xref: /netbsd-src/sys/dev/ata/wd.c (revision e5548b402ae4c44fb816de42c7bba9581ce23ef5)
1 /*	$NetBSD: wd.c,v 1.314 2005/12/11 12:21:14 christos 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 1.314 2005/12/11 12:21:14 christos 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_EINTR|B_ERROR));
649 	obp->b_error = bp->b_error;
650 	obp->b_resid = bp->b_resid;
651 	pool_put(&bufpool, 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 = pool_get(&bufpool, PR_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 		BUF_INIT(nbp);
688 		nbp->b_error = 0;
689 		nbp->b_proc = bp->b_proc;
690 		nbp->b_vp = NULLVP;
691 		nbp->b_dev = bp->b_dev;
692 
693 		nbp->b_bcount = bp->b_bcount / 2;
694 		nbp->b_bufsize = bp->b_bcount / 2;
695 		nbp->b_data = bp->b_data;
696 
697 		nbp->b_blkno = bp->b_blkno;
698 		nbp->b_rawblkno = bp->b_rawblkno;
699 
700 		nbp->b_flags = bp->b_flags | B_CALL;
701 		nbp->b_iodone = wd_split_mod15_write;
702 
703 		/* Put ptr to orig buf in b_private and use new buf */
704 		nbp->b_private = bp;
705 
706 		BIO_COPYPRIO(nbp, bp);
707 
708 		bp = nbp;
709 	}
710 
711 	wd->sc_wdc_bio.blkno = bp->b_rawblkno;
712 	wd->sc_wdc_bio.blkdone =0;
713 	wd->sc_bp = bp;
714 	/*
715 	 * If we're retrying, retry in single-sector mode. This will give us
716 	 * the sector number of the problem, and will eventually allow the
717 	 * transfer to succeed.
718 	 */
719 	if (wd->retries >= WDIORETRIES_SINGLE)
720 		wd->sc_wdc_bio.flags = ATA_SINGLE;
721 	else
722 		wd->sc_wdc_bio.flags = 0;
723 	if (wd->sc_flags & WDF_LBA48 &&
724 	    (wd->sc_wdc_bio.blkno > LBA48_THRESHOLD ||
725 	    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) != 0))
726 		wd->sc_wdc_bio.flags |= ATA_LBA48;
727 	if (wd->sc_flags & WDF_LBA)
728 		wd->sc_wdc_bio.flags |= ATA_LBA;
729 	if (bp->b_flags & B_READ)
730 		wd->sc_wdc_bio.flags |= ATA_READ;
731 	wd->sc_wdc_bio.bcount = bp->b_bcount;
732 	wd->sc_wdc_bio.databuf = bp->b_data;
733 	/* Instrumentation. */
734 	disk_busy(&wd->sc_dk);
735 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
736 	case ATACMD_TRY_AGAIN:
737 		callout_reset(&wd->sc_restart_ch, hz, wdrestart, wd);
738 		break;
739 	case ATACMD_QUEUED:
740 	case ATACMD_COMPLETE:
741 		break;
742 	default:
743 		panic("__wdstart: bad return code from ata_bio()");
744 	}
745 }
746 
747 void
748 wddone(void *v)
749 {
750 	struct wd_softc *wd = v;
751 	struct buf *bp = wd->sc_bp;
752 	const char *errmsg;
753 	int do_perror = 0;
754 	ATADEBUG_PRINT(("wddone %s\n", wd->sc_dev.dv_xname),
755 	    DEBUG_XFERS);
756 
757 	if (bp == NULL)
758 		return;
759 	bp->b_resid = wd->sc_wdc_bio.bcount;
760 	switch (wd->sc_wdc_bio.error) {
761 	case ERR_DMA:
762 		errmsg = "DMA error";
763 		goto retry;
764 	case ERR_DF:
765 		errmsg = "device fault";
766 		goto retry;
767 	case TIMEOUT:
768 		errmsg = "device timeout";
769 		goto retry;
770 	case ERR_RESET:
771 		errmsg = "channel reset";
772 		goto retry2;
773 	case ERROR:
774 		/* Don't care about media change bits */
775 		if (wd->sc_wdc_bio.r_error != 0 &&
776 		    (wd->sc_wdc_bio.r_error & ~(WDCE_MC | WDCE_MCR)) == 0)
777 			goto noerror;
778 		errmsg = "error";
779 		do_perror = 1;
780 retry:		/* Just reset and retry. Can we do more ? */
781 		(*wd->atabus->ata_reset_drive)(wd->drvp, AT_RST_NOCMD);
782 retry2:
783 		diskerr(bp, "wd", errmsg, LOG_PRINTF,
784 		    wd->sc_wdc_bio.blkdone, wd->sc_dk.dk_label);
785 		if (wd->retries < WDIORETRIES)
786 			printf(", retrying\n");
787 		if (do_perror)
788 			wdperror(wd);
789 		if (wd->retries < WDIORETRIES) {
790 			wd->retries++;
791 			callout_reset(&wd->sc_restart_ch, RECOVERYTIME,
792 			    wdrestart, wd);
793 			return;
794 		}
795 		printf("\n");
796 
797 #ifdef WD_SOFTBADSECT
798 		/*
799 		 * Not all errors indicate a failed block but those that do,
800 		 * put the block on the bad-block list for the device.  Only
801 		 * do this for reads because the drive should do it for writes,
802 		 * itself, according to Manuel.
803 		 */
804 		if ((bp->b_flags & B_READ) &&
805 		    ((wd->drvp->ata_vers >= 4 && wd->sc_wdc_bio.r_error & 64) ||
806 	     	     (wd->drvp->ata_vers < 4 && wd->sc_wdc_bio.r_error & 192))) {
807 			struct disk_badsectors *dbs;
808 
809 			dbs = malloc(sizeof *dbs, M_TEMP, M_WAITOK);
810 			dbs->dbs_min = bp->b_rawblkno;
811 			dbs->dbs_max = dbs->dbs_min + (bp->b_bcount >> DEV_BSHIFT) - 1;
812 			microtime(&dbs->dbs_failedat);
813 			SLIST_INSERT_HEAD(&wd->sc_bslist, dbs, dbs_next);
814 			wd->sc_bscount++;
815 		}
816 #endif
817 		bp->b_flags |= B_ERROR;
818 		bp->b_error = EIO;
819 		break;
820 	case NOERROR:
821 noerror:	if ((wd->sc_wdc_bio.flags & ATA_CORR) || wd->retries > 0)
822 			printf("%s: soft error (corrected)\n",
823 			    wd->sc_dev.dv_xname);
824 		break;
825 	case ERR_NODEV:
826 		bp->b_flags |= B_ERROR;
827 		bp->b_error = EIO;
828 		break;
829 	}
830 	disk_unbusy(&wd->sc_dk, (bp->b_bcount - bp->b_resid),
831 	    (bp->b_flags & B_READ));
832 #if NRND > 0
833 	rnd_add_uint32(&wd->rnd_source, bp->b_blkno);
834 #endif
835 	/* XXX Yuck, but we don't want to increment openings in this case */
836 	if (__predict_false((bp->b_flags & B_CALL) != 0 &&
837 			    bp->b_iodone == wd_split_mod15_write))
838 		biodone(bp);
839 	else {
840 		biodone(bp);
841 		wd->openings++;
842 	}
843 	wdstart(wd);
844 }
845 
846 void
847 wdrestart(void *v)
848 {
849 	struct wd_softc *wd = v;
850 	struct buf *bp = wd->sc_bp;
851 	int s;
852 	ATADEBUG_PRINT(("wdrestart %s\n", wd->sc_dev.dv_xname),
853 	    DEBUG_XFERS);
854 
855 	s = splbio();
856 	__wdstart(v, bp);
857 	splx(s);
858 }
859 
860 int
861 wdread(dev_t dev, struct uio *uio, int flags)
862 {
863 
864 	ATADEBUG_PRINT(("wdread\n"), DEBUG_XFERS);
865 	return (physio(wdstrategy, NULL, dev, B_READ, minphys, uio));
866 }
867 
868 int
869 wdwrite(dev_t dev, struct uio *uio, int flags)
870 {
871 
872 	ATADEBUG_PRINT(("wdwrite\n"), DEBUG_XFERS);
873 	return (physio(wdstrategy, NULL, dev, B_WRITE, minphys, uio));
874 }
875 
876 int
877 wdopen(dev_t dev, int flag, int fmt, struct lwp *l)
878 {
879 	struct wd_softc *wd;
880 	int part, error;
881 
882 	ATADEBUG_PRINT(("wdopen\n"), DEBUG_FUNCS);
883 	wd = device_lookup(&wd_cd, WDUNIT(dev));
884 	if (wd == NULL)
885 		return (ENXIO);
886 
887 	if ((wd->sc_dev.dv_flags & DVF_ACTIVE) == 0)
888 		return (ENODEV);
889 
890 	part = WDPART(dev);
891 
892 	if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
893 		return (error);
894 
895 	/*
896 	 * If there are wedges, and this is not RAW_PART, then we
897 	 * need to fail.
898 	 */
899 	if (wd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
900 		error = EBUSY;
901 		goto bad1;
902 	}
903 
904 	/*
905 	 * If this is the first open of this device, add a reference
906 	 * to the adapter.
907 	 */
908 	if (wd->sc_dk.dk_openmask == 0 &&
909 	    (error = wd->atabus->ata_addref(wd->drvp)) != 0)
910 		goto bad1;
911 
912 	if (wd->sc_dk.dk_openmask != 0) {
913 		/*
914 		 * If any partition is open, but the disk has been invalidated,
915 		 * disallow further opens.
916 		 */
917 		if ((wd->sc_flags & WDF_LOADED) == 0) {
918 			error = EIO;
919 			goto bad2;
920 		}
921 	} else {
922 		if ((wd->sc_flags & WDF_LOADED) == 0) {
923 			wd->sc_flags |= WDF_LOADED;
924 
925 			/* Load the physical device parameters. */
926 			wd_get_params(wd, AT_WAIT, &wd->sc_params);
927 
928 			/* Load the partition info if not already loaded. */
929 			wdgetdisklabel(wd);
930 		}
931 	}
932 
933 	/* Check that the partition exists. */
934 	if (part != RAW_PART &&
935 	    (part >= wd->sc_dk.dk_label->d_npartitions ||
936 	     wd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
937 		error = ENXIO;
938 		goto bad2;
939 	}
940 
941 	/* Insure only one open at a time. */
942 	switch (fmt) {
943 	case S_IFCHR:
944 		wd->sc_dk.dk_copenmask |= (1 << part);
945 		break;
946 	case S_IFBLK:
947 		wd->sc_dk.dk_bopenmask |= (1 << part);
948 		break;
949 	}
950 	wd->sc_dk.dk_openmask =
951 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
952 
953 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
954 	return 0;
955 
956  bad2:
957 	if (wd->sc_dk.dk_openmask == 0)
958 		wd->atabus->ata_delref(wd->drvp);
959  bad1:
960 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
961 	return error;
962 }
963 
964 int
965 wdclose(dev_t dev, int flag, int fmt, struct lwp *l)
966 {
967 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
968 	int part = WDPART(dev);
969 	int error;
970 
971 	ATADEBUG_PRINT(("wdclose\n"), DEBUG_FUNCS);
972 
973 	if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
974 		return error;
975 
976 	switch (fmt) {
977 	case S_IFCHR:
978 		wd->sc_dk.dk_copenmask &= ~(1 << part);
979 		break;
980 	case S_IFBLK:
981 		wd->sc_dk.dk_bopenmask &= ~(1 << part);
982 		break;
983 	}
984 	wd->sc_dk.dk_openmask =
985 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
986 
987 	if (wd->sc_dk.dk_openmask == 0) {
988 		wd_flushcache(wd, AT_WAIT);
989 
990 		if (! (wd->sc_flags & WDF_KLABEL))
991 			wd->sc_flags &= ~WDF_LOADED;
992 
993 		wd->atabus->ata_delref(wd->drvp);
994 	}
995 
996 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
997 	return 0;
998 }
999 
1000 void
1001 wdgetdefaultlabel(struct wd_softc *wd, struct disklabel *lp)
1002 {
1003 
1004 	ATADEBUG_PRINT(("wdgetdefaultlabel\n"), DEBUG_FUNCS);
1005 	memset(lp, 0, sizeof(struct disklabel));
1006 
1007 	lp->d_secsize = DEV_BSIZE;
1008 	lp->d_ntracks = wd->sc_params.atap_heads;
1009 	lp->d_nsectors = wd->sc_params.atap_sectors;
1010 	lp->d_ncylinders = (wd->sc_flags & WDF_LBA) ? wd->sc_capacity /
1011 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors) :
1012 		wd->sc_params.atap_cylinders;
1013 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1014 
1015 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
1016 		lp->d_type = DTYPE_ST506;
1017 	else
1018 		lp->d_type = DTYPE_ESDI;
1019 
1020 	strncpy(lp->d_typename, wd->sc_params.atap_model, 16);
1021 	strncpy(lp->d_packname, "fictitious", 16);
1022 	if (wd->sc_capacity > UINT32_MAX)
1023 		lp->d_secperunit = UINT32_MAX;
1024 	else
1025 		lp->d_secperunit = wd->sc_capacity;
1026 	lp->d_rpm = 3600;
1027 	lp->d_interleave = 1;
1028 	lp->d_flags = 0;
1029 
1030 	lp->d_partitions[RAW_PART].p_offset = 0;
1031 	lp->d_partitions[RAW_PART].p_size =
1032 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1033 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1034 	lp->d_npartitions = RAW_PART + 1;
1035 
1036 	lp->d_magic = DISKMAGIC;
1037 	lp->d_magic2 = DISKMAGIC;
1038 	lp->d_checksum = dkcksum(lp);
1039 }
1040 
1041 /*
1042  * Fabricate a default disk label, and try to read the correct one.
1043  */
1044 void
1045 wdgetdisklabel(struct wd_softc *wd)
1046 {
1047 	struct disklabel *lp = wd->sc_dk.dk_label;
1048 	const char *errstring;
1049 	int s;
1050 
1051 	ATADEBUG_PRINT(("wdgetdisklabel\n"), DEBUG_FUNCS);
1052 
1053 	memset(wd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1054 
1055 	wdgetdefaultlabel(wd, lp);
1056 
1057 	wd->sc_badsect[0] = -1;
1058 
1059 	if (wd->drvp->state > RESET) {
1060 		s = splbio();
1061 		wd->drvp->drive_flags |= DRIVE_RESET;
1062 		splx(s);
1063 	}
1064 	errstring = readdisklabel(MAKEWDDEV(0, wd->sc_dev.dv_unit, RAW_PART),
1065 	    wdstrategy, lp, wd->sc_dk.dk_cpulabel);
1066 	if (errstring) {
1067 		/*
1068 		 * This probably happened because the drive's default
1069 		 * geometry doesn't match the DOS geometry.  We
1070 		 * assume the DOS geometry is now in the label and try
1071 		 * again.  XXX This is a kluge.
1072 		 */
1073 		if (wd->drvp->state > RESET) {
1074 			s = splbio();
1075 			wd->drvp->drive_flags |= DRIVE_RESET;
1076 			splx(s);
1077 		}
1078 		errstring = readdisklabel(MAKEWDDEV(0, wd->sc_dev.dv_unit,
1079 		    RAW_PART), wdstrategy, lp, wd->sc_dk.dk_cpulabel);
1080 	}
1081 	if (errstring) {
1082 		printf("%s: %s\n", wd->sc_dev.dv_xname, errstring);
1083 		return;
1084 	}
1085 
1086 	if (wd->drvp->state > RESET) {
1087 		s = splbio();
1088 		wd->drvp->drive_flags |= DRIVE_RESET;
1089 		splx(s);
1090 	}
1091 #ifdef HAS_BAD144_HANDLING
1092 	if ((lp->d_flags & D_BADSECT) != 0)
1093 		bad144intern(wd);
1094 #endif
1095 }
1096 
1097 void
1098 wdperror(const struct wd_softc *wd)
1099 {
1100 	static const char *const errstr0_3[] = {"address mark not found",
1101 	    "track 0 not found", "aborted command", "media change requested",
1102 	    "id not found", "media changed", "uncorrectable data error",
1103 	    "bad block detected"};
1104 	static const char *const errstr4_5[] = {
1105 	    "obsolete (address mark not found)",
1106 	    "no media/write protected", "aborted command",
1107 	    "media change requested", "id not found", "media changed",
1108 	    "uncorrectable data error", "interface CRC error"};
1109 	const char *const *errstr;
1110 	int i;
1111 	const char *sep = "";
1112 
1113 	const char *devname = wd->sc_dev.dv_xname;
1114 	struct ata_drive_datas *drvp = wd->drvp;
1115 	int errno = wd->sc_wdc_bio.r_error;
1116 
1117 	if (drvp->ata_vers >= 4)
1118 		errstr = errstr4_5;
1119 	else
1120 		errstr = errstr0_3;
1121 
1122 	printf("%s: (", devname);
1123 
1124 	if (errno == 0)
1125 		printf("error not notified");
1126 
1127 	for (i = 0; i < 8; i++) {
1128 		if (errno & (1 << i)) {
1129 			printf("%s%s", sep, errstr[i]);
1130 			sep = ", ";
1131 		}
1132 	}
1133 	printf(")\n");
1134 }
1135 
1136 int
1137 wdioctl(dev_t dev, u_long xfer, caddr_t addr, int flag, struct lwp *l)
1138 {
1139 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
1140 	int error = 0, s;
1141 #ifdef __HAVE_OLD_DISKLABEL
1142 	struct disklabel *newlabel = NULL;
1143 #endif
1144 
1145 	ATADEBUG_PRINT(("wdioctl\n"), DEBUG_FUNCS);
1146 
1147 	if ((wd->sc_flags & WDF_LOADED) == 0)
1148 		return EIO;
1149 
1150 	switch (xfer) {
1151 #ifdef HAS_BAD144_HANDLING
1152 	case DIOCSBAD:
1153 		if ((flag & FWRITE) == 0)
1154 			return EBADF;
1155 		wd->sc_dk.dk_cpulabel->bad = *(struct dkbad *)addr;
1156 		wd->sc_dk.dk_label->d_flags |= D_BADSECT;
1157 		bad144intern(wd);
1158 		return 0;
1159 #endif
1160 #ifdef WD_SOFTBADSECT
1161 	case DIOCBSLIST :
1162 	{
1163 		u_int32_t count, missing, skip;
1164 		struct disk_badsecinfo dbsi;
1165 		struct disk_badsectors *dbs;
1166 		size_t available;
1167 		caddr_t laddr;
1168 
1169 		dbsi = *(struct disk_badsecinfo *)addr;
1170 		missing = wd->sc_bscount;
1171 		count = 0;
1172 		available = dbsi.dbsi_bufsize;
1173 		skip = dbsi.dbsi_skip;
1174 		laddr = dbsi.dbsi_buffer;
1175 
1176 		/*
1177 		 * We start this loop with the expectation that all of the
1178 		 * entries will be missed and decrement this counter each
1179 		 * time we either skip over one (already copied out) or
1180 		 * we actually copy it back to user space.  The structs
1181 		 * holding the bad sector information are copied directly
1182 		 * back to user space whilst the summary is returned via
1183 		 * the struct passed in via the ioctl.
1184 		 */
1185 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next) {
1186 			if (skip > 0) {
1187 				missing--;
1188 				skip--;
1189 				continue;
1190 			}
1191 			if (available < sizeof(*dbs))
1192 				break;
1193 			available -= sizeof(*dbs);
1194 			copyout(dbs, laddr, sizeof(*dbs));
1195 			laddr += sizeof(*dbs);
1196 			missing--;
1197 			count++;
1198 		}
1199 		dbsi.dbsi_left = missing;
1200 		dbsi.dbsi_copied = count;
1201 		*(struct disk_badsecinfo *)addr = dbsi;
1202 		return 0;
1203 	}
1204 
1205 	case DIOCBSFLUSH :
1206 		/* Clean out the bad sector list */
1207 		while (!SLIST_EMPTY(&wd->sc_bslist)) {
1208 			void *head = SLIST_FIRST(&wd->sc_bslist);
1209 			SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
1210 			free(head, M_TEMP);
1211 		}
1212 		wd->sc_bscount = 0;
1213 		return 0;
1214 #endif
1215 	case DIOCGDINFO:
1216 		*(struct disklabel *)addr = *(wd->sc_dk.dk_label);
1217 		return 0;
1218 #ifdef __HAVE_OLD_DISKLABEL
1219 	case ODIOCGDINFO:
1220 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1221 		if (newlabel == NULL)
1222 			return EIO;
1223 		*newlabel = *(wd->sc_dk.dk_label);
1224 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1225 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1226 		else
1227 			error = ENOTTY;
1228 		free(newlabel, M_TEMP);
1229 		return error;
1230 #endif
1231 
1232 	case DIOCGPART:
1233 		((struct partinfo *)addr)->disklab = wd->sc_dk.dk_label;
1234 		((struct partinfo *)addr)->part =
1235 		    &wd->sc_dk.dk_label->d_partitions[WDPART(dev)];
1236 		return 0;
1237 
1238 	case DIOCWDINFO:
1239 	case DIOCSDINFO:
1240 #ifdef __HAVE_OLD_DISKLABEL
1241 	case ODIOCWDINFO:
1242 	case ODIOCSDINFO:
1243 #endif
1244 	{
1245 		struct disklabel *lp;
1246 
1247 		if ((flag & FWRITE) == 0)
1248 			return EBADF;
1249 
1250 #ifdef __HAVE_OLD_DISKLABEL
1251 		if (xfer == ODIOCSDINFO || xfer == ODIOCWDINFO) {
1252 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1253 			if (newlabel == NULL)
1254 				return EIO;
1255 			memset(newlabel, 0, sizeof newlabel);
1256 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1257 			lp = newlabel;
1258 		} else
1259 #endif
1260 		lp = (struct disklabel *)addr;
1261 
1262 		if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE,
1263 				     NULL)) != 0)
1264 			goto bad;
1265 		wd->sc_flags |= WDF_LABELLING;
1266 
1267 		error = setdisklabel(wd->sc_dk.dk_label,
1268 		    lp, /*wd->sc_dk.dk_openmask : */0,
1269 		    wd->sc_dk.dk_cpulabel);
1270 		if (error == 0) {
1271 			if (wd->drvp->state > RESET) {
1272 				s = splbio();
1273 				wd->drvp->drive_flags |= DRIVE_RESET;
1274 				splx(s);
1275 			}
1276 			if (xfer == DIOCWDINFO
1277 #ifdef __HAVE_OLD_DISKLABEL
1278 			    || xfer == ODIOCWDINFO
1279 #endif
1280 			    )
1281 				error = writedisklabel(WDLABELDEV(dev),
1282 				    wdstrategy, wd->sc_dk.dk_label,
1283 				    wd->sc_dk.dk_cpulabel);
1284 		}
1285 
1286 		wd->sc_flags &= ~WDF_LABELLING;
1287 		(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
1288 bad:
1289 #ifdef __HAVE_OLD_DISKLABEL
1290 		if (newlabel != NULL)
1291 			free(newlabel, M_TEMP);
1292 #endif
1293 		return error;
1294 	}
1295 
1296 	case DIOCKLABEL:
1297 		if (*(int *)addr)
1298 			wd->sc_flags |= WDF_KLABEL;
1299 		else
1300 			wd->sc_flags &= ~WDF_KLABEL;
1301 		return 0;
1302 
1303 	case DIOCWLABEL:
1304 		if ((flag & FWRITE) == 0)
1305 			return EBADF;
1306 		if (*(int *)addr)
1307 			wd->sc_flags |= WDF_WLABEL;
1308 		else
1309 			wd->sc_flags &= ~WDF_WLABEL;
1310 		return 0;
1311 
1312 	case DIOCGDEFLABEL:
1313 		wdgetdefaultlabel(wd, (struct disklabel *)addr);
1314 		return 0;
1315 #ifdef __HAVE_OLD_DISKLABEL
1316 	case ODIOCGDEFLABEL:
1317 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1318 		if (newlabel == NULL)
1319 			return EIO;
1320 		wdgetdefaultlabel(wd, newlabel);
1321 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1322 			memcpy(addr, &newlabel, sizeof (struct olddisklabel));
1323 		else
1324 			error = ENOTTY;
1325 		free(newlabel, M_TEMP);
1326 		return error;
1327 #endif
1328 
1329 #ifdef notyet
1330 	case DIOCWFORMAT:
1331 		if ((flag & FWRITE) == 0)
1332 			return EBADF;
1333 		{
1334 		register struct format_op *fop;
1335 		struct iovec aiov;
1336 		struct uio auio;
1337 
1338 		fop = (struct format_op *)addr;
1339 		aiov.iov_base = fop->df_buf;
1340 		aiov.iov_len = fop->df_count;
1341 		auio.uio_iov = &aiov;
1342 		auio.uio_iovcnt = 1;
1343 		auio.uio_resid = fop->df_count;
1344 		auio.uio_segflg = 0;
1345 		auio.uio_offset =
1346 			fop->df_startblk * wd->sc_dk.dk_label->d_secsize;
1347 		auio.uio_lwp = l;
1348 		error = physio(wdformat, NULL, dev, B_WRITE, minphys,
1349 		    &auio);
1350 		fop->df_count -= auio.uio_resid;
1351 		fop->df_reg[0] = wdc->sc_status;
1352 		fop->df_reg[1] = wdc->sc_error;
1353 		return error;
1354 		}
1355 #endif
1356 	case DIOCGCACHE:
1357 		return wd_getcache(wd, (int *)addr);
1358 
1359 	case DIOCSCACHE:
1360 		return wd_setcache(wd, *(int *)addr);
1361 
1362 	case DIOCCACHESYNC:
1363 		return wd_flushcache(wd, AT_WAIT);
1364 
1365 	case ATAIOCCOMMAND:
1366 		/*
1367 		 * Make sure this command is (relatively) safe first
1368 		 */
1369 		if ((((atareq_t *) addr)->flags & ATACMD_READ) == 0 &&
1370 		    (flag & FWRITE) == 0)
1371 			return (EBADF);
1372 		{
1373 		struct wd_ioctl *wi;
1374 		atareq_t *atareq = (atareq_t *) addr;
1375 		int error1;
1376 
1377 		wi = wi_get();
1378 		wi->wi_softc = wd;
1379 		wi->wi_atareq = *atareq;
1380 
1381 		if (atareq->datalen && atareq->flags &
1382 		    (ATACMD_READ | ATACMD_WRITE)) {
1383 			wi->wi_iov.iov_base = atareq->databuf;
1384 			wi->wi_iov.iov_len = atareq->datalen;
1385 			wi->wi_uio.uio_iov = &wi->wi_iov;
1386 			wi->wi_uio.uio_iovcnt = 1;
1387 			wi->wi_uio.uio_resid = atareq->datalen;
1388 			wi->wi_uio.uio_offset = 0;
1389 			wi->wi_uio.uio_segflg = UIO_USERSPACE;
1390 			wi->wi_uio.uio_rw =
1391 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE;
1392 			wi->wi_uio.uio_lwp = l;
1393 			error1 = physio(wdioctlstrategy, &wi->wi_bp, dev,
1394 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE,
1395 			    minphys, &wi->wi_uio);
1396 		} else {
1397 			/* No need to call physio if we don't have any
1398 			   user data */
1399 			wi->wi_bp.b_flags = 0;
1400 			wi->wi_bp.b_data = 0;
1401 			wi->wi_bp.b_bcount = 0;
1402 			wi->wi_bp.b_dev = 0;
1403 			wi->wi_bp.b_proc = l->l_proc;
1404 			wdioctlstrategy(&wi->wi_bp);
1405 			error1 = wi->wi_bp.b_error;
1406 		}
1407 		*atareq = wi->wi_atareq;
1408 		wi_free(wi);
1409 		return(error1);
1410 		}
1411 
1412 	case DIOCAWEDGE:
1413 	    {
1414 	    	struct dkwedge_info *dkw = (void *) addr;
1415 
1416 		if ((flag & FWRITE) == 0)
1417 			return (EBADF);
1418 
1419 		/* If the ioctl happens here, the parent is us. */
1420 		strcpy(dkw->dkw_parent, wd->sc_dev.dv_xname);
1421 		return (dkwedge_add(dkw));
1422 	    }
1423 
1424 	case DIOCDWEDGE:
1425 	    {
1426 	    	struct dkwedge_info *dkw = (void *) addr;
1427 
1428 		if ((flag & FWRITE) == 0)
1429 			return (EBADF);
1430 
1431 		/* If the ioctl happens here, the parent is us. */
1432 		strcpy(dkw->dkw_parent, wd->sc_dev.dv_xname);
1433 		return (dkwedge_del(dkw));
1434 	    }
1435 
1436 	case DIOCLWEDGES:
1437 	    {
1438 	    	struct dkwedge_list *dkwl = (void *) addr;
1439 
1440 		return (dkwedge_list(&wd->sc_dk, dkwl, l));
1441 	    }
1442 
1443 	default:
1444 		return ENOTTY;
1445 	}
1446 
1447 #ifdef DIAGNOSTIC
1448 	panic("wdioctl: impossible");
1449 #endif
1450 }
1451 
1452 #ifdef B_FORMAT
1453 int
1454 wdformat(struct buf *bp)
1455 {
1456 
1457 	bp->b_flags |= B_FORMAT;
1458 	return wdstrategy(bp);
1459 }
1460 #endif
1461 
1462 int
1463 wdsize(dev_t dev)
1464 {
1465 	struct wd_softc *wd;
1466 	int part, omask;
1467 	int size;
1468 
1469 	ATADEBUG_PRINT(("wdsize\n"), DEBUG_FUNCS);
1470 
1471 	wd = device_lookup(&wd_cd, WDUNIT(dev));
1472 	if (wd == NULL)
1473 		return (-1);
1474 
1475 	part = WDPART(dev);
1476 	omask = wd->sc_dk.dk_openmask & (1 << part);
1477 
1478 	if (omask == 0 && wdopen(dev, 0, S_IFBLK, NULL) != 0)
1479 		return (-1);
1480 	if (wd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1481 		size = -1;
1482 	else
1483 		size = wd->sc_dk.dk_label->d_partitions[part].p_size *
1484 		    (wd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1485 	if (omask == 0 && wdclose(dev, 0, S_IFBLK, NULL) != 0)
1486 		return (-1);
1487 	return (size);
1488 }
1489 
1490 /* #define WD_DUMP_NOT_TRUSTED if you just want to watch */
1491 static int wddoingadump = 0;
1492 static int wddumprecalibrated = 0;
1493 
1494 /*
1495  * Dump core after a system crash.
1496  */
1497 int
1498 wddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1499 {
1500 	struct wd_softc *wd;	/* disk unit to do the I/O */
1501 	struct disklabel *lp;   /* disk's disklabel */
1502 	int part, err;
1503 	int nblks;	/* total number of sectors left to write */
1504 
1505 	/* Check if recursive dump; if so, punt. */
1506 	if (wddoingadump)
1507 		return EFAULT;
1508 	wddoingadump = 1;
1509 
1510 	wd = device_lookup(&wd_cd, WDUNIT(dev));
1511 	if (wd == NULL)
1512 		return (ENXIO);
1513 
1514 	part = WDPART(dev);
1515 
1516 	/* Convert to disk sectors.  Request must be a multiple of size. */
1517 	lp = wd->sc_dk.dk_label;
1518 	if ((size % lp->d_secsize) != 0)
1519 		return EFAULT;
1520 	nblks = size / lp->d_secsize;
1521 	blkno = blkno / (lp->d_secsize / DEV_BSIZE);
1522 
1523 	/* Check transfer bounds against partition size. */
1524 	if ((blkno < 0) || ((blkno + nblks) > lp->d_partitions[part].p_size))
1525 		return EINVAL;
1526 
1527 	/* Offset block number to start of partition. */
1528 	blkno += lp->d_partitions[part].p_offset;
1529 
1530 	/* Recalibrate, if first dump transfer. */
1531 	if (wddumprecalibrated == 0) {
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 	wd->sc_bp = NULL;
1539 	wd->sc_wdc_bio.blkno = blkno;
1540 	wd->sc_wdc_bio.flags = ATA_POLL;
1541 	if (wd->sc_flags & WDF_LBA48 &&
1542 	    (blkno > LBA48_THRESHOLD ||
1543     	    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) != 0))
1544 		wd->sc_wdc_bio.flags |= ATA_LBA48;
1545 	if (wd->sc_flags & WDF_LBA)
1546 		wd->sc_wdc_bio.flags |= ATA_LBA;
1547 	wd->sc_wdc_bio.bcount = nblks * lp->d_secsize;
1548 	wd->sc_wdc_bio.databuf = va;
1549 #ifndef WD_DUMP_NOT_TRUSTED
1550 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
1551 	case ATACMD_TRY_AGAIN:
1552 		panic("wddump: try again");
1553 		break;
1554 	case ATACMD_QUEUED:
1555 		panic("wddump: polled command has been queued");
1556 		break;
1557 	case ATACMD_COMPLETE:
1558 		break;
1559 	}
1560 	switch(wd->sc_wdc_bio.error) {
1561 	case TIMEOUT:
1562 		printf("wddump: device timed out");
1563 		err = EIO;
1564 		break;
1565 	case ERR_DF:
1566 		printf("wddump: drive fault");
1567 		err = EIO;
1568 		break;
1569 	case ERR_DMA:
1570 		printf("wddump: DMA error");
1571 		err = EIO;
1572 		break;
1573 	case ERROR:
1574 		printf("wddump: ");
1575 		wdperror(wd);
1576 		err = EIO;
1577 		break;
1578 	case NOERROR:
1579 		err = 0;
1580 		break;
1581 	default:
1582 		panic("wddump: unknown error type");
1583 	}
1584 	if (err != 0) {
1585 		printf("\n");
1586 		return err;
1587 	}
1588 #else	/* WD_DUMP_NOT_TRUSTED */
1589 	/* Let's just talk about this first... */
1590 	printf("wd%d: dump addr 0x%x, cylin %d, head %d, sector %d\n",
1591 	    unit, va, cylin, head, sector);
1592 	delay(500 * 1000);	/* half a second */
1593 #endif
1594 
1595 	wddoingadump = 0;
1596 	return 0;
1597 }
1598 
1599 #ifdef HAS_BAD144_HANDLING
1600 /*
1601  * Internalize the bad sector table.
1602  */
1603 void
1604 bad144intern(struct wd_softc *wd)
1605 {
1606 	struct dkbad *bt = &wd->sc_dk.dk_cpulabel->bad;
1607 	struct disklabel *lp = wd->sc_dk.dk_label;
1608 	int i = 0;
1609 
1610 	ATADEBUG_PRINT(("bad144intern\n"), DEBUG_XFERS);
1611 
1612 	for (; i < NBT_BAD; i++) {
1613 		if (bt->bt_bad[i].bt_cyl == 0xffff)
1614 			break;
1615 		wd->sc_badsect[i] =
1616 		    bt->bt_bad[i].bt_cyl * lp->d_secpercyl +
1617 		    (bt->bt_bad[i].bt_trksec >> 8) * lp->d_nsectors +
1618 		    (bt->bt_bad[i].bt_trksec & 0xff);
1619 	}
1620 	for (; i < NBT_BAD+1; i++)
1621 		wd->sc_badsect[i] = -1;
1622 }
1623 #endif
1624 
1625 int
1626 wd_get_params(struct wd_softc *wd, u_int8_t flags, struct ataparams *params)
1627 {
1628 	switch (wd->atabus->ata_get_params(wd->drvp, flags, params)) {
1629 	case CMD_AGAIN:
1630 		return 1;
1631 	case CMD_ERR:
1632 		/*
1633 		 * We `know' there's a drive here; just assume it's old.
1634 		 * This geometry is only used to read the MBR and print a
1635 		 * (false) attach message.
1636 		 */
1637 		strncpy(params->atap_model, "ST506",
1638 		    sizeof params->atap_model);
1639 		params->atap_config = ATA_CFG_FIXED;
1640 		params->atap_cylinders = 1024;
1641 		params->atap_heads = 8;
1642 		params->atap_sectors = 17;
1643 		params->atap_multi = 1;
1644 		params->atap_capabilities1 = params->atap_capabilities2 = 0;
1645 		wd->drvp->ata_vers = -1; /* Mark it as pre-ATA */
1646 		return 0;
1647 	case CMD_OK:
1648 		return 0;
1649 	default:
1650 		panic("wd_get_params: bad return code from ata_get_params");
1651 		/* NOTREACHED */
1652 	}
1653 }
1654 
1655 int
1656 wd_getcache(struct wd_softc *wd, int *bitsp)
1657 {
1658 	struct ataparams params;
1659 
1660 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
1661 		return EIO;
1662 	if (params.atap_cmd_set1 == 0x0000 ||
1663 	    params.atap_cmd_set1 == 0xffff ||
1664 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0) {
1665 		*bitsp = 0;
1666 		return 0;
1667 	}
1668 	*bitsp = DKCACHE_WCHANGE | DKCACHE_READ;
1669 	if (params.atap_cmd1_en & WDC_CMD1_CACHE)
1670 		*bitsp |= DKCACHE_WRITE;
1671 
1672 	return 0;
1673 }
1674 
1675 const char at_errbits[] = "\20\10ERROR\11TIMEOU\12DF";
1676 
1677 int
1678 wd_setcache(struct wd_softc *wd, int bits)
1679 {
1680 	struct ataparams params;
1681 	struct ata_command ata_c;
1682 
1683 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
1684 		return EIO;
1685 
1686 	if (params.atap_cmd_set1 == 0x0000 ||
1687 	    params.atap_cmd_set1 == 0xffff ||
1688 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0)
1689 		return EOPNOTSUPP;
1690 
1691 	if ((bits & DKCACHE_READ) == 0 ||
1692 	    (bits & DKCACHE_SAVE) != 0)
1693 		return EOPNOTSUPP;
1694 
1695 	memset(&ata_c, 0, sizeof(struct ata_command));
1696 	ata_c.r_command = SET_FEATURES;
1697 	ata_c.r_st_bmask = 0;
1698 	ata_c.r_st_pmask = 0;
1699 	ata_c.timeout = 30000; /* 30s timeout */
1700 	ata_c.flags = AT_WAIT;
1701 	if (bits & DKCACHE_WRITE)
1702 		ata_c.r_features = WDSF_WRITE_CACHE_EN;
1703 	else
1704 		ata_c.r_features = WDSF_WRITE_CACHE_DS;
1705 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1706 		printf("%s: wd_setcache command not complete\n",
1707 		    wd->sc_dev.dv_xname);
1708 		return EIO;
1709 	}
1710 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1711 		char sbuf[sizeof(at_errbits) + 64];
1712 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
1713 		printf("%s: wd_setcache: status=%s\n", wd->sc_dev.dv_xname,
1714 		    sbuf);
1715 		return EIO;
1716 	}
1717 	return 0;
1718 }
1719 
1720 int
1721 wd_standby(struct wd_softc *wd, int flags)
1722 {
1723 	struct ata_command ata_c;
1724 
1725 	memset(&ata_c, 0, sizeof(struct ata_command));
1726 	ata_c.r_command = WDCC_STANDBY_IMMED;
1727 	ata_c.r_st_bmask = WDCS_DRDY;
1728 	ata_c.r_st_pmask = WDCS_DRDY;
1729 	ata_c.flags = flags;
1730 	ata_c.timeout = 30000; /* 30s timeout */
1731 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1732 		printf("%s: standby immediate command didn't complete\n",
1733 		    wd->sc_dev.dv_xname);
1734 		return EIO;
1735 	}
1736 	if (ata_c.flags & AT_ERROR) {
1737 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
1738 			return ENODEV;
1739 	}
1740 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1741 		char sbuf[sizeof(at_errbits) + 64];
1742 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
1743 		printf("%s: wd_standby: status=%s\n", wd->sc_dev.dv_xname,
1744 		    sbuf);
1745 		return EIO;
1746 	}
1747 	return 0;
1748 }
1749 
1750 int
1751 wd_flushcache(struct wd_softc *wd, int flags)
1752 {
1753 	struct ata_command ata_c;
1754 
1755 	/*
1756 	 * WDCC_FLUSHCACHE is here since ATA-4, but some drives report
1757 	 * only ATA-2 and still support it.
1758 	 */
1759 	if (wd->drvp->ata_vers < 4 &&
1760 	    ((wd->sc_params.atap_cmd_set2 & WDC_CMD2_FC) == 0 ||
1761 	    wd->sc_params.atap_cmd_set2 == 0xffff))
1762 		return ENODEV;
1763 	memset(&ata_c, 0, sizeof(struct ata_command));
1764 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0 &&
1765 	    (wd->sc_params.atap_cmd2_en & ATA_CMD2_FCE) != 0)
1766 		ata_c.r_command = WDCC_FLUSHCACHE_EXT;
1767 	else
1768 		ata_c.r_command = WDCC_FLUSHCACHE;
1769 	ata_c.r_st_bmask = WDCS_DRDY;
1770 	ata_c.r_st_pmask = WDCS_DRDY;
1771 	ata_c.flags = flags;
1772 	ata_c.timeout = 30000; /* 30s timeout */
1773 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1774 		printf("%s: flush cache command didn't complete\n",
1775 		    wd->sc_dev.dv_xname);
1776 		return EIO;
1777 	}
1778 	if (ata_c.flags & AT_ERROR) {
1779 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
1780 			return ENODEV;
1781 	}
1782 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1783 		char sbuf[sizeof(at_errbits) + 64];
1784 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
1785 		printf("%s: wd_flushcache: status=%s\n", wd->sc_dev.dv_xname,
1786 		    sbuf);
1787 		return EIO;
1788 	}
1789 	return 0;
1790 }
1791 
1792 void
1793 wd_shutdown(void *arg)
1794 {
1795 	struct wd_softc *wd = arg;
1796 	wd_flushcache(wd, AT_POLL);
1797 }
1798 
1799 /*
1800  * Allocate space for a ioctl queue structure.  Mostly taken from
1801  * scsipi_ioctl.c
1802  */
1803 struct wd_ioctl *
1804 wi_get(void)
1805 {
1806 	struct wd_ioctl *wi;
1807 	int s;
1808 
1809 	wi = malloc(sizeof(struct wd_ioctl), M_TEMP, M_WAITOK|M_ZERO);
1810 	simple_lock_init(&wi->wi_bp.b_interlock);
1811 	s = splbio();
1812 	LIST_INSERT_HEAD(&wi_head, wi, wi_list);
1813 	splx(s);
1814 	return (wi);
1815 }
1816 
1817 /*
1818  * Free an ioctl structure and remove it from our list
1819  */
1820 
1821 void
1822 wi_free(struct wd_ioctl *wi)
1823 {
1824 	int s;
1825 
1826 	s = splbio();
1827 	LIST_REMOVE(wi, wi_list);
1828 	splx(s);
1829 	free(wi, M_TEMP);
1830 }
1831 
1832 /*
1833  * Find a wd_ioctl structure based on the struct buf.
1834  */
1835 
1836 struct wd_ioctl *
1837 wi_find(struct buf *bp)
1838 {
1839 	struct wd_ioctl *wi;
1840 	int s;
1841 
1842 	s = splbio();
1843 	for (wi = wi_head.lh_first; wi != 0; wi = wi->wi_list.le_next)
1844 		if (bp == &wi->wi_bp)
1845 			break;
1846 	splx(s);
1847 	return (wi);
1848 }
1849 
1850 /*
1851  * Ioctl pseudo strategy routine
1852  *
1853  * This is mostly stolen from scsipi_ioctl.c:scsistrategy().  What
1854  * happens here is:
1855  *
1856  * - wdioctl() queues a wd_ioctl structure.
1857  *
1858  * - wdioctl() calls physio/wdioctlstrategy based on whether or not
1859  *   user space I/O is required.  If physio() is called, physio() eventually
1860  *   calls wdioctlstrategy().
1861  *
1862  * - In either case, wdioctlstrategy() calls wd->atabus->ata_exec_command()
1863  *   to perform the actual command
1864  *
1865  * The reason for the use of the pseudo strategy routine is because
1866  * when doing I/O to/from user space, physio _really_ wants to be in
1867  * the loop.  We could put the entire buffer into the ioctl request
1868  * structure, but that won't scale if we want to do things like download
1869  * microcode.
1870  */
1871 
1872 void
1873 wdioctlstrategy(struct buf *bp)
1874 {
1875 	struct wd_ioctl *wi;
1876 	struct ata_command ata_c;
1877 	int error = 0;
1878 
1879 	wi = wi_find(bp);
1880 	if (wi == NULL) {
1881 		printf("wdioctlstrategy: "
1882 		    "No matching ioctl request found in queue\n");
1883 		error = EINVAL;
1884 		goto bad;
1885 	}
1886 
1887 	memset(&ata_c, 0, sizeof(ata_c));
1888 
1889 	/*
1890 	 * Abort if physio broke up the transfer
1891 	 */
1892 
1893 	if (bp->b_bcount != wi->wi_atareq.datalen) {
1894 		printf("physio split wd ioctl request... cannot proceed\n");
1895 		error = EIO;
1896 		goto bad;
1897 	}
1898 
1899 	/*
1900 	 * Abort if we didn't get a buffer size that was a multiple of
1901 	 * our sector size (or was larger than NBBY)
1902 	 */
1903 
1904 	if ((bp->b_bcount % wi->wi_softc->sc_dk.dk_label->d_secsize) != 0 ||
1905 	    (bp->b_bcount / wi->wi_softc->sc_dk.dk_label->d_secsize) >=
1906 	     (1 << NBBY)) {
1907 		error = EINVAL;
1908 		goto bad;
1909 	}
1910 
1911 	/*
1912 	 * Make sure a timeout was supplied in the ioctl request
1913 	 */
1914 
1915 	if (wi->wi_atareq.timeout == 0) {
1916 		error = EINVAL;
1917 		goto bad;
1918 	}
1919 
1920 	if (wi->wi_atareq.flags & ATACMD_READ)
1921 		ata_c.flags |= AT_READ;
1922 	else if (wi->wi_atareq.flags & ATACMD_WRITE)
1923 		ata_c.flags |= AT_WRITE;
1924 
1925 	if (wi->wi_atareq.flags & ATACMD_READREG)
1926 		ata_c.flags |= AT_READREG;
1927 
1928 	ata_c.flags |= AT_WAIT;
1929 
1930 	ata_c.timeout = wi->wi_atareq.timeout;
1931 	ata_c.r_command = wi->wi_atareq.command;
1932 	ata_c.r_head = wi->wi_atareq.head & 0x0f;
1933 	ata_c.r_cyl = wi->wi_atareq.cylinder;
1934 	ata_c.r_sector = wi->wi_atareq.sec_num;
1935 	ata_c.r_count = wi->wi_atareq.sec_count;
1936 	ata_c.r_features = wi->wi_atareq.features;
1937 	ata_c.r_st_bmask = WDCS_DRDY;
1938 	ata_c.r_st_pmask = WDCS_DRDY;
1939 	ata_c.data = wi->wi_bp.b_data;
1940 	ata_c.bcount = wi->wi_bp.b_bcount;
1941 
1942 	if (wi->wi_softc->atabus->ata_exec_command(wi->wi_softc->drvp, &ata_c)
1943 	    != ATACMD_COMPLETE) {
1944 		wi->wi_atareq.retsts = ATACMD_ERROR;
1945 		goto bad;
1946 	}
1947 
1948 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1949 		if (ata_c.flags & AT_ERROR) {
1950 			wi->wi_atareq.retsts = ATACMD_ERROR;
1951 			wi->wi_atareq.error = ata_c.r_error;
1952 		} else if (ata_c.flags & AT_DF)
1953 			wi->wi_atareq.retsts = ATACMD_DF;
1954 		else
1955 			wi->wi_atareq.retsts = ATACMD_TIMEOUT;
1956 	} else {
1957 		wi->wi_atareq.retsts = ATACMD_OK;
1958 		if (wi->wi_atareq.flags & ATACMD_READREG) {
1959 			wi->wi_atareq.head = ata_c.r_head ;
1960 			wi->wi_atareq.cylinder = ata_c.r_cyl;
1961 			wi->wi_atareq.sec_num = ata_c.r_sector;
1962 			wi->wi_atareq.sec_count = ata_c.r_count;
1963 			wi->wi_atareq.features = ata_c.r_features;
1964 			wi->wi_atareq.error = ata_c.r_error;
1965 		}
1966 	}
1967 
1968 	bp->b_error = 0;
1969 	biodone(bp);
1970 	return;
1971 bad:
1972 	bp->b_flags |= B_ERROR;
1973 	bp->b_error = error;
1974 	biodone(bp);
1975 }
1976