xref: /netbsd-src/sys/dev/ata/wd.c (revision 1ad9454efb13a65cd7535ccf867508cb14d9d30e)
1 /*	$NetBSD: wd.c,v 1.330 2006/09/22 04:48:38 thorpej 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 1.330 2006/09/22 04:48:38 thorpej 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 #include <prop/proplib.h>
109 
110 #define	LBA48_THRESHOLD		(0xfffffff)	/* 128GB / DEV_BSIZE */
111 
112 #define	WDIORETRIES_SINGLE 4	/* number of retries before single-sector */
113 #define	WDIORETRIES	5	/* number of retries before giving up */
114 #define	RECOVERYTIME hz/2	/* time to wait before retrying a cmd */
115 
116 #define	WDUNIT(dev)		DISKUNIT(dev)
117 #define	WDPART(dev)		DISKPART(dev)
118 #define	WDMINOR(unit, part)	DISKMINOR(unit, part)
119 #define	MAKEWDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
120 
121 #define	WDLABELDEV(dev)	(MAKEWDDEV(major(dev), WDUNIT(dev), RAW_PART))
122 
123 #define DEBUG_INTR   0x01
124 #define DEBUG_XFERS  0x02
125 #define DEBUG_STATUS 0x04
126 #define DEBUG_FUNCS  0x08
127 #define DEBUG_PROBE  0x10
128 #ifdef ATADEBUG
129 int wdcdebug_wd_mask = 0x0;
130 #define ATADEBUG_PRINT(args, level) \
131 	if (wdcdebug_wd_mask & (level)) \
132 		printf args
133 #else
134 #define ATADEBUG_PRINT(args, level)
135 #endif
136 
137 int	wdprobe(struct device *, struct cfdata *, void *);
138 void	wdattach(struct device *, struct device *, void *);
139 int	wddetach(struct device *, int);
140 int	wdactivate(struct device *, enum devact);
141 int	wdprint(void *, char *);
142 void	wdperror(const struct wd_softc *);
143 
144 CFATTACH_DECL(wd, sizeof(struct wd_softc),
145     wdprobe, wdattach, wddetach, wdactivate);
146 
147 extern struct cfdriver wd_cd;
148 
149 dev_type_open(wdopen);
150 dev_type_close(wdclose);
151 dev_type_read(wdread);
152 dev_type_write(wdwrite);
153 dev_type_ioctl(wdioctl);
154 dev_type_strategy(wdstrategy);
155 dev_type_dump(wddump);
156 dev_type_size(wdsize);
157 
158 const struct bdevsw wd_bdevsw = {
159 	wdopen, wdclose, wdstrategy, wdioctl, wddump, wdsize, D_DISK
160 };
161 
162 const struct cdevsw wd_cdevsw = {
163 	wdopen, wdclose, wdread, wdwrite, wdioctl,
164 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
165 };
166 
167 /*
168  * Glue necessary to hook WDCIOCCOMMAND into physio
169  */
170 
171 struct wd_ioctl {
172 	LIST_ENTRY(wd_ioctl) wi_list;
173 	struct buf wi_bp;
174 	struct uio wi_uio;
175 	struct iovec wi_iov;
176 	atareq_t wi_atareq;
177 	struct wd_softc *wi_softc;
178 };
179 
180 LIST_HEAD(, wd_ioctl) wi_head;
181 
182 struct	wd_ioctl *wi_find(struct buf *);
183 void	wi_free(struct wd_ioctl *);
184 struct	wd_ioctl *wi_get(void);
185 void	wdioctlstrategy(struct buf *);
186 
187 void  wdgetdefaultlabel(struct wd_softc *, struct disklabel *);
188 void  wdgetdisklabel(struct wd_softc *);
189 void  wdstart(void *);
190 void  __wdstart(struct wd_softc*, struct buf *);
191 void  wdrestart(void *);
192 void  wddone(void *);
193 int   wd_get_params(struct wd_softc *, u_int8_t, struct ataparams *);
194 int   wd_standby(struct wd_softc *, int);
195 int   wd_flushcache(struct wd_softc *, int);
196 void  wd_shutdown(void *);
197 
198 int   wd_getcache(struct wd_softc *, int *);
199 int   wd_setcache(struct wd_softc *, int);
200 
201 struct dkdriver wddkdriver = { wdstrategy, minphys };
202 
203 #ifdef HAS_BAD144_HANDLING
204 static void bad144intern(struct wd_softc *);
205 #endif
206 
207 #define	WD_QUIRK_SPLIT_MOD15_WRITE	0x0001	/* must split certain writes */
208 #define	WD_QUIRK_FORCE_LBA48		0x0002	/* must use LBA48 commands */
209 
210 #define	WD_QUIRK_FMT "\20\1SPLIT_MOD15_WRITE\2FORCE_LBA48"
211 
212 /*
213  * Quirk table for IDE drives.  Put more-specific matches first, since
214  * a simple globbing routine is used for matching.
215  */
216 static const struct wd_quirk {
217 	const char *wdq_match;		/* inquiry pattern to match */
218 	int wdq_quirks;			/* drive quirks */
219 } wd_quirk_table[] = {
220 	/*
221 	 * Some Seagate S-ATA drives have a PHY which can get confused
222 	 * with the way data is packetized by some S-ATA controllers.
223 	 *
224 	 * The work-around is to split in two any write transfer whose
225 	 * sector count % 15 == 1 (assuming 512 byte sectors).
226 	 *
227 	 * XXX This is an incomplete list.  There are at least a couple
228 	 * XXX more model numbers.  If you have trouble with such transfers
229 	 * XXX (8K is the most common) on Seagate S-ATA drives, please
230 	 * XXX notify thorpej@NetBSD.org.
231 	 */
232 	{ "ST3120023AS",
233 	  WD_QUIRK_SPLIT_MOD15_WRITE },
234 	{ "ST380023AS",
235 	  WD_QUIRK_SPLIT_MOD15_WRITE },
236 
237 	/*
238 	 * These seagate drives seems to have issue addressing sector 0xfffffff
239 	 * (aka LBA48_THRESHOLD) in LBA mode. The workaround is to force
240 	 * LBA48
241 	 * Note that we can't just change the code to always use LBA48 for
242 	 * sector 0xfffffff, because this would break valid and working
243 	 * setups using LBA48 drives on non-LBA48-capable controllers
244 	 * (and it's hard to get a list of such controllers)
245 	 */
246 	{ "ST3160021A*",
247 	  WD_QUIRK_FORCE_LBA48 },
248 	{ "ST3160812A*",
249 	  WD_QUIRK_FORCE_LBA48 },
250 	{ "ST3160023A*",
251 	  WD_QUIRK_FORCE_LBA48 },
252 	{ "ST3160827A*",
253 	  WD_QUIRK_FORCE_LBA48 },
254 	/* Attempt to catch all seagate drives larger than 200GB */
255 	{ "ST3[2-9][0-9][0-9][0-9][0-9][0-9][A-Z]*",
256 	  WD_QUIRK_FORCE_LBA48 },
257 	{ NULL,
258 	  0 }
259 };
260 
261 static const struct wd_quirk *
262 wd_lookup_quirks(const char *name)
263 {
264 	const struct wd_quirk *wdq;
265 	const char *estr;
266 
267 	for (wdq = wd_quirk_table; wdq->wdq_match != NULL; wdq++) {
268 		/*
269 		 * We only want exact matches (which include matches
270 		 * against globbing characters).
271 		 */
272 		if (pmatch(name, wdq->wdq_match, &estr) == 2)
273 			return (wdq);
274 	}
275 	return (NULL);
276 }
277 
278 int
279 wdprobe(struct device *parent, struct cfdata *match, void *aux)
280 {
281 	struct ata_device *adev = aux;
282 
283 	if (adev == NULL)
284 		return 0;
285 	if (adev->adev_bustype->bustype_type != SCSIPI_BUSTYPE_ATA)
286 		return 0;
287 
288 	if (match->cf_loc[ATA_HLCF_DRIVE] != ATA_HLCF_DRIVE_DEFAULT &&
289 	    match->cf_loc[ATA_HLCF_DRIVE] != adev->adev_drv_data->drive)
290 		return 0;
291 	return 1;
292 }
293 
294 void
295 wdattach(struct device *parent, struct device *self, void *aux)
296 {
297 	struct wd_softc *wd = (void *)self;
298 	struct ata_device *adev= aux;
299 	int i, blank;
300 	char tbuf[41], pbuf[9], c, *p, *q;
301 	const struct wd_quirk *wdq;
302 	ATADEBUG_PRINT(("wdattach\n"), DEBUG_FUNCS | DEBUG_PROBE);
303 
304 	callout_init(&wd->sc_restart_ch);
305 	bufq_alloc(&wd->sc_q, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
306 #ifdef WD_SOFTBADSECT
307 	SLIST_INIT(&wd->sc_bslist);
308 #endif
309 	wd->atabus = adev->adev_bustype;
310 	wd->openings = adev->adev_openings;
311 	wd->drvp = adev->adev_drv_data;
312 
313 	wd->drvp->drv_done = wddone;
314 	wd->drvp->drv_softc = &wd->sc_dev;
315 
316 	aprint_naive("\n");
317 
318 	/* read our drive info */
319 	if (wd_get_params(wd, AT_WAIT, &wd->sc_params) != 0) {
320 		aprint_error("\n%s: IDENTIFY failed\n", wd->sc_dev.dv_xname);
321 		return;
322 	}
323 
324 	for (blank = 0, p = wd->sc_params.atap_model, q = tbuf, i = 0;
325 	    i < sizeof(wd->sc_params.atap_model); i++) {
326 		c = *p++;
327 		if (c == '\0')
328 			break;
329 		if (c != ' ') {
330 			if (blank) {
331 				*q++ = ' ';
332 				blank = 0;
333 			}
334 			*q++ = c;
335 		} else
336 			blank = 1;
337 	}
338 	*q++ = '\0';
339 
340 	aprint_normal(": <%s>\n", tbuf);
341 
342 	wdq = wd_lookup_quirks(tbuf);
343 	if (wdq != NULL)
344 		wd->sc_quirks = wdq->wdq_quirks;
345 
346 	if (wd->sc_quirks != 0) {
347 		char sbuf[sizeof(WD_QUIRK_FMT) + 64];
348 		bitmask_snprintf(wd->sc_quirks, WD_QUIRK_FMT,
349 		    sbuf, sizeof(sbuf));
350 		aprint_normal("%s: quirks %s\n", wd->sc_dev.dv_xname, sbuf);
351 	}
352 
353 	if ((wd->sc_params.atap_multi & 0xff) > 1) {
354 		wd->sc_multi = wd->sc_params.atap_multi & 0xff;
355 	} else {
356 		wd->sc_multi = 1;
357 	}
358 
359 	aprint_normal("%s: drive supports %d-sector PIO transfers,",
360 	    wd->sc_dev.dv_xname, wd->sc_multi);
361 
362 	/* 48-bit LBA addressing */
363 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0)
364 		wd->sc_flags |= WDF_LBA48;
365 
366 	/* Prior to ATA-4, LBA was optional. */
367 	if ((wd->sc_params.atap_capabilities1 & WDC_CAP_LBA) != 0)
368 		wd->sc_flags |= WDF_LBA;
369 #if 0
370 	/* ATA-4 requires LBA. */
371 	if (wd->sc_params.atap_ataversion != 0xffff &&
372 	    wd->sc_params.atap_ataversion >= WDC_VER_ATA4)
373 		wd->sc_flags |= WDF_LBA;
374 #endif
375 
376 	if ((wd->sc_flags & WDF_LBA48) != 0) {
377 		aprint_normal(" LBA48 addressing\n");
378 		wd->sc_capacity =
379 		    ((u_int64_t) wd->sc_params.__reserved6[11] << 48) |
380 		    ((u_int64_t) wd->sc_params.__reserved6[10] << 32) |
381 		    ((u_int64_t) wd->sc_params.__reserved6[9]  << 16) |
382 		    ((u_int64_t) wd->sc_params.__reserved6[8]  << 0);
383 	} else if ((wd->sc_flags & WDF_LBA) != 0) {
384 		aprint_normal(" LBA addressing\n");
385 		wd->sc_capacity =
386 		    ((u_int64_t)wd->sc_params.atap_capacity[1] << 16) |
387 		    wd->sc_params.atap_capacity[0];
388 	} else {
389 		aprint_normal(" chs addressing\n");
390 		wd->sc_capacity =
391 		    wd->sc_params.atap_cylinders *
392 		    wd->sc_params.atap_heads *
393 		    wd->sc_params.atap_sectors;
394 	}
395 	format_bytes(pbuf, sizeof(pbuf), wd->sc_capacity * DEV_BSIZE);
396 	aprint_normal("%s: %s, %d cyl, %d head, %d sec, "
397 	    "%d bytes/sect x %llu sectors\n",
398 	    self->dv_xname, pbuf,
399 	    (wd->sc_flags & WDF_LBA) ? (int)(wd->sc_capacity /
400 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors)) :
401 		wd->sc_params.atap_cylinders,
402 	    wd->sc_params.atap_heads, wd->sc_params.atap_sectors,
403 	    DEV_BSIZE, (unsigned long long)wd->sc_capacity);
404 
405 	ATADEBUG_PRINT(("%s: atap_dmatiming_mimi=%d, atap_dmatiming_recom=%d\n",
406 	    self->dv_xname, wd->sc_params.atap_dmatiming_mimi,
407 	    wd->sc_params.atap_dmatiming_recom), DEBUG_PROBE);
408 	/*
409 	 * Initialize and attach the disk structure.
410 	 */
411 	wd->sc_dk.dk_driver = &wddkdriver;
412 	wd->sc_dk.dk_name = wd->sc_dev.dv_xname;
413 	disk_attach(&wd->sc_dk);
414 	wd->sc_wdc_bio.lp = wd->sc_dk.dk_label;
415 	wd->sc_sdhook = shutdownhook_establish(wd_shutdown, wd);
416 	if (wd->sc_sdhook == NULL)
417 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
418 		    wd->sc_dev.dv_xname);
419 #if NRND > 0
420 	rnd_attach_source(&wd->rnd_source, wd->sc_dev.dv_xname,
421 			  RND_TYPE_DISK, 0);
422 #endif
423 
424 	/* Discover wedges on this disk. */
425 	dkwedge_discover(&wd->sc_dk);
426 }
427 
428 int
429 wdactivate(struct device *self, enum devact act)
430 {
431 	int rv = 0;
432 
433 	switch (act) {
434 	case DVACT_ACTIVATE:
435 		rv = EOPNOTSUPP;
436 		break;
437 
438 	case DVACT_DEACTIVATE:
439 		/*
440 		 * Nothing to do; we key off the device's DVF_ACTIVATE.
441 		 */
442 		break;
443 	}
444 	return (rv);
445 }
446 
447 int
448 wddetach(struct device *self, int flags)
449 {
450 	struct wd_softc *sc = (struct wd_softc *)self;
451 	int s, bmaj, cmaj, i, mn;
452 
453 	/* locate the major number */
454 	bmaj = bdevsw_lookup_major(&wd_bdevsw);
455 	cmaj = cdevsw_lookup_major(&wd_cdevsw);
456 
457 	/* Nuke the vnodes for any open instances. */
458 	for (i = 0; i < MAXPARTITIONS; i++) {
459 		mn = WDMINOR(device_unit(self), i);
460 		vdevgone(bmaj, mn, mn, VBLK);
461 		vdevgone(cmaj, mn, mn, VCHR);
462 	}
463 
464 	/* Delete all of our wedges. */
465 	dkwedge_delall(&sc->sc_dk);
466 
467 	s = splbio();
468 
469 	/* Kill off any queued buffers. */
470 	bufq_drain(sc->sc_q);
471 
472 	bufq_free(sc->sc_q);
473 	sc->atabus->ata_killpending(sc->drvp);
474 
475 	splx(s);
476 
477 	/* Detach disk. */
478 	disk_detach(&sc->sc_dk);
479 
480 #ifdef WD_SOFTBADSECT
481 	/* Clean out the bad sector list */
482 	while (!SLIST_EMPTY(&sc->sc_bslist)) {
483 		void *head = SLIST_FIRST(&sc->sc_bslist);
484 		SLIST_REMOVE_HEAD(&sc->sc_bslist, dbs_next);
485 		free(head, M_TEMP);
486 	}
487 	sc->sc_bscount = 0;
488 #endif
489 
490 	/* Get rid of the shutdown hook. */
491 	if (sc->sc_sdhook != NULL)
492 		shutdownhook_disestablish(sc->sc_sdhook);
493 
494 #if NRND > 0
495 	/* Unhook the entropy source. */
496 	rnd_detach_source(&sc->rnd_source);
497 #endif
498 
499 	sc->drvp->drive_flags = 0; /* no drive any more here */
500 
501 	return (0);
502 }
503 
504 /*
505  * Read/write routine for a buffer.  Validates the arguments and schedules the
506  * transfer.  Does not wait for the transfer to complete.
507  */
508 void
509 wdstrategy(struct buf *bp)
510 {
511 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(bp->b_dev));
512 	struct disklabel *lp = wd->sc_dk.dk_label;
513 	daddr_t blkno;
514 	int s;
515 
516 	ATADEBUG_PRINT(("wdstrategy (%s)\n", wd->sc_dev.dv_xname),
517 	    DEBUG_XFERS);
518 
519 	/* Valid request?  */
520 	if (bp->b_blkno < 0 ||
521 	    (bp->b_bcount % lp->d_secsize) != 0 ||
522 	    (bp->b_bcount / lp->d_secsize) >= (1 << NBBY)) {
523 		bp->b_error = EINVAL;
524 		goto bad;
525 	}
526 
527 	/* If device invalidated (e.g. media change, door open), error. */
528 	if ((wd->sc_flags & WDF_LOADED) == 0) {
529 		bp->b_error = EIO;
530 		goto bad;
531 	}
532 
533 	/* If it's a null transfer, return immediately. */
534 	if (bp->b_bcount == 0)
535 		goto done;
536 
537 	/*
538 	 * Do bounds checking, adjust transfer. if error, process.
539 	 * If end of partition, just return.
540 	 */
541 	if (WDPART(bp->b_dev) == RAW_PART) {
542 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
543 		    wd->sc_capacity) <= 0)
544 			goto done;
545 	} else {
546 		if (bounds_check_with_label(&wd->sc_dk, bp,
547 		    (wd->sc_flags & (WDF_WLABEL|WDF_LABELLING)) != 0) <= 0)
548 			goto done;
549 	}
550 
551 	/*
552 	 * Now convert the block number to absolute and put it in
553 	 * terms of the device's logical block size.
554 	 */
555 	if (lp->d_secsize >= DEV_BSIZE)
556 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
557 	else
558 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
559 
560 	if (WDPART(bp->b_dev) != RAW_PART)
561 		blkno += lp->d_partitions[WDPART(bp->b_dev)].p_offset;
562 
563 	bp->b_rawblkno = blkno;
564 
565 #ifdef WD_SOFTBADSECT
566 	/*
567 	 * If the transfer about to be attempted contains only a block that
568 	 * is known to be bad then return an error for the transfer without
569 	 * even attempting to start a transfer up under the premis that we
570 	 * will just end up doing more retries for a transfer that will end
571 	 * up failing again.
572 	 * XXX:SMP - mutex required to protect with DIOCBSFLUSH
573 	 */
574 	if (__predict_false(!SLIST_EMPTY(&wd->sc_bslist))) {
575 		struct disk_badsectors *dbs;
576 		daddr_t maxblk = blkno + (bp->b_bcount >> DEV_BSHIFT) - 1;
577 
578 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next)
579 			if ((dbs->dbs_min <= blkno && blkno <= dbs->dbs_max) ||
580 			    (dbs->dbs_min <= maxblk && maxblk <= dbs->dbs_max)){
581 				bp->b_error = EIO;
582 				goto bad;
583 			}
584 	}
585 #endif
586 
587 	/* Queue transfer on drive, activate drive and controller if idle. */
588 	s = splbio();
589 	BUFQ_PUT(wd->sc_q, bp);
590 	wdstart(wd);
591 	splx(s);
592 	return;
593 bad:
594 	bp->b_flags |= B_ERROR;
595 done:
596 	/* Toss transfer; we're done early. */
597 	bp->b_resid = bp->b_bcount;
598 	biodone(bp);
599 }
600 
601 /*
602  * Queue a drive for I/O.
603  */
604 void
605 wdstart(void *arg)
606 {
607 	struct wd_softc *wd = arg;
608 	struct buf *bp = NULL;
609 
610 	ATADEBUG_PRINT(("wdstart %s\n", wd->sc_dev.dv_xname),
611 	    DEBUG_XFERS);
612 	while (wd->openings > 0) {
613 
614 		/* Is there a buf for us ? */
615 		if ((bp = BUFQ_GET(wd->sc_q)) == NULL)
616 			return;
617 
618 		/*
619 		 * Make the command. First lock the device
620 		 */
621 		wd->openings--;
622 
623 		wd->retries = 0;
624 		__wdstart(wd, bp);
625 	}
626 }
627 
628 static void
629 wd_split_mod15_write(struct buf *bp)
630 {
631 	struct buf *obp = bp->b_private;
632 	struct wd_softc *sc = wd_cd.cd_devs[DISKUNIT(obp->b_dev)];
633 
634 	if (__predict_false(bp->b_flags & B_ERROR) != 0) {
635 		/*
636 		 * Propagate the error.  If this was the first half of
637 		 * the original transfer, make sure to account for that
638 		 * in the residual.
639 		 */
640 		if (bp->b_data == obp->b_data)
641 			bp->b_resid += bp->b_bcount;
642 		goto done;
643 	}
644 
645 	/*
646 	 * If this was the second half of the transfer, we're all done!
647 	 */
648 	if (bp->b_data != obp->b_data)
649 		goto done;
650 
651 	/*
652 	 * Advance the pointer to the second half and issue that command
653 	 * using the same opening.
654 	 */
655 	bp->b_flags = obp->b_flags | B_CALL;
656 	bp->b_data += bp->b_bcount;
657 	bp->b_blkno += (bp->b_bcount / 512);
658 	bp->b_rawblkno += (bp->b_bcount / 512);
659 	__wdstart(sc, bp);
660 	return;
661 
662  done:
663 	obp->b_flags |= bp->b_flags & B_ERROR;
664 	obp->b_error = bp->b_error;
665 	obp->b_resid = bp->b_resid;
666 	putiobuf(bp);
667 	biodone(obp);
668 	sc->openings++;
669 	/* wddone() will call wdstart() */
670 }
671 
672 void
673 __wdstart(struct wd_softc *wd, struct buf *bp)
674 {
675 
676 	/*
677 	 * Deal with the "split mod15 write" quirk.  We just divide the
678 	 * transfer in two, doing the first half and then then second half
679 	 * with the same command opening.
680 	 *
681 	 * Note we MUST do this here, because we can't let insertion
682 	 * into the bufq cause the transfers to be re-merged.
683 	 */
684 	if (__predict_false((wd->sc_quirks & WD_QUIRK_SPLIT_MOD15_WRITE) != 0 &&
685 			    (bp->b_flags & B_READ) == 0 &&
686 			    bp->b_bcount > 512 &&
687 			    ((bp->b_bcount / 512) % 15) == 1)) {
688 		struct buf *nbp;
689 
690 		/* already at splbio */
691 		nbp = getiobuf_nowait();
692 		if (__predict_false(nbp == NULL)) {
693 			/* No memory -- fail the iop. */
694 			bp->b_error = ENOMEM;
695 			bp->b_flags |= B_ERROR;
696 			bp->b_resid = bp->b_bcount;
697 			biodone(bp);
698 			wd->openings++;
699 			return;
700 		}
701 
702 		nbp->b_error = 0;
703 		nbp->b_proc = bp->b_proc;
704 		nbp->b_vp = NULLVP;
705 		nbp->b_dev = bp->b_dev;
706 
707 		nbp->b_bcount = bp->b_bcount / 2;
708 		nbp->b_bufsize = bp->b_bcount / 2;
709 		nbp->b_data = bp->b_data;
710 
711 		nbp->b_blkno = bp->b_blkno;
712 		nbp->b_rawblkno = bp->b_rawblkno;
713 
714 		nbp->b_flags = bp->b_flags | B_CALL;
715 		nbp->b_iodone = wd_split_mod15_write;
716 
717 		/* Put ptr to orig buf in b_private and use new buf */
718 		nbp->b_private = bp;
719 
720 		BIO_COPYPRIO(nbp, bp);
721 
722 		bp = nbp;
723 	}
724 
725 	wd->sc_wdc_bio.blkno = bp->b_rawblkno;
726 	wd->sc_wdc_bio.blkdone =0;
727 	wd->sc_bp = bp;
728 	/*
729 	 * If we're retrying, retry in single-sector mode. This will give us
730 	 * the sector number of the problem, and will eventually allow the
731 	 * transfer to succeed.
732 	 */
733 	if (wd->retries >= WDIORETRIES_SINGLE)
734 		wd->sc_wdc_bio.flags = ATA_SINGLE;
735 	else
736 		wd->sc_wdc_bio.flags = 0;
737 	if (wd->sc_flags & WDF_LBA48 &&
738 	    (wd->sc_wdc_bio.blkno > LBA48_THRESHOLD ||
739 	    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) != 0))
740 		wd->sc_wdc_bio.flags |= ATA_LBA48;
741 	if (wd->sc_flags & WDF_LBA)
742 		wd->sc_wdc_bio.flags |= ATA_LBA;
743 	if (bp->b_flags & B_READ)
744 		wd->sc_wdc_bio.flags |= ATA_READ;
745 	wd->sc_wdc_bio.bcount = bp->b_bcount;
746 	wd->sc_wdc_bio.databuf = bp->b_data;
747 	/* Instrumentation. */
748 	disk_busy(&wd->sc_dk);
749 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
750 	case ATACMD_TRY_AGAIN:
751 		callout_reset(&wd->sc_restart_ch, hz, wdrestart, wd);
752 		break;
753 	case ATACMD_QUEUED:
754 	case ATACMD_COMPLETE:
755 		break;
756 	default:
757 		panic("__wdstart: bad return code from ata_bio()");
758 	}
759 }
760 
761 void
762 wddone(void *v)
763 {
764 	struct wd_softc *wd = v;
765 	struct buf *bp = wd->sc_bp;
766 	const char *errmsg;
767 	int do_perror = 0;
768 	ATADEBUG_PRINT(("wddone %s\n", wd->sc_dev.dv_xname),
769 	    DEBUG_XFERS);
770 
771 	if (bp == NULL)
772 		return;
773 	bp->b_resid = wd->sc_wdc_bio.bcount;
774 	switch (wd->sc_wdc_bio.error) {
775 	case ERR_DMA:
776 		errmsg = "DMA error";
777 		goto retry;
778 	case ERR_DF:
779 		errmsg = "device fault";
780 		goto retry;
781 	case TIMEOUT:
782 		errmsg = "device timeout";
783 		goto retry;
784 	case ERR_RESET:
785 		errmsg = "channel reset";
786 		goto retry2;
787 	case ERROR:
788 		/* Don't care about media change bits */
789 		if (wd->sc_wdc_bio.r_error != 0 &&
790 		    (wd->sc_wdc_bio.r_error & ~(WDCE_MC | WDCE_MCR)) == 0)
791 			goto noerror;
792 		errmsg = "error";
793 		do_perror = 1;
794 retry:		/* Just reset and retry. Can we do more ? */
795 		(*wd->atabus->ata_reset_drive)(wd->drvp, AT_RST_NOCMD);
796 retry2:
797 		diskerr(bp, "wd", errmsg, LOG_PRINTF,
798 		    wd->sc_wdc_bio.blkdone, wd->sc_dk.dk_label);
799 		if (wd->retries < WDIORETRIES)
800 			printf(", retrying\n");
801 		if (do_perror)
802 			wdperror(wd);
803 		if (wd->retries < WDIORETRIES) {
804 			wd->retries++;
805 			callout_reset(&wd->sc_restart_ch, RECOVERYTIME,
806 			    wdrestart, wd);
807 			return;
808 		}
809 		printf("\n");
810 
811 #ifdef WD_SOFTBADSECT
812 		/*
813 		 * Not all errors indicate a failed block but those that do,
814 		 * put the block on the bad-block list for the device.  Only
815 		 * do this for reads because the drive should do it for writes,
816 		 * itself, according to Manuel.
817 		 */
818 		if ((bp->b_flags & B_READ) &&
819 		    ((wd->drvp->ata_vers >= 4 && wd->sc_wdc_bio.r_error & 64) ||
820 	     	     (wd->drvp->ata_vers < 4 && wd->sc_wdc_bio.r_error & 192))) {
821 			struct disk_badsectors *dbs;
822 
823 			dbs = malloc(sizeof *dbs, M_TEMP, M_WAITOK);
824 			dbs->dbs_min = bp->b_rawblkno;
825 			dbs->dbs_max = dbs->dbs_min + (bp->b_bcount >> DEV_BSHIFT) - 1;
826 			microtime(&dbs->dbs_failedat);
827 			SLIST_INSERT_HEAD(&wd->sc_bslist, dbs, dbs_next);
828 			wd->sc_bscount++;
829 		}
830 #endif
831 		bp->b_flags |= B_ERROR;
832 		bp->b_error = EIO;
833 		break;
834 	case NOERROR:
835 noerror:	if ((wd->sc_wdc_bio.flags & ATA_CORR) || wd->retries > 0)
836 			printf("%s: soft error (corrected)\n",
837 			    wd->sc_dev.dv_xname);
838 		break;
839 	case ERR_NODEV:
840 		bp->b_flags |= B_ERROR;
841 		bp->b_error = EIO;
842 		break;
843 	}
844 	disk_unbusy(&wd->sc_dk, (bp->b_bcount - bp->b_resid),
845 	    (bp->b_flags & B_READ));
846 #if NRND > 0
847 	rnd_add_uint32(&wd->rnd_source, bp->b_blkno);
848 #endif
849 	/* XXX Yuck, but we don't want to increment openings in this case */
850 	if (__predict_false((bp->b_flags & B_CALL) != 0 &&
851 			    bp->b_iodone == wd_split_mod15_write))
852 		biodone(bp);
853 	else {
854 		biodone(bp);
855 		wd->openings++;
856 	}
857 	wdstart(wd);
858 }
859 
860 void
861 wdrestart(void *v)
862 {
863 	struct wd_softc *wd = v;
864 	struct buf *bp = wd->sc_bp;
865 	int s;
866 	ATADEBUG_PRINT(("wdrestart %s\n", wd->sc_dev.dv_xname),
867 	    DEBUG_XFERS);
868 
869 	s = splbio();
870 	__wdstart(v, bp);
871 	splx(s);
872 }
873 
874 int
875 wdread(dev_t dev, struct uio *uio, int flags)
876 {
877 
878 	ATADEBUG_PRINT(("wdread\n"), DEBUG_XFERS);
879 	return (physio(wdstrategy, NULL, dev, B_READ, minphys, uio));
880 }
881 
882 int
883 wdwrite(dev_t dev, struct uio *uio, int flags)
884 {
885 
886 	ATADEBUG_PRINT(("wdwrite\n"), DEBUG_XFERS);
887 	return (physio(wdstrategy, NULL, dev, B_WRITE, minphys, uio));
888 }
889 
890 int
891 wdopen(dev_t dev, int flag, int fmt, struct lwp *l)
892 {
893 	struct wd_softc *wd;
894 	int part, error;
895 
896 	ATADEBUG_PRINT(("wdopen\n"), DEBUG_FUNCS);
897 	wd = device_lookup(&wd_cd, WDUNIT(dev));
898 	if (wd == NULL)
899 		return (ENXIO);
900 
901 	if (! device_is_active(&wd->sc_dev))
902 		return (ENODEV);
903 
904 	part = WDPART(dev);
905 
906 	if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
907 		return (error);
908 
909 	/*
910 	 * If there are wedges, and this is not RAW_PART, then we
911 	 * need to fail.
912 	 */
913 	if (wd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
914 		error = EBUSY;
915 		goto bad1;
916 	}
917 
918 	/*
919 	 * If this is the first open of this device, add a reference
920 	 * to the adapter.
921 	 */
922 	if (wd->sc_dk.dk_openmask == 0 &&
923 	    (error = wd->atabus->ata_addref(wd->drvp)) != 0)
924 		goto bad1;
925 
926 	if (wd->sc_dk.dk_openmask != 0) {
927 		/*
928 		 * If any partition is open, but the disk has been invalidated,
929 		 * disallow further opens.
930 		 */
931 		if ((wd->sc_flags & WDF_LOADED) == 0) {
932 			error = EIO;
933 			goto bad2;
934 		}
935 	} else {
936 		if ((wd->sc_flags & WDF_LOADED) == 0) {
937 			wd->sc_flags |= WDF_LOADED;
938 
939 			/* Load the physical device parameters. */
940 			wd_get_params(wd, AT_WAIT, &wd->sc_params);
941 
942 			/* Load the partition info if not already loaded. */
943 			wdgetdisklabel(wd);
944 		}
945 	}
946 
947 	/* Check that the partition exists. */
948 	if (part != RAW_PART &&
949 	    (part >= wd->sc_dk.dk_label->d_npartitions ||
950 	     wd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
951 		error = ENXIO;
952 		goto bad2;
953 	}
954 
955 	/* Insure only one open at a time. */
956 	switch (fmt) {
957 	case S_IFCHR:
958 		wd->sc_dk.dk_copenmask |= (1 << part);
959 		break;
960 	case S_IFBLK:
961 		wd->sc_dk.dk_bopenmask |= (1 << part);
962 		break;
963 	}
964 	wd->sc_dk.dk_openmask =
965 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
966 
967 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
968 	return 0;
969 
970  bad2:
971 	if (wd->sc_dk.dk_openmask == 0)
972 		wd->atabus->ata_delref(wd->drvp);
973  bad1:
974 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
975 	return error;
976 }
977 
978 int
979 wdclose(dev_t dev, int flag, int fmt, struct lwp *l)
980 {
981 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
982 	int part = WDPART(dev);
983 	int error;
984 
985 	ATADEBUG_PRINT(("wdclose\n"), DEBUG_FUNCS);
986 
987 	if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE, NULL)) != 0)
988 		return error;
989 
990 	switch (fmt) {
991 	case S_IFCHR:
992 		wd->sc_dk.dk_copenmask &= ~(1 << part);
993 		break;
994 	case S_IFBLK:
995 		wd->sc_dk.dk_bopenmask &= ~(1 << part);
996 		break;
997 	}
998 	wd->sc_dk.dk_openmask =
999 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
1000 
1001 	if (wd->sc_dk.dk_openmask == 0) {
1002 		wd_flushcache(wd, AT_WAIT);
1003 
1004 		if (! (wd->sc_flags & WDF_KLABEL))
1005 			wd->sc_flags &= ~WDF_LOADED;
1006 
1007 		wd->atabus->ata_delref(wd->drvp);
1008 	}
1009 
1010 	(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
1011 	return 0;
1012 }
1013 
1014 void
1015 wdgetdefaultlabel(struct wd_softc *wd, struct disklabel *lp)
1016 {
1017 
1018 	ATADEBUG_PRINT(("wdgetdefaultlabel\n"), DEBUG_FUNCS);
1019 	memset(lp, 0, sizeof(struct disklabel));
1020 
1021 	lp->d_secsize = DEV_BSIZE;
1022 	lp->d_ntracks = wd->sc_params.atap_heads;
1023 	lp->d_nsectors = wd->sc_params.atap_sectors;
1024 	lp->d_ncylinders = (wd->sc_flags & WDF_LBA) ? wd->sc_capacity /
1025 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors) :
1026 		wd->sc_params.atap_cylinders;
1027 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1028 
1029 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
1030 		lp->d_type = DTYPE_ST506;
1031 	else
1032 		lp->d_type = DTYPE_ESDI;
1033 
1034 	strncpy(lp->d_typename, wd->sc_params.atap_model, 16);
1035 	strncpy(lp->d_packname, "fictitious", 16);
1036 	if (wd->sc_capacity > UINT32_MAX)
1037 		lp->d_secperunit = UINT32_MAX;
1038 	else
1039 		lp->d_secperunit = wd->sc_capacity;
1040 	lp->d_rpm = 3600;
1041 	lp->d_interleave = 1;
1042 	lp->d_flags = 0;
1043 
1044 	lp->d_partitions[RAW_PART].p_offset = 0;
1045 	lp->d_partitions[RAW_PART].p_size =
1046 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1047 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1048 	lp->d_npartitions = RAW_PART + 1;
1049 
1050 	lp->d_magic = DISKMAGIC;
1051 	lp->d_magic2 = DISKMAGIC;
1052 	lp->d_checksum = dkcksum(lp);
1053 }
1054 
1055 /*
1056  * Fabricate a default disk label, and try to read the correct one.
1057  */
1058 void
1059 wdgetdisklabel(struct wd_softc *wd)
1060 {
1061 	struct disklabel *lp = wd->sc_dk.dk_label;
1062 	const char *errstring;
1063 	int s;
1064 
1065 	ATADEBUG_PRINT(("wdgetdisklabel\n"), DEBUG_FUNCS);
1066 
1067 	memset(wd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1068 
1069 	wdgetdefaultlabel(wd, lp);
1070 
1071 	wd->sc_badsect[0] = -1;
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, device_unit(&wd->sc_dev),
1079 				  RAW_PART), wdstrategy, lp,
1080 				  wd->sc_dk.dk_cpulabel);
1081 	if (errstring) {
1082 		/*
1083 		 * This probably happened because the drive's default
1084 		 * geometry doesn't match the DOS geometry.  We
1085 		 * assume the DOS geometry is now in the label and try
1086 		 * again.  XXX This is a kluge.
1087 		 */
1088 		if (wd->drvp->state > RESET) {
1089 			s = splbio();
1090 			wd->drvp->drive_flags |= DRIVE_RESET;
1091 			splx(s);
1092 		}
1093 		errstring = readdisklabel(MAKEWDDEV(0, device_unit(&wd->sc_dev),
1094 		    RAW_PART), wdstrategy, lp, wd->sc_dk.dk_cpulabel);
1095 	}
1096 	if (errstring) {
1097 		printf("%s: %s\n", wd->sc_dev.dv_xname, errstring);
1098 		return;
1099 	}
1100 
1101 	if (wd->drvp->state > RESET) {
1102 		s = splbio();
1103 		wd->drvp->drive_flags |= DRIVE_RESET;
1104 		splx(s);
1105 	}
1106 #ifdef HAS_BAD144_HANDLING
1107 	if ((lp->d_flags & D_BADSECT) != 0)
1108 		bad144intern(wd);
1109 #endif
1110 }
1111 
1112 void
1113 wdperror(const struct wd_softc *wd)
1114 {
1115 	static const char *const errstr0_3[] = {"address mark not found",
1116 	    "track 0 not found", "aborted command", "media change requested",
1117 	    "id not found", "media changed", "uncorrectable data error",
1118 	    "bad block detected"};
1119 	static const char *const errstr4_5[] = {
1120 	    "obsolete (address mark not found)",
1121 	    "no media/write protected", "aborted command",
1122 	    "media change requested", "id not found", "media changed",
1123 	    "uncorrectable data error", "interface CRC error"};
1124 	const char *const *errstr;
1125 	int i;
1126 	const char *sep = "";
1127 
1128 	const char *devname = wd->sc_dev.dv_xname;
1129 	struct ata_drive_datas *drvp = wd->drvp;
1130 	int errno = wd->sc_wdc_bio.r_error;
1131 
1132 	if (drvp->ata_vers >= 4)
1133 		errstr = errstr4_5;
1134 	else
1135 		errstr = errstr0_3;
1136 
1137 	printf("%s: (", devname);
1138 
1139 	if (errno == 0)
1140 		printf("error not notified");
1141 
1142 	for (i = 0; i < 8; i++) {
1143 		if (errno & (1 << i)) {
1144 			printf("%s%s", sep, errstr[i]);
1145 			sep = ", ";
1146 		}
1147 	}
1148 	printf(")\n");
1149 }
1150 
1151 int
1152 wdioctl(dev_t dev, u_long xfer, caddr_t addr, int flag, struct lwp *l)
1153 {
1154 	struct wd_softc *wd = device_lookup(&wd_cd, WDUNIT(dev));
1155 	int error = 0, s;
1156 #ifdef __HAVE_OLD_DISKLABEL
1157 	struct disklabel *newlabel = NULL;
1158 #endif
1159 
1160 	ATADEBUG_PRINT(("wdioctl\n"), DEBUG_FUNCS);
1161 
1162 	if ((wd->sc_flags & WDF_LOADED) == 0)
1163 		return EIO;
1164 
1165 	switch (xfer) {
1166 #ifdef HAS_BAD144_HANDLING
1167 	case DIOCSBAD:
1168 		if ((flag & FWRITE) == 0)
1169 			return EBADF;
1170 		wd->sc_dk.dk_cpulabel->bad = *(struct dkbad *)addr;
1171 		wd->sc_dk.dk_label->d_flags |= D_BADSECT;
1172 		bad144intern(wd);
1173 		return 0;
1174 #endif
1175 #ifdef WD_SOFTBADSECT
1176 	case DIOCBSLIST :
1177 	{
1178 		u_int32_t count, missing, skip;
1179 		struct disk_badsecinfo dbsi;
1180 		struct disk_badsectors *dbs;
1181 		size_t available;
1182 		caddr_t laddr;
1183 
1184 		dbsi = *(struct disk_badsecinfo *)addr;
1185 		missing = wd->sc_bscount;
1186 		count = 0;
1187 		available = dbsi.dbsi_bufsize;
1188 		skip = dbsi.dbsi_skip;
1189 		laddr = dbsi.dbsi_buffer;
1190 
1191 		/*
1192 		 * We start this loop with the expectation that all of the
1193 		 * entries will be missed and decrement this counter each
1194 		 * time we either skip over one (already copied out) or
1195 		 * we actually copy it back to user space.  The structs
1196 		 * holding the bad sector information are copied directly
1197 		 * back to user space whilst the summary is returned via
1198 		 * the struct passed in via the ioctl.
1199 		 */
1200 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next) {
1201 			if (skip > 0) {
1202 				missing--;
1203 				skip--;
1204 				continue;
1205 			}
1206 			if (available < sizeof(*dbs))
1207 				break;
1208 			available -= sizeof(*dbs);
1209 			copyout(dbs, laddr, sizeof(*dbs));
1210 			laddr += sizeof(*dbs);
1211 			missing--;
1212 			count++;
1213 		}
1214 		dbsi.dbsi_left = missing;
1215 		dbsi.dbsi_copied = count;
1216 		*(struct disk_badsecinfo *)addr = dbsi;
1217 		return 0;
1218 	}
1219 
1220 	case DIOCBSFLUSH :
1221 		/* Clean out the bad sector list */
1222 		while (!SLIST_EMPTY(&wd->sc_bslist)) {
1223 			void *head = SLIST_FIRST(&wd->sc_bslist);
1224 			SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
1225 			free(head, M_TEMP);
1226 		}
1227 		wd->sc_bscount = 0;
1228 		return 0;
1229 #endif
1230 	case DIOCGDINFO:
1231 		*(struct disklabel *)addr = *(wd->sc_dk.dk_label);
1232 		return 0;
1233 #ifdef __HAVE_OLD_DISKLABEL
1234 	case ODIOCGDINFO:
1235 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1236 		if (newlabel == NULL)
1237 			return EIO;
1238 		*newlabel = *(wd->sc_dk.dk_label);
1239 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1240 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1241 		else
1242 			error = ENOTTY;
1243 		free(newlabel, M_TEMP);
1244 		return error;
1245 #endif
1246 
1247 	case DIOCGPART:
1248 		((struct partinfo *)addr)->disklab = wd->sc_dk.dk_label;
1249 		((struct partinfo *)addr)->part =
1250 		    &wd->sc_dk.dk_label->d_partitions[WDPART(dev)];
1251 		return 0;
1252 
1253 	case DIOCWDINFO:
1254 	case DIOCSDINFO:
1255 #ifdef __HAVE_OLD_DISKLABEL
1256 	case ODIOCWDINFO:
1257 	case ODIOCSDINFO:
1258 #endif
1259 	{
1260 		struct disklabel *lp;
1261 
1262 		if ((flag & FWRITE) == 0)
1263 			return EBADF;
1264 
1265 #ifdef __HAVE_OLD_DISKLABEL
1266 		if (xfer == ODIOCSDINFO || xfer == ODIOCWDINFO) {
1267 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1268 			if (newlabel == NULL)
1269 				return EIO;
1270 			memset(newlabel, 0, sizeof newlabel);
1271 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1272 			lp = newlabel;
1273 		} else
1274 #endif
1275 		lp = (struct disklabel *)addr;
1276 
1277 		if ((error = lockmgr(&wd->sc_dk.dk_openlock, LK_EXCLUSIVE,
1278 				     NULL)) != 0)
1279 			goto bad;
1280 		wd->sc_flags |= WDF_LABELLING;
1281 
1282 		error = setdisklabel(wd->sc_dk.dk_label,
1283 		    lp, /*wd->sc_dk.dk_openmask : */0,
1284 		    wd->sc_dk.dk_cpulabel);
1285 		if (error == 0) {
1286 			if (wd->drvp->state > RESET) {
1287 				s = splbio();
1288 				wd->drvp->drive_flags |= DRIVE_RESET;
1289 				splx(s);
1290 			}
1291 			if (xfer == DIOCWDINFO
1292 #ifdef __HAVE_OLD_DISKLABEL
1293 			    || xfer == ODIOCWDINFO
1294 #endif
1295 			    )
1296 				error = writedisklabel(WDLABELDEV(dev),
1297 				    wdstrategy, wd->sc_dk.dk_label,
1298 				    wd->sc_dk.dk_cpulabel);
1299 		}
1300 
1301 		wd->sc_flags &= ~WDF_LABELLING;
1302 		(void) lockmgr(&wd->sc_dk.dk_openlock, LK_RELEASE, NULL);
1303 bad:
1304 #ifdef __HAVE_OLD_DISKLABEL
1305 		if (newlabel != NULL)
1306 			free(newlabel, M_TEMP);
1307 #endif
1308 		return error;
1309 	}
1310 
1311 	case DIOCKLABEL:
1312 		if (*(int *)addr)
1313 			wd->sc_flags |= WDF_KLABEL;
1314 		else
1315 			wd->sc_flags &= ~WDF_KLABEL;
1316 		return 0;
1317 
1318 	case DIOCWLABEL:
1319 		if ((flag & FWRITE) == 0)
1320 			return EBADF;
1321 		if (*(int *)addr)
1322 			wd->sc_flags |= WDF_WLABEL;
1323 		else
1324 			wd->sc_flags &= ~WDF_WLABEL;
1325 		return 0;
1326 
1327 	case DIOCGDEFLABEL:
1328 		wdgetdefaultlabel(wd, (struct disklabel *)addr);
1329 		return 0;
1330 #ifdef __HAVE_OLD_DISKLABEL
1331 	case ODIOCGDEFLABEL:
1332 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1333 		if (newlabel == NULL)
1334 			return EIO;
1335 		wdgetdefaultlabel(wd, newlabel);
1336 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1337 			memcpy(addr, &newlabel, sizeof (struct olddisklabel));
1338 		else
1339 			error = ENOTTY;
1340 		free(newlabel, M_TEMP);
1341 		return error;
1342 #endif
1343 
1344 #ifdef notyet
1345 	case DIOCWFORMAT:
1346 		if ((flag & FWRITE) == 0)
1347 			return EBADF;
1348 		{
1349 		register struct format_op *fop;
1350 		struct iovec aiov;
1351 		struct uio auio;
1352 
1353 		fop = (struct format_op *)addr;
1354 		aiov.iov_base = fop->df_buf;
1355 		aiov.iov_len = fop->df_count;
1356 		auio.uio_iov = &aiov;
1357 		auio.uio_iovcnt = 1;
1358 		auio.uio_resid = fop->df_count;
1359 		auio.uio_offset =
1360 			fop->df_startblk * wd->sc_dk.dk_label->d_secsize;
1361 		auio.uio_vmspace = l->l_proc->p_vmspace;
1362 		error = physio(wdformat, NULL, dev, B_WRITE, minphys,
1363 		    &auio);
1364 		fop->df_count -= auio.uio_resid;
1365 		fop->df_reg[0] = wdc->sc_status;
1366 		fop->df_reg[1] = wdc->sc_error;
1367 		return error;
1368 		}
1369 #endif
1370 	case DIOCGCACHE:
1371 		return wd_getcache(wd, (int *)addr);
1372 
1373 	case DIOCSCACHE:
1374 		return wd_setcache(wd, *(int *)addr);
1375 
1376 	case DIOCCACHESYNC:
1377 		return wd_flushcache(wd, AT_WAIT);
1378 
1379 	case ATAIOCCOMMAND:
1380 		/*
1381 		 * Make sure this command is (relatively) safe first
1382 		 */
1383 		if ((((atareq_t *) addr)->flags & ATACMD_READ) == 0 &&
1384 		    (flag & FWRITE) == 0)
1385 			return (EBADF);
1386 		{
1387 		struct wd_ioctl *wi;
1388 		atareq_t *atareq = (atareq_t *) addr;
1389 		int error1;
1390 
1391 		wi = wi_get();
1392 		wi->wi_softc = wd;
1393 		wi->wi_atareq = *atareq;
1394 
1395 		if (atareq->datalen && atareq->flags &
1396 		    (ATACMD_READ | ATACMD_WRITE)) {
1397 			wi->wi_iov.iov_base = atareq->databuf;
1398 			wi->wi_iov.iov_len = atareq->datalen;
1399 			wi->wi_uio.uio_iov = &wi->wi_iov;
1400 			wi->wi_uio.uio_iovcnt = 1;
1401 			wi->wi_uio.uio_resid = atareq->datalen;
1402 			wi->wi_uio.uio_offset = 0;
1403 			wi->wi_uio.uio_rw =
1404 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE;
1405 			wi->wi_uio.uio_vmspace = l->l_proc->p_vmspace;
1406 			error1 = physio(wdioctlstrategy, &wi->wi_bp, dev,
1407 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE,
1408 			    minphys, &wi->wi_uio);
1409 		} else {
1410 			/* No need to call physio if we don't have any
1411 			   user data */
1412 			wi->wi_bp.b_flags = 0;
1413 			wi->wi_bp.b_data = 0;
1414 			wi->wi_bp.b_bcount = 0;
1415 			wi->wi_bp.b_dev = 0;
1416 			wi->wi_bp.b_proc = l->l_proc;
1417 			wdioctlstrategy(&wi->wi_bp);
1418 			error1 = wi->wi_bp.b_error;
1419 		}
1420 		*atareq = wi->wi_atareq;
1421 		wi_free(wi);
1422 		return(error1);
1423 		}
1424 
1425 	case DIOCAWEDGE:
1426 	    {
1427 	    	struct dkwedge_info *dkw = (void *) addr;
1428 
1429 		if ((flag & FWRITE) == 0)
1430 			return (EBADF);
1431 
1432 		/* If the ioctl happens here, the parent is us. */
1433 		strcpy(dkw->dkw_parent, wd->sc_dev.dv_xname);
1434 		return (dkwedge_add(dkw));
1435 	    }
1436 
1437 	case DIOCDWEDGE:
1438 	    {
1439 	    	struct dkwedge_info *dkw = (void *) addr;
1440 
1441 		if ((flag & FWRITE) == 0)
1442 			return (EBADF);
1443 
1444 		/* If the ioctl happens here, the parent is us. */
1445 		strcpy(dkw->dkw_parent, wd->sc_dev.dv_xname);
1446 		return (dkwedge_del(dkw));
1447 	    }
1448 
1449 	case DIOCLWEDGES:
1450 	    {
1451 	    	struct dkwedge_list *dkwl = (void *) addr;
1452 
1453 		return (dkwedge_list(&wd->sc_dk, dkwl, l));
1454 	    }
1455 
1456 	case DIOCGSTRATEGY:
1457 	    {
1458 		struct disk_strategy *dks = (void *)addr;
1459 
1460 		s = splbio();
1461 		strlcpy(dks->dks_name, bufq_getstrategyname(wd->sc_q),
1462 		    sizeof(dks->dks_name));
1463 		splx(s);
1464 		dks->dks_paramlen = 0;
1465 
1466 		return 0;
1467 	    }
1468 
1469 	case DIOCSSTRATEGY:
1470 	    {
1471 		struct disk_strategy *dks = (void *)addr;
1472 		struct bufq_state *new;
1473 		struct bufq_state *old;
1474 
1475 		if ((flag & FWRITE) == 0) {
1476 			return EBADF;
1477 		}
1478 		if (dks->dks_param != NULL) {
1479 			return EINVAL;
1480 		}
1481 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
1482 		error = bufq_alloc(&new, dks->dks_name,
1483 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
1484 		if (error) {
1485 			return error;
1486 		}
1487 		s = splbio();
1488 		old = wd->sc_q;
1489 		bufq_move(new, old);
1490 		wd->sc_q = new;
1491 		splx(s);
1492 		bufq_free(old);
1493 
1494 		return 0;
1495 	    }
1496 
1497 	default:
1498 		return ENOTTY;
1499 	}
1500 
1501 #ifdef DIAGNOSTIC
1502 	panic("wdioctl: impossible");
1503 #endif
1504 }
1505 
1506 #ifdef B_FORMAT
1507 int
1508 wdformat(struct buf *bp)
1509 {
1510 
1511 	bp->b_flags |= B_FORMAT;
1512 	return wdstrategy(bp);
1513 }
1514 #endif
1515 
1516 int
1517 wdsize(dev_t dev)
1518 {
1519 	struct wd_softc *wd;
1520 	int part, omask;
1521 	int size;
1522 
1523 	ATADEBUG_PRINT(("wdsize\n"), DEBUG_FUNCS);
1524 
1525 	wd = device_lookup(&wd_cd, WDUNIT(dev));
1526 	if (wd == NULL)
1527 		return (-1);
1528 
1529 	part = WDPART(dev);
1530 	omask = wd->sc_dk.dk_openmask & (1 << part);
1531 
1532 	if (omask == 0 && wdopen(dev, 0, S_IFBLK, NULL) != 0)
1533 		return (-1);
1534 	if (wd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1535 		size = -1;
1536 	else
1537 		size = wd->sc_dk.dk_label->d_partitions[part].p_size *
1538 		    (wd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1539 	if (omask == 0 && wdclose(dev, 0, S_IFBLK, NULL) != 0)
1540 		return (-1);
1541 	return (size);
1542 }
1543 
1544 /* #define WD_DUMP_NOT_TRUSTED if you just want to watch */
1545 static int wddoingadump = 0;
1546 static int wddumprecalibrated = 0;
1547 
1548 /*
1549  * Dump core after a system crash.
1550  */
1551 int
1552 wddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1553 {
1554 	struct wd_softc *wd;	/* disk unit to do the I/O */
1555 	struct disklabel *lp;   /* disk's disklabel */
1556 	int part, err;
1557 	int nblks;	/* total number of sectors left to write */
1558 
1559 	/* Check if recursive dump; if so, punt. */
1560 	if (wddoingadump)
1561 		return EFAULT;
1562 	wddoingadump = 1;
1563 
1564 	wd = device_lookup(&wd_cd, WDUNIT(dev));
1565 	if (wd == NULL)
1566 		return (ENXIO);
1567 
1568 	part = WDPART(dev);
1569 
1570 	/* Convert to disk sectors.  Request must be a multiple of size. */
1571 	lp = wd->sc_dk.dk_label;
1572 	if ((size % lp->d_secsize) != 0)
1573 		return EFAULT;
1574 	nblks = size / lp->d_secsize;
1575 	blkno = blkno / (lp->d_secsize / DEV_BSIZE);
1576 
1577 	/* Check transfer bounds against partition size. */
1578 	if ((blkno < 0) || ((blkno + nblks) > lp->d_partitions[part].p_size))
1579 		return EINVAL;
1580 
1581 	/* Offset block number to start of partition. */
1582 	blkno += lp->d_partitions[part].p_offset;
1583 
1584 	/* Recalibrate, if first dump transfer. */
1585 	if (wddumprecalibrated == 0) {
1586 		wddumprecalibrated = 1;
1587 		(*wd->atabus->ata_reset_drive)(wd->drvp,
1588 					       AT_POLL | AT_RST_EMERG);
1589 		wd->drvp->state = RESET;
1590 	}
1591 
1592 	wd->sc_bp = NULL;
1593 	wd->sc_wdc_bio.blkno = blkno;
1594 	wd->sc_wdc_bio.flags = ATA_POLL;
1595 	if (wd->sc_flags & WDF_LBA48 &&
1596 	    (blkno > LBA48_THRESHOLD ||
1597     	    (wd->sc_quirks & WD_QUIRK_FORCE_LBA48) != 0))
1598 		wd->sc_wdc_bio.flags |= ATA_LBA48;
1599 	if (wd->sc_flags & WDF_LBA)
1600 		wd->sc_wdc_bio.flags |= ATA_LBA;
1601 	wd->sc_wdc_bio.bcount = nblks * lp->d_secsize;
1602 	wd->sc_wdc_bio.databuf = va;
1603 #ifndef WD_DUMP_NOT_TRUSTED
1604 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
1605 	case ATACMD_TRY_AGAIN:
1606 		panic("wddump: try again");
1607 		break;
1608 	case ATACMD_QUEUED:
1609 		panic("wddump: polled command has been queued");
1610 		break;
1611 	case ATACMD_COMPLETE:
1612 		break;
1613 	}
1614 	switch(wd->sc_wdc_bio.error) {
1615 	case TIMEOUT:
1616 		printf("wddump: device timed out");
1617 		err = EIO;
1618 		break;
1619 	case ERR_DF:
1620 		printf("wddump: drive fault");
1621 		err = EIO;
1622 		break;
1623 	case ERR_DMA:
1624 		printf("wddump: DMA error");
1625 		err = EIO;
1626 		break;
1627 	case ERROR:
1628 		printf("wddump: ");
1629 		wdperror(wd);
1630 		err = EIO;
1631 		break;
1632 	case NOERROR:
1633 		err = 0;
1634 		break;
1635 	default:
1636 		panic("wddump: unknown error type");
1637 	}
1638 	if (err != 0) {
1639 		printf("\n");
1640 		return err;
1641 	}
1642 #else	/* WD_DUMP_NOT_TRUSTED */
1643 	/* Let's just talk about this first... */
1644 	printf("wd%d: dump addr 0x%x, cylin %d, head %d, sector %d\n",
1645 	    unit, va, cylin, head, sector);
1646 	delay(500 * 1000);	/* half a second */
1647 #endif
1648 
1649 	wddoingadump = 0;
1650 	return 0;
1651 }
1652 
1653 #ifdef HAS_BAD144_HANDLING
1654 /*
1655  * Internalize the bad sector table.
1656  */
1657 void
1658 bad144intern(struct wd_softc *wd)
1659 {
1660 	struct dkbad *bt = &wd->sc_dk.dk_cpulabel->bad;
1661 	struct disklabel *lp = wd->sc_dk.dk_label;
1662 	int i = 0;
1663 
1664 	ATADEBUG_PRINT(("bad144intern\n"), DEBUG_XFERS);
1665 
1666 	for (; i < NBT_BAD; i++) {
1667 		if (bt->bt_bad[i].bt_cyl == 0xffff)
1668 			break;
1669 		wd->sc_badsect[i] =
1670 		    bt->bt_bad[i].bt_cyl * lp->d_secpercyl +
1671 		    (bt->bt_bad[i].bt_trksec >> 8) * lp->d_nsectors +
1672 		    (bt->bt_bad[i].bt_trksec & 0xff);
1673 	}
1674 	for (; i < NBT_BAD+1; i++)
1675 		wd->sc_badsect[i] = -1;
1676 }
1677 #endif
1678 
1679 static void
1680 wd_params_to_properties(struct wd_softc *wd, struct ataparams *params)
1681 {
1682 	prop_dictionary_t disk_info, geom;
1683 	prop_string_t string;
1684 	prop_number_t number;
1685 
1686 	disk_info = prop_dictionary_create();
1687 
1688 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
1689 		string = prop_string_create_cstring_nocopy("ST506");
1690 	else {
1691 		/* XXX Should have a case for ATA here, too. */
1692 		string = prop_string_create_cstring_nocopy("ESDI");
1693 	}
1694 	prop_dictionary_set(disk_info, "type", string);
1695 	prop_object_release(string);
1696 
1697 	geom = prop_dictionary_create();
1698 
1699 	number = prop_number_create_integer(wd->sc_capacity);
1700 	prop_dictionary_set(geom, "sectors-per-unit", number);
1701 	prop_object_release(number);
1702 
1703 	number = prop_number_create_integer(DEV_BSIZE /* XXX 512? */);
1704 	prop_dictionary_set(geom, "sector-size", number);
1705 	prop_object_release(number);
1706 
1707 	number = prop_number_create_integer(wd->sc_params.atap_sectors);
1708 	prop_dictionary_set(geom, "sectors-per-track", number);
1709 	prop_object_release(number);
1710 
1711 	number = prop_number_create_integer(wd->sc_params.atap_heads);
1712 	prop_dictionary_set(geom, "tracks-per-cylinder", number);
1713 	prop_object_release(number);
1714 
1715 	number = prop_number_create_integer(
1716 	    (wd->sc_flags & WDF_LBA) ?
1717 	        wd->sc_capacity / (wd->sc_params.atap_heads *
1718 				   wd->sc_params.atap_sectors)
1719 				     :
1720 		wd->sc_params.atap_cylinders);
1721 	prop_dictionary_set(geom, "cylinders-per-unit", number);
1722 	prop_object_release(number);
1723 
1724 	prop_dictionary_set(disk_info, "geometry", geom);
1725 	prop_object_release(geom);
1726 
1727 	prop_dictionary_set(device_properties(&wd->sc_dev),
1728 			    "disk-info", disk_info);
1729 	prop_object_release(disk_info);
1730 }
1731 
1732 int
1733 wd_get_params(struct wd_softc *wd, u_int8_t flags, struct ataparams *params)
1734 {
1735 	switch (wd->atabus->ata_get_params(wd->drvp, flags, params)) {
1736 	case CMD_AGAIN:
1737 		return 1;
1738 	case CMD_ERR:
1739 		/*
1740 		 * We `know' there's a drive here; just assume it's old.
1741 		 * This geometry is only used to read the MBR and print a
1742 		 * (false) attach message.
1743 		 */
1744 		strncpy(params->atap_model, "ST506",
1745 		    sizeof params->atap_model);
1746 		params->atap_config = ATA_CFG_FIXED;
1747 		params->atap_cylinders = 1024;
1748 		params->atap_heads = 8;
1749 		params->atap_sectors = 17;
1750 		params->atap_multi = 1;
1751 		params->atap_capabilities1 = params->atap_capabilities2 = 0;
1752 		wd->drvp->ata_vers = -1; /* Mark it as pre-ATA */
1753 		/* FALLTHROUGH */
1754 	case CMD_OK:
1755 		wd_params_to_properties(wd, params);
1756 		return 0;
1757 	default:
1758 		panic("wd_get_params: bad return code from ata_get_params");
1759 		/* NOTREACHED */
1760 	}
1761 }
1762 
1763 int
1764 wd_getcache(struct wd_softc *wd, int *bitsp)
1765 {
1766 	struct ataparams params;
1767 
1768 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
1769 		return EIO;
1770 	if (params.atap_cmd_set1 == 0x0000 ||
1771 	    params.atap_cmd_set1 == 0xffff ||
1772 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0) {
1773 		*bitsp = 0;
1774 		return 0;
1775 	}
1776 	*bitsp = DKCACHE_WCHANGE | DKCACHE_READ;
1777 	if (params.atap_cmd1_en & WDC_CMD1_CACHE)
1778 		*bitsp |= DKCACHE_WRITE;
1779 
1780 	return 0;
1781 }
1782 
1783 const char at_errbits[] = "\20\10ERROR\11TIMEOU\12DF";
1784 
1785 int
1786 wd_setcache(struct wd_softc *wd, int bits)
1787 {
1788 	struct ataparams params;
1789 	struct ata_command ata_c;
1790 
1791 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
1792 		return EIO;
1793 
1794 	if (params.atap_cmd_set1 == 0x0000 ||
1795 	    params.atap_cmd_set1 == 0xffff ||
1796 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0)
1797 		return EOPNOTSUPP;
1798 
1799 	if ((bits & DKCACHE_READ) == 0 ||
1800 	    (bits & DKCACHE_SAVE) != 0)
1801 		return EOPNOTSUPP;
1802 
1803 	memset(&ata_c, 0, sizeof(struct ata_command));
1804 	ata_c.r_command = SET_FEATURES;
1805 	ata_c.r_st_bmask = 0;
1806 	ata_c.r_st_pmask = 0;
1807 	ata_c.timeout = 30000; /* 30s timeout */
1808 	ata_c.flags = AT_WAIT;
1809 	if (bits & DKCACHE_WRITE)
1810 		ata_c.r_features = WDSF_WRITE_CACHE_EN;
1811 	else
1812 		ata_c.r_features = WDSF_WRITE_CACHE_DS;
1813 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1814 		printf("%s: wd_setcache command not complete\n",
1815 		    wd->sc_dev.dv_xname);
1816 		return EIO;
1817 	}
1818 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1819 		char sbuf[sizeof(at_errbits) + 64];
1820 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
1821 		printf("%s: wd_setcache: status=%s\n", wd->sc_dev.dv_xname,
1822 		    sbuf);
1823 		return EIO;
1824 	}
1825 	return 0;
1826 }
1827 
1828 int
1829 wd_standby(struct wd_softc *wd, int flags)
1830 {
1831 	struct ata_command ata_c;
1832 
1833 	memset(&ata_c, 0, sizeof(struct ata_command));
1834 	ata_c.r_command = WDCC_STANDBY_IMMED;
1835 	ata_c.r_st_bmask = WDCS_DRDY;
1836 	ata_c.r_st_pmask = WDCS_DRDY;
1837 	ata_c.flags = flags;
1838 	ata_c.timeout = 30000; /* 30s timeout */
1839 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1840 		printf("%s: standby immediate command didn't complete\n",
1841 		    wd->sc_dev.dv_xname);
1842 		return EIO;
1843 	}
1844 	if (ata_c.flags & AT_ERROR) {
1845 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
1846 			return ENODEV;
1847 	}
1848 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1849 		char sbuf[sizeof(at_errbits) + 64];
1850 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
1851 		printf("%s: wd_standby: status=%s\n", wd->sc_dev.dv_xname,
1852 		    sbuf);
1853 		return EIO;
1854 	}
1855 	return 0;
1856 }
1857 
1858 int
1859 wd_flushcache(struct wd_softc *wd, int flags)
1860 {
1861 	struct ata_command ata_c;
1862 
1863 	/*
1864 	 * WDCC_FLUSHCACHE is here since ATA-4, but some drives report
1865 	 * only ATA-2 and still support it.
1866 	 */
1867 	if (wd->drvp->ata_vers < 4 &&
1868 	    ((wd->sc_params.atap_cmd_set2 & WDC_CMD2_FC) == 0 ||
1869 	    wd->sc_params.atap_cmd_set2 == 0xffff))
1870 		return ENODEV;
1871 	memset(&ata_c, 0, sizeof(struct ata_command));
1872 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0 &&
1873 	    (wd->sc_params.atap_cmd2_en & ATA_CMD2_FCE) != 0)
1874 		ata_c.r_command = WDCC_FLUSHCACHE_EXT;
1875 	else
1876 		ata_c.r_command = WDCC_FLUSHCACHE;
1877 	ata_c.r_st_bmask = WDCS_DRDY;
1878 	ata_c.r_st_pmask = WDCS_DRDY;
1879 	ata_c.flags = flags;
1880 	ata_c.timeout = 30000; /* 30s timeout */
1881 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1882 		printf("%s: flush cache command didn't complete\n",
1883 		    wd->sc_dev.dv_xname);
1884 		return EIO;
1885 	}
1886 	if (ata_c.flags & AT_ERROR) {
1887 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
1888 			return ENODEV;
1889 	}
1890 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1891 		char sbuf[sizeof(at_errbits) + 64];
1892 		bitmask_snprintf(ata_c.flags, at_errbits, sbuf, sizeof(sbuf));
1893 		printf("%s: wd_flushcache: status=%s\n", wd->sc_dev.dv_xname,
1894 		    sbuf);
1895 		return EIO;
1896 	}
1897 	return 0;
1898 }
1899 
1900 void
1901 wd_shutdown(void *arg)
1902 {
1903 	struct wd_softc *wd = arg;
1904 	wd_flushcache(wd, AT_POLL);
1905 }
1906 
1907 /*
1908  * Allocate space for a ioctl queue structure.  Mostly taken from
1909  * scsipi_ioctl.c
1910  */
1911 struct wd_ioctl *
1912 wi_get(void)
1913 {
1914 	struct wd_ioctl *wi;
1915 	int s;
1916 
1917 	wi = malloc(sizeof(struct wd_ioctl), M_TEMP, M_WAITOK|M_ZERO);
1918 	simple_lock_init(&wi->wi_bp.b_interlock);
1919 	s = splbio();
1920 	LIST_INSERT_HEAD(&wi_head, wi, wi_list);
1921 	splx(s);
1922 	return (wi);
1923 }
1924 
1925 /*
1926  * Free an ioctl structure and remove it from our list
1927  */
1928 
1929 void
1930 wi_free(struct wd_ioctl *wi)
1931 {
1932 	int s;
1933 
1934 	s = splbio();
1935 	LIST_REMOVE(wi, wi_list);
1936 	splx(s);
1937 	free(wi, M_TEMP);
1938 }
1939 
1940 /*
1941  * Find a wd_ioctl structure based on the struct buf.
1942  */
1943 
1944 struct wd_ioctl *
1945 wi_find(struct buf *bp)
1946 {
1947 	struct wd_ioctl *wi;
1948 	int s;
1949 
1950 	s = splbio();
1951 	for (wi = wi_head.lh_first; wi != 0; wi = wi->wi_list.le_next)
1952 		if (bp == &wi->wi_bp)
1953 			break;
1954 	splx(s);
1955 	return (wi);
1956 }
1957 
1958 /*
1959  * Ioctl pseudo strategy routine
1960  *
1961  * This is mostly stolen from scsipi_ioctl.c:scsistrategy().  What
1962  * happens here is:
1963  *
1964  * - wdioctl() queues a wd_ioctl structure.
1965  *
1966  * - wdioctl() calls physio/wdioctlstrategy based on whether or not
1967  *   user space I/O is required.  If physio() is called, physio() eventually
1968  *   calls wdioctlstrategy().
1969  *
1970  * - In either case, wdioctlstrategy() calls wd->atabus->ata_exec_command()
1971  *   to perform the actual command
1972  *
1973  * The reason for the use of the pseudo strategy routine is because
1974  * when doing I/O to/from user space, physio _really_ wants to be in
1975  * the loop.  We could put the entire buffer into the ioctl request
1976  * structure, but that won't scale if we want to do things like download
1977  * microcode.
1978  */
1979 
1980 void
1981 wdioctlstrategy(struct buf *bp)
1982 {
1983 	struct wd_ioctl *wi;
1984 	struct ata_command ata_c;
1985 	int error = 0;
1986 
1987 	wi = wi_find(bp);
1988 	if (wi == NULL) {
1989 		printf("wdioctlstrategy: "
1990 		    "No matching ioctl request found in queue\n");
1991 		error = EINVAL;
1992 		goto bad;
1993 	}
1994 
1995 	memset(&ata_c, 0, sizeof(ata_c));
1996 
1997 	/*
1998 	 * Abort if physio broke up the transfer
1999 	 */
2000 
2001 	if (bp->b_bcount != wi->wi_atareq.datalen) {
2002 		printf("physio split wd ioctl request... cannot proceed\n");
2003 		error = EIO;
2004 		goto bad;
2005 	}
2006 
2007 	/*
2008 	 * Abort if we didn't get a buffer size that was a multiple of
2009 	 * our sector size (or was larger than NBBY)
2010 	 */
2011 
2012 	if ((bp->b_bcount % wi->wi_softc->sc_dk.dk_label->d_secsize) != 0 ||
2013 	    (bp->b_bcount / wi->wi_softc->sc_dk.dk_label->d_secsize) >=
2014 	     (1 << NBBY)) {
2015 		error = EINVAL;
2016 		goto bad;
2017 	}
2018 
2019 	/*
2020 	 * Make sure a timeout was supplied in the ioctl request
2021 	 */
2022 
2023 	if (wi->wi_atareq.timeout == 0) {
2024 		error = EINVAL;
2025 		goto bad;
2026 	}
2027 
2028 	if (wi->wi_atareq.flags & ATACMD_READ)
2029 		ata_c.flags |= AT_READ;
2030 	else if (wi->wi_atareq.flags & ATACMD_WRITE)
2031 		ata_c.flags |= AT_WRITE;
2032 
2033 	if (wi->wi_atareq.flags & ATACMD_READREG)
2034 		ata_c.flags |= AT_READREG;
2035 
2036 	ata_c.flags |= AT_WAIT;
2037 
2038 	ata_c.timeout = wi->wi_atareq.timeout;
2039 	ata_c.r_command = wi->wi_atareq.command;
2040 	ata_c.r_head = wi->wi_atareq.head & 0x0f;
2041 	ata_c.r_cyl = wi->wi_atareq.cylinder;
2042 	ata_c.r_sector = wi->wi_atareq.sec_num;
2043 	ata_c.r_count = wi->wi_atareq.sec_count;
2044 	ata_c.r_features = wi->wi_atareq.features;
2045 	ata_c.r_st_bmask = WDCS_DRDY;
2046 	ata_c.r_st_pmask = WDCS_DRDY;
2047 	ata_c.data = wi->wi_bp.b_data;
2048 	ata_c.bcount = wi->wi_bp.b_bcount;
2049 
2050 	if (wi->wi_softc->atabus->ata_exec_command(wi->wi_softc->drvp, &ata_c)
2051 	    != ATACMD_COMPLETE) {
2052 		wi->wi_atareq.retsts = ATACMD_ERROR;
2053 		goto bad;
2054 	}
2055 
2056 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
2057 		if (ata_c.flags & AT_ERROR) {
2058 			wi->wi_atareq.retsts = ATACMD_ERROR;
2059 			wi->wi_atareq.error = ata_c.r_error;
2060 		} else if (ata_c.flags & AT_DF)
2061 			wi->wi_atareq.retsts = ATACMD_DF;
2062 		else
2063 			wi->wi_atareq.retsts = ATACMD_TIMEOUT;
2064 	} else {
2065 		wi->wi_atareq.retsts = ATACMD_OK;
2066 		if (wi->wi_atareq.flags & ATACMD_READREG) {
2067 			wi->wi_atareq.head = ata_c.r_head ;
2068 			wi->wi_atareq.cylinder = ata_c.r_cyl;
2069 			wi->wi_atareq.sec_num = ata_c.r_sector;
2070 			wi->wi_atareq.sec_count = ata_c.r_count;
2071 			wi->wi_atareq.features = ata_c.r_features;
2072 			wi->wi_atareq.error = ata_c.r_error;
2073 		}
2074 	}
2075 
2076 	bp->b_error = 0;
2077 	biodone(bp);
2078 	return;
2079 bad:
2080 	bp->b_flags |= B_ERROR;
2081 	bp->b_error = error;
2082 	biodone(bp);
2083 }
2084