simple globbing routine is used for matching.
211  */
212 static const struct wd_quirk {
213 	const char *wdq_match;		/* inquiry pattern to match */
214 	int wdq_quirks;			/* drive quirks */
215 } wd_quirk_table[] = {
216 	/*
217 	 * Some Seagate S-ATA drives have a PHY which can get confused
218 	 * with the way data is packetized by some S-ATA controllers.
219 	 *
220 	 * The work-around is to split in two any write transfer whose
221 	 * sector count % 15 == 1 (assuming 512 byte sectors).
222 	 *
223 	 * XXX This is an incomplete list.  There are at least a couple
224 	 * XXX more model numbers.  If you have trouble with such transfers
225 	 * XXX (8K is the most common) on Seagate S-ATA drives, please
226 	 * XXX notify thorpej@NetBSD.org.
227 	 */
228 	{ "ST3120023AS",
229 	  WD_QUIRK_SPLIT_MOD15_WRITE },
230 	{ "ST380023AS",
231 	  WD_QUIRK_SPLIT_MOD15_WRITE },
232 
233 	/*
234 	 * This seagate drive seems to have issue addressing sector 0xfffffff
235 	 * (aka LBA48_THRESHOLD) in LBA mode. The workaround is to force
236 	 * LBA48
237 	 */
238 	{ "ST3160023A*",
239 	  WD_QUIRK_FORCE_LBA48 },
240 	{ "ST3160827A*",
241 	  WD_QUIRK_FORCE_LBA48 },
242 	{ "ST3200822A*",
243 	  WD_QUIRK_FORCE_LBA48 },
244 	{ "ST3250823A*",
245 	  WD_QUIRK_FORCE_LBA48 },
246 
247 	{ NULL,
248 	  0 }
249 };
250 
251 static const struct wd_quirk *
252 wd_lookup_quirks(const char *name)
253 {
254 	const struct wd_quirk *wdq;
255 	const char *estr;
256 
257 	for (wdq = wd_quirk_table; wdq->wdq_match != NULL; wdq++) {
258 		/*
259 		 * We only want exact matches (which include matches
260 		 * against globbing characters).
261 		 */
262 		if (pmatch(name, wdq->wdq_match, &estr) == 2)
263 			return (wdq);
264 	}
265 	return (NULL);
266 }
267 
268 int
269 wdprobe(struct device *parent, struct cfdata *match, void *aux)
270 {
271 	struct ata_device *adev = aux;
272 
273 	if (adev == NULL)
274 		return 0;
275 	if (adev->adev_bustype->bustype_type != SCSIPI_BUSTYPE_ATA)
276 		return 0;
277 
278 	if (match->cf_loc[ATA_HLCF_DRIVE] != ATA_HLCF_DRIVE_DEFAULT &&
279 	    match->cf_loc[ATA_HLCF_DRIVE] != adev->adev_drv_data->drive)
280 		return 0;
281 	return 1;
282 }
283 
284 void
285 wdattach(struct device *parent, struct device *self, void *aux)
286 {
287 	struct wd_softc *wd = (void *)self;
288 	struct ata_device *adev= aux;
289 	int i, blank;
290 	char tbuf[41], pbuf[9], c, *p, *q;
291 	const struct wd_quirk *wdq;
292 	ATADEBUG_PRINT(("wdattach\n"), DEBUG_FUNCS | DEBUG_PROBE);
293 
294 	callout_init(&wd->sc_restart_ch);
295 	bufq_alloc(&wd->sc_q, BUFQ_DISK_DEFAULT_STRAT()|BUFQ_SORT_RAWBLOCK);
296 #ifdef WD_SOFTBADSECT
297 	SLIST_INIT(&wd->sc_bslist);
298 #endif
299 	wd->atabus = adev->adev_bustype;
300 	wd->openings = adev->adev_openings;
301 	wd->drvp = adev->adev_drv_data;
302 
303 	wd->drvp->drv_done = wddone;
304 	wd->drvp->drv_softc = &wd->sc_dev;
305 
306 	aprint_naive("\n");
307 
308 	/* read our drive info */
309 	if (wd_get_params(wd, AT_WAIT, &wd->sc_params) != 0) {
310 		aprint_error("\n%s: IDENTIFY failed\n", wd->sc_dev.dv_xname);
311 		return;
312 	}
313 
314 	for (blank = 0, p = wd->sc_params.atap_model, q = tbuf, i = 0;
315 	    i < sizeof(wd->sc_params.atap_model); i++) {
316 		c = *p++;
317 		if (c == '\0')
318 			break;
319 		if (c != ' ') {
320 			if (blank) {
321 				*q++ = ' ';
322 				blank = 0;
323 			}
324 			*q++ = c;
325 		} else
326 			blank = 1;
327 	}
328 	*q++ = '\0';
329 
330 	aprint_normal(": <%s>\n", tbuf);
331 
332 	wdq = wd_lookup_quirks(tbuf);
333 	if (wdq != NULL)
334 		wd->sc_quirks = wdq->wdq_quirks;
335 
336 	if ((wd->sc_params.atap_multi & 0xff) > 1) {
337 		wd->sc_multi = wd->sc_params.atap_multi & 0xff;
338 	} else {
339 		wd->sc_multi = 1;
340 	}
341 
342 	aprint_normal("%s: drive supports %d-sector PIO transfers,",
343 	    wd->sc_dev.dv_xname, wd->sc_multi);
344 
345 	/* 48-bit LBA addressing */
346 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0)
347 		wd->sc_flags |= WDF_LBA48;
348 
349 	/* Prior to ATA-4, LBA was optional. */
350 	if ((wd->sc_params.atap_capabilities1 & WDC_CAP_LBA) != 0)
351 		wd->sc_flags |= WDF_LBA;
352 #if 0
353 	/* ATA-4 requires LBA. */
354 	if (wd->sc_params.atap_ataversion != 0xffff &&
355 	    wd->sc_params.atap_ataversion >= WDC_VER_ATA4)
356 		wd->sc_flags |= WDF_LBA;
357 #endif
358 
359 	if ((wd->sc_flags & WDF_LBA48) != 0) {
360 		aprint_normal(" LBA48 addressing\n");
361 		wd->sc_capacity =
362 		    ((u_int64_t) wd->sc_params.__reserved6[11] << 48) |
363 		    ((u_int64_t) wd->sc_params.__reserved6[10] << 32) |
364 		    ((u_int64_t) wd->sc_params.__reserved6[9]  << 16) |
365 		    ((u_int64_t) wd->sc_params.__reserved6[8]  << 0);
366 	} else if ((wd->sc_flags & WDF_LBA) != 0) {
367 		aprint_normal(" LBA addressing\n");
368 		wd->sc_capacity =
369 		    (wd->sc_params.atap_capacity[1] << 16) |
370 		    wd->sc_params.atap_capacity[0];
371 	} else {
372 		aprint_normal(" chs addressing\n");
373 		wd->sc_capacity =
374 		    wd->sc_params.atap_cylinders *
375 		    wd->sc_params.atap_heads *
376 		    wd->sc_params.atap_sectors;
377 	}
378 	format_bytes(pbuf, sizeof(pbuf), wd->sc_capacity * DEV_BSIZE);
379 	aprint_normal("%s: %s, %d cyl, %d head, %d sec, "
380 	    "%d bytes/sect x %llu sectors\n",
381 	    self->dv_xname, pbuf,
382 	    (wd->sc_flags & WDF_LBA) ? (int)(wd->sc_capacity /
383 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors)) :
384 		wd->sc_params.atap_cylinders,
385 	    wd->sc_params.atap_heads, wd->sc_params.atap_sectors,
386 	    DEV_BSIZE, (unsigned long long)wd->sc_capacity);
387 
388 	ATADEBUG_PRINT(("%s: atap_dmatiming_mimi=%d, atap_dmatiming_recom=%d\n",
389 	    self->dv_xname, wd->sc_params.atap_dmatiming_mimi,
390 	    wd->sc_params.atap_dmatiming_recom), DEBUG_PROBE);
391 	/*
392 	 * Initialize and attach the disk structure.
393 	 */
394 	wd->sc_dk.dk_driver = &wddkdriver;
395 	wd->sc_dk.dk_name = wd->sc_dev.dv_xname;
396 	disk_attach(&wd->sc_dk);
397 	wd->sc_wdc_bio.lp = wd->sc_dk.dk_label;
398 	wd->sc_sdhook = shutdownhook_establish(wd_shutdown, wd);
399 	if (wd->sc_sdhook == NULL)
400 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
401 		    wd->sc_dev.dv_xname);
402 #if NRND > 0
403 	rnd_attach_source(&wd->rnd_source, wd->sc_dev.dv_xname,
404 			  RND_TYPE_DISK, 0);
405 #endif
406 
407 	/* Discover wedges on this disk. */
408 	dkwedge_discover(&wd->sc_dk);
409 }
410 
411 int
412 wdactivate(struct device *self, enum devact act)
413 {
414 	int rv = 0;
415 
416 	switch (act) {
417 	case DVACT_ACTIVATE:
418 		rv = EOPNOTSUPP;
419 		break;
420 
421 	case DVACT_DEACTIVATE:
422 		/*
423 		 * Nothing to do; we key off the device's DVF_ACTIVATE.
424 		 */
425 		break;
426 	}
427 	return (rv);
428 }
429 
430 int
431 wddetach(struct device *self, int flags)
432 {
433 	struct wd_softc *sc = (struct wd_softc *)self;
434 	int s, bmaj, cmaj, i, mn;
435 
436 	/* locate the major number */
437 	bmaj = bdevsw_lookup_major(&wd_bdevsw);
438 	cmaj = cdevsw_lookup_major(&wd_cdevsw);
439 
440 	/* Nuke the vnodes for any open instances. */
441 	for (i = 0; i < MAXPARTITIONS; i++) {
442 		mn = WDMINOR(self->dv_unit, i);
443 		vdevgone(bmaj, mn, mn, VBLK);
444 		vdevgone(cmaj, mn, mn, VCHR);
445 	}
446 
447 	/* Delete all of our wedges. */
448 	dkwedge_delall(&sc->sc_dk);
449 
450 	s = splbio();
451 
452 	/* Kill off any queued buffers. */
453 	bufq_drain(&sc->sc_q);
454 
455 	bufq_free(&sc->sc_q);
456 	sc->atabus->ata_killpending(sc->drvp);
457 
458 	splx(s);
459 
460 	/* Detach disk. */
461 	disk_detach(&sc->sc_dk);
462 
463 #ifdef WD_SOFTBADSECT
464 	/* Clean out the bad sector list */
465 	while (!SLIST_EMPTY(&sc->sc_bslist)) {
466 		void *head = SLIST_FIRST(&sc->sc_bslist);
467 		SLIST_REMOVE_HEAD(&sc->sc_bslist, dbs_next);
468 		free(head, M_TEMP);
469 	}
470 	sc->sc_bscount = 0;
471 #endif
472 
473 	/* Get rid of the shutdown hook. */
474 	if (sc->sc_sdhook != NULL)
475 		shutdownhook_disestablish(sc->sc_sdhook);
476 
477 #if NRND > 0
478 	/* Unhook the entropy source. */
479 	rnd_detach_source(&sc->rnd_source);
480 #endif
481 
482 	sc->drvp->drive_flags = 0; /* no drive any more here */
483 
484 	return (0);
485 }
486 
487 /*
488  * Read/write routine for a buffer.  Validates the arguments and schedules the
489  * transfer.  Does not wait for the transfer to complete.
490  */
491 void
492 wdstrategy(struct buf *bp)
493 {
494 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(bp->b_dev));
495 	struct disklabel *lp = wd->sc_dk.dk_label;
496 	daddr_t blkno;
497 	int s;
498 
499 	ATADEBUG_PRINT(("wdstrategy (%s)\n", wd->sc_dev.dv_xname),
500 	    DEBUG_XFERS);
501 
502 	/* Valid request?  */
503 	if (bp->b_blkno < 0 ||
504 	    (bp->b_bcount % lp->d_secsize) != 0 ||
505 	    (bp->b_bcount / lp->d_secsize) >= (1 << NBBY)) {
506 		bp->b_error = EINVAL;
507 		goto bad;
508 	}
509 
510 	/* If device invalidated (e.g. media change, door open), error. */
511 	if ((wd->sc_flags & WDF_LOADED) == 0) {
512 		bp->b_error = EIO;
513 		goto bad;
514 	}
515 
516 	/* If it's a null transfer, return immediately. */
517 	if (bp->b_bcount == 0)
518 		goto done;
519 
520 	/*
521 	 * Do bounds checking, adjust transfer. if error, process.
522 	 * If end of partition, just return.
523 	 */
524 	if (WDPART(bp->b_dev) == RAW_PART) {
525 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
526 		    wd->sc_capacity) <= 0)
527 			goto done;
528 	} else {
529 		if (bounds_check_with_label(&wd->sc_dk, bp,
530 		    (wd->sc_flags & (WDF_WLABEL|WDF_LABELLING)) != 0) <= 0)
531 			goto done;
532 	}
533 
534 	/*
535 	 * Now convert the block number to absolute and put it in
536 	 * terms of the device's logical block size.
537 	 */
538 	if (lp->d_secsize >= DEV_BSIZE)
539 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
540 	else
541 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
542 
543 	if (WDPART(bp->b_dev) != RAW_PART)
544 		blkno += lp->d_partitions[WDPART(bp->b_dev)].p_offset;
545 
546 	bp->b_rawblkno = blkno;
547 
548 #ifdef WD_SOFTBADSECT
549 	/*
550 	 * If the transfer about to be attempted contains only a block that
551 	 * is known to be bad then return an error for the transfer without
552 	 * even attempting to start a transfer up under the premis that we
553 	 * will just end up doing more retries for a transfer that will end
554 	 * up failing again.
555 	 * XXX:SMP - mutex required to protect with DIOCBSFLUSH
556 	 */
557 	if (__predict_false(!SLIST_EMPTY(&wd->sc_bslist))) {
558 		struct disk_badsectors *dbs;
559 		daddr_t maxblk = blkno + (bp->b_bcount >> DEV_BSHIFT) - 1;
560 
561 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next)
562 			if ((dbs->dbs_min <= blkno && blkno <= dbs->dbs_max) ||
563 			    (dbs->dbs_min <= maxblk && maxblk <= dbs->dbs_max)){
564 				bp->b_error = EIO;
565 				goto bad;
566 			}
567 	}
568 #endif
569 
570 	/* Queue transfer on drive, activate drive and controller if idle. */
571 	s = splbio();
572 	BUFQ_PUT(&wd->sc_q, bp);
573 	wdstart(wd);
574 	splx(s);
575 	return;
576 bad:
577 	bp->b_flags |= B_ERROR;
578 done:
579 	/* Toss transfer; we're done early. */
580 	bp->b_resid = bp->b_bcount;
581 	biodone(bp);
582 }
583 
584 /*
585  * Queue a drive for I/O.
586  */
587 void
588 wdstart(void *arg)
589 {
590 	struct wd_softc *wd = arg;
591 	struct buf *bp = NULL;
592 
593 	ATADEBUG_PRINT(("wdstart %s\n", wd->sc_dev.dv_xname),
594 	    DEBUG_XFERS);
595 	while (wd->openings > 0) {
596 
597 		/* Is there a buf for us ? */
598 		if ((bp = BUFQ_GET(&wd->sc_q)) == NULL)
599 			return;
600 
601 		/*
602 		 * Make the command. First lock the device
603 		 */
604 		wd->openings--;
605 
606 		wd->retries = 0;
607 		__wdstart(wd, bp);
608 	}
609 }
610 
611 static void
612 wd_split_mod15_write(struct buf *bp)
613 {
614 	struct buf *obp = bp->b_private;
615 	struct wd_softc *sc = wd_cd.cd_devs[DISKUNIT(obp->b_dev)];
616 
617 	if (__predict_false(bp->b_flags & B_ERROR) != 0) {
618 		/*
619 		 * Propagate the error.  If this was the first half of
620 		 * the original transfer, make sure to account for that
621 		 * in the residual.
622 		 */
623 		if (bp->b_data == obp->b_data)
624 			bp->b_resid += bp->b_bcount;
625 		goto done;
626 	}
627 
628 	/*
629 	 * If this was the second half of the transfer, we're all done!
630 	 */
631 	if (bp->b_data != obp->b_data)
632 		goto done;
633 
634 	/*
635 	 * Advance the pointer to the second half and issue that command
636 	 * using the same opening.
637 	 */
638 	bp->b_flags = obp->b_flags | B_CALL;
639 	bp->b_data += bp->b_bcount;
640 	bp->b_blkno += (bp->b_bcount / 512);
641 	bp->b_rawblkno += (bp->b_bcount / 512);
642 	__wdstart(sc, bp);
643 	return;
644 
645  done:
646 	obp->b_flags |= (bp->b_flags & (B_EINTR|B_ERROR));
647 	obp->b_error = bp->b_error;
648 	obp->b_resid = bp->b_resid;
649 	pool_put(&bufpool, bp);
650 	biodone(obp);
651 	sc->openings++;
652 	/* wddone() will call wdstart() */
653 }
654 
655 void
656 __wdstart(struct wd_softc *wd, struct buf *bp)
657 {
658 
659 	/*
660 	 * Deal with the "split mod15 write" quirk.  We just divide the
661 	 * transfer in two, doing the first half and then then second half
662 	 * with the same command opening.
663 	 *
664 	 * Note we MUST do this here, because we can't let insertion
665 	 * into the bufq cause the transfers to be re-merged.
666 	 */
667 	if (__predict_false((wd->sc_quirks & WD_QUIRK_SPLIT_MOD15_WRITE) != 0 &&
668 			    (bp->b_flags & B_READ) == 0 &&
669 			    bp->b_bcount > 512 &&
670 			    ((bp->b_bcount / 512) % 15) == 1)) {
671 		struct buf *nbp;
672 
673 		/* already at splbio */
674 		nbp = pool_get(&bufpool, PR_NOWAIT);
675 		if (__predict_false(nbp == NULL)) {
676 			/* No memory -- fail the iop. */
677 			bp->b_error = ENOMEM;
678 			bp->b_flags |= B_ERROR;
679 			bp->b_resid = bp->b_bcount;
680 			biodone(bp);
681 			wd->openings++;
682 			return;
683 		}
684 
685 		BUF_INIT(nbp);
686 		nbp->b_error = 0;
687 		nbp->b_proc = bp->b_proc;
688 		nbp->b_vp = NULLVP;
689 		nbp->b_dev = bp->b_dev;
690 
691 		nbp->b_bcount = bp->b_bcount / 2;
692 		nbp->b_bufsize = bp->b_bcount / 2;
693 		nbp->b_data = bp->b_data;
694 
695 		nbp->b_blkno = bp->b_blkno;
696 		nbp->b_rawblkno = bp->b_rawblkno;
697 
698 		nbp->b_flags = bp->b_flags | B_CALL;
699 		nbp->b_iodone = wd_split_mod15_write;
700 
701 		/* Put ptr to orig buf in b_private and use new buf */
702 		nbp->b_private = bp;
703 
704 		BIO_COPYPRIO(nbp, bp);
705 
706 		bp = nbp;
707 	}
708 
709 	wd->sc_wdc_bio.blkno = bp->b_rawblkno;
710 	wd->sc_wdc_bio.blkdone =0;
711 	wd->sc_bp = bp;
712 	/*
713 	 * If we're retrying, retry in single-sector mode. This will give us
714 	 * the sector number of the problem, and will eventually allow the
715 	 * transfer to succeed.
716 	 */
717 	if (wd->retries >= WDIORETRIES_SINGLE)
718 		wd->sc_wdc_bio.flags = ATA_SINGLE;
719 	else
720 		wd->sc_wdc_bio.flags = 0;
721 	if (wd->sc_flags & WDF_LBA48 &&
722 	    (wd->sc_wdc_bio.blkno > LBA48_THRESHOLD ||
723 	    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) != 0))
724 		wd->sc_wdc_bio.flags |= ATA_LBA48;
725 	if (wd->sc_flags & WDF_LBA)
726 		wd->sc_wdc_bio.flags |= ATA_LBA;
727 	if (bp->b_flags & B_READ)
728 		wd->sc_wdc_bio.flags |= ATA_READ;
729 	wd->sc_wdc_bio.bcount = bp->b_bcount;
730 	wd->sc_wdc_bio.databuf = bp->b_data;
731 	/* Instrumentation. */
732 	disk_busy(&wd->sc_dk);
733 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
734 	case ATACMD_TRY_AGAIN:
735 		callout_reset(&wd->sc_restart_ch, hz, wdrestart, wd);
736 		break;
737 	case ATACMD_QUEUED:
738 	case ATACMD_COMPLETE:
739 		break;
740 	default:
741 		panic("__wdstart: bad return code from ata_bio()");
742 	}
743 }
744 
745 void
746 wddone(void *v)
747 {
748 	struct wd_softc *wd = v;
749 	struct buf *bp = wd->sc_bp;
750 	const char *errmsg;
751 	int do_perror = 0;
752 	ATADEBUG_PRINT(("wddone %s\n", wd->sc_dev.dv_xname),
753 	    DEBUG_XFERS);
754 
755 	if (bp == NULL)
756 		return;
757 	bp->b_resid = wd->sc_wdc_bio.bcount;
758 	switch (wd->sc_wdc_bio.error) {
759 	case ERR_DMA:
760 		errmsg = "DMA error";
761 		goto retry;
762 	case ERR_DF:
763 		errmsg = "device fault";
764 		goto retry;
765 	case TIMEOUT:
766 		errmsg = "device timeout";
767 		goto retry;
768 	case ERR_RESET:
769 		errmsg = "channel reset";
770 		goto retry2;
771 	case ERROR:
772 		/* Don't care about media change bits */
773 		if (wd->sc_wdc_bio.r_error != 0 &&
774 		    (wd->sc_wdc_bio.r_error & ~(WDCE_MC | WDCE_MCR)) == 0)
775 			goto noerror;
776 		errmsg = "error";
777 		do_perror = 1;
778 retry:		/* Just reset and retry. Can we do more ? */
779 		(*wd->atabus->ata_reset_drive)(wd->drvp, AT_RST_NOCMD);
780 retry2:
781 		diskerr(bp, "wd", errmsg, LOG_PRINTF,
782 		    wd->sc_wdc_bio.blkdone, wd->sc_dk.dk_label);
783 		if (wd->retries < WDIORETRIES)
784 			printf(", retrying\n");
785 		if (do_perror)
786 			wdperror(wd);
787 		if (wd->retries < WDIORETRIES) {
788 			wd->retries++;
789 			callout_reset(&wd->sc_restart_ch, RECOVERYTIME,
790 			    wdrestart, wd);
791 			return;
792 		}
793 		printf("\n");
794 
795 #ifdef WD_SOFTBADSECT
796 		/*
797 		 * Not all errors indicate a failed block but those that do,
798 		 * put the block on the bad-block list for the device.  Only
799 		 * do this for reads because the drive should do it for writes,
800 		 * itself, according to Manuel.
801 		 */
802 		if ((bp->b_flags & B_READ) &&
803 		    ((wd->drvp->ata_vers >= 4 && wd->sc_wdc_bio.r_error & 64) ||
804 	     	     (wd->drvp->ata_vers < 4 && wd->sc_wdc_bio.r_error & 192))) {
805 			struct disk_badsectors *dbs;
806 
807 			dbs = malloc(sizeof *dbs, M_TEMP, M_WAITOK);
808 			dbs->dbs_min = bp->b_rawblkno;
809 			dbs->dbs_max = dbs->dbs_min + (bp->b_bcount >> DEV_BSHIFT) - 1;
810 			microtime(&dbs->dbs_failedat);
811 			SLIST_INSERT_HEAD(&wd->sc_bslist, dbs, dbs_next);
812 			wd->sc_bscount++;
813 		}
814 #endif
815 		bp->b_flags |= B_ERROR;
816 		bp->b_error = EIO;
817 		break;
818 	case NOERROR:
819 noerror:	if ((wd->sc_wdc_bio.flags & ATA_CORR) || wd->retries > 0)
820 			printf("%s: soft error (corrected)\n",
821 			    wd->sc_dev.dv_xname);
822 		break;
823 	case ERR_NODEV:
824 		bp->b_flags |= B_ERROR;
825 		bp->b_error = EIO;
826 		break;
827 	}
828 	disk_unbusy(&wd->sc_dk, (bp->b_bcount - bp->b_resid),
829 	    (bp->b_flags & B_READ));
830 #if NRND > 0
831 	rnd_add_uint32(&wd->rnd_source, bp->b_blkno);
832 #endif
833 	/* XXX Yuck, but we don't want to increment openings in this case */
834 	if (__predict_false((bp->b_flags & B_CALL) != 0 &&
835 			    bp->b_iodone == wd_split_mod15_write))
836 		biodone(bp);
837 	else {
838 		biodone(bp);
839 		wd->openings++;
840 	}
841 	wdstart(wd);
842 }
843 
844 void
845 wdrestart(void *v)
846 {
847 	struct wd_softc *wd = v;
848 	struct buf *bp = wd->sc_bp;
849 	int s;
850 	ATADEBUG_PRINT(("wdrestart %s\n", wd->sc_dev.dv_xname),
851 	    DEBUG_XFERS);
852 
853 	s = splbio();
854 	__wdstart(v, bp);
855 	splx(s);
856 }
857 
858 int
859 wdread(dev_t dev, struct uio *uio, int flags)
860 {
861 
862 	ATADEBUG_PRINT(("wdread\n"), DEBUG_XFERS);
863 	return (physio(wdstrategy, NULL, dev, B_READ, minphys, uio));
864 }
865 
866 int
867 wdwrite(dev_t dev, struct uio *uio, int flags)
868 {
869 
870 	ATADEBUG_PRINT(("wdwrite\n"), DEBUG_XFERS);
871 	return (physio(wdstrategy, NULL, dev, B_WRITE, minphys, uio));
872 }
873 
874 int
875 wdopen(dev_t dev, int flag, int fmt, struct proc *p)
876 {
877 	struct wd_softc *wd;
878 	int part, error;
879 
880 	ATADEBUG_PRINT(("wdopen\n"), DEBUG_FUNCS);
881 	wd = device_lookup(&wd_cd, WDUNIT(dev));
882 	if (wd == NULL)
883 		return (ENXIO);
884 
885 	if ((wd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
886 		return (ENODEV);
887 
888 	part = WDPART(dev);
889 
890 	if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
891 		return (error);
892 
893 	/*
894 	 * If there are wedges, and this is not RAW_PART, then we
895 	 * need to fail.
896 	 */
897 	if (wd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
898 		error = EBUSY;
899 		goto bad1;
900 	}
901 
902 	/*
903 	 * If this is the first open of this device, add a reference
904 	 * to the adapter.
905 	 */
906 	if (wd->sc_dk.dk_openmask == 0 &&
907 	    (error = wd->atabus->ata_addref(wd->drvp)) != 0)
908 		goto bad1;
909 
910 	if (wd->sc_dk.dk_openmask != 0) {
911 		/*
912 		 * If any partition is open, but the disk has been invalidated,
913 		 * disallow further opens.
914 		 */
915 		if ((wd->sc_flags & WDF_LOADED) == 0) {
916 			error = EIO;
917 			goto bad2;
918 		}
919 	} else {
920 		if ((wd->sc_flags & WDF_LOADED) == 0) {
921 			wd->sc_flags |= WDF_LOADED;
922 
923 			/* Load the physical device parameters. */
924 			wd_get_params(wd, AT_WAIT, &wd->sc_params);
925 
926 			/* Load the partition info if not already loaded. */
927 			wdgetdisklabel(wd);
928 		}
929 	}
930 
931 	/* Check that the partition exists. */
932 	if (part != RAW_PART &&
933 	    (part >= wd->sc_dk.dk_label->d_npartitions ||
934 	     wd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
935 		error = ENXIO;
936 		goto bad2;
937 	}
938 
939 	/* Insure only one open at a time. */
940 	switch (fmt) {
941 	case S_IFCHR:
942 		wd->sc_dk.dk_copenmask |= (1 << part);
943 		break;
944 	case S_IFBLK:
945 		wd->sc_dk.dk_bopenmask |= (1 << part);
946 		break;
947 	}
948 	wd->sc_dk.dk_openmask =
949 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
950 
951 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
952 	return 0;
953 
954  bad2:
955 	if (wd->sc_dk.dk_openmask == 0)
956 		wd->atabus->ata_delref(wd->drvp);
957  bad1:
958 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
959 	return error;
960 }
961 
962 int
963 wdclose(dev_t dev, int flag, int fmt, struct proc *p)
964 {
965 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
966 	int part = WDPART(dev);
967 	int error;
968 
969 	ATADEBUG_PRINT(("wdclose\n"), DEBUG_FUNCS);
970 
971 	if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
972 		return error;
973 
974 	switch (fmt) {
975 	case S_IFCHR:
976 		wd->sc_dk.dk_copenmask &= ~(1 << part);
977 		break;
978 	case S_IFBLK:
979 		wd->sc_dk.dk_bopenmask &= ~(1 << part);
980 		break;
981 	}
982 	wd->sc_dk.dk_openmask =
983 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
984 
985 	if (wd->sc_dk.dk_openmask == 0) {
986 		wd_flushcache(wd, AT_WAIT);
987 
988 		if (! (wd->sc_flags & WDF_KLABEL))
989 			wd->sc_flags &= ~WDF_LOADED;
990 
991 		wd->atabus->ata_delref(wd->drvp);
992 	}
993 
994 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
995 	return 0;
996 }
997 
998 void
999 wdgetdefaultlabel(struct wd_softc *wd, struct disklabel *lp)
1000 {
1001 
1002 	ATADEBUG_PRINT(("wdgetdefaultlabel\n"), DEBUG_FUNCS);
1003 	memset(lp, 0, sizeof(struct disklabel));
1004 
1005 	lp->d_secsize = DEV_BSIZE;
1006 	lp->d_ntracks = wd->sc_params.atap_heads;
1007 	lp->d_nsectors = wd->sc_params.atap_sectors;
1008 	lp->d_ncylinders = (wd->sc_flags & WDF_LBA) ? wd->sc_capacity /
1009 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors) :
1010 		wd->sc_params.atap_cylinders;
1011 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1012 
1013 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
1014 		lp->d_type = DTYPE_ST506;
1015 	else
1016 		lp->d_type = DTYPE_ESDI;
1017 
1018 	strncpy(lp->d_typename, wd->sc_params.atap_model, 16);
1019 	strncpy(lp->d_packname, "fictitious", 16);
1020 	if (wd->sc_capacity > UINT32_MAX)
1021 		lp->d_secperunit = UINT32_MAX;
1022 	else
1023 		lp->d_secperunit = wd->sc_capacity;
1024 	lp->d_rpm = 3600;
1025 	lp->d_interleave = 1;
1026 	lp->d_flags = 0;
1027 
1028 	lp->d_partitions[RAW_PART].p_offset = 0;
1029 	lp->d_partitions[RAW_PART].p_size =
1030 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1031 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1032 	lp->d_npartitions = RAW_PART + 1;
1033 
1034 	lp->d_magic = DISKMAGIC;
1035 	lp->d_magic2 = DISKMAGIC;
1036 	lp->d_checksum = dkcksum(lp);
1037 }
1038 
1039 /*
1040  * Fabricate a default disk label, and try to read the correct one.
1041  */
1042 void
1043 wdgetdisklabel(struct wd_softc *wd)
1044 {
1045 	struct disklabel *lp = wd->sc_dk.dk_label;
1046 	const char *errstring;
1047 	int s;
1048 
1049 	ATADEBUG_PRINT(("wdgetdisklabel\n"), DEBUG_FUNCS);
1050 
1051 	memset(wd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1052 
1053 	wdgetdefaultlabel(wd, lp);
1054 
1055 	wd->sc_badsect[0] = -1;
1056 
1057 	if (wd->drvp->state > RESET) {
1058 		s = splbio();
1059 		wd->drvp->drive_flags |= DRIVE_RESET;
1060 		splx(s);
1061 	}
1062 	errstring = readdisklabel(MAKEWDDEV(0, wd->sc_dev.dv_unit, RAW_PART),
1063 	    wdstrategy, lp, wd->sc_dk.dk_cpulabel);
1064 	if (errstring) {
1065 		/*
1066 		 * This probably happened because the drive's default
1067 		 * geometry doesn't match the DOS geometry.  We
1068 		 * assume the DOS geometry is now in the label and try
1069 		 * again.  XXX This is a kluge.
1070 		 */
1071 		if (wd->drvp->state > RESET) {
1072 			s = splbio();
1073 			wd->drvp->drive_flags |= DRIVE_RESET;
1074 			splx(s);
1075 		}
1076 		errstring = readdisklabel(MAKEWDDEV(0, wd->sc_dev.dv_unit,
1077 		    RAW_PART), wdstrategy, lp, wd->sc_dk.dk_cpulabel);
1078 	}
1079 	if (errstring) {
1080 		printf("%s: %s\n", wd->sc_dev.dv_xname, errstring);
1081 		return;
1082 	}
1083 
1084 	if (wd->drvp->state > RESET) {
1085 		s = splbio();
1086 		wd->drvp->drive_flags |= DRIVE_RESET;
1087 		splx(s);
1088 	}
1089 #ifdef HAS_BAD144_HANDLING
1090 	if ((lp->d_flags & D_BADSECT) != 0)
1091 		bad144intern(wd);
1092 #endif
1093 }
1094 
1095 void
1096 wdperror(const struct wd_softc *wd)
1097 {
1098 	static const char *const errstr0_3[] = {"address mark not found",
1099 	    "track 0 not found", "aborted command", "media change requested",
1100 	    "id not found", "media changed", "uncorrectable data error",
1101 	    "bad block detected"};
1102 	static const char *const errstr4_5[] = {
1103 	    "obsolete (address mark not found)",
1104 	    "no media/write protected", "aborted command",
1105 	    "media change requested", "id not found", "media changed",
1106 	    "uncorrectable data error", "interface CRC error"};
1107 	const char *const *errstr;
1108 	int i;
1109 	const char *sep = "";
1110 
1111 	const char *devname = wd->sc_dev.dv_xname;
1112 	struct ata_drive_datas *drvp = wd->drvp;
1113 	int errno = wd->sc_wdc_bio.r_error;
1114 
1115 	if (drvp->ata_vers >= 4)
1116 		errstr = errstr4_5;
1117 	else
1118 		errstr = errstr0_3;
1119 
1120 	printf("%s: (", devname);
1121 
1122 	if (errno == 0)
1123 		printf("error not notified");
1124 
1125 	for (i = 0; i < 8; i++) {
1126 		if (errno & (1 << i)) {
1127 			printf("%s%s", sep, errstr[i]);
1128 			sep = ", ";
1129 		}
1130 	}
1131 	printf(")\n");
1132 }
1133 
1134 int
1135 wdioctl(dev_t dev, u_long xfer, caddr_t addr, int flag, struct proc *p)
1136 {
1137 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
1138 	int error = 0, s;
1139 #ifdef __HAVE_OLD_DISKLABEL
1140 	struct disklabel *newlabel = NULL;
1141 #endif
1142 
1143 	ATADEBUG_PRINT(("wdioctl\n"), DEBUG_FUNCS);
1144 
1145 	if ((wd->sc_flags & WDF_LOADED) == 0)
1146 		return EIO;
1147 
1148 	switch (xfer) {
1149 #ifdef HAS_BAD144_HANDLING
1150 	case DIOCSBAD:
1151 		if ((flag & FWRITE) == 0)
1152 			return EBADF;
1153 		wd->sc_dk.dk_cpulabel->bad = *(struct dkbad *)addr;
1154 		wd->sc_dk.dk_label->d_flags |= D_BADSECT;
1155 		bad144intern(wd);
1156 		return 0;
1157 #endif
1158 #ifdef WD_SOFTBADSECT
1159 	case DIOCBSLIST :
1160 	{
1161 		u_int32_t count, missing, skip;
1162 		struct disk_badsecinfo dbsi;
1163 		struct disk_badsectors *dbs;
1164 		size_t available;
1165 		caddr_t laddr;
1166 
1167 		dbsi = *(struct disk_badsecinfo *)addr;
1168 		missing = wd->sc_bscount;
1169 		count = 0;
1170 		available = dbsi.dbsi_bufsize;
1171 		skip = dbsi.dbsi_skip;
1172 		laddr = dbsi.dbsi_buffer;
1173 
1174 		/*
1175 		 * We start this loop with the expectation that all of the
1176 		 * entries will be missed and decrement this counter each
1177 		 * time we either skip over one (already copied out) or
1178 		 * we actually copy it back to user space.  The structs
1179 		 * holding the bad sector information are copied directly
1180 		 * back to user space whilst the summary is returned via
1181 		 * the struct passed in via the ioctl.
1182 		 */
1183 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next) {
1184 			if (skip > 0) {
1185 				missing--;
1186 				skip--;
1187 				continue;
1188 			}
1189 			if (available < sizeof(*dbs))
1190 				break;
1191 			available -= sizeof(*dbs);
1192 			copyout(dbs, laddr, sizeof(*dbs));
1193 			laddr += sizeof(*dbs);
1194 			missing--;
1195 			count++;
1196 		}
1197 		dbsi.dbsi_left = missing;
1198 		dbsi.dbsi_copied = count;
1199 		*(struct disk_badsecinfo *)addr = dbsi;
1200 		return 0;
1201 	}
1202 
1203 	case DIOCBSFLUSH :
1204 		/* Clean out the bad sector list */
1205 		while (!SLIST_EMPTY(&wd->sc_bslist)) {
1206 			void *head = SLIST_FIRST(&wd->sc_bslist);
1207 			SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
1208 			free(head, M_TEMP);
1209 		}
1210 		wd->sc_bscount = 0;
1211 		return 0;
1212 #endif
1213 	case DIOCGDINFO:
1214 		*(struct disklabel *)addr = *(wd->sc_dk.dk_label);
1215 		return 0;
1216 #ifdef __HAVE_OLD_DISKLABEL
1217 	case ODIOCGDINFO:
1218 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1219 		if (newlabel == NULL)
1220 			return EIO;
1221 		*newlabel = *(wd->sc_dk.dk_label);
1222 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1223 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1224 		else
1225 			error = ENOTTY;
1226 		free(newlabel, M_TEMP);
1227 		return error;
1228 #endif
1229 
1230 	case DIOCGPART:
1231 		((struct partinfo *)addr)->disklab = wd->sc_dk.dk_label;
1232 		((struct partinfo *)addr)->part =
1233 		    &wd->sc_dk.dk_label->d_partitions[WDPART(dev)];
1234 		return 0;
1235 
1236 	case DIOCWDINFO:
1237 	case DIOCSDINFO:
1238 #ifdef __HAVE_OLD_DISKLABEL
1239 	case ODIOCWDINFO:
1240 	case ODIOCSDINFO:
1241 #endif
1242 	{
1243 		struct disklabel *lp;
1244 
1245 		if ((flag & FWRITE) == 0)
1246 			return EBADF;
1247 
1248 #ifdef __HAVE_OLD_DISKLABEL
1249 		if (xfer == ODIOCSDINFO || xfer == ODIOCWDINFO) {
1250 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1251 			if (newlabel == NULL)
1252 				return EIO;
1253 			memset(newlabel, 0, sizeof newlabel);
1254 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1255 			lp = newlabel;
1256 		} else
1257 #endif
1258 		lp = (struct disklabel *)addr;
1259 
1260 		if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE,
1261 				     NULL)) != 0)
1262 			goto bad;
1263 		wd->sc_flags |= WDF_LABELLING;
1264 
1265 		error = setdisklabel(wd->sc_dk.dk_label,
1266 		    lp, /*wd->sc_dk.dk_openmask : */0,
1267 		    wd->sc_dk.dk_cpulabel);
1268 		if (error == 0) {
1269 			if (wd->drvp->state > RESET) {
1270 				s = splbio();
1271 				wd->drvp->drive_flags |= DRIVE_RESET;
1272 				splx(s);
1273 			}
1274 			if (xfer == DIOCWDINFO
1275 #ifdef __HAVE_OLD_DISKLABEL
1276 			    || xfer == ODIOCWDINFO
1277 #endif
1278 			    )
1279 				error = writedisklabel(WDLABELDEV(dev),
1280 				    wdstrategy, wd->sc_dk.dk_label,
1281 				    wd->sc_dk.dk_cpulabel);
1282 		}
1283 
1284 		wd->sc_flags &= ~WDF_LABELLING;
1285 		(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
1286 bad:
1287 #ifdef __HAVE_OLD_DISKLABEL
1288 		if (newlabel != NULL)
1289 			free(newlabel, M_TEMP);
1290 #endif
1291 		return error;
1292 	}
1293 
1294 	case DIOCKLABEL:
1295 		if (*(int *)addr)
1296 			wd->sc_flags |= WDF_KLABEL;
1297 		else
1298 			wd->sc_flags &= ~WDF_KLABEL;
1299 		return 0;
1300 
1301 	case DIOCWLABEL:
1302 		if ((flag & FWRITE) == 0)
1303 			return EBADF;
1304 		if (*(int *)addr)
1305 			wd->sc_flags |= WDF_WLABEL;
1306 		else
1307 			wd->sc_flags &= ~WDF_WLABEL;
1308 		return 0;
1309 
1310 	case DIOCGDEFLABEL:
1311 		wdgetdefaultlabel(wd, (struct disklabel *)addr);
1312 		return 0;
1313 #ifdef __HAVE_OLD_DISKLABEL
1314 	case ODIOCGDEFLABEL:
1315 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1316 		if (newlabel == NULL)
1317 			return EIO;
1318 		wdgetdefaultlabel(wd, newlabel);
1319 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1320 			memcpy(addr, &newlabel, sizeof (struct olddisklabel));
1321 		else
1322 			error = ENOTTY;
1323 		free(newlabel, M_TEMP);
1324 		return error;
1325 #endif
1326 
1327 #ifdef notyet
1328 	case DIOCWFORMAT:
1329 		if ((flag & FWRITE) == 0)
1330 			return EBADF;
1331 		{
1332 		register struct format_op *fop;
1333 		struct iovec aiov;
1334 		struct uio auio;
1335 
1336 		fop = (struct format_op *)addr;
1337 		aiov.iov_base = fop->df_buf;
1338 		aiov.iov_len = fop->df_count;
1339 		auio.uio_iov = &aiov;
1340 		auio.uio_iovcnt = 1;
1341 		auio.uio_resid = fop->df_count;
1342 		auio.uio_segflg = 0;
1343 		auio.uio_offset =
1344 			fop->df_startblk * wd->sc_dk.dk_label->d_secsize;
1345 		auio.uio_procp = p;
1346 		error = physio(wdformat, NULL, dev, B_WRITE, minphys,
1347 		    &auio);
1348 		fop->df_count -= auio.uio_resid;
1349 		fop->df_reg[0] = wdc->sc_status;
1350 		fop->df_reg[1] = wdc->sc_error;
1351 		return error;
1352 		}
1353 #endif
1354 	case DIOCGCACHE:
1355 		return wd_getcache(wd, (int *)addr);
1356 
1357 	case DIOCSCACHE:
1358 		return wd_setcache(wd, *(int *)addr);
1359 
1360 	case DIOCCACHESYNC:
1361 		return wd_flushcache(wd, AT_WAIT);
1362 
1363 	case ATAIOCCOMMAND:
1364 		/*
1365 		 * Make sure this command is (relatively) safe first
1366 		 */
1367 		if ((((atareq_t *) addr)->flags & ATACMD_READ) == 0 &&
1368 		    (flag & FWRITE) == 0)
1369 			return (EBADF);
1370 		{
1371 		struct wd_ioctl *wi;
1372 		atareq_t *atareq = (atareq_t *) addr;
1373 		int error1;
1374 
1375 		wi = wi_get();
1376 		wi->wi_softc = wd;
1377 		wi->wi_atareq = *atareq;
1378 
1379 		if (atareq->datalen && atareq->flags &
1380 		    (ATACMD_READ | ATACMD_WRITE)) {
1381 			wi->wi_iov.iov_base = atareq->databuf;
1382 			wi->wi_iov.iov_len = atareq->datalen;
1383 			wi->wi_uio.uio_iov = &wi->wi_iov;
1384 			wi->wi_uio.uio_iovcnt = 1;
1385 			wi->wi_uio.uio_resid = atareq->datalen;
1386 			wi->wi_uio.uio_offset = 0;
1387 			wi->wi_uio.uio_segflg = UIO_USERSPACE;
1388 			wi->wi_uio.uio_rw =
1389 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE;
1390 			wi->wi_uio.uio_procp = p;
1391 			error1 = physio(wdioctlstrategy, &wi->wi_bp, dev,
1392 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE,
1393 			    minphys, &wi->wi_uio);
1394 		} else {
1395 			/* No need to call physio if we don't have any
1396 			   user data */
1397 			wi->wi_bp.b_flags = 0;
1398 			wi->wi_bp.b_data = 0;
1399 			wi->wi_bp.b_bcount = 0;
1400 			wi->wi_bp.b_dev = 0;
1401 			wi->wi_bp.b_proc = p;
1402 			wdioctlstrategy(&wi->wi_bp);
1403 			error1 = wi->wi_bp.b_error;
1404 		}
1405 		*atareq = wi->wi_atareq;
1406 		wi_free(wi);
1407 		return(error1);
1408 		}
1409 
1410 	case DIOCAWEDGE:
1411 	    {
1412 	    	struct dkwedge_info *dkw = (void *) addr;
1413 
1414 		if ((flag & FWRITE) == 0)
1415 			return (EBADF);
1416 
1417 		/* If the ioctl happens here, the parent is us. */
1418 		strcpy(dkw->dkw_parent, wd->sc_dev.dv_xname);
1419 		return (dkwedge_add(dkw));
1420 	    }
1421 
1422 	case DIOCDWEDGE:
1423 	    {
1424 	    	struct dkwedge_info *dkw = (void *) addr;
1425 
1426 		if ((flag & FWRITE) == 0)
1427 			return (EBADF);
1428 
1429 		/* If the ioctl happens here, the parent is us. */
1430 		strcpy(dkw->dkw_parent, wd->sc_dev.dv_xname);
1431 		return (dkwedge_del(dkw));
1432 	    }
1433 
1434 	case DIOCLWEDGES:
1435 	    {
1436 	    	struct dkwedge_list *dkwl = (void *) addr;
1437 
1438 		return (dkwedge_list(&wd->sc_dk, dkwl, p));
1439 	    }
1440 
1441 	default:
1442 		return ENOTTY;
1443 	}
1444 
1445 #ifdef DIAGNOSTIC
1446 	panic("wdioctl: impossible");
1447 #endif
1448 }
1449 
1450 #ifdef B_FORMAT
1451 int
1452 wdformat(struct buf *bp)
1453 {
1454 
1455 	bp->b_flags |= B_FORMAT;
1456 	return wdstrategy(bp);
1457 }
1458 #endif
1459 
1460 int
1461 wdsize(dev_t dev)
1462 {
1463 	struct wd_softc *wd;
1464 	int part, omask;
1465 	int size;
1466 
1467 	ATADEBUG_PRINT(("wdsize\n"), DEBUG_FUNCS);
1468 
1469 	wd = device_lookup(&wd_cd, WDUNIT(dev));
1470 	if (wd == NULL)
1471 		return (-1);
1472 
1473 	part = WDPART(dev);
1474 	omask = wd->sc_dk.dk_openmask & (1 << part);
1475 
1476 	if (omask == 0 && wdopen(dev, 0, S_IFBLK, NULL) != 0)
1477 		return (-1);
1478 	if (wd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1479 		size = -1;
1480 	else
1481 		size = wd->sc_dk.dk_label->d_partitions[part].p_size *
1482 		    (wd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1483 	if (omask == 0 && wdclose(dev, 0, S_IFBLK, NULL) != 0)
1484 		return (-1);
1485 	return (size);
1486 }
1487 
1488 /* #define WD_DUMP_NOT_TRUSTED if you just want to watch */
1489 static int wddoingadump = 0;
1490 static int wddumprecalibrated = 0;
1491 static int wddumpmulti = 1;
1492 
1493 /*
1494  * Dump core after a system crash.
1495  */
1496 int
1497 wddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1498 {
1499 	struct wd_softc *wd;	/* disk unit to do the I/O */
1500 	struct disklabel *lp;   /* disk's disklabel */
1501 	int part, err;
1502 	int nblks;	/* total number of sectors left to write */
1503 
1504 	/* Check if recursive dump; if so, punt. */
1505 	if (wddoingadump)
1506 		return EFAULT;
1507 	wddoingadump = 1;
1508 
1509 	wd = device_lookup(&wd_cd, WDUNIT(dev));
1510 	if (wd == NULL)
1511 		return (ENXIO);
1512 
1513 	part = WDPART(dev);
1514 
1515 	/* Convert to disk sectors.  Request must be a multiple of size. */
1516 	lp = wd->sc_dk.dk_label;
1517 	if ((size % lp->d_secsize) != 0)
1518 		return EFAULT;
1519 	nblks = size / lp->d_secsize;
1520 	blkno = blkno / (lp->d_secsize / DEV_BSIZE);
1521 
1522 	/* Check transfer bounds against partition size. */
1523 	if ((blkno < 0) || ((blkno + nblks) > lp->d_partitions[part].p_size))
1524 		return EINVAL;
1525 
1526 	/* Offset block number to start of partition. */
1527 	blkno += lp->d_partitions[part].p_offset;
1528 
1529 	/* Recalibrate, if first dump transfer. */
1530 	if (wddumprecalibrated == 0) {
1531 		wddumpmulti = wd->sc_multi;
1532 		wddumprecalibrated = 1;
1533 		(*wd->atabus->ata_reset_drive)(wd->drvp,
1534 					       AT_POLL | AT_RST_EMERG);
1535 		wd->drvp->state = RESET;
1536 	}
1537 
1538 	while (nblks > 0) {
1539 		wd->sc_bp = NULL;
1540 		wd->sc_wdc_bio.blkno = blkno;
1541 		wd->sc_wdc_bio.flags = ATA_POLL;
1542 		if (wd->sc_flags & WDF_LBA48 &&
1543 		    (blkno > LBA48_THRESHOLD ||
1544 	    	    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) != 0))
1545 			wd->sc_wdc_bio.flags |= ATA_LBA48;
1546 		if (wd->sc_flags & WDF_LBA)
1547 			wd->sc_wdc_bio.flags |= ATA_LBA;
1548 		wd->sc_wdc_bio.bcount =
1549 			min(nblks, wddumpmulti) * lp->d_secsize;
1550 		wd->sc_wdc_bio.databuf = va;
1551 #ifndef WD_DUMP_NOT_TRUSTED
1552 		switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
1553 		case ATACMD_TRY_AGAIN:
1554 			panic("wddump: try again");
1555 			break;
1556 		case ATACMD_QUEUED:
1557 			panic("wddump: polled command has been queued");
1558 			break;
1559 		case ATACMD_COMPLETE:
1560 			break;
1561 		}
1562 		switch(wd->sc_wdc_bio.error) {
1563 		case TIMEOUT:
1564 			printf("wddump: device timed out");
1565 			err = EIO;
1566 			break;
1567 		case ERR_DF:
1568 			printf("wddump: drive fault");
1569 			err = EIO;
1570 			break;
1571 		case ERR_DMA:
1572 			printf("wddump: DMA error");
1573 			err = EIO;
1574 			break;
1575 		case ERROR:
1576 			printf("wddump: ");
1577 			wdperror(wd);
1578 			err = EIO;
1579 			break;
1580 		case NOERROR:
1581 			err = 0;
1582 			break;
1583 		default:
1584 			panic("wddump: unknown error type");
1585 		}
1586 		if (err != 0) {
1587 			printf("\n");
1588 			return err;
1589 		}
1590 #else	/* WD_DUMP_NOT_TRUSTED */
1591 		/* Let's just talk about this first... */
1592 		printf("wd%d: dump addr 0x%x, cylin %d, head %d, sector %d\n",
1593 		    unit, va, cylin, head, sector);
1594 		delay(500 * 1000);	/* half a second */
1595 #endif
1596 
1597 		/* update block count */
1598 		nblks -= min(nblks, wddumpmulti);
1599 		blkno += min(nblks, wddumpmulti);
1600 		va += min(nblks, wddumpmulti) * lp->d_secsize;
1601 	}
1602 
1603 	wddoingadump = 0;
1604 	return 0;
1605 }
1606 
1607 #ifdef HAS_BAD144_HANDLING
1608 /*
1609  * Internalize the bad sector table.
1610  */
1611 void
1612 bad144intern(struct wd_softc *wd)
1613 {
1614 	struct dkbad *bt = &wd->sc_dk.dk_cpulabel->bad;
1615 	struct disklabel *lp = wd->sc_dk.dk_label;
1616 	int i = 0;
1617 
1618 	ATADEBUG_PRINT(("bad144intern\n"), DEBUG_XFERS);
1619 
1620 	for (; i < NBT_BAD; i++) {
1621 		if (bt->bt_bad[i].bt_cyl == 0xffff)
1622 			break;
1623 		wd->sc_badsect[i] =
1624 		    bt->bt_bad[i].bt_cyl * lp->d_secpercyl +
1625 		    (bt->bt_bad[i].bt_trksec >> 8) * lp->d_nsectors +
1626 		    (bt->bt_bad[i].bt_trksec & 0xff);
1627 	}
1628 	for (; i < NBT_BAD+1; i++)
1629 		wd->sc_badsect[i] = -1;
1630 }
1631 #endif
1632 
1633 int
1634 wd_get_params(struct wd_softc *wd, u_int8_t flags, struct ataparams *params)
1635 {
1636 	switch (wd->atabus->ata_get_params(wd->drvp, flags, params)) {
1637 	case CMD_AGAIN:
1638 		return 1;
1639 	case CMD_ERR:
1640 		/*
1641 		 * We `know' there's a drive here; just assume it's old.
1642 		 * This geometry is only used to read the MBR and print a
1643 		 * (false) attach message.
1644 		 */
1645 		strncpy(params->atap_model, "ST506",
1646 		    sizeof params->atap_model);
1647 		params->atap_config = ATA_CFG_FIXED;
1648 		params->atap_cylinders = 1024;
1649 		params->atap_heads = 8;
1650 		params->atap_sectors = 17;
1651 		params->atap_multi = 1;
1652 		params->atap_capabilities1 = params->atap_capabilities2 = 0;
1653 		wd->drvp->ata_vers = -1; /* Mark it as pre-ATA */
1654 		return 0;
1655 	case CMD_OK:
1656 		return 0;
1657 	default:
1658 		panic("wd_get_params: bad return code from ata_get_params");
1659 		/* NOTREACHED */
1660 	}
1661 }
1662 
1663 int
1664 wd_getcache(struct wd_softc *wd, int *bitsp)
1665 {
1666 	struct ataparams params;
1667 
1668 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
1669 		return EIO;
1670 	if (params.atap_cmd_set1 == 0x0000 ||
1671 	    params.atap_cmd_set1 == 0xffff ||
1672 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0) {
1673 		*bitsp = 0;
1674 		return 0;
1675 	}
1676 	*bitsp = DKCACHE_WCHANGE | DKCACHE_READ;
1677 	if (params.atap_cmd1_en & WDC_CMD1_CACHE)
1678 		*bitsp |= DKCACHE_WRITE;
1679 
1680 	return 0;
1681 }
1682 
1683 const char at_errbits[] = "\20\10ERROR\11TIMEOU\12DF";
1684 
1685 int
1686 wd_setcache(struct wd_softc *wd, int bits)
1687 {
1688 	struct ataparams params;
1689 	struct ata_command ata_c;
1690 
1691 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
1692 		return EIO;
1693 
1694 	if (params.atap_cmd_set1 == 0x0000 ||
1695 	    params.atap_cmd_set1 == 0xffff ||
1696 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0)
1697 		return EOPNOTSUPP;
1698 
1699 	if ((bits & DKCACHE_READ) == 0 ||
1700 	    (bits & DKCACHE_SAVE) != 0)
1701 		return EOPNOTSUPP;
1702 
1703 	memset(&ata_c, 0, sizeof(struct ata_command));
1704 	ata_c.r_command = SET_FEATURES;
1705 	ata_c.r_st_bmask = 0;
1706 	ata_c.r_st_pmask = 0;
1707 	ata_c.timeout = 30000; /* 30s timeout */
1708 	ata_c.flags = AT_WAIT;
1709 	if (bits & DKCACHE_WRITE)
1710 		ata_c.r_features = WDSF_WRITE_CACHE_EN;
1711 	else
1712 		ata_c.r_features = WDSF_WRITE_CACHE_DS;
1713 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1714 		printf("%s: wd_setcache command not complete\n",
1715 		    wd->sc_dev.dv_xname);
1716 		return EIO;
1717 	}
1718 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1719 		char sbuf[sizeof(at_errbits) + 64];
1720 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
1721 		printf("%s: wd_setcache: status=%s\n", wd->sc_dev.dv_xname,
1722 		    sbuf);
1723 		return EIO;
1724 	}
1725 	return 0;
1726 }
1727 
1728 int
1729 wd_standby(struct wd_softc *wd, int flags)
1730 {
1731 	struct ata_command ata_c;
1732 
1733 	memset(&ata_c, 0, sizeof(struct ata_command));
1734 	ata_c.r_command = WDCC_STANDBY_IMMED;
1735 	ata_c.r_st_bmask = WDCS_DRDY;
1736 	ata_c.r_st_pmask = WDCS_DRDY;
1737 	ata_c.flags = flags;
1738 	ata_c.timeout = 30000; /* 30s timeout */
1739 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1740 		printf("%s: standby immediate command didn't complete\n",
1741 		    wd->sc_dev.dv_xname);
1742 		return EIO;
1743 	}
1744 	if (ata_c.flags & AT_ERROR) {
1745 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
1746 			return ENODEV;
1747 	}
1748 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1749 		char sbuf[sizeof(at_errbits) + 64];
1750 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
1751 		printf("%s: wd_standby: status=%s\n", wd->sc_dev.dv_xname,
1752 		    sbuf);
1753 		return EIO;
1754 	}
1755 	return 0;
1756 }
1757 
1758 int
1759 wd_flushcache(struct wd_softc *wd, int flags)
1760 {
1761 	struct ata_command ata_c;
1762 
1763 	/*
1764 	 * WDCC_FLUSHCACHE is here since ATA-4, but some drives report
1765 	 * only ATA-2 and still support it.
1766 	 */
1767 	if (wd->drvp->ata_vers < 4 &&
1768 	    ((wd->sc_params.atap_cmd_set2 & WDC_CMD2_FC) == 0 ||
1769 	    wd->sc_params.atap_cmd_set2 == 0xffff))
1770 		return ENODEV;
1771 	memset(&ata_c, 0, sizeof(struct ata_command));
1772 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0 &&
1773 	    (wd->sc_params.atap_cmd2_en & ATA_CMD2_FCE) != 0)
1774 		ata_c.r_command = WDCC_FLUSHCACHE_EXT;
1775 	else
1776 		ata_c.r_command = WDCC_FLUSHCACHE;
1777 	ata_c.r_st_bmask = WDCS_DRDY;
1778 	ata_c.r_st_pmask = WDCS_DRDY;
1779 	ata_c.flags = flags;
1780 	ata_c.timeout = 30000; /* 30s timeout */
1781 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1782 		printf("%s: flush cache command didn't complete\n",
1783 		    wd->sc_dev.dv_xname);
1784 		return EIO;
1785 	}
1786 	if (ata_c.flags & AT_ERROR) {
1787 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
1788 			return ENODEV;
1789 	}
1790 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1791 		char sbuf[sizeof(at_errbits) + 64];
1792 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
1793 		printf("%s: wd_flushcache: status=%s\n", wd->sc_dev.dv_xname,
1794 		    sbuf);
1795 		return EIO;
1796 	}
1797 	return 0;
1798 }
1799 
1800 void
1801 wd_shutdown(void *arg)
1802 {
1803 	struct wd_softc *wd = arg;
1804 	wd_flushcache(wd, AT_POLL);
1805 }
1806 
1807 /*
1808  * Allocate space for a ioctl queue structure.  Mostly taken from
1809  * scsipi_ioctl.c
1810  */
1811 struct wd_ioctl *
1812 wi_get(void)
1813 {
1814 	struct wd_ioctl *wi;
1815 	int s;
1816 
1817 	wi = malloc(sizeof(struct wd_ioctl), M_TEMP, M_WAITOK|M_ZERO);
1818 	simple_lock_init(&wi->wi_bp.b_interlock);
1819 	s = splbio();
1820 	LIST_INSERT_HEAD(&wi_head, wi, wi_list);
1821 	splx(s);
1822 	return (wi);
1823 }
1824 
1825 /*
1826  * Free an ioctl structure and remove it from our list
1827  */
1828 
1829 void
1830 wi_free(struct wd_ioctl *wi)
1831 {
1832 	int s;
1833 
1834 	s = splbio();
1835 	LIST_REMOVE(wi, wi_list);
1836 	splx(s);
1837 	free(wi, M_TEMP);
1838 }
1839 
1840 /*
1841  * Find a wd_ioctl structure based on the struct buf.
1842  */
1843 
1844 struct wd_ioctl *
1845 wi_find(struct buf *bp)
1846 {
1847 	struct wd_ioctl *wi;
1848 	int s;
1849 
1850 	s = splbio();
1851 	for (wi = wi_head.lh_first; wi != 0; wi = wi->wi_list.le_next)
1852 		if (bp == &wi->wi_bp)
1853 			break;
1854 	splx(s);
1855 	return (wi);
1856 }
1857 
1858 /*
1859  * Ioctl pseudo strategy routine
1860  *
1861  * This is mostly stolen from scsipi_ioctl.c:scsistrategy().  What
1862  * happens here is:
1863  *
1864  * - wdioctl() queues a wd_ioctl structure.
1865  *
1866  * - wdioctl() calls physio/wdioctlstrategy based on whether or not
1867  *   user space I/O is required.  If physio() is called, physio() eventually
1868  *   calls wdioctlstrategy().
1869  *
1870  * - In either case, wdioctlstrategy() calls wd->atabus->ata_exec_command()
1871  *   to perform the actual command
1872  *
1873  * The reason for the use of the pseudo strategy routine is because
1874  * when doing I/O to/from user space, physio _really_ wants to be in
1875  * the loop.  We could put the entire buffer into the ioctl request
1876  * structure, but that won't scale if we want to do things like download
1877  * microcode.
1878  */
1879 
1880 void
1881 wdioctlstrategy(struct buf *bp)
1882 {
1883 	struct wd_ioctl *wi;
1884 	struct ata_command ata_c;
1885 	int error = 0;
1886 
1887 	wi = wi_find(bp);
1888 	if (wi == NULL) {
1889 		printf("user_strat: No ioctl\n");
1890 		error = EINVAL;
1891 		goto bad;
1892 	}
1893 
1894 	memset(&ata_c, 0, sizeof(ata_c));
1895 
1896 	/*
1897 	 * Abort if physio broke up the transfer
1898 	 */
1899 
1900 	if (bp->b_bcount != wi->wi_atareq.datalen) {
1901 		printf("physio split wd ioctl request... cannot proceed\n");
1902 		error = EIO;
1903 		goto bad;
1904 	}
1905 
1906 	/*
1907 	 * Abort if we didn't get a buffer size that was a multiple of
1908 	 * our sector size (or was larger than NBBY)
1909 	 */
1910 
1911 	if ((bp->b_bcount % wi->wi_softc->sc_dk.dk_label->d_secsize) != 0 ||
1912 	    (bp->b_bcount / wi->wi_softc->sc_dk.dk_label->d_secsize) >=
1913 	     (1 << NBBY)) {
1914 		error = EINVAL;
1915 		goto bad;
1916 	}
1917 
1918 	/*
1919 	 * Make sure a timeout was supplied in the ioctl request
1920 	 */
1921 
1922 	if (wi->wi_atareq.timeout == 0) {
1923 		error = EINVAL;
1924 		goto bad;
1925 	}
1926 
1927 	if (wi->wi_atareq.flags & ATACMD_READ)
1928 		ata_c.flags |= AT_READ;
1929 	else if (wi->wi_atareq.flags & ATACMD_WRITE)
1930 		ata_c.flags |= AT_WRITE;
1931 
1932 	if (wi->wi_atareq.flags & ATACMD_READREG)
1933 		ata_c.flags |= AT_READREG;
1934 
1935 	ata_c.flags |= AT_WAIT;
1936 
1937 	ata_c.timeout = wi->wi_atareq.timeout;
1938 	ata_c.r_command = wi->wi_atareq.command;
1939 	ata_c.r_head = wi->wi_atareq.head & 0x0f;
1940 	ata_c.r_cyl = wi->wi_atareq.cylinder;
1941 	ata_c.r_sector = wi->wi_atareq.sec_num;
1942 	ata_c.r_count = wi->wi_atareq.sec_count;
1943 	ata_c.r_features = wi->wi_atareq.features;
1944 	ata_c.r_st_bmask = WDCS_DRDY;
1945 	ata_c.r_st_pmask = WDCS_DRDY;
1946 	ata_c.data = wi->wi_bp.b_data;
1947 	ata_c.bcount = wi->wi_bp.b_bcount;
1948 
1949 	if (wi->wi_softc->atabus->ata_exec_command(wi->wi_softc->drvp, &ata_c)
1950 	    != ATACMD_COMPLETE) {
1951 		wi->wi_atareq.retsts = ATACMD_ERROR;
1952 		goto bad;
1953 	}
1954 
1955 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1956 		if (ata_c.flags & AT_ERROR) {
1957 			wi->wi_atareq.retsts = ATACMD_ERROR;
1958 			wi->wi_atareq.error = ata_c.r_error;
1959 		} else if (ata_c.flags & AT_DF)
1960 			wi->wi_atareq.retsts = ATACMD_DF;
1961 		else
1962 			wi->wi_atareq.retsts = ATACMD_TIMEOUT;
1963 	} else {
1964 		wi->wi_atareq.retsts = ATACMD_OK;
1965 		if (wi->wi_atareq.flags & ATACMD_READREG) {
1966 			wi->wi_atareq.head = ata_c.r_head ;
1967 			wi->wi_atareq.cylinder = ata_c.r_cyl;
1968 			wi->wi_atareq.sec_num = ata_c.r_sector;
1969 			wi->wi_atareq.sec_count = ata_c.r_count;
1970 			wi->wi_atareq.features = ata_c.r_features;
1971 			wi->wi_atareq.error = ata_c.r_error;
1972 		}
1973 	}
1974 
1975 	bp->b_error = 0;
1976 	biodone(bp);
1977 	return;
1978 bad:
1979 	bp->b_flags |= B_ERROR;
1980 	bp->b_error = error;
1981 	biodone(bp);
1982 }
1983