xref: /netbsd-src/sys/dev/ata/wd.c (revision aaf4ece63a859a04e37cf3a7229b5fab0157cc06)
1 /*	$NetBSD: wd.c,v 1.317 2006/01/07 00:26:58 yamt 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 1.317 2006/01/07 00:26:58 yamt 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 	 * These seagate drives seems to have issue addressing sector 0xfffffff
235 	 * (aka LBA48_THRESHOLD) in LBA mode. The workaround is to force
236 	 * LBA48
237 	 * Note that we can't just change the code to always use LBA48 for
238 	 * sector 0xfffffff, because this would break valid and working
239 	 * setups using LBA48 drives on non-LBA48-capable controllers
240 	 * (and it's hard to get a list of such controllers)
241 	 */
242 	{ "ST3160023A*",
243 	  WD_QUIRK_FORCE_LBA48 },
244 	{ "ST3160827A*",
245 	  WD_QUIRK_FORCE_LBA48 },
246 	/* Attempt to catch all seagate drives larger than 200GB */
247 	{ "ST3[2-9][0-9][0-9][0-9][0-9][0-9]A*",
248 	  WD_QUIRK_FORCE_LBA48 },
249 	{ NULL,
250 	  0 }
251 };
252 
253 static const struct wd_quirk *
254 wd_lookup_quirks(const char *name)
255 {
256 	const struct wd_quirk *wdq;
257 	const char *estr;
258 
259 	for (wdq = wd_quirk_table; wdq->wdq_match != NULL; wdq++) {
260 		/*
261 		 * We only want exact matches (which include matches
262 		 * against globbing characters).
263 		 */
264 		if (pmatch(name, wdq->wdq_match, &estr) == 2)
265 			return (wdq);
266 	}
267 	return (NULL);
268 }
269 
270 int
271 wdprobe(struct device *parent, struct cfdata *match, void *aux)
272 {
273 	struct ata_device *adev = aux;
274 
275 	if (adev == NULL)
276 		return 0;
277 	if (adev->adev_bustype->bustype_type != SCSIPI_BUSTYPE_ATA)
278 		return 0;
279 
280 	if (match->cf_loc[ATA_HLCF_DRIVE] != ATA_HLCF_DRIVE_DEFAULT &&
281 	    match->cf_loc[ATA_HLCF_DRIVE] != adev->adev_drv_data->drive)
282 		return 0;
283 	return 1;
284 }
285 
286 void
287 wdattach(struct device *parent, struct device *self, void *aux)
288 {
289 	struct wd_softc *wd = (void *)self;
290 	struct ata_device *adev= aux;
291 	int i, blank;
292 	char tbuf[41], pbuf[9], c, *p, *q;
293 	const struct wd_quirk *wdq;
294 	ATADEBUG_PRINT(("wdattach\n"), DEBUG_FUNCS | DEBUG_PROBE);
295 
296 	callout_init(&wd->sc_restart_ch);
297 	bufq_alloc(&wd->sc_q, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
298 #ifdef WD_SOFTBADSECT
299 	SLIST_INIT(&wd->sc_bslist);
300 #endif
301 	wd->atabus = adev->adev_bustype;
302 	wd->openings = adev->adev_openings;
303 	wd->drvp = adev->adev_drv_data;
304 
305 	wd->drvp->drv_done = wddone;
306 	wd->drvp->drv_softc = &wd->sc_dev;
307 
308 	aprint_naive("\n");
309 
310 	/* read our drive info */
311 	if (wd_get_params(wd, AT_WAIT, &wd->sc_params) != 0) {
312 		aprint_error("\n%s: IDENTIFY failed\n", wd->sc_dev.dv_xname);
313 		return;
314 	}
315 
316 	for (blank = 0, p = wd->sc_params.atap_model, q = tbuf, i = 0;
317 	    i < sizeof(wd->sc_params.atap_model); i++) {
318 		c = *p++;
319 		if (c == '\0')
320 			break;
321 		if (c != ' ') {
322 			if (blank) {
323 				*q++ = ' ';
324 				blank = 0;
325 			}
326 			*q++ = c;
327 		} else
328 			blank = 1;
329 	}
330 	*q++ = '\0';
331 
332 	aprint_normal(": <%s>\n", tbuf);
333 
334 	wdq = wd_lookup_quirks(tbuf);
335 	if (wdq != NULL)
336 		wd->sc_quirks = wdq->wdq_quirks;
337 
338 	if ((wd->sc_params.atap_multi & 0xff) > 1) {
339 		wd->sc_multi = wd->sc_params.atap_multi & 0xff;
340 	} else {
341 		wd->sc_multi = 1;
342 	}
343 
344 	aprint_normal("%s: drive supports %d-sector PIO transfers,",
345 	    wd->sc_dev.dv_xname, wd->sc_multi);
346 
347 	/* 48-bit LBA addressing */
348 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0)
349 		wd->sc_flags |= WDF_LBA48;
350 
351 	/* Prior to ATA-4, LBA was optional. */
352 	if ((wd->sc_params.atap_capabilities1 & WDC_CAP_LBA) != 0)
353 		wd->sc_flags |= WDF_LBA;
354 #if 0
355 	/* ATA-4 requires LBA. */
356 	if (wd->sc_params.atap_ataversion != 0xffff &&
357 	    wd->sc_params.atap_ataversion >= WDC_VER_ATA4)
358 		wd->sc_flags |= WDF_LBA;
359 #endif
360 
361 	if ((wd->sc_flags & WDF_LBA48) != 0) {
362 		aprint_normal(" LBA48 addressing\n");
363 		wd->sc_capacity =
364 		    ((u_int64_t) wd->sc_params.__reserved6[11] << 48) |
365 		    ((u_int64_t) wd->sc_params.__reserved6[10] << 32) |
366 		    ((u_int64_t) wd->sc_params.__reserved6[9]  << 16) |
367 		    ((u_int64_t) wd->sc_params.__reserved6[8]  << 0);
368 	} else if ((wd->sc_flags & WDF_LBA) != 0) {
369 		aprint_normal(" LBA addressing\n");
370 		wd->sc_capacity =
371 		    (wd->sc_params.atap_capacity[1] << 16) |
372 		    wd->sc_params.atap_capacity[0];
373 	} else {
374 		aprint_normal(" chs addressing\n");
375 		wd->sc_capacity =
376 		    wd->sc_params.atap_cylinders *
377 		    wd->sc_params.atap_heads *
378 		    wd->sc_params.atap_sectors;
379 	}
380 	format_bytes(pbuf, sizeof(pbuf), wd->sc_capacity * DEV_BSIZE);
381 	aprint_normal("%s: %s, %d cyl, %d head, %d sec, "
382 	    "%d bytes/sect x %llu sectors\n",
383 	    self->dv_xname, pbuf,
384 	    (wd->sc_flags & WDF_LBA) ? (int)(wd->sc_capacity /
385 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors)) :
386 		wd->sc_params.atap_cylinders,
387 	    wd->sc_params.atap_heads, wd->sc_params.atap_sectors,
388 	    DEV_BSIZE, (unsigned long long)wd->sc_capacity);
389 
390 	ATADEBUG_PRINT(("%s: atap_dmatiming_mimi=%d, atap_dmatiming_recom=%d\n",
391 	    self->dv_xname, wd->sc_params.atap_dmatiming_mimi,
392 	    wd->sc_params.atap_dmatiming_recom), DEBUG_PROBE);
393 	/*
394 	 * Initialize and attach the disk structure.
395 	 */
396 	wd->sc_dk.dk_driver = &wddkdriver;
397 	wd->sc_dk.dk_name = wd->sc_dev.dv_xname;
398 	disk_attach(&wd->sc_dk);
399 	wd->sc_wdc_bio.lp = wd->sc_dk.dk_label;
400 	wd->sc_sdhook = shutdownhook_establish(wd_shutdown, wd);
401 	if (wd->sc_sdhook == NULL)
402 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
403 		    wd->sc_dev.dv_xname);
404 #if NRND > 0
405 	rnd_attach_source(&wd->rnd_source, wd->sc_dev.dv_xname,
406 			  RND_TYPE_DISK, 0);
407 #endif
408 
409 	/* Discover wedges on this disk. */
410 	dkwedge_discover(&wd->sc_dk);
411 }
412 
413 int
414 wdactivate(struct device *self, enum devact act)
415 {
416 	int rv = 0;
417 
418 	switch (act) {
419 	case DVACT_ACTIVATE:
420 		rv = EOPNOTSUPP;
421 		break;
422 
423 	case DVACT_DEACTIVATE:
424 		/*
425 		 * Nothing to do; we key off the device's DVF_ACTIVATE.
426 		 */
427 		break;
428 	}
429 	return (rv);
430 }
431 
432 int
433 wddetach(struct device *self, int flags)
434 {
435 	struct wd_softc *sc = (struct wd_softc *)self;
436 	int s, bmaj, cmaj, i, mn;
437 
438 	/* locate the major number */
439 	bmaj = bdevsw_lookup_major(&wd_bdevsw);
440 	cmaj = cdevsw_lookup_major(&wd_cdevsw);
441 
442 	/* Nuke the vnodes for any open instances. */
443 	for (i = 0; i < MAXPARTITIONS; i++) {
444 		mn = WDMINOR(self->dv_unit, i);
445 		vdevgone(bmaj, mn, mn, VBLK);
446 		vdevgone(cmaj, mn, mn, VCHR);
447 	}
448 
449 	/* Delete all of our wedges. */
450 	dkwedge_delall(&sc->sc_dk);
451 
452 	s = splbio();
453 
454 	/* Kill off any queued buffers. */
455 	bufq_drain(sc->sc_q);
456 
457 	bufq_free(sc->sc_q);
458 	sc->atabus->ata_killpending(sc->drvp);
459 
460 	splx(s);
461 
462 	/* Detach disk. */
463 	disk_detach(&sc->sc_dk);
464 
465 #ifdef WD_SOFTBADSECT
466 	/* Clean out the bad sector list */
467 	while (!SLIST_EMPTY(&sc->sc_bslist)) {
468 		void *head = SLIST_FIRST(&sc->sc_bslist);
469 		SLIST_REMOVE_HEAD(&sc->sc_bslist, dbs_next);
470 		free(head, M_TEMP);
471 	}
472 	sc->sc_bscount = 0;
473 #endif
474 
475 	/* Get rid of the shutdown hook. */
476 	if (sc->sc_sdhook != NULL)
477 		shutdownhook_disestablish(sc->sc_sdhook);
478 
479 #if NRND > 0
480 	/* Unhook the entropy source. */
481 	rnd_detach_source(&sc->rnd_source);
482 #endif
483 
484 	sc->drvp->drive_flags = 0; /* no drive any more here */
485 
486 	return (0);
487 }
488 
489 /*
490  * Read/write routine for a buffer.  Validates the arguments and schedules the
491  * transfer.  Does not wait for the transfer to complete.
492  */
493 void
494 wdstrategy(struct buf *bp)
495 {
496 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(bp->b_dev));
497 	struct disklabel *lp = wd->sc_dk.dk_label;
498 	daddr_t blkno;
499 	int s;
500 
501 	ATADEBUG_PRINT(("wdstrategy (%s)\n", wd->sc_dev.dv_xname),
502 	    DEBUG_XFERS);
503 
504 	/* Valid request?  */
505 	if (bp->b_blkno < 0 ||
506 	    (bp->b_bcount % lp->d_secsize) != 0 ||
507 	    (bp->b_bcount / lp->d_secsize) >= (1 << NBBY)) {
508 		bp->b_error = EINVAL;
509 		goto bad;
510 	}
511 
512 	/* If device invalidated (e.g. media change, door open), error. */
513 	if ((wd->sc_flags & WDF_LOADED) == 0) {
514 		bp->b_error = EIO;
515 		goto bad;
516 	}
517 
518 	/* If it's a null transfer, return immediately. */
519 	if (bp->b_bcount == 0)
520 		goto done;
521 
522 	/*
523 	 * Do bounds checking, adjust transfer. if error, process.
524 	 * If end of partition, just return.
525 	 */
526 	if (WDPART(bp->b_dev) == RAW_PART) {
527 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
528 		    wd->sc_capacity) <= 0)
529 			goto done;
530 	} else {
531 		if (bounds_check_with_label(&wd->sc_dk, bp,
532 		    (wd->sc_flags & (WDF_WLABEL|WDF_LABELLING)) != 0) <= 0)
533 			goto done;
534 	}
535 
536 	/*
537 	 * Now convert the block number to absolute and put it in
538 	 * terms of the device's logical block size.
539 	 */
540 	if (lp->d_secsize >= DEV_BSIZE)
541 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
542 	else
543 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
544 
545 	if (WDPART(bp->b_dev) != RAW_PART)
546 		blkno += lp->d_partitions[WDPART(bp->b_dev)].p_offset;
547 
548 	bp->b_rawblkno = blkno;
549 
550 #ifdef WD_SOFTBADSECT
551 	/*
552 	 * If the transfer about to be attempted contains only a block that
553 	 * is known to be bad then return an error for the transfer without
554 	 * even attempting to start a transfer up under the premis that we
555 	 * will just end up doing more retries for a transfer that will end
556 	 * up failing again.
557 	 * XXX:SMP - mutex required to protect with DIOCBSFLUSH
558 	 */
559 	if (__predict_false(!SLIST_EMPTY(&wd->sc_bslist))) {
560 		struct disk_badsectors *dbs;
561 		daddr_t maxblk = blkno + (bp->b_bcount >> DEV_BSHIFT) - 1;
562 
563 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next)
564 			if ((dbs->dbs_min <= blkno && blkno <= dbs->dbs_max) ||
565 			    (dbs->dbs_min <= maxblk && maxblk <= dbs->dbs_max)){
566 				bp->b_error = EIO;
567 				goto bad;
568 			}
569 	}
570 #endif
571 
572 	/* Queue transfer on drive, activate drive and controller if idle. */
573 	s = splbio();
574 	BUFQ_PUT(wd->sc_q, bp);
575 	wdstart(wd);
576 	splx(s);
577 	return;
578 bad:
579 	bp->b_flags |= B_ERROR;
580 done:
581 	/* Toss transfer; we're done early. */
582 	bp->b_resid = bp->b_bcount;
583 	biodone(bp);
584 }
585 
586 /*
587  * Queue a drive for I/O.
588  */
589 void
590 wdstart(void *arg)
591 {
592 	struct wd_softc *wd = arg;
593 	struct buf *bp = NULL;
594 
595 	ATADEBUG_PRINT(("wdstart %s\n", wd->sc_dev.dv_xname),
596 	    DEBUG_XFERS);
597 	while (wd->openings > 0) {
598 
599 		/* Is there a buf for us ? */
600 		if ((bp = BUFQ_GET(wd->sc_q)) == NULL)
601 			return;
602 
603 		/*
604 		 * Make the command. First lock the device
605 		 */
606 		wd->openings--;
607 
608 		wd->retries = 0;
609 		__wdstart(wd, bp);
610 	}
611 }
612 
613 static void
614 wd_split_mod15_write(struct buf *bp)
615 {
616 	struct buf *obp = bp->b_private;
617 	struct wd_softc *sc = wd_cd.cd_devs[DISKUNIT(obp->b_dev)];
618 
619 	if (__predict_false(bp->b_flags & B_ERROR) != 0) {
620 		/*
621 		 * Propagate the error.  If this was the first half of
622 		 * the original transfer, make sure to account for that
623 		 * in the residual.
624 		 */
625 		if (bp->b_data == obp->b_data)
626 			bp->b_resid += bp->b_bcount;
627 		goto done;
628 	}
629 
630 	/*
631 	 * If this was the second half of the transfer, we're all done!
632 	 */
633 	if (bp->b_data != obp->b_data)
634 		goto done;
635 
636 	/*
637 	 * Advance the pointer to the second half and issue that command
638 	 * using the same opening.
639 	 */
640 	bp->b_flags = obp->b_flags | B_CALL;
641 	bp->b_data += bp->b_bcount;
642 	bp->b_blkno += (bp->b_bcount / 512);
643 	bp->b_rawblkno += (bp->b_bcount / 512);
644 	__wdstart(sc, bp);
645 	return;
646 
647  done:
648 	obp->b_flags |= bp->b_flags & B_ERROR;
649 	obp->b_error = bp->b_error;
650 	obp->b_resid = bp->b_resid;
651 	putiobuf(bp);
652 	biodone(obp);
653 	sc->openings++;
654 	/* wddone() will call wdstart() */
655 }
656 
657 void
658 __wdstart(struct wd_softc *wd, struct buf *bp)
659 {
660 
661 	/*
662 	 * Deal with the "split mod15 write" quirk.  We just divide the
663 	 * transfer in two, doing the first half and then then second half
664 	 * with the same command opening.
665 	 *
666 	 * Note we MUST do this here, because we can't let insertion
667 	 * into the bufq cause the transfers to be re-merged.
668 	 */
669 	if (__predict_false((wd->sc_quirks & WD_QUIRK_SPLIT_MOD15_WRITE) != 0 &&
670 			    (bp->b_flags & B_READ) == 0 &&
671 			    bp->b_bcount > 512 &&
672 			    ((bp->b_bcount / 512) % 15) == 1)) {
673 		struct buf *nbp;
674 
675 		/* already at splbio */
676 		nbp = getiobuf_nowait();
677 		if (__predict_false(nbp == NULL)) {
678 			/* No memory -- fail the iop. */
679 			bp->b_error = ENOMEM;
680 			bp->b_flags |= B_ERROR;
681 			bp->b_resid = bp->b_bcount;
682 			biodone(bp);
683 			wd->openings++;
684 			return;
685 		}
686 
687 		nbp->b_error = 0;
688 		nbp->b_proc = bp->b_proc;
689 		nbp->b_vp = NULLVP;
690 		nbp->b_dev = bp->b_dev;
691 
692 		nbp->b_bcount = bp->b_bcount / 2;
693 		nbp->b_bufsize = bp->b_bcount / 2;
694 		nbp->b_data = bp->b_data;
695 
696 		nbp->b_blkno = bp->b_blkno;
697 		nbp->b_rawblkno = bp->b_rawblkno;
698 
699 		nbp->b_flags = bp->b_flags | B_CALL;
700 		nbp->b_iodone = wd_split_mod15_write;
701 
702 		/* Put ptr to orig buf in b_private and use new buf */
703 		nbp->b_private = bp;
704 
705 		BIO_COPYPRIO(nbp, bp);
706 
707 		bp = nbp;
708 	}
709 
710 	wd->sc_wdc_bio.blkno = bp->b_rawblkno;
711 	wd->sc_wdc_bio.blkdone =0;
712 	wd->sc_bp = bp;
713 	/*
714 	 * If we're retrying, retry in single-sector mode. This will give us
715 	 * the sector number of the problem, and will eventually allow the
716 	 * transfer to succeed.
717 	 */
718 	if (wd->retries >= WDIORETRIES_SINGLE)
719 		wd->sc_wdc_bio.flags = ATA_SINGLE;
720 	else
721 		wd->sc_wdc_bio.flags = 0;
722 	if (wd->sc_flags & WDF_LBA48 &&
723 	    (wd->sc_wdc_bio.blkno > LBA48_THRESHOLD ||
724 	    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) != 0))
725 		wd->sc_wdc_bio.flags |= ATA_LBA48;
726 	if (wd->sc_flags & WDF_LBA)
727 		wd->sc_wdc_bio.flags |= ATA_LBA;
728 	if (bp->b_flags & B_READ)
729 		wd->sc_wdc_bio.flags |= ATA_READ;
730 	wd->sc_wdc_bio.bcount = bp->b_bcount;
731 	wd->sc_wdc_bio.databuf = bp->b_data;
732 	/* Instrumentation. */
733 	disk_busy(&wd->sc_dk);
734 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
735 	case ATACMD_TRY_AGAIN:
736 		callout_reset(&wd->sc_restart_ch, hz, wdrestart, wd);
737 		break;
738 	case ATACMD_QUEUED:
739 	case ATACMD_COMPLETE:
740 		break;
741 	default:
742 		panic("__wdstart: bad return code from ata_bio()");
743 	}
744 }
745 
746 void
747 wddone(void *v)
748 {
749 	struct wd_softc *wd = v;
750 	struct buf *bp = wd->sc_bp;
751 	const char *errmsg;
752 	int do_perror = 0;
753 	ATADEBUG_PRINT(("wddone %s\n", wd->sc_dev.dv_xname),
754 	    DEBUG_XFERS);
755 
756 	if (bp == NULL)
757 		return;
758 	bp->b_resid = wd->sc_wdc_bio.bcount;
759 	switch (wd->sc_wdc_bio.error) {
760 	case ERR_DMA:
761 		errmsg = "DMA error";
762 		goto retry;
763 	case ERR_DF:
764 		errmsg = "device fault";
765 		goto retry;
766 	case TIMEOUT:
767 		errmsg = "device timeout";
768 		goto retry;
769 	case ERR_RESET:
770 		errmsg = "channel reset";
771 		goto retry2;
772 	case ERROR:
773 		/* Don't care about media change bits */
774 		if (wd->sc_wdc_bio.r_error != 0 &&
775 		    (wd->sc_wdc_bio.r_error & ~(WDCE_MC | WDCE_MCR)) == 0)
776 			goto noerror;
777 		errmsg = "error";
778 		do_perror = 1;
779 retry:		/* Just reset and retry. Can we do more ? */
780 		(*wd->atabus->ata_reset_drive)(wd->drvp, AT_RST_NOCMD);
781 retry2:
782 		diskerr(bp, "wd", errmsg, LOG_PRINTF,
783 		    wd->sc_wdc_bio.blkdone, wd->sc_dk.dk_label);
784 		if (wd->retries < WDIORETRIES)
785 			printf(", retrying\n");
786 		if (do_perror)
787 			wdperror(wd);
788 		if (wd->retries < WDIORETRIES) {
789 			wd->retries++;
790 			callout_reset(&wd->sc_restart_ch, RECOVERYTIME,
791 			    wdrestart, wd);
792 			return;
793 		}
794 		printf("\n");
795 
796 #ifdef WD_SOFTBADSECT
797 		/*
798 		 * Not all errors indicate a failed block but those that do,
799 		 * put the block on the bad-block list for the device.  Only
800 		 * do this for reads because the drive should do it for writes,
801 		 * itself, according to Manuel.
802 		 */
803 		if ((bp->b_flags & B_READ) &&
804 		    ((wd->drvp->ata_vers >= 4 && wd->sc_wdc_bio.r_error & 64) ||
805 	     	     (wd->drvp->ata_vers < 4 && wd->sc_wdc_bio.r_error & 192))) {
806 			struct disk_badsectors *dbs;
807 
808 			dbs = malloc(sizeof *dbs, M_TEMP, M_WAITOK);
809 			dbs->dbs_min = bp->b_rawblkno;
810 			dbs->dbs_max = dbs->dbs_min + (bp->b_bcount >> DEV_BSHIFT) - 1;
811 			microtime(&dbs->dbs_failedat);
812 			SLIST_INSERT_HEAD(&wd->sc_bslist, dbs, dbs_next);
813 			wd->sc_bscount++;
814 		}
815 #endif
816 		bp->b_flags |= B_ERROR;
817 		bp->b_error = EIO;
818 		break;
819 	case NOERROR:
820 noerror:	if ((wd->sc_wdc_bio.flags & ATA_CORR) || wd->retries > 0)
821 			printf("%s: soft error (corrected)\n",
822 			    wd->sc_dev.dv_xname);
823 		break;
824 	case ERR_NODEV:
825 		bp->b_flags |= B_ERROR;
826 		bp->b_error = EIO;
827 		break;
828 	}
829 	disk_unbusy(&wd->sc_dk, (bp->b_bcount - bp->b_resid),
830 	    (bp->b_flags & B_READ));
831 #if NRND > 0
832 	rnd_add_uint32(&wd->rnd_source, bp->b_blkno);
833 #endif
834 	/* XXX Yuck, but we don't want to increment openings in this case */
835 	if (__predict_false((bp->b_flags & B_CALL) != 0 &&
836 			    bp->b_iodone == wd_split_mod15_write))
837 		biodone(bp);
838 	else {
839 		biodone(bp);
840 		wd->openings++;
841 	}
842 	wdstart(wd);
843 }
844 
845 void
846 wdrestart(void *v)
847 {
848 	struct wd_softc *wd = v;
849 	struct buf *bp = wd->sc_bp;
850 	int s;
851 	ATADEBUG_PRINT(("wdrestart %s\n", wd->sc_dev.dv_xname),
852 	    DEBUG_XFERS);
853 
854 	s = splbio();
855 	__wdstart(v, bp);
856 	splx(s);
857 }
858 
859 int
860 wdread(dev_t dev, struct uio *uio, int flags)
861 {
862 
863 	ATADEBUG_PRINT(("wdread\n"), DEBUG_XFERS);
864 	return (physio(wdstrategy, NULL, dev, B_READ, minphys, uio));
865 }
866 
867 int
868 wdwrite(dev_t dev, struct uio *uio, int flags)
869 {
870 
871 	ATADEBUG_PRINT(("wdwrite\n"), DEBUG_XFERS);
872 	return (physio(wdstrategy, NULL, dev, B_WRITE, minphys, uio));
873 }
874 
875 int
876 wdopen(dev_t dev, int flag, int fmt, struct lwp *l)
877 {
878 	struct wd_softc *wd;
879 	int part, error;
880 
881 	ATADEBUG_PRINT(("wdopen\n"), DEBUG_FUNCS);
882 	wd = device_lookup(&wd_cd, WDUNIT(dev));
883 	if (wd == NULL)
884 		return (ENXIO);
885 
886 	if ((wd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
887 		return (ENODEV);
888 
889 	part = WDPART(dev);
890 
891 	if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
892 		return (error);
893 
894 	/*
895 	 * If there are wedges, and this is not RAW_PART, then we
896 	 * need to fail.
897 	 */
898 	if (wd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
899 		error = EBUSY;
900 		goto bad1;
901 	}
902 
903 	/*
904 	 * If this is the first open of this device, add a reference
905 	 * to the adapter.
906 	 */
907 	if (wd->sc_dk.dk_openmask == 0 &&
908 	    (error = wd->atabus->ata_addref(wd->drvp)) != 0)
909 		goto bad1;
910 
911 	if (wd->sc_dk.dk_openmask != 0) {
912 		/*
913 		 * If any partition is open, but the disk has been invalidated,
914 		 * disallow further opens.
915 		 */
916 		if ((wd->sc_flags & WDF_LOADED) == 0) {
917 			error = EIO;
918 			goto bad2;
919 		}
920 	} else {
921 		if ((wd->sc_flags & WDF_LOADED) == 0) {
922 			wd->sc_flags |= WDF_LOADED;
923 
924 			/* Load the physical device parameters. */
925 			wd_get_params(wd, AT_WAIT, &wd->sc_params);
926 
927 			/* Load the partition info if not already loaded. */
928 			wdgetdisklabel(wd);
929 		}
930 	}
931 
932 	/* Check that the partition exists. */
933 	if (part != RAW_PART &&
934 	    (part >= wd->sc_dk.dk_label->d_npartitions ||
935 	     wd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
936 		error = ENXIO;
937 		goto bad2;
938 	}
939 
940 	/* Insure only one open at a time. */
941 	switch (fmt) {
942 	case S_IFCHR:
943 		wd->sc_dk.dk_copenmask |= (1 << part);
944 		break;
945 	case S_IFBLK:
946 		wd->sc_dk.dk_bopenmask |= (1 << part);
947 		break;
948 	}
949 	wd->sc_dk.dk_openmask =
950 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
951 
952 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
953 	return 0;
954 
955  bad2:
956 	if (wd->sc_dk.dk_openmask == 0)
957 		wd->atabus->ata_delref(wd->drvp);
958  bad1:
959 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
960 	return error;
961 }
962 
963 int
964 wdclose(dev_t dev, int flag, int fmt, struct lwp *l)
965 {
966 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
967 	int part = WDPART(dev);
968 	int error;
969 
970 	ATADEBUG_PRINT(("wdclose\n"), DEBUG_FUNCS);
971 
972 	if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
973 		return error;
974 
975 	switch (fmt) {
976 	case S_IFCHR:
977 		wd->sc_dk.dk_copenmask &= ~(1 << part);
978 		break;
979 	case S_IFBLK:
980 		wd->sc_dk.dk_bopenmask &= ~(1 << part);
981 		break;
982 	}
983 	wd->sc_dk.dk_openmask =
984 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
985 
986 	if (wd->sc_dk.dk_openmask == 0) {
987 		wd_flushcache(wd, AT_WAIT);
988 
989 		if (! (wd->sc_flags & WDF_KLABEL))
990 			wd->sc_flags &= ~WDF_LOADED;
991 
992 		wd->atabus->ata_delref(wd->drvp);
993 	}
994 
995 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
996 	return 0;
997 }
998 
999 void
1000 wdgetdefaultlabel(struct wd_softc *wd, struct disklabel *lp)
1001 {
1002 
1003 	ATADEBUG_PRINT(("wdgetdefaultlabel\n"), DEBUG_FUNCS);
1004 	memset(lp, 0, sizeof(struct disklabel));
1005 
1006 	lp->d_secsize = DEV_BSIZE;
1007 	lp->d_ntracks = wd->sc_params.atap_heads;
1008 	lp->d_nsectors = wd->sc_params.atap_sectors;
1009 	lp->d_ncylinders = (wd->sc_flags & WDF_LBA) ? wd->sc_capacity /
1010 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors) :
1011 		wd->sc_params.atap_cylinders;
1012 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1013 
1014 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
1015 		lp->d_type = DTYPE_ST506;
1016 	else
1017 		lp->d_type = DTYPE_ESDI;
1018 
1019 	strncpy(lp->d_typename, wd->sc_params.atap_model, 16);
1020 	strncpy(lp->d_packname, "fictitious", 16);
1021 	if (wd->sc_capacity > UINT32_MAX)
1022 		lp->d_secperunit = UINT32_MAX;
1023 	else
1024 		lp->d_secperunit = wd->sc_capacity;
1025 	lp->d_rpm = 3600;
1026 	lp->d_interleave = 1;
1027 	lp->d_flags = 0;
1028 
1029 	lp->d_partitions[RAW_PART].p_offset = 0;
1030 	lp->d_partitions[RAW_PART].p_size =
1031 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1032 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1033 	lp->d_npartitions = RAW_PART + 1;
1034 
1035 	lp->d_magic = DISKMAGIC;
1036 	lp->d_magic2 = DISKMAGIC;
1037 	lp->d_checksum = dkcksum(lp);
1038 }
1039 
1040 /*
1041  * Fabricate a default disk label, and try to read the correct one.
1042  */
1043 void
1044 wdgetdisklabel(struct wd_softc *wd)
1045 {
1046 	struct disklabel *lp = wd->sc_dk.dk_label;
1047 	const char *errstring;
1048 	int s;
1049 
1050 	ATADEBUG_PRINT(("wdgetdisklabel\n"), DEBUG_FUNCS);
1051 
1052 	memset(wd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1053 
1054 	wdgetdefaultlabel(wd, lp);
1055 
1056 	wd->sc_badsect[0] = -1;
1057 
1058 	if (wd->drvp->state > RESET) {
1059 		s = splbio();
1060 		wd->drvp->drive_flags |= DRIVE_RESET;
1061 		splx(s);
1062 	}
1063 	errstring = readdisklabel(MAKEWDDEV(0, wd->sc_dev.dv_unit, RAW_PART),
1064 	    wdstrategy, lp, wd->sc_dk.dk_cpulabel);
1065 	if (errstring) {
1066 		/*
1067 		 * This probably happened because the drive's default
1068 		 * geometry doesn't match the DOS geometry.  We
1069 		 * assume the DOS geometry is now in the label and try
1070 		 * again.  XXX This is a kluge.
1071 		 */
1072 		if (wd->drvp->state > RESET) {
1073 			s = splbio();
1074 			wd->drvp->drive_flags |= DRIVE_RESET;
1075 			splx(s);
1076 		}
1077 		errstring = readdisklabel(MAKEWDDEV(0, wd->sc_dev.dv_unit,
1078 		    RAW_PART), wdstrategy, lp, wd->sc_dk.dk_cpulabel);
1079 	}
1080 	if (errstring) {
1081 		printf("%s: %s\n", wd->sc_dev.dv_xname, errstring);
1082 		return;
1083 	}
1084 
1085 	if (wd->drvp->state > RESET) {
1086 		s = splbio();
1087 		wd->drvp->drive_flags |= DRIVE_RESET;
1088 		splx(s);
1089 	}
1090 #ifdef HAS_BAD144_HANDLING
1091 	if ((lp->d_flags & D_BADSECT) != 0)
1092 		bad144intern(wd);
1093 #endif
1094 }
1095 
1096 void
1097 wdperror(const struct wd_softc *wd)
1098 {
1099 	static const char *const errstr0_3[] = {"address mark not found",
1100 	    "track 0 not found", "aborted command", "media change requested",
1101 	    "id not found", "media changed", "uncorrectable data error",
1102 	    "bad block detected"};
1103 	static const char *const errstr4_5[] = {
1104 	    "obsolete (address mark not found)",
1105 	    "no media/write protected", "aborted command",
1106 	    "media change requested", "id not found", "media changed",
1107 	    "uncorrectable data error", "interface CRC error"};
1108 	const char *const *errstr;
1109 	int i;
1110 	const char *sep = "";
1111 
1112 	const char *devname = wd->sc_dev.dv_xname;
1113 	struct ata_drive_datas *drvp = wd->drvp;
1114 	int errno = wd->sc_wdc_bio.r_error;
1115 
1116 	if (drvp->ata_vers >= 4)
1117 		errstr = errstr4_5;
1118 	else
1119 		errstr = errstr0_3;
1120 
1121 	printf("%s: (", devname);
1122 
1123 	if (errno == 0)
1124 		printf("error not notified");
1125 
1126 	for (i = 0; i < 8; i++) {
1127 		if (errno & (1 << i)) {
1128 			printf("%s%s", sep, errstr[i]);
1129 			sep = ", ";
1130 		}
1131 	}
1132 	printf(")\n");
1133 }
1134 
1135 int
1136 wdioctl(dev_t dev, u_long xfer, caddr_t addr, int flag, struct lwp *l)
1137 {
1138 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
1139 	int error = 0, s;
1140 #ifdef __HAVE_OLD_DISKLABEL
1141 	struct disklabel *newlabel = NULL;
1142 #endif
1143 
1144 	ATADEBUG_PRINT(("wdioctl\n"), DEBUG_FUNCS);
1145 
1146 	if ((wd->sc_flags & WDF_LOADED) == 0)
1147 		return EIO;
1148 
1149 	switch (xfer) {
1150 #ifdef HAS_BAD144_HANDLING
1151 	case DIOCSBAD:
1152 		if ((flag & FWRITE) == 0)
1153 			return EBADF;
1154 		wd->sc_dk.dk_cpulabel->bad = *(struct dkbad *)addr;
1155 		wd->sc_dk.dk_label->d_flags |= D_BADSECT;
1156 		bad144intern(wd);
1157 		return 0;
1158 #endif
1159 #ifdef WD_SOFTBADSECT
1160 	case DIOCBSLIST :
1161 	{
1162 		u_int32_t count, missing, skip;
1163 		struct disk_badsecinfo dbsi;
1164 		struct disk_badsectors *dbs;
1165 		size_t available;
1166 		caddr_t laddr;
1167 
1168 		dbsi = *(struct disk_badsecinfo *)addr;
1169 		missing = wd->sc_bscount;
1170 		count = 0;
1171 		available = dbsi.dbsi_bufsize;
1172 		skip = dbsi.dbsi_skip;
1173 		laddr = dbsi.dbsi_buffer;
1174 
1175 		/*
1176 		 * We start this loop with the expectation that all of the
1177 		 * entries will be missed and decrement this counter each
1178 		 * time we either skip over one (already copied out) or
1179 		 * we actually copy it back to user space.  The structs
1180 		 * holding the bad sector information are copied directly
1181 		 * back to user space whilst the summary is returned via
1182 		 * the struct passed in via the ioctl.
1183 		 */
1184 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next) {
1185 			if (skip > 0) {
1186 				missing--;
1187 				skip--;
1188 				continue;
1189 			}
1190 			if (available < sizeof(*dbs))
1191 				break;
1192 			available -= sizeof(*dbs);
1193 			copyout(dbs, laddr, sizeof(*dbs));
1194 			laddr += sizeof(*dbs);
1195 			missing--;
1196 			count++;
1197 		}
1198 		dbsi.dbsi_left = missing;
1199 		dbsi.dbsi_copied = count;
1200 		*(struct disk_badsecinfo *)addr = dbsi;
1201 		return 0;
1202 	}
1203 
1204 	case DIOCBSFLUSH :
1205 		/* Clean out the bad sector list */
1206 		while (!SLIST_EMPTY(&wd->sc_bslist)) {
1207 			void *head = SLIST_FIRST(&wd->sc_bslist);
1208 			SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
1209 			free(head, M_TEMP);
1210 		}
1211 		wd->sc_bscount = 0;
1212 		return 0;
1213 #endif
1214 	case DIOCGDINFO:
1215 		*(struct disklabel *)addr = *(wd->sc_dk.dk_label);
1216 		return 0;
1217 #ifdef __HAVE_OLD_DISKLABEL
1218 	case ODIOCGDINFO:
1219 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1220 		if (newlabel == NULL)
1221 			return EIO;
1222 		*newlabel = *(wd->sc_dk.dk_label);
1223 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1224 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1225 		else
1226 			error = ENOTTY;
1227 		free(newlabel, M_TEMP);
1228 		return error;
1229 #endif
1230 
1231 	case DIOCGPART:
1232 		((struct partinfo *)addr)->disklab = wd->sc_dk.dk_label;
1233 		((struct partinfo *)addr)->part =
1234 		    &wd->sc_dk.dk_label->d_partitions[WDPART(dev)];
1235 		return 0;
1236 
1237 	case DIOCWDINFO:
1238 	case DIOCSDINFO:
1239 #ifdef __HAVE_OLD_DISKLABEL
1240 	case ODIOCWDINFO:
1241 	case ODIOCSDINFO:
1242 #endif
1243 	{
1244 		struct disklabel *lp;
1245 
1246 		if ((flag & FWRITE) == 0)
1247 			return EBADF;
1248 
1249 #ifdef __HAVE_OLD_DISKLABEL
1250 		if (xfer == ODIOCSDINFO || xfer == ODIOCWDINFO) {
1251 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1252 			if (newlabel == NULL)
1253 				return EIO;
1254 			memset(newlabel, 0, sizeof newlabel);
1255 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1256 			lp = newlabel;
1257 		} else
1258 #endif
1259 		lp = (struct disklabel *)addr;
1260 
1261 		if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE,
1262 				     NULL)) != 0)
1263 			goto bad;
1264 		wd->sc_flags |= WDF_LABELLING;
1265 
1266 		error = setdisklabel(wd->sc_dk.dk_label,
1267 		    lp, /*wd->sc_dk.dk_openmask : */0,
1268 		    wd->sc_dk.dk_cpulabel);
1269 		if (error == 0) {
1270 			if (wd->drvp->state > RESET) {
1271 				s = splbio();
1272 				wd->drvp->drive_flags |= DRIVE_RESET;
1273 				splx(s);
1274 			}
1275 			if (xfer == DIOCWDINFO
1276 #ifdef __HAVE_OLD_DISKLABEL
1277 			    || xfer == ODIOCWDINFO
1278 #endif
1279 			    )
1280 				error = writedisklabel(WDLABELDEV(dev),
1281 				    wdstrategy, wd->sc_dk.dk_label,
1282 				    wd->sc_dk.dk_cpulabel);
1283 		}
1284 
1285 		wd->sc_flags &= ~WDF_LABELLING;
1286 		(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
1287 bad:
1288 #ifdef __HAVE_OLD_DISKLABEL
1289 		if (newlabel != NULL)
1290 			free(newlabel, M_TEMP);
1291 #endif
1292 		return error;
1293 	}
1294 
1295 	case DIOCKLABEL:
1296 		if (*(int *)addr)
1297 			wd->sc_flags |= WDF_KLABEL;
1298 		else
1299 			wd->sc_flags &= ~WDF_KLABEL;
1300 		return 0;
1301 
1302 	case DIOCWLABEL:
1303 		if ((flag & FWRITE) == 0)
1304 			return EBADF;
1305 		if (*(int *)addr)
1306 			wd->sc_flags |= WDF_WLABEL;
1307 		else
1308 			wd->sc_flags &= ~WDF_WLABEL;
1309 		return 0;
1310 
1311 	case DIOCGDEFLABEL:
1312 		wdgetdefaultlabel(wd, (struct disklabel *)addr);
1313 		return 0;
1314 #ifdef __HAVE_OLD_DISKLABEL
1315 	case ODIOCGDEFLABEL:
1316 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1317 		if (newlabel == NULL)
1318 			return EIO;
1319 		wdgetdefaultlabel(wd, newlabel);
1320 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1321 			memcpy(addr, &newlabel, sizeof (struct olddisklabel));
1322 		else
1323 			error = ENOTTY;
1324 		free(newlabel, M_TEMP);
1325 		return error;
1326 #endif
1327 
1328 #ifdef notyet
1329 	case DIOCWFORMAT:
1330 		if ((flag & FWRITE) == 0)
1331 			return EBADF;
1332 		{
1333 		register struct format_op *fop;
1334 		struct iovec aiov;
1335 		struct uio auio;
1336 
1337 		fop = (struct format_op *)addr;
1338 		aiov.iov_base = fop->df_buf;
1339 		aiov.iov_len = fop->df_count;
1340 		auio.uio_iov = &aiov;
1341 		auio.uio_iovcnt = 1;
1342 		auio.uio_resid = fop->df_count;
1343 		auio.uio_segflg = 0;
1344 		auio.uio_offset =
1345 			fop->df_startblk * wd->sc_dk.dk_label->d_secsize;
1346 		auio.uio_lwp = l;
1347 		error = physio(wdformat, NULL, dev, B_WRITE, minphys,
1348 		    &auio);
1349 		fop->df_count -= auio.uio_resid;
1350 		fop->df_reg[0] = wdc->sc_status;
1351 		fop->df_reg[1] = wdc->sc_error;
1352 		return error;
1353 		}
1354 #endif
1355 	case DIOCGCACHE:
1356 		return wd_getcache(wd, (int *)addr);
1357 
1358 	case DIOCSCACHE:
1359 		return wd_setcache(wd, *(int *)addr);
1360 
1361 	case DIOCCACHESYNC:
1362 		return wd_flushcache(wd, AT_WAIT);
1363 
1364 	case ATAIOCCOMMAND:
1365 		/*
1366 		 * Make sure this command is (relatively) safe first
1367 		 */
1368 		if ((((atareq_t *) addr)->flags & ATACMD_READ) == 0 &&
1369 		    (flag & FWRITE) == 0)
1370 			return (EBADF);
1371 		{
1372 		struct wd_ioctl *wi;
1373 		atareq_t *atareq = (atareq_t *) addr;
1374 		int error1;
1375 
1376 		wi = wi_get();
1377 		wi->wi_softc = wd;
1378 		wi->wi_atareq = *atareq;
1379 
1380 		if (atareq->datalen && atareq->flags &
1381 		    (ATACMD_READ | ATACMD_WRITE)) {
1382 			wi->wi_iov.iov_base = atareq->databuf;
1383 			wi->wi_iov.iov_len = atareq->datalen;
1384 			wi->wi_uio.uio_iov = &wi->wi_iov;
1385 			wi->wi_uio.uio_iovcnt = 1;
1386 			wi->wi_uio.uio_resid = atareq->datalen;
1387 			wi->wi_uio.uio_offset = 0;
1388 			wi->wi_uio.uio_segflg = UIO_USERSPACE;
1389 			wi->wi_uio.uio_rw =
1390 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE;
1391 			wi->wi_uio.uio_lwp = l;
1392 			error1 = physio(wdioctlstrategy, &wi->wi_bp, dev,
1393 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE,
1394 			    minphys, &wi->wi_uio);
1395 		} else {
1396 			/* No need to call physio if we don't have any
1397 			   user data */
1398 			wi->wi_bp.b_flags = 0;
1399 			wi->wi_bp.b_data = 0;
1400 			wi->wi_bp.b_bcount = 0;
1401 			wi->wi_bp.b_dev = 0;
1402 			wi->wi_bp.b_proc = l->l_proc;
1403 			wdioctlstrategy(&wi->wi_bp);
1404 			error1 = wi->wi_bp.b_error;
1405 		}
1406 		*atareq = wi->wi_atareq;
1407 		wi_free(wi);
1408 		return(error1);
1409 		}
1410 
1411 	case DIOCAWEDGE:
1412 	    {
1413 	    	struct dkwedge_info *dkw = (void *) addr;
1414 
1415 		if ((flag & FWRITE) == 0)
1416 			return (EBADF);
1417 
1418 		/* If the ioctl happens here, the parent is us. */
1419 		strcpy(dkw->dkw_parent, wd->sc_dev.dv_xname);
1420 		return (dkwedge_add(dkw));
1421 	    }
1422 
1423 	case DIOCDWEDGE:
1424 	    {
1425 	    	struct dkwedge_info *dkw = (void *) addr;
1426 
1427 		if ((flag & FWRITE) == 0)
1428 			return (EBADF);
1429 
1430 		/* If the ioctl happens here, the parent is us. */
1431 		strcpy(dkw->dkw_parent, wd->sc_dev.dv_xname);
1432 		return (dkwedge_del(dkw));
1433 	    }
1434 
1435 	case DIOCLWEDGES:
1436 	    {
1437 	    	struct dkwedge_list *dkwl = (void *) addr;
1438 
1439 		return (dkwedge_list(&wd->sc_dk, dkwl, l));
1440 	    }
1441 
1442 	case DIOCGSTRATEGY:
1443 	    {
1444 		struct disk_strategy *dks = (void *)addr;
1445 
1446 		s = splbio();
1447 		strlcpy(dks->dks_name, bufq_getstrategyname(wd->sc_q),
1448 		    sizeof(dks->dks_name));
1449 		splx(s);
1450 		dks->dks_paramlen = 0;
1451 
1452 		return 0;
1453 	    }
1454 
1455 	case DIOCSSTRATEGY:
1456 	    {
1457 		struct disk_strategy *dks = (void *)addr;
1458 		struct bufq_state *new;
1459 		struct bufq_state *old;
1460 
1461 		if ((flag & FWRITE) == 0) {
1462 			return EBADF;
1463 		}
1464 		if (dks->dks_param != NULL) {
1465 			return EINVAL;
1466 		}
1467 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
1468 		error = bufq_alloc(&new, dks->dks_name,
1469 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
1470 		if (error) {
1471 			return error;
1472 		}
1473 		s = splbio();
1474 		old = wd->sc_q;
1475 		bufq_move(new, old);
1476 		wd->sc_q = new;
1477 		splx(s);
1478 		bufq_free(old);
1479 
1480 		return 0;
1481 	    }
1482 
1483 	default:
1484 		return ENOTTY;
1485 	}
1486 
1487 #ifdef DIAGNOSTIC
1488 	panic("wdioctl: impossible");
1489 #endif
1490 }
1491 
1492 #ifdef B_FORMAT
1493 int
1494 wdformat(struct buf *bp)
1495 {
1496 
1497 	bp->b_flags |= B_FORMAT;
1498 	return wdstrategy(bp);
1499 }
1500 #endif
1501 
1502 int
1503 wdsize(dev_t dev)
1504 {
1505 	struct wd_softc *wd;
1506 	int part, omask;
1507 	int size;
1508 
1509 	ATADEBUG_PRINT(("wdsize\n"), DEBUG_FUNCS);
1510 
1511 	wd = device_lookup(&wd_cd, WDUNIT(dev));
1512 	if (wd == NULL)
1513 		return (-1);
1514 
1515 	part = WDPART(dev);
1516 	omask = wd->sc_dk.dk_openmask & (1 << part);
1517 
1518 	if (omask == 0 && wdopen(dev, 0, S_IFBLK, NULL) != 0)
1519 		return (-1);
1520 	if (wd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1521 		size = -1;
1522 	else
1523 		size = wd->sc_dk.dk_label->d_partitions[part].p_size *
1524 		    (wd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1525 	if (omask == 0 && wdclose(dev, 0, S_IFBLK, NULL) != 0)
1526 		return (-1);
1527 	return (size);
1528 }
1529 
1530 /* #define WD_DUMP_NOT_TRUSTED if you just want to watch */
1531 static int wddoingadump = 0;
1532 static int wddumprecalibrated = 0;
1533 
1534 /*
1535  * Dump core after a system crash.
1536  */
1537 int
1538 wddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1539 {
1540 	struct wd_softc *wd;	/* disk unit to do the I/O */
1541 	struct disklabel *lp;   /* disk's disklabel */
1542 	int part, err;
1543 	int nblks;	/* total number of sectors left to write */
1544 
1545 	/* Check if recursive dump; if so, punt. */
1546 	if (wddoingadump)
1547 		return EFAULT;
1548 	wddoingadump = 1;
1549 
1550 	wd = device_lookup(&wd_cd, WDUNIT(dev));
1551 	if (wd == NULL)
1552 		return (ENXIO);
1553 
1554 	part = WDPART(dev);
1555 
1556 	/* Convert to disk sectors.  Request must be a multiple of size. */
1557 	lp = wd->sc_dk.dk_label;
1558 	if ((size % lp->d_secsize) != 0)
1559 		return EFAULT;
1560 	nblks = size / lp->d_secsize;
1561 	blkno = blkno / (lp->d_secsize / DEV_BSIZE);
1562 
1563 	/* Check transfer bounds against partition size. */
1564 	if ((blkno < 0) || ((blkno + nblks) > lp->d_partitions[part].p_size))
1565 		return EINVAL;
1566 
1567 	/* Offset block number to start of partition. */
1568 	blkno += lp->d_partitions[part].p_offset;
1569 
1570 	/* Recalibrate, if first dump transfer. */
1571 	if (wddumprecalibrated == 0) {
1572 		wddumprecalibrated = 1;
1573 		(*wd->atabus->ata_reset_drive)(wd->drvp,
1574 					       AT_POLL | AT_RST_EMERG);
1575 		wd->drvp->state = RESET;
1576 	}
1577 
1578 	wd->sc_bp = NULL;
1579 	wd->sc_wdc_bio.blkno = blkno;
1580 	wd->sc_wdc_bio.flags = ATA_POLL;
1581 	if (wd->sc_flags & WDF_LBA48 &&
1582 	    (blkno > LBA48_THRESHOLD ||
1583     	    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) != 0))
1584 		wd->sc_wdc_bio.flags |= ATA_LBA48;
1585 	if (wd->sc_flags & WDF_LBA)
1586 		wd->sc_wdc_bio.flags |= ATA_LBA;
1587 	wd->sc_wdc_bio.bcount = nblks * lp->d_secsize;
1588 	wd->sc_wdc_bio.databuf = va;
1589 #ifndef WD_DUMP_NOT_TRUSTED
1590 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
1591 	case ATACMD_TRY_AGAIN:
1592 		panic("wddump: try again");
1593 		break;
1594 	case ATACMD_QUEUED:
1595 		panic("wddump: polled command has been queued");
1596 		break;
1597 	case ATACMD_COMPLETE:
1598 		break;
1599 	}
1600 	switch(wd->sc_wdc_bio.error) {
1601 	case TIMEOUT:
1602 		printf("wddump: device timed out");
1603 		err = EIO;
1604 		break;
1605 	case ERR_DF:
1606 		printf("wddump: drive fault");
1607 		err = EIO;
1608 		break;
1609 	case ERR_DMA:
1610 		printf("wddump: DMA error");
1611 		err = EIO;
1612 		break;
1613 	case ERROR:
1614 		printf("wddump: ");
1615 		wdperror(wd);
1616 		err = EIO;
1617 		break;
1618 	case NOERROR:
1619 		err = 0;
1620 		break;
1621 	default:
1622 		panic("wddump: unknown error type");
1623 	}
1624 	if (err != 0) {
1625 		printf("\n");
1626 		return err;
1627 	}
1628 #else	/* WD_DUMP_NOT_TRUSTED */
1629 	/* Let's just talk about this first... */
1630 	printf("wd%d: dump addr 0x%x, cylin %d, head %d, sector %d\n",
1631 	    unit, va, cylin, head, sector);
1632 	delay(500 * 1000);	/* half a second */
1633 #endif
1634 
1635 	wddoingadump = 0;
1636 	return 0;
1637 }
1638 
1639 #ifdef HAS_BAD144_HANDLING
1640 /*
1641  * Internalize the bad sector table.
1642  */
1643 void
1644 bad144intern(struct wd_softc *wd)
1645 {
1646 	struct dkbad *bt = &wd->sc_dk.dk_cpulabel->bad;
1647 	struct disklabel *lp = wd->sc_dk.dk_label;
1648 	int i = 0;
1649 
1650 	ATADEBUG_PRINT(("bad144intern\n"), DEBUG_XFERS);
1651 
1652 	for (; i < NBT_BAD; i++) {
1653 		if (bt->bt_bad[i].bt_cyl == 0xffff)
1654 			break;
1655 		wd->sc_badsect[i] =
1656 		    bt->bt_bad[i].bt_cyl * lp->d_secpercyl +
1657 		    (bt->bt_bad[i].bt_trksec >> 8) * lp->d_nsectors +
1658 		    (bt->bt_bad[i].bt_trksec & 0xff);
1659 	}
1660 	for (; i < NBT_BAD+1; i++)
1661 		wd->sc_badsect[i] = -1;
1662 }
1663 #endif
1664 
1665 int
1666 wd_get_params(struct wd_softc *wd, u_int8_t flags, struct ataparams *params)
1667 {
1668 	switch (wd->atabus->ata_get_params(wd->drvp, flags, params)) {
1669 	case CMD_AGAIN:
1670 		return 1;
1671 	case CMD_ERR:
1672 		/*
1673 		 * We `know' there's a drive here; just assume it's old.
1674 		 * This geometry is only used to read the MBR and print a
1675 		 * (false) attach message.
1676 		 */
1677 		strncpy(params->atap_model, "ST506",
1678 		    sizeof params->atap_model);
1679 		params->atap_config = ATA_CFG_FIXED;
1680 		params->atap_cylinders = 1024;
1681 		params->atap_heads = 8;
1682 		params->atap_sectors = 17;
1683 		params->atap_multi = 1;
1684 		params->atap_capabilities1 = params->atap_capabilities2 = 0;
1685 		wd->drvp->ata_vers = -1; /* Mark it as pre-ATA */
1686 		return 0;
1687 	case CMD_OK:
1688 		return 0;
1689 	default:
1690 		panic("wd_get_params: bad return code from ata_get_params");
1691 		/* NOTREACHED */
1692 	}
1693 }
1694 
1695 int
1696 wd_getcache(struct wd_softc *wd, int *bitsp)
1697 {
1698 	struct ataparams params;
1699 
1700 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
1701 		return EIO;
1702 	if (params.atap_cmd_set1 == 0x0000 ||
1703 	    params.atap_cmd_set1 == 0xffff ||
1704 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0) {
1705 		*bitsp = 0;
1706 		return 0;
1707 	}
1708 	*bitsp = DKCACHE_WCHANGE | DKCACHE_READ;
1709 	if (params.atap_cmd1_en & WDC_CMD1_CACHE)
1710 		*bitsp |= DKCACHE_WRITE;
1711 
1712 	return 0;
1713 }
1714 
1715 const char at_errbits[] = "\20\10ERROR\11TIMEOU\12DF";
1716 
1717 int
1718 wd_setcache(struct wd_softc *wd, int bits)
1719 {
1720 	struct ataparams params;
1721 	struct ata_command ata_c;
1722 
1723 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
1724 		return EIO;
1725 
1726 	if (params.atap_cmd_set1 == 0x0000 ||
1727 	    params.atap_cmd_set1 == 0xffff ||
1728 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0)
1729 		return EOPNOTSUPP;
1730 
1731 	if ((bits & DKCACHE_READ) == 0 ||
1732 	    (bits & DKCACHE_SAVE) != 0)
1733 		return EOPNOTSUPP;
1734 
1735 	memset(&ata_c, 0, sizeof(struct ata_command));
1736 	ata_c.r_command = SET_FEATURES;
1737 	ata_c.r_st_bmask = 0;
1738 	ata_c.r_st_pmask = 0;
1739 	ata_c.timeout = 30000; /* 30s timeout */
1740 	ata_c.flags = AT_WAIT;
1741 	if (bits & DKCACHE_WRITE)
1742 		ata_c.r_features = WDSF_WRITE_CACHE_EN;
1743 	else
1744 		ata_c.r_features = WDSF_WRITE_CACHE_DS;
1745 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1746 		printf("%s: wd_setcache command not complete\n",
1747 		    wd->sc_dev.dv_xname);
1748 		return EIO;
1749 	}
1750 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1751 		char sbuf[sizeof(at_errbits) + 64];
1752 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
1753 		printf("%s: wd_setcache: status=%s\n", wd->sc_dev.dv_xname,
1754 		    sbuf);
1755 		return EIO;
1756 	}
1757 	return 0;
1758 }
1759 
1760 int
1761 wd_standby(struct wd_softc *wd, int flags)
1762 {
1763 	struct ata_command ata_c;
1764 
1765 	memset(&ata_c, 0, sizeof(struct ata_command));
1766 	ata_c.r_command = WDCC_STANDBY_IMMED;
1767 	ata_c.r_st_bmask = WDCS_DRDY;
1768 	ata_c.r_st_pmask = WDCS_DRDY;
1769 	ata_c.flags = flags;
1770 	ata_c.timeout = 30000; /* 30s timeout */
1771 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1772 		printf("%s: standby immediate command didn't complete\n",
1773 		    wd->sc_dev.dv_xname);
1774 		return EIO;
1775 	}
1776 	if (ata_c.flags & AT_ERROR) {
1777 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
1778 			return ENODEV;
1779 	}
1780 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1781 		char sbuf[sizeof(at_errbits) + 64];
1782 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
1783 		printf("%s: wd_standby: status=%s\n", wd->sc_dev.dv_xname,
1784 		    sbuf);
1785 		return EIO;
1786 	}
1787 	return 0;
1788 }
1789 
1790 int
1791 wd_flushcache(struct wd_softc *wd, int flags)
1792 {
1793 	struct ata_command ata_c;
1794 
1795 	/*
1796 	 * WDCC_FLUSHCACHE is here since ATA-4, but some drives report
1797 	 * only ATA-2 and still support it.
1798 	 */
1799 	if (wd->drvp->ata_vers < 4 &&
1800 	    ((wd->sc_params.atap_cmd_set2 & WDC_CMD2_FC) == 0 ||
1801 	    wd->sc_params.atap_cmd_set2 == 0xffff))
1802 		return ENODEV;
1803 	memset(&ata_c, 0, sizeof(struct ata_command));
1804 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0 &&
1805 	    (wd->sc_params.atap_cmd2_en & ATA_CMD2_FCE) != 0)
1806 		ata_c.r_command = WDCC_FLUSHCACHE_EXT;
1807 	else
1808 		ata_c.r_command = WDCC_FLUSHCACHE;
1809 	ata_c.r_st_bmask = WDCS_DRDY;
1810 	ata_c.r_st_pmask = WDCS_DRDY;
1811 	ata_c.flags = flags;
1812 	ata_c.timeout = 30000; /* 30s timeout */
1813 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1814 		printf("%s: flush cache command didn't complete\n",
1815 		    wd->sc_dev.dv_xname);
1816 		return EIO;
1817 	}
1818 	if (ata_c.flags & AT_ERROR) {
1819 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
1820 			return ENODEV;
1821 	}
1822 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1823 		char sbuf[sizeof(at_errbits) + 64];
1824 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
1825 		printf("%s: wd_flushcache: status=%s\n", wd->sc_dev.dv_xname,
1826 		    sbuf);
1827 		return EIO;
1828 	}
1829 	return 0;
1830 }
1831 
1832 void
1833 wd_shutdown(void *arg)
1834 {
1835 	struct wd_softc *wd = arg;
1836 	wd_flushcache(wd, AT_POLL);
1837 }
1838 
1839 /*
1840  * Allocate space for a ioctl queue structure.  Mostly taken from
1841  * scsipi_ioctl.c
1842  */
1843 struct wd_ioctl *
1844 wi_get(void)
1845 {
1846 	struct wd_ioctl *wi;
1847 	int s;
1848 
1849 	wi = malloc(sizeof(struct wd_ioctl), M_TEMP, M_WAITOK|M_ZERO);
1850 	simple_lock_init(&wi->wi_bp.b_interlock);
1851 	s = splbio();
1852 	LIST_INSERT_HEAD(&wi_head, wi, wi_list);
1853 	splx(s);
1854 	return (wi);
1855 }
1856 
1857 /*
1858  * Free an ioctl structure and remove it from our list
1859  */
1860 
1861 void
1862 wi_free(struct wd_ioctl *wi)
1863 {
1864 	int s;
1865 
1866 	s = splbio();
1867 	LIST_REMOVE(wi, wi_list);
1868 	splx(s);
1869 	free(wi, M_TEMP);
1870 }
1871 
1872 /*
1873  * Find a wd_ioctl structure based on the struct buf.
1874  */
1875 
1876 struct wd_ioctl *
1877 wi_find(struct buf *bp)
1878 {
1879 	struct wd_ioctl *wi;
1880 	int s;
1881 
1882 	s = splbio();
1883 	for (wi = wi_head.lh_first; wi != 0; wi = wi->wi_list.le_next)
1884 		if (bp == &wi->wi_bp)
1885 			break;
1886 	splx(s);
1887 	return (wi);
1888 }
1889 
1890 /*
1891  * Ioctl pseudo strategy routine
1892  *
1893  * This is mostly stolen from scsipi_ioctl.c:scsistrategy().  What
1894  * happens here is:
1895  *
1896  * - wdioctl() queues a wd_ioctl structure.
1897  *
1898  * - wdioctl() calls physio/wdioctlstrategy based on whether or not
1899  *   user space I/O is required.  If physio() is called, physio() eventually
1900  *   calls wdioctlstrategy().
1901  *
1902  * - In either case, wdioctlstrategy() calls wd->atabus->ata_exec_command()
1903  *   to perform the actual command
1904  *
1905  * The reason for the use of the pseudo strategy routine is because
1906  * when doing I/O to/from user space, physio _really_ wants to be in
1907  * the loop.  We could put the entire buffer into the ioctl request
1908  * structure, but that won't scale if we want to do things like download
1909  * microcode.
1910  */
1911 
1912 void
1913 wdioctlstrategy(struct buf *bp)
1914 {
1915 	struct wd_ioctl *wi;
1916 	struct ata_command ata_c;
1917 	int error = 0;
1918 
1919 	wi = wi_find(bp);
1920 	if (wi == NULL) {
1921 		printf("wdioctlstrategy: "
1922 		    "No matching ioctl request found in queue\n");
1923 		error = EINVAL;
1924 		goto bad;
1925 	}
1926 
1927 	memset(&ata_c, 0, sizeof(ata_c));
1928 
1929 	/*
1930 	 * Abort if physio broke up the transfer
1931 	 */
1932 
1933 	if (bp->b_bcount != wi->wi_atareq.datalen) {
1934 		printf("physio split wd ioctl request... cannot proceed\n");
1935 		error = EIO;
1936 		goto bad;
1937 	}
1938 
1939 	/*
1940 	 * Abort if we didn't get a buffer size that was a multiple of
1941 	 * our sector size (or was larger than NBBY)
1942 	 */
1943 
1944 	if ((bp->b_bcount % wi->wi_softc->sc_dk.dk_label->d_secsize) != 0 ||
1945 	    (bp->b_bcount / wi->wi_softc->sc_dk.dk_label->d_secsize) >=
1946 	     (1 << NBBY)) {
1947 		error = EINVAL;
1948 		goto bad;
1949 	}
1950 
1951 	/*
1952 	 * Make sure a timeout was supplied in the ioctl request
1953 	 */
1954 
1955 	if (wi->wi_atareq.timeout == 0) {
1956 		error = EINVAL;
1957 		goto bad;
1958 	}
1959 
1960 	if (wi->wi_atareq.flags & ATACMD_READ)
1961 		ata_c.flags |= AT_READ;
1962 	else if (wi->wi_atareq.flags & ATACMD_WRITE)
1963 		ata_c.flags |= AT_WRITE;
1964 
1965 	if (wi->wi_atareq.flags & ATACMD_READREG)
1966 		ata_c.flags |= AT_READREG;
1967 
1968 	ata_c.flags |= AT_WAIT;
1969 
1970 	ata_c.timeout = wi->wi_atareq.timeout;
1971 	ata_c.r_command = wi->wi_atareq.command;
1972 	ata_c.r_head = wi->wi_atareq.head & 0x0f;
1973 	ata_c.r_cyl = wi->wi_atareq.cylinder;
1974 	ata_c.r_sector = wi->wi_atareq.sec_num;
1975 	ata_c.r_count = wi->wi_atareq.sec_count;
1976 	ata_c.r_features = wi->wi_atareq.features;
1977 	ata_c.r_st_bmask = WDCS_DRDY;
1978 	ata_c.r_st_pmask = WDCS_DRDY;
1979 	ata_c.data = wi->wi_bp.b_data;
1980 	ata_c.bcount = wi->wi_bp.b_bcount;
1981 
1982 	if (wi->wi_softc->atabus->ata_exec_command(wi->wi_softc->drvp, &ata_c)
1983 	    != ATACMD_COMPLETE) {
1984 		wi->wi_atareq.retsts = ATACMD_ERROR;
1985 		goto bad;
1986 	}
1987 
1988 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1989 		if (ata_c.flags & AT_ERROR) {
1990 			wi->wi_atareq.retsts = ATACMD_ERROR;
1991 			wi->wi_atareq.error = ata_c.r_error;
1992 		} else if (ata_c.flags & AT_DF)
1993 			wi->wi_atareq.retsts = ATACMD_DF;
1994 		else
1995 			wi->wi_atareq.retsts = ATACMD_TIMEOUT;
1996 	} else {
1997 		wi->wi_atareq.retsts = ATACMD_OK;
1998 		if (wi->wi_atareq.flags & ATACMD_READREG) {
1999 			wi->wi_atareq.head = ata_c.r_head ;
2000 			wi->wi_atareq.cylinder = ata_c.r_cyl;
2001 			wi->wi_atareq.sec_num = ata_c.r_sector;
2002 			wi->wi_atareq.sec_count = ata_c.r_count;
2003 			wi->wi_atareq.features = ata_c.r_features;
2004 			wi->wi_atareq.error = ata_c.r_error;
2005 		}
2006 	}
2007 
2008 	bp->b_error = 0;
2009 	biodone(bp);
2010 	return;
2011 bad:
2012 	bp->b_flags |= B_ERROR;
2013 	bp->b_error = error;
2014 	biodone(bp);
2015 }
2016