xref: /netbsd/sys/dev/ata/wd.c (revision 6550d01e)
1 /*	$NetBSD: wd.c,v 1.385 2010/11/05 15:49:37 dyoung Exp $ */
2 
3 /*
4  * Copyright (c) 1998, 2001 Manuel Bouyer.  All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *	notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *	notice, this list of conditions and the following disclaimer in the
13  *	documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  */
26 
27 /*-
28  * Copyright (c) 1998, 2003, 2004 The NetBSD Foundation, Inc.
29  * All rights reserved.
30  *
31  * This code is derived from software contributed to The NetBSD Foundation
32  * by Charles M. Hannum and by Onno van der Linden.
33  *
34  * Redistribution and use in source and binary forms, with or without
35  * modification, are permitted provided that the following conditions
36  * are met:
37  * 1. Redistributions of source code must retain the above copyright
38  *    notice, this list of conditions and the following disclaimer.
39  * 2. Redistributions in binary form must reproduce the above copyright
40  *    notice, this list of conditions and the following disclaimer in the
41  *    documentation and/or other materials provided with the distribution.
42  *
43  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
44  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
45  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
46  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
47  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
48  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
49  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
50  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
51  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
52  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
53  * POSSIBILITY OF SUCH DAMAGE.
54  */
55 
56 #include <sys/cdefs.h>
57 __KERNEL_RCSID(0, "$NetBSD: wd.c,v 1.385 2010/11/05 15:49:37 dyoung Exp $");
58 
59 #include "opt_ata.h"
60 
61 #include "rnd.h"
62 
63 #include <sys/param.h>
64 #include <sys/systm.h>
65 #include <sys/kernel.h>
66 #include <sys/conf.h>
67 #include <sys/file.h>
68 #include <sys/stat.h>
69 #include <sys/ioctl.h>
70 #include <sys/buf.h>
71 #include <sys/bufq.h>
72 #include <sys/uio.h>
73 #include <sys/malloc.h>
74 #include <sys/device.h>
75 #include <sys/disklabel.h>
76 #include <sys/disk.h>
77 #include <sys/syslog.h>
78 #include <sys/proc.h>
79 #include <sys/reboot.h>
80 #include <sys/vnode.h>
81 #if NRND > 0
82 #include <sys/rnd.h>
83 #endif
84 
85 #include <sys/intr.h>
86 #include <sys/bus.h>
87 
88 #include <dev/ata/atareg.h>
89 #include <dev/ata/atavar.h>
90 #include <dev/ata/wdvar.h>
91 #include <dev/ic/wdcreg.h>
92 #include <sys/ataio.h>
93 #include "locators.h"
94 
95 #include <prop/proplib.h>
96 
97 #define	WDIORETRIES_SINGLE 4	/* number of retries before single-sector */
98 #define	WDIORETRIES	5	/* number of retries before giving up */
99 #define	RECOVERYTIME hz/2	/* time to wait before retrying a cmd */
100 
101 #define	WDUNIT(dev)		DISKUNIT(dev)
102 #define	WDPART(dev)		DISKPART(dev)
103 #define	WDMINOR(unit, part)	DISKMINOR(unit, part)
104 #define	MAKEWDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
105 
106 #define	WDLABELDEV(dev)	(MAKEWDDEV(major(dev), WDUNIT(dev), RAW_PART))
107 
108 #define DEBUG_INTR   0x01
109 #define DEBUG_XFERS  0x02
110 #define DEBUG_STATUS 0x04
111 #define DEBUG_FUNCS  0x08
112 #define DEBUG_PROBE  0x10
113 #ifdef ATADEBUG
114 int wdcdebug_wd_mask = 0x0;
115 #define ATADEBUG_PRINT(args, level) \
116 	if (wdcdebug_wd_mask & (level)) \
117 		printf args
118 #else
119 #define ATADEBUG_PRINT(args, level)
120 #endif
121 
122 int	wdprobe(device_t, cfdata_t, void *);
123 void	wdattach(device_t, device_t, void *);
124 int	wddetach(device_t, int);
125 int	wdprint(void *, char *);
126 void	wdperror(const struct wd_softc *);
127 
128 static int	wdlastclose(device_t);
129 static bool	wd_suspend(device_t, const pmf_qual_t *);
130 static int	wd_standby(struct wd_softc *, int);
131 
132 CFATTACH_DECL3_NEW(wd, sizeof(struct wd_softc),
133     wdprobe, wdattach, wddetach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
134 
135 extern struct cfdriver wd_cd;
136 
137 dev_type_open(wdopen);
138 dev_type_close(wdclose);
139 dev_type_read(wdread);
140 dev_type_write(wdwrite);
141 dev_type_ioctl(wdioctl);
142 dev_type_strategy(wdstrategy);
143 dev_type_dump(wddump);
144 dev_type_size(wdsize);
145 
146 const struct bdevsw wd_bdevsw = {
147 	wdopen, wdclose, wdstrategy, wdioctl, wddump, wdsize, D_DISK
148 };
149 
150 const struct cdevsw wd_cdevsw = {
151 	wdopen, wdclose, wdread, wdwrite, wdioctl,
152 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
153 };
154 
155 /*
156  * Glue necessary to hook WDCIOCCOMMAND into physio
157  */
158 
159 struct wd_ioctl {
160 	LIST_ENTRY(wd_ioctl) wi_list;
161 	struct buf wi_bp;
162 	struct uio wi_uio;
163 	struct iovec wi_iov;
164 	atareq_t wi_atareq;
165 	struct wd_softc *wi_softc;
166 };
167 
168 LIST_HEAD(, wd_ioctl) wi_head;
169 
170 struct	wd_ioctl *wi_find(struct buf *);
171 void	wi_free(struct wd_ioctl *);
172 struct	wd_ioctl *wi_get(void);
173 void	wdioctlstrategy(struct buf *);
174 
175 void  wdgetdefaultlabel(struct wd_softc *, struct disklabel *);
176 void  wdgetdisklabel(struct wd_softc *);
177 void  wdstart(void *);
178 void  wdstart1(struct wd_softc*, struct buf *);
179 void  wdrestart(void *);
180 void  wddone(void *);
181 int   wd_get_params(struct wd_softc *, u_int8_t, struct ataparams *);
182 int   wd_flushcache(struct wd_softc *, int);
183 bool  wd_shutdown(device_t, int);
184 
185 int   wd_getcache(struct wd_softc *, int *);
186 int   wd_setcache(struct wd_softc *, int);
187 
188 struct dkdriver wddkdriver = { wdstrategy, minphys };
189 
190 #ifdef HAS_BAD144_HANDLING
191 static void bad144intern(struct wd_softc *);
192 #endif
193 
194 #define	WD_QUIRK_SPLIT_MOD15_WRITE	0x0001	/* must split certain writes */
195 
196 #define	WD_QUIRK_FMT "\20\1SPLIT_MOD15_WRITE\2FORCE_LBA48"
197 
198 /*
199  * Quirk table for IDE drives.  Put more-specific matches first, since
200  * a simple globbing routine is used for matching.
201  */
202 static const struct wd_quirk {
203 	const char *wdq_match;		/* inquiry pattern to match */
204 	int wdq_quirks;			/* drive quirks */
205 } wd_quirk_table[] = {
206 	/*
207 	 * Some Seagate S-ATA drives have a PHY which can get confused
208 	 * with the way data is packetized by some S-ATA controllers.
209 	 *
210 	 * The work-around is to split in two any write transfer whose
211 	 * sector count % 15 == 1 (assuming 512 byte sectors).
212 	 *
213 	 * XXX This is an incomplete list.  There are at least a couple
214 	 * XXX more model numbers.  If you have trouble with such transfers
215 	 * XXX (8K is the most common) on Seagate S-ATA drives, please
216 	 * XXX notify thorpej@NetBSD.org.
217 	 */
218 	{ "ST3120023AS",
219 	  WD_QUIRK_SPLIT_MOD15_WRITE },
220 	{ "ST380023AS",
221 	  WD_QUIRK_SPLIT_MOD15_WRITE },
222 	{ NULL,
223 	  0 }
224 };
225 
226 static const struct wd_quirk *
227 wd_lookup_quirks(const char *name)
228 {
229 	const struct wd_quirk *wdq;
230 	const char *estr;
231 
232 	for (wdq = wd_quirk_table; wdq->wdq_match != NULL; wdq++) {
233 		/*
234 		 * We only want exact matches (which include matches
235 		 * against globbing characters).
236 		 */
237 		if (pmatch(name, wdq->wdq_match, &estr) == 2)
238 			return (wdq);
239 	}
240 	return (NULL);
241 }
242 
243 int
244 wdprobe(device_t parent, cfdata_t match, void *aux)
245 {
246 	struct ata_device *adev = aux;
247 
248 	if (adev == NULL)
249 		return 0;
250 	if (adev->adev_bustype->bustype_type != SCSIPI_BUSTYPE_ATA)
251 		return 0;
252 
253 	if (match->cf_loc[ATA_HLCF_DRIVE] != ATA_HLCF_DRIVE_DEFAULT &&
254 	    match->cf_loc[ATA_HLCF_DRIVE] != adev->adev_drv_data->drive)
255 		return 0;
256 	return 1;
257 }
258 
259 void
260 wdattach(device_t parent, device_t self, void *aux)
261 {
262 	struct wd_softc *wd = device_private(self);
263 	struct ata_device *adev= aux;
264 	int i, blank;
265 	char tbuf[41], pbuf[9], c, *p, *q;
266 	const struct wd_quirk *wdq;
267 
268 	wd->sc_dev = self;
269 
270 	ATADEBUG_PRINT(("wdattach\n"), DEBUG_FUNCS | DEBUG_PROBE);
271 	callout_init(&wd->sc_restart_ch, 0);
272 	bufq_alloc(&wd->sc_q, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
273 #ifdef WD_SOFTBADSECT
274 	SLIST_INIT(&wd->sc_bslist);
275 #endif
276 	wd->atabus = adev->adev_bustype;
277 	wd->openings = adev->adev_openings;
278 	wd->drvp = adev->adev_drv_data;
279 
280 	wd->drvp->drv_done = wddone;
281 	wd->drvp->drv_softc = wd->sc_dev;
282 
283 	aprint_naive("\n");
284 
285 	/* read our drive info */
286 	if (wd_get_params(wd, AT_WAIT, &wd->sc_params) != 0) {
287 		aprint_error("\n%s: IDENTIFY failed\n", device_xname(self));
288 		return;
289 	}
290 
291 	for (blank = 0, p = wd->sc_params.atap_model, q = tbuf, i = 0;
292 	    i < sizeof(wd->sc_params.atap_model); i++) {
293 		c = *p++;
294 		if (c == '\0')
295 			break;
296 		if (c != ' ') {
297 			if (blank) {
298 				*q++ = ' ';
299 				blank = 0;
300 			}
301 			*q++ = c;
302 		} else
303 			blank = 1;
304 	}
305 	*q++ = '\0';
306 
307 	aprint_normal(": <%s>\n", tbuf);
308 
309 	wdq = wd_lookup_quirks(tbuf);
310 	if (wdq != NULL)
311 		wd->sc_quirks = wdq->wdq_quirks;
312 
313 	if (wd->sc_quirks != 0) {
314 		char sbuf[sizeof(WD_QUIRK_FMT) + 64];
315 		snprintb(sbuf, sizeof(sbuf), WD_QUIRK_FMT, wd->sc_quirks);
316 		aprint_normal_dev(self, "quirks %s\n", sbuf);
317 	}
318 
319 	if ((wd->sc_params.atap_multi & 0xff) > 1) {
320 		wd->sc_multi = wd->sc_params.atap_multi & 0xff;
321 	} else {
322 		wd->sc_multi = 1;
323 	}
324 
325 	aprint_verbose_dev(self, "drive supports %d-sector PIO transfers,",
326 	    wd->sc_multi);
327 
328 	/* 48-bit LBA addressing */
329 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0)
330 		wd->sc_flags |= WDF_LBA48;
331 
332 	/* Prior to ATA-4, LBA was optional. */
333 	if ((wd->sc_params.atap_capabilities1 & WDC_CAP_LBA) != 0)
334 		wd->sc_flags |= WDF_LBA;
335 #if 0
336 	/* ATA-4 requires LBA. */
337 	if (wd->sc_params.atap_ataversion != 0xffff &&
338 	    wd->sc_params.atap_ataversion >= WDC_VER_ATA4)
339 		wd->sc_flags |= WDF_LBA;
340 #endif
341 
342 	if ((wd->sc_flags & WDF_LBA48) != 0) {
343 		aprint_verbose(" LBA48 addressing\n");
344 		wd->sc_capacity =
345 		    ((u_int64_t) wd->sc_params.atap_max_lba[3] << 48) |
346 		    ((u_int64_t) wd->sc_params.atap_max_lba[2] << 32) |
347 		    ((u_int64_t) wd->sc_params.atap_max_lba[1] << 16) |
348 		    ((u_int64_t) wd->sc_params.atap_max_lba[0] <<  0);
349 		wd->sc_capacity28 =
350 		    (wd->sc_params.atap_capacity[1] << 16) |
351 		    wd->sc_params.atap_capacity[0];
352 	} else if ((wd->sc_flags & WDF_LBA) != 0) {
353 		aprint_verbose(" LBA addressing\n");
354 		wd->sc_capacity28 = wd->sc_capacity =
355 		    (wd->sc_params.atap_capacity[1] << 16) |
356 		    wd->sc_params.atap_capacity[0];
357 	} else {
358 		aprint_verbose(" chs addressing\n");
359 		wd->sc_capacity28 = wd->sc_capacity =
360 		    wd->sc_params.atap_cylinders *
361 		    wd->sc_params.atap_heads *
362 		    wd->sc_params.atap_sectors;
363 	}
364 	format_bytes(pbuf, sizeof(pbuf), wd->sc_capacity * DEV_BSIZE);
365 	aprint_normal_dev(self, "%s, %d cyl, %d head, %d sec, "
366 	    "%d bytes/sect x %llu sectors\n",
367 	    pbuf,
368 	    (wd->sc_flags & WDF_LBA) ? (int)(wd->sc_capacity /
369 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors)) :
370 		wd->sc_params.atap_cylinders,
371 	    wd->sc_params.atap_heads, wd->sc_params.atap_sectors,
372 	    DEV_BSIZE, (unsigned long long)wd->sc_capacity);
373 
374 	ATADEBUG_PRINT(("%s: atap_dmatiming_mimi=%d, atap_dmatiming_recom=%d\n",
375 	    device_xname(self), wd->sc_params.atap_dmatiming_mimi,
376 	    wd->sc_params.atap_dmatiming_recom), DEBUG_PROBE);
377 	/*
378 	 * Initialize and attach the disk structure.
379 	 */
380 	/* we fill in dk_info later */
381 	disk_init(&wd->sc_dk, device_xname(wd->sc_dev), &wddkdriver);
382 	disk_attach(&wd->sc_dk);
383 	wd->sc_wdc_bio.lp = wd->sc_dk.dk_label;
384 #if NRND > 0
385 	rnd_attach_source(&wd->rnd_source, device_xname(wd->sc_dev),
386 			  RND_TYPE_DISK, 0);
387 #endif
388 
389 	/* Discover wedges on this disk. */
390 	dkwedge_discover(&wd->sc_dk);
391 
392 	if (!pmf_device_register1(self, wd_suspend, NULL, wd_shutdown))
393 		aprint_error_dev(self, "couldn't establish power handler\n");
394 }
395 
396 static bool
397 wd_suspend(device_t dv, const pmf_qual_t *qual)
398 {
399 	struct wd_softc *sc = device_private(dv);
400 
401 	wd_flushcache(sc, AT_WAIT);
402 	wd_standby(sc, AT_WAIT);
403 	return true;
404 }
405 
406 int
407 wddetach(device_t self, int flags)
408 {
409 	struct wd_softc *sc = device_private(self);
410 	int bmaj, cmaj, i, mn, rc, s;
411 
412 	if ((rc = disk_begindetach(&sc->sc_dk, wdlastclose, self, flags)) != 0)
413 		return rc;
414 
415 	/* locate the major number */
416 	bmaj = bdevsw_lookup_major(&wd_bdevsw);
417 	cmaj = cdevsw_lookup_major(&wd_cdevsw);
418 
419 	/* Nuke the vnodes for any open instances. */
420 	for (i = 0; i < MAXPARTITIONS; i++) {
421 		mn = WDMINOR(device_unit(self), i);
422 		vdevgone(bmaj, mn, mn, VBLK);
423 		vdevgone(cmaj, mn, mn, VCHR);
424 	}
425 
426 	/* Delete all of our wedges. */
427 	dkwedge_delall(&sc->sc_dk);
428 
429 	s = splbio();
430 
431 	/* Kill off any queued buffers. */
432 	bufq_drain(sc->sc_q);
433 
434 	bufq_free(sc->sc_q);
435 	sc->atabus->ata_killpending(sc->drvp);
436 
437 	splx(s);
438 
439 	/* Detach disk. */
440 	disk_detach(&sc->sc_dk);
441 	disk_destroy(&sc->sc_dk);
442 
443 #ifdef WD_SOFTBADSECT
444 	/* Clean out the bad sector list */
445 	while (!SLIST_EMPTY(&sc->sc_bslist)) {
446 		void *head = SLIST_FIRST(&sc->sc_bslist);
447 		SLIST_REMOVE_HEAD(&sc->sc_bslist, dbs_next);
448 		free(head, M_TEMP);
449 	}
450 	sc->sc_bscount = 0;
451 #endif
452 
453 	pmf_device_deregister(self);
454 
455 #if NRND > 0
456 	/* Unhook the entropy source. */
457 	rnd_detach_source(&sc->rnd_source);
458 #endif
459 
460 	callout_destroy(&sc->sc_restart_ch);
461 
462 	sc->drvp->drive_flags = 0; /* no drive any more here */
463 
464 	return (0);
465 }
466 
467 /*
468  * Read/write routine for a buffer.  Validates the arguments and schedules the
469  * transfer.  Does not wait for the transfer to complete.
470  */
471 void
472 wdstrategy(struct buf *bp)
473 {
474 	struct wd_softc *wd =
475 	    device_lookup_private(&wd_cd, WDUNIT(bp->b_dev));
476 	struct disklabel *lp = wd->sc_dk.dk_label;
477 	daddr_t blkno;
478 	int s;
479 
480 	ATADEBUG_PRINT(("wdstrategy (%s)\n", device_xname(wd->sc_dev)),
481 	    DEBUG_XFERS);
482 
483 	/* Valid request?  */
484 	if (bp->b_blkno < 0 ||
485 	    (bp->b_bcount % lp->d_secsize) != 0 ||
486 	    (bp->b_bcount / lp->d_secsize) >= (1 << NBBY)) {
487 		bp->b_error = EINVAL;
488 		goto done;
489 	}
490 
491 	/* If device invalidated (e.g. media change, door open,
492 	 * device detachment), then error.
493 	 */
494 	if ((wd->sc_flags & WDF_LOADED) == 0 ||
495 	    !device_is_enabled(wd->sc_dev)) {
496 		bp->b_error = EIO;
497 		goto done;
498 	}
499 
500 	/* If it's a null transfer, return immediately. */
501 	if (bp->b_bcount == 0)
502 		goto done;
503 
504 	/*
505 	 * Do bounds checking, adjust transfer. if error, process.
506 	 * If end of partition, just return.
507 	 */
508 	if (WDPART(bp->b_dev) == RAW_PART) {
509 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
510 		    wd->sc_capacity) <= 0)
511 			goto done;
512 	} else {
513 		if (bounds_check_with_label(&wd->sc_dk, bp,
514 		    (wd->sc_flags & (WDF_WLABEL|WDF_LABELLING)) != 0) <= 0)
515 			goto done;
516 	}
517 
518 	/*
519 	 * Now convert the block number to absolute and put it in
520 	 * terms of the device's logical block size.
521 	 */
522 	if (lp->d_secsize >= DEV_BSIZE)
523 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
524 	else
525 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
526 
527 	if (WDPART(bp->b_dev) != RAW_PART)
528 		blkno += lp->d_partitions[WDPART(bp->b_dev)].p_offset;
529 
530 	bp->b_rawblkno = blkno;
531 
532 #ifdef WD_SOFTBADSECT
533 	/*
534 	 * If the transfer about to be attempted contains only a block that
535 	 * is known to be bad then return an error for the transfer without
536 	 * even attempting to start a transfer up under the premis that we
537 	 * will just end up doing more retries for a transfer that will end
538 	 * up failing again.
539 	 * XXX:SMP - mutex required to protect with DIOCBSFLUSH
540 	 */
541 	if (__predict_false(!SLIST_EMPTY(&wd->sc_bslist))) {
542 		struct disk_badsectors *dbs;
543 		daddr_t maxblk = blkno + (bp->b_bcount >> DEV_BSHIFT) - 1;
544 
545 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next)
546 			if ((dbs->dbs_min <= blkno && blkno <= dbs->dbs_max) ||
547 			    (dbs->dbs_min <= maxblk && maxblk <= dbs->dbs_max)){
548 				bp->b_error = EIO;
549 				goto done;
550 			}
551 	}
552 #endif
553 
554 	/* Queue transfer on drive, activate drive and controller if idle. */
555 	s = splbio();
556 	bufq_put(wd->sc_q, bp);
557 	wdstart(wd);
558 	splx(s);
559 	return;
560 done:
561 	/* Toss transfer; we're done early. */
562 	bp->b_resid = bp->b_bcount;
563 	biodone(bp);
564 }
565 
566 /*
567  * Queue a drive for I/O.
568  */
569 void
570 wdstart(void *arg)
571 {
572 	struct wd_softc *wd = arg;
573 	struct buf *bp = NULL;
574 
575 	ATADEBUG_PRINT(("wdstart %s\n", device_xname(wd->sc_dev)),
576 	    DEBUG_XFERS);
577 
578 	if (!device_is_active(wd->sc_dev))
579 		return;
580 
581 	while (wd->openings > 0) {
582 
583 		/* Is there a buf for us ? */
584 		if ((bp = bufq_get(wd->sc_q)) == NULL)
585 			return;
586 
587 		/*
588 		 * Make the command. First lock the device
589 		 */
590 		wd->openings--;
591 
592 		wd->retries = 0;
593 		wdstart1(wd, bp);
594 	}
595 }
596 
597 static void
598 wd_split_mod15_write(struct buf *bp)
599 {
600 	struct buf *obp = bp->b_private;
601 	struct wd_softc *sc =
602 	    device_lookup_private(&wd_cd, DISKUNIT(obp->b_dev));
603 	int s;
604 
605 	if (__predict_false(bp->b_error != 0)) {
606 		/*
607 		 * Propagate the error.  If this was the first half of
608 		 * the original transfer, make sure to account for that
609 		 * in the residual.
610 		 */
611 		if (bp->b_data == obp->b_data)
612 			bp->b_resid += bp->b_bcount;
613 		goto done;
614 	}
615 
616 	/*
617 	 * If this was the second half of the transfer, we're all done!
618 	 */
619 	if (bp->b_data != obp->b_data)
620 		goto done;
621 
622 	/*
623 	 * Advance the pointer to the second half and issue that command
624 	 * using the same opening.
625 	 */
626 	bp->b_flags = obp->b_flags;
627 	bp->b_oflags = obp->b_oflags;
628 	bp->b_cflags = obp->b_cflags;
629 	bp->b_data = (char *)bp->b_data + bp->b_bcount;
630 	bp->b_blkno += (bp->b_bcount / 512);
631 	bp->b_rawblkno += (bp->b_bcount / 512);
632 	s = splbio();
633 	wdstart1(sc, bp);
634 	splx(s);
635 	return;
636 
637  done:
638 	obp->b_error = bp->b_error;
639 	obp->b_resid = bp->b_resid;
640 	s = splbio();
641 	putiobuf(bp);
642 	biodone(obp);
643 	sc->openings++;
644 	splx(s);
645 	/* wddone() will call wdstart() */
646 }
647 
648 void
649 wdstart1(struct wd_softc *wd, struct buf *bp)
650 {
651 
652 	/*
653 	 * Deal with the "split mod15 write" quirk.  We just divide the
654 	 * transfer in two, doing the first half and then then second half
655 	 * with the same command opening.
656 	 *
657 	 * Note we MUST do this here, because we can't let insertion
658 	 * into the bufq cause the transfers to be re-merged.
659 	 */
660 	if (__predict_false((wd->sc_quirks & WD_QUIRK_SPLIT_MOD15_WRITE) != 0 &&
661 			    (bp->b_flags & B_READ) == 0 &&
662 			    bp->b_bcount > 512 &&
663 			    ((bp->b_bcount / 512) % 15) == 1)) {
664 		struct buf *nbp;
665 
666 		/* already at splbio */
667 		nbp = getiobuf(NULL, false);
668 		if (__predict_false(nbp == NULL)) {
669 			/* No memory -- fail the iop. */
670 			bp->b_error = ENOMEM;
671 			bp->b_resid = bp->b_bcount;
672 			biodone(bp);
673 			wd->openings++;
674 			return;
675 		}
676 
677 		nbp->b_error = 0;
678 		nbp->b_proc = bp->b_proc;
679 		nbp->b_dev = bp->b_dev;
680 
681 		nbp->b_bcount = bp->b_bcount / 2;
682 		nbp->b_bufsize = bp->b_bcount / 2;
683 		nbp->b_data = bp->b_data;
684 
685 		nbp->b_blkno = bp->b_blkno;
686 		nbp->b_rawblkno = bp->b_rawblkno;
687 
688 		nbp->b_flags = bp->b_flags;
689 		nbp->b_oflags = bp->b_oflags;
690 		nbp->b_cflags = bp->b_cflags;
691 		nbp->b_iodone = wd_split_mod15_write;
692 
693 		/* Put ptr to orig buf in b_private and use new buf */
694 		nbp->b_private = bp;
695 
696 		BIO_COPYPRIO(nbp, bp);
697 
698 		bp = nbp;
699 	}
700 
701 	wd->sc_wdc_bio.blkno = bp->b_rawblkno;
702 	wd->sc_wdc_bio.bcount = bp->b_bcount;
703 	wd->sc_wdc_bio.databuf = bp->b_data;
704 	wd->sc_wdc_bio.blkdone =0;
705 	wd->sc_bp = bp;
706 	/*
707 	 * If we're retrying, retry in single-sector mode. This will give us
708 	 * the sector number of the problem, and will eventually allow the
709 	 * transfer to succeed.
710 	 */
711 	if (wd->retries >= WDIORETRIES_SINGLE)
712 		wd->sc_wdc_bio.flags = ATA_SINGLE;
713 	else
714 		wd->sc_wdc_bio.flags = 0;
715 	if (wd->sc_flags & WDF_LBA48 &&
716 	    (wd->sc_wdc_bio.blkno +
717 	     wd->sc_wdc_bio.bcount / wd->sc_dk.dk_label->d_secsize) >
718 	    wd->sc_capacity28)
719 		wd->sc_wdc_bio.flags |= ATA_LBA48;
720 	if (wd->sc_flags & WDF_LBA)
721 		wd->sc_wdc_bio.flags |= ATA_LBA;
722 	if (bp->b_flags & B_READ)
723 		wd->sc_wdc_bio.flags |= ATA_READ;
724 	/* Instrumentation. */
725 	disk_busy(&wd->sc_dk);
726 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
727 	case ATACMD_TRY_AGAIN:
728 		callout_reset(&wd->sc_restart_ch, hz, wdrestart, wd);
729 		break;
730 	case ATACMD_QUEUED:
731 	case ATACMD_COMPLETE:
732 		break;
733 	default:
734 		panic("wdstart1: bad return code from ata_bio()");
735 	}
736 }
737 
738 void
739 wddone(void *v)
740 {
741 	struct wd_softc *wd = device_private(v);
742 	struct buf *bp = wd->sc_bp;
743 	const char *errmsg;
744 	int do_perror = 0;
745 
746 	ATADEBUG_PRINT(("wddone %s\n", device_xname(wd->sc_dev)),
747 	    DEBUG_XFERS);
748 	if (bp == NULL)
749 		return;
750 	bp->b_resid = wd->sc_wdc_bio.bcount;
751 	switch (wd->sc_wdc_bio.error) {
752 	case ERR_DMA:
753 		errmsg = "DMA error";
754 		goto retry;
755 	case ERR_DF:
756 		errmsg = "device fault";
757 		goto retry;
758 	case TIMEOUT:
759 		errmsg = "device timeout";
760 		goto retry;
761 	case ERR_RESET:
762 		errmsg = "channel reset";
763 		goto retry2;
764 	case ERROR:
765 		/* Don't care about media change bits */
766 		if (wd->sc_wdc_bio.r_error != 0 &&
767 		    (wd->sc_wdc_bio.r_error & ~(WDCE_MC | WDCE_MCR)) == 0)
768 			goto noerror;
769 		errmsg = "error";
770 		do_perror = 1;
771 retry:		/* Just reset and retry. Can we do more ? */
772 		(*wd->atabus->ata_reset_drive)(wd->drvp, AT_RST_NOCMD);
773 retry2:
774 		diskerr(bp, "wd", errmsg, LOG_PRINTF,
775 		    wd->sc_wdc_bio.blkdone, wd->sc_dk.dk_label);
776 		if (wd->retries < WDIORETRIES)
777 			printf(", retrying");
778 		printf("\n");
779 		if (do_perror)
780 			wdperror(wd);
781 		if (wd->retries < WDIORETRIES) {
782 			wd->retries++;
783 			callout_reset(&wd->sc_restart_ch, RECOVERYTIME,
784 			    wdrestart, wd);
785 			return;
786 		}
787 
788 #ifdef WD_SOFTBADSECT
789 		/*
790 		 * Not all errors indicate a failed block but those that do,
791 		 * put the block on the bad-block list for the device.  Only
792 		 * do this for reads because the drive should do it for writes,
793 		 * itself, according to Manuel.
794 		 */
795 		if ((bp->b_flags & B_READ) &&
796 		    ((wd->drvp->ata_vers >= 4 && wd->sc_wdc_bio.r_error & 64) ||
797 	     	     (wd->drvp->ata_vers < 4 && wd->sc_wdc_bio.r_error & 192))) {
798 			struct disk_badsectors *dbs;
799 
800 			dbs = malloc(sizeof *dbs, M_TEMP, M_WAITOK);
801 			dbs->dbs_min = bp->b_rawblkno;
802 			dbs->dbs_max = dbs->dbs_min + (bp->b_bcount >> DEV_BSHIFT) - 1;
803 			microtime(&dbs->dbs_failedat);
804 			SLIST_INSERT_HEAD(&wd->sc_bslist, dbs, dbs_next);
805 			wd->sc_bscount++;
806 		}
807 #endif
808 		bp->b_error = EIO;
809 		break;
810 	case NOERROR:
811 noerror:	if ((wd->sc_wdc_bio.flags & ATA_CORR) || wd->retries > 0)
812 			aprint_error_dev(wd->sc_dev,
813 			    "soft error (corrected)\n");
814 		break;
815 	case ERR_NODEV:
816 		bp->b_error = EIO;
817 		break;
818 	}
819 	disk_unbusy(&wd->sc_dk, (bp->b_bcount - bp->b_resid),
820 	    (bp->b_flags & B_READ));
821 #if NRND > 0
822 	rnd_add_uint32(&wd->rnd_source, bp->b_blkno);
823 #endif
824 	/* XXX Yuck, but we don't want to increment openings in this case */
825 	if (__predict_false(bp->b_iodone == wd_split_mod15_write))
826 		biodone(bp);
827 	else {
828 		biodone(bp);
829 		wd->openings++;
830 	}
831 	wdstart(wd);
832 }
833 
834 void
835 wdrestart(void *v)
836 {
837 	struct wd_softc *wd = v;
838 	struct buf *bp = wd->sc_bp;
839 	int s;
840 
841 	ATADEBUG_PRINT(("wdrestart %s\n", device_xname(wd->sc_dev)),
842 	    DEBUG_XFERS);
843 	s = splbio();
844 	wdstart1(v, bp);
845 	splx(s);
846 }
847 
848 int
849 wdread(dev_t dev, struct uio *uio, int flags)
850 {
851 
852 	ATADEBUG_PRINT(("wdread\n"), DEBUG_XFERS);
853 	return (physio(wdstrategy, NULL, dev, B_READ, minphys, uio));
854 }
855 
856 int
857 wdwrite(dev_t dev, struct uio *uio, int flags)
858 {
859 
860 	ATADEBUG_PRINT(("wdwrite\n"), DEBUG_XFERS);
861 	return (physio(wdstrategy, NULL, dev, B_WRITE, minphys, uio));
862 }
863 
864 int
865 wdopen(dev_t dev, int flag, int fmt, struct lwp *l)
866 {
867 	struct wd_softc *wd;
868 	int part, error;
869 
870 	ATADEBUG_PRINT(("wdopen\n"), DEBUG_FUNCS);
871 	wd = device_lookup_private(&wd_cd, WDUNIT(dev));
872 	if (wd == NULL)
873 		return (ENXIO);
874 
875 	if (! device_is_active(wd->sc_dev))
876 		return (ENODEV);
877 
878 	part = WDPART(dev);
879 
880 	mutex_enter(&wd->sc_dk.dk_openlock);
881 
882 	/*
883 	 * If there are wedges, and this is not RAW_PART, then we
884 	 * need to fail.
885 	 */
886 	if (wd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
887 		error = EBUSY;
888 		goto bad1;
889 	}
890 
891 	/*
892 	 * If this is the first open of this device, add a reference
893 	 * to the adapter.
894 	 */
895 	if (wd->sc_dk.dk_openmask == 0 &&
896 	    (error = wd->atabus->ata_addref(wd->drvp)) != 0)
897 		goto bad1;
898 
899 	if (wd->sc_dk.dk_openmask != 0) {
900 		/*
901 		 * If any partition is open, but the disk has been invalidated,
902 		 * disallow further opens.
903 		 */
904 		if ((wd->sc_flags & WDF_LOADED) == 0) {
905 			error = EIO;
906 			goto bad2;
907 		}
908 	} else {
909 		if ((wd->sc_flags & WDF_LOADED) == 0) {
910 			wd->sc_flags |= WDF_LOADED;
911 
912 			/* Load the physical device parameters. */
913 			wd_get_params(wd, AT_WAIT, &wd->sc_params);
914 
915 			/* Load the partition info if not already loaded. */
916 			wdgetdisklabel(wd);
917 		}
918 	}
919 
920 	/* Check that the partition exists. */
921 	if (part != RAW_PART &&
922 	    (part >= wd->sc_dk.dk_label->d_npartitions ||
923 	     wd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
924 		error = ENXIO;
925 		goto bad2;
926 	}
927 
928 	/* Insure only one open at a time. */
929 	switch (fmt) {
930 	case S_IFCHR:
931 		wd->sc_dk.dk_copenmask |= (1 << part);
932 		break;
933 	case S_IFBLK:
934 		wd->sc_dk.dk_bopenmask |= (1 << part);
935 		break;
936 	}
937 	wd->sc_dk.dk_openmask =
938 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
939 
940 	mutex_exit(&wd->sc_dk.dk_openlock);
941 	return 0;
942 
943  bad2:
944 	if (wd->sc_dk.dk_openmask == 0)
945 		wd->atabus->ata_delref(wd->drvp);
946  bad1:
947 	mutex_exit(&wd->sc_dk.dk_openlock);
948 	return error;
949 }
950 
951 /*
952  * Caller must hold wd->sc_dk.dk_openlock.
953  */
954 static int
955 wdlastclose(device_t self)
956 {
957 	struct wd_softc *wd = device_private(self);
958 
959 	wd_flushcache(wd, AT_WAIT);
960 
961 	if (! (wd->sc_flags & WDF_KLABEL))
962 		wd->sc_flags &= ~WDF_LOADED;
963 
964 	wd->atabus->ata_delref(wd->drvp);
965 
966 	return 0;
967 }
968 
969 int
970 wdclose(dev_t dev, int flag, int fmt, struct lwp *l)
971 {
972 	struct wd_softc *wd =
973 	    device_lookup_private(&wd_cd, WDUNIT(dev));
974 	int part = WDPART(dev);
975 
976 	ATADEBUG_PRINT(("wdclose\n"), DEBUG_FUNCS);
977 
978 	mutex_enter(&wd->sc_dk.dk_openlock);
979 
980 	switch (fmt) {
981 	case S_IFCHR:
982 		wd->sc_dk.dk_copenmask &= ~(1 << part);
983 		break;
984 	case S_IFBLK:
985 		wd->sc_dk.dk_bopenmask &= ~(1 << part);
986 		break;
987 	}
988 	wd->sc_dk.dk_openmask =
989 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
990 
991 	if (wd->sc_dk.dk_openmask == 0)
992 		wdlastclose(wd->sc_dev);
993 
994 	mutex_exit(&wd->sc_dk.dk_openlock);
995 	return 0;
996 }
997 
998 void
999 wdgetdefaultlabel(struct wd_softc *wd, struct disklabel *lp)
1000 {
1001 
1002 	ATADEBUG_PRINT(("wdgetdefaultlabel\n"), DEBUG_FUNCS);
1003 	memset(lp, 0, sizeof(struct disklabel));
1004 
1005 	lp->d_secsize = DEV_BSIZE;
1006 	lp->d_ntracks = wd->sc_params.atap_heads;
1007 	lp->d_nsectors = wd->sc_params.atap_sectors;
1008 	lp->d_ncylinders = (wd->sc_flags & WDF_LBA) ? wd->sc_capacity /
1009 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors) :
1010 		wd->sc_params.atap_cylinders;
1011 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1012 
1013 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
1014 		lp->d_type = DTYPE_ST506;
1015 	else
1016 		lp->d_type = DTYPE_ESDI;
1017 
1018 	strncpy(lp->d_typename, wd->sc_params.atap_model, 16);
1019 	strncpy(lp->d_packname, "fictitious", 16);
1020 	if (wd->sc_capacity > UINT32_MAX)
1021 		lp->d_secperunit = UINT32_MAX;
1022 	else
1023 		lp->d_secperunit = wd->sc_capacity;
1024 	lp->d_rpm = 3600;
1025 	lp->d_interleave = 1;
1026 	lp->d_flags = 0;
1027 
1028 	lp->d_partitions[RAW_PART].p_offset = 0;
1029 	lp->d_partitions[RAW_PART].p_size =
1030 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1031 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1032 	lp->d_npartitions = RAW_PART + 1;
1033 
1034 	lp->d_magic = DISKMAGIC;
1035 	lp->d_magic2 = DISKMAGIC;
1036 	lp->d_checksum = dkcksum(lp);
1037 }
1038 
1039 /*
1040  * Fabricate a default disk label, and try to read the correct one.
1041  */
1042 void
1043 wdgetdisklabel(struct wd_softc *wd)
1044 {
1045 	struct disklabel *lp = wd->sc_dk.dk_label;
1046 	const char *errstring;
1047 	int s;
1048 
1049 	ATADEBUG_PRINT(("wdgetdisklabel\n"), DEBUG_FUNCS);
1050 
1051 	memset(wd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1052 
1053 	wdgetdefaultlabel(wd, lp);
1054 
1055 	wd->sc_badsect[0] = -1;
1056 
1057 	if (wd->drvp->state > RESET) {
1058 		s = splbio();
1059 		wd->drvp->drive_flags |= DRIVE_RESET;
1060 		splx(s);
1061 	}
1062 	errstring = readdisklabel(MAKEWDDEV(0, device_unit(wd->sc_dev),
1063 				  RAW_PART), wdstrategy, lp,
1064 				  wd->sc_dk.dk_cpulabel);
1065 	if (errstring) {
1066 		/*
1067 		 * This probably happened because the drive's default
1068 		 * geometry doesn't match the DOS geometry.  We
1069 		 * assume the DOS geometry is now in the label and try
1070 		 * again.  XXX This is a kluge.
1071 		 */
1072 		if (wd->drvp->state > RESET) {
1073 			s = splbio();
1074 			wd->drvp->drive_flags |= DRIVE_RESET;
1075 			splx(s);
1076 		}
1077 		errstring = readdisklabel(MAKEWDDEV(0, device_unit(wd->sc_dev),
1078 		    RAW_PART), wdstrategy, lp, wd->sc_dk.dk_cpulabel);
1079 	}
1080 	if (errstring) {
1081 		aprint_error_dev(wd->sc_dev, "%s\n", errstring);
1082 		return;
1083 	}
1084 
1085 	if (wd->drvp->state > RESET) {
1086 		s = splbio();
1087 		wd->drvp->drive_flags |= DRIVE_RESET;
1088 		splx(s);
1089 	}
1090 #ifdef HAS_BAD144_HANDLING
1091 	if ((lp->d_flags & D_BADSECT) != 0)
1092 		bad144intern(wd);
1093 #endif
1094 }
1095 
1096 void
1097 wdperror(const struct wd_softc *wd)
1098 {
1099 	static const char *const errstr0_3[] = {"address mark not found",
1100 	    "track 0 not found", "aborted command", "media change requested",
1101 	    "id not found", "media changed", "uncorrectable data error",
1102 	    "bad block detected"};
1103 	static const char *const errstr4_5[] = {
1104 	    "obsolete (address mark not found)",
1105 	    "no media/write protected", "aborted command",
1106 	    "media change requested", "id not found", "media changed",
1107 	    "uncorrectable data error", "interface CRC error"};
1108 	const char *const *errstr;
1109 	int i;
1110 	const char *sep = "";
1111 
1112 	const char *devname = device_xname(wd->sc_dev);
1113 	struct ata_drive_datas *drvp = wd->drvp;
1114 	int errno = wd->sc_wdc_bio.r_error;
1115 
1116 	if (drvp->ata_vers >= 4)
1117 		errstr = errstr4_5;
1118 	else
1119 		errstr = errstr0_3;
1120 
1121 	printf("%s: (", devname);
1122 
1123 	if (errno == 0)
1124 		printf("error not notified");
1125 
1126 	for (i = 0; i < 8; i++) {
1127 		if (errno & (1 << i)) {
1128 			printf("%s%s", sep, errstr[i]);
1129 			sep = ", ";
1130 		}
1131 	}
1132 	printf(")\n");
1133 }
1134 
1135 int
1136 wdioctl(dev_t dev, u_long xfer, void *addr, int flag, struct lwp *l)
1137 {
1138 	struct wd_softc *wd =
1139 	    device_lookup_private(&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 	error = disk_ioctl(&wd->sc_dk, xfer, addr, flag, l);
1151 	if (error != EPASSTHROUGH)
1152 		return (error);
1153 
1154 	switch (xfer) {
1155 #ifdef HAS_BAD144_HANDLING
1156 	case DIOCSBAD:
1157 		if ((flag & FWRITE) == 0)
1158 			return EBADF;
1159 		wd->sc_dk.dk_cpulabel->bad = *(struct dkbad *)addr;
1160 		wd->sc_dk.dk_label->d_flags |= D_BADSECT;
1161 		bad144intern(wd);
1162 		return 0;
1163 #endif
1164 #ifdef WD_SOFTBADSECT
1165 	case DIOCBSLIST :
1166 	{
1167 		u_int32_t count, missing, skip;
1168 		struct disk_badsecinfo dbsi;
1169 		struct disk_badsectors *dbs;
1170 		size_t available;
1171 		uint8_t *laddr;
1172 
1173 		dbsi = *(struct disk_badsecinfo *)addr;
1174 		missing = wd->sc_bscount;
1175 		count = 0;
1176 		available = dbsi.dbsi_bufsize;
1177 		skip = dbsi.dbsi_skip;
1178 		laddr = (uint8_t *)dbsi.dbsi_buffer;
1179 
1180 		/*
1181 		 * We start this loop with the expectation that all of the
1182 		 * entries will be missed and decrement this counter each
1183 		 * time we either skip over one (already copied out) or
1184 		 * we actually copy it back to user space.  The structs
1185 		 * holding the bad sector information are copied directly
1186 		 * back to user space whilst the summary is returned via
1187 		 * the struct passed in via the ioctl.
1188 		 */
1189 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next) {
1190 			if (skip > 0) {
1191 				missing--;
1192 				skip--;
1193 				continue;
1194 			}
1195 			if (available < sizeof(*dbs))
1196 				break;
1197 			available -= sizeof(*dbs);
1198 			copyout(dbs, laddr, sizeof(*dbs));
1199 			laddr += sizeof(*dbs);
1200 			missing--;
1201 			count++;
1202 		}
1203 		dbsi.dbsi_left = missing;
1204 		dbsi.dbsi_copied = count;
1205 		*(struct disk_badsecinfo *)addr = dbsi;
1206 		return 0;
1207 	}
1208 
1209 	case DIOCBSFLUSH :
1210 		/* Clean out the bad sector list */
1211 		while (!SLIST_EMPTY(&wd->sc_bslist)) {
1212 			void *head = SLIST_FIRST(&wd->sc_bslist);
1213 			SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
1214 			free(head, M_TEMP);
1215 		}
1216 		wd->sc_bscount = 0;
1217 		return 0;
1218 #endif
1219 	case DIOCGDINFO:
1220 		*(struct disklabel *)addr = *(wd->sc_dk.dk_label);
1221 		return 0;
1222 #ifdef __HAVE_OLD_DISKLABEL
1223 	case ODIOCGDINFO:
1224 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1225 		if (newlabel == NULL)
1226 			return EIO;
1227 		*newlabel = *(wd->sc_dk.dk_label);
1228 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1229 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1230 		else
1231 			error = ENOTTY;
1232 		free(newlabel, M_TEMP);
1233 		return error;
1234 #endif
1235 
1236 	case DIOCGPART:
1237 		((struct partinfo *)addr)->disklab = wd->sc_dk.dk_label;
1238 		((struct partinfo *)addr)->part =
1239 		    &wd->sc_dk.dk_label->d_partitions[WDPART(dev)];
1240 		return 0;
1241 
1242 	case DIOCWDINFO:
1243 	case DIOCSDINFO:
1244 #ifdef __HAVE_OLD_DISKLABEL
1245 	case ODIOCWDINFO:
1246 	case ODIOCSDINFO:
1247 #endif
1248 	{
1249 		struct disklabel *lp;
1250 
1251 		if ((flag & FWRITE) == 0)
1252 			return EBADF;
1253 
1254 #ifdef __HAVE_OLD_DISKLABEL
1255 		if (xfer == ODIOCSDINFO || xfer == ODIOCWDINFO) {
1256 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1257 			if (newlabel == NULL)
1258 				return EIO;
1259 			memset(newlabel, 0, sizeof newlabel);
1260 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1261 			lp = newlabel;
1262 		} else
1263 #endif
1264 		lp = (struct disklabel *)addr;
1265 
1266 		mutex_enter(&wd->sc_dk.dk_openlock);
1267 		wd->sc_flags |= WDF_LABELLING;
1268 
1269 		error = setdisklabel(wd->sc_dk.dk_label,
1270 		    lp, /*wd->sc_dk.dk_openmask : */0,
1271 		    wd->sc_dk.dk_cpulabel);
1272 		if (error == 0) {
1273 			if (wd->drvp->state > RESET) {
1274 				s = splbio();
1275 				wd->drvp->drive_flags |= DRIVE_RESET;
1276 				splx(s);
1277 			}
1278 			if (xfer == DIOCWDINFO
1279 #ifdef __HAVE_OLD_DISKLABEL
1280 			    || xfer == ODIOCWDINFO
1281 #endif
1282 			    )
1283 				error = writedisklabel(WDLABELDEV(dev),
1284 				    wdstrategy, wd->sc_dk.dk_label,
1285 				    wd->sc_dk.dk_cpulabel);
1286 		}
1287 
1288 		wd->sc_flags &= ~WDF_LABELLING;
1289 		mutex_exit(&wd->sc_dk.dk_openlock);
1290 #ifdef __HAVE_OLD_DISKLABEL
1291 		if (newlabel != NULL)
1292 			free(newlabel, M_TEMP);
1293 #endif
1294 		return error;
1295 	}
1296 
1297 	case DIOCKLABEL:
1298 		if (*(int *)addr)
1299 			wd->sc_flags |= WDF_KLABEL;
1300 		else
1301 			wd->sc_flags &= ~WDF_KLABEL;
1302 		return 0;
1303 
1304 	case DIOCWLABEL:
1305 		if ((flag & FWRITE) == 0)
1306 			return EBADF;
1307 		if (*(int *)addr)
1308 			wd->sc_flags |= WDF_WLABEL;
1309 		else
1310 			wd->sc_flags &= ~WDF_WLABEL;
1311 		return 0;
1312 
1313 	case DIOCGDEFLABEL:
1314 		wdgetdefaultlabel(wd, (struct disklabel *)addr);
1315 		return 0;
1316 #ifdef __HAVE_OLD_DISKLABEL
1317 	case ODIOCGDEFLABEL:
1318 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1319 		if (newlabel == NULL)
1320 			return EIO;
1321 		wdgetdefaultlabel(wd, newlabel);
1322 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1323 			memcpy(addr, &newlabel, sizeof (struct olddisklabel));
1324 		else
1325 			error = ENOTTY;
1326 		free(newlabel, M_TEMP);
1327 		return error;
1328 #endif
1329 
1330 #ifdef notyet
1331 	case DIOCWFORMAT:
1332 		if ((flag & FWRITE) == 0)
1333 			return EBADF;
1334 		{
1335 		register struct format_op *fop;
1336 		struct iovec aiov;
1337 		struct uio auio;
1338 
1339 		fop = (struct format_op *)addr;
1340 		aiov.iov_base = fop->df_buf;
1341 		aiov.iov_len = fop->df_count;
1342 		auio.uio_iov = &aiov;
1343 		auio.uio_iovcnt = 1;
1344 		auio.uio_resid = fop->df_count;
1345 		auio.uio_offset =
1346 			fop->df_startblk * wd->sc_dk.dk_label->d_secsize;
1347 		auio.uio_vmspace = l->l_proc->p_vmspace;
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 			void *tbuf;
1384 			if (atareq->datalen < DEV_BSIZE
1385 			    && atareq->command == WDCC_IDENTIFY) {
1386 				tbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1387 				wi->wi_iov.iov_base = tbuf;
1388 				wi->wi_iov.iov_len = DEV_BSIZE;
1389 				UIO_SETUP_SYSSPACE(&wi->wi_uio);
1390 			} else {
1391 				tbuf = NULL;
1392 				wi->wi_iov.iov_base = atareq->databuf;
1393 				wi->wi_iov.iov_len = atareq->datalen;
1394 				wi->wi_uio.uio_vmspace = l->l_proc->p_vmspace;
1395 			}
1396 			wi->wi_uio.uio_iov = &wi->wi_iov;
1397 			wi->wi_uio.uio_iovcnt = 1;
1398 			wi->wi_uio.uio_resid = atareq->datalen;
1399 			wi->wi_uio.uio_offset = 0;
1400 			wi->wi_uio.uio_rw =
1401 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE;
1402 			error1 = physio(wdioctlstrategy, &wi->wi_bp, dev,
1403 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE,
1404 			    minphys, &wi->wi_uio);
1405 			if (tbuf != NULL && error1 == 0) {
1406 				error1 = copyout(tbuf, atareq->databuf,
1407 				    atareq->datalen);
1408 				free(tbuf, M_TEMP);
1409 			}
1410 		} else {
1411 			/* No need to call physio if we don't have any
1412 			   user data */
1413 			wi->wi_bp.b_flags = 0;
1414 			wi->wi_bp.b_data = 0;
1415 			wi->wi_bp.b_bcount = 0;
1416 			wi->wi_bp.b_dev = 0;
1417 			wi->wi_bp.b_proc = l->l_proc;
1418 			wdioctlstrategy(&wi->wi_bp);
1419 			error1 = wi->wi_bp.b_error;
1420 		}
1421 		*atareq = wi->wi_atareq;
1422 		wi_free(wi);
1423 		return(error1);
1424 		}
1425 
1426 	case DIOCAWEDGE:
1427 	    {
1428 	    	struct dkwedge_info *dkw = (void *) addr;
1429 
1430 		if ((flag & FWRITE) == 0)
1431 			return (EBADF);
1432 
1433 		/* If the ioctl happens here, the parent is us. */
1434 		strcpy(dkw->dkw_parent, device_xname(wd->sc_dev));
1435 		return (dkwedge_add(dkw));
1436 	    }
1437 
1438 	case DIOCDWEDGE:
1439 	    {
1440 	    	struct dkwedge_info *dkw = (void *) addr;
1441 
1442 		if ((flag & FWRITE) == 0)
1443 			return (EBADF);
1444 
1445 		/* If the ioctl happens here, the parent is us. */
1446 		strcpy(dkw->dkw_parent, device_xname(wd->sc_dev));
1447 		return (dkwedge_del(dkw));
1448 	    }
1449 
1450 	case DIOCLWEDGES:
1451 	    {
1452 	    	struct dkwedge_list *dkwl = (void *) addr;
1453 
1454 		return (dkwedge_list(&wd->sc_dk, dkwl, l));
1455 	    }
1456 
1457 	case DIOCGSTRATEGY:
1458 	    {
1459 		struct disk_strategy *dks = (void *)addr;
1460 
1461 		s = splbio();
1462 		strlcpy(dks->dks_name, bufq_getstrategyname(wd->sc_q),
1463 		    sizeof(dks->dks_name));
1464 		splx(s);
1465 		dks->dks_paramlen = 0;
1466 
1467 		return 0;
1468 	    }
1469 
1470 	case DIOCSSTRATEGY:
1471 	    {
1472 		struct disk_strategy *dks = (void *)addr;
1473 		struct bufq_state *new;
1474 		struct bufq_state *old;
1475 
1476 		if ((flag & FWRITE) == 0) {
1477 			return EBADF;
1478 		}
1479 		if (dks->dks_param != NULL) {
1480 			return EINVAL;
1481 		}
1482 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
1483 		error = bufq_alloc(&new, dks->dks_name,
1484 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
1485 		if (error) {
1486 			return error;
1487 		}
1488 		s = splbio();
1489 		old = wd->sc_q;
1490 		bufq_move(new, old);
1491 		wd->sc_q = new;
1492 		splx(s);
1493 		bufq_free(old);
1494 
1495 		return 0;
1496 	    }
1497 
1498 	default:
1499 		return ENOTTY;
1500 	}
1501 
1502 #ifdef DIAGNOSTIC
1503 	panic("wdioctl: impossible");
1504 #endif
1505 }
1506 
1507 #ifdef B_FORMAT
1508 int
1509 wdformat(struct buf *bp)
1510 {
1511 
1512 	bp->b_flags |= B_FORMAT;
1513 	return wdstrategy(bp);
1514 }
1515 #endif
1516 
1517 int
1518 wdsize(dev_t dev)
1519 {
1520 	struct wd_softc *wd;
1521 	int part, omask;
1522 	int size;
1523 
1524 	ATADEBUG_PRINT(("wdsize\n"), DEBUG_FUNCS);
1525 
1526 	wd = device_lookup_private(&wd_cd, WDUNIT(dev));
1527 	if (wd == NULL)
1528 		return (-1);
1529 
1530 	part = WDPART(dev);
1531 	omask = wd->sc_dk.dk_openmask & (1 << part);
1532 
1533 	if (omask == 0 && wdopen(dev, 0, S_IFBLK, NULL) != 0)
1534 		return (-1);
1535 	if (wd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1536 		size = -1;
1537 	else
1538 		size = wd->sc_dk.dk_label->d_partitions[part].p_size *
1539 		    (wd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1540 	if (omask == 0 && wdclose(dev, 0, S_IFBLK, NULL) != 0)
1541 		return (-1);
1542 	return (size);
1543 }
1544 
1545 /* #define WD_DUMP_NOT_TRUSTED if you just want to watch */
1546 static int wddoingadump = 0;
1547 static int wddumprecalibrated = 0;
1548 
1549 /*
1550  * Dump core after a system crash.
1551  */
1552 int
1553 wddump(dev_t dev, daddr_t blkno, void *va, size_t size)
1554 {
1555 	struct wd_softc *wd;	/* disk unit to do the I/O */
1556 	struct disklabel *lp;   /* disk's disklabel */
1557 	int part, err;
1558 	int nblks;	/* total number of sectors left to write */
1559 
1560 	/* Check if recursive dump; if so, punt. */
1561 	if (wddoingadump)
1562 		return EFAULT;
1563 	wddoingadump = 1;
1564 
1565 	wd = device_lookup_private(&wd_cd, WDUNIT(dev));
1566 	if (wd == NULL)
1567 		return (ENXIO);
1568 
1569 	part = WDPART(dev);
1570 
1571 	/* Convert to disk sectors.  Request must be a multiple of size. */
1572 	lp = wd->sc_dk.dk_label;
1573 	if ((size % lp->d_secsize) != 0)
1574 		return EFAULT;
1575 	nblks = size / lp->d_secsize;
1576 	blkno = blkno / (lp->d_secsize / DEV_BSIZE);
1577 
1578 	/* Check transfer bounds against partition size. */
1579 	if ((blkno < 0) || ((blkno + nblks) > lp->d_partitions[part].p_size))
1580 		return EINVAL;
1581 
1582 	/* Offset block number to start of partition. */
1583 	blkno += lp->d_partitions[part].p_offset;
1584 
1585 	/* Recalibrate, if first dump transfer. */
1586 	if (wddumprecalibrated == 0) {
1587 		wddumprecalibrated = 1;
1588 		(*wd->atabus->ata_reset_drive)(wd->drvp,
1589 					       AT_POLL | AT_RST_EMERG);
1590 		wd->drvp->state = RESET;
1591 	}
1592 
1593 	wd->sc_bp = NULL;
1594 	wd->sc_wdc_bio.blkno = blkno;
1595 	wd->sc_wdc_bio.flags = ATA_POLL;
1596 	if (wd->sc_flags & WDF_LBA48 &&
1597 	    (wd->sc_wdc_bio.blkno + nblks) > wd->sc_capacity28)
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 (err = 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 	default:
1614 		panic("wddump: unknown atacmd code %d", err);
1615 	}
1616 	switch(err = wd->sc_wdc_bio.error) {
1617 	case TIMEOUT:
1618 		printf("wddump: device timed out");
1619 		err = EIO;
1620 		break;
1621 	case ERR_DF:
1622 		printf("wddump: drive fault");
1623 		err = EIO;
1624 		break;
1625 	case ERR_DMA:
1626 		printf("wddump: DMA error");
1627 		err = EIO;
1628 		break;
1629 	case ERROR:
1630 		printf("wddump: ");
1631 		wdperror(wd);
1632 		err = EIO;
1633 		break;
1634 	case NOERROR:
1635 		err = 0;
1636 		break;
1637 	default:
1638 		panic("wddump: unknown error type %d", err);
1639 	}
1640 	if (err != 0) {
1641 		printf("\n");
1642 		return err;
1643 	}
1644 #else	/* WD_DUMP_NOT_TRUSTED */
1645 	/* Let's just talk about this first... */
1646 	printf("wd%d: dump addr 0x%x, cylin %d, head %d, sector %d\n",
1647 	    unit, va, cylin, head, sector);
1648 	delay(500 * 1000);	/* half a second */
1649 #endif
1650 
1651 	wddoingadump = 0;
1652 	return 0;
1653 }
1654 
1655 #ifdef HAS_BAD144_HANDLING
1656 /*
1657  * Internalize the bad sector table.
1658  */
1659 void
1660 bad144intern(struct wd_softc *wd)
1661 {
1662 	struct dkbad *bt = &wd->sc_dk.dk_cpulabel->bad;
1663 	struct disklabel *lp = wd->sc_dk.dk_label;
1664 	int i = 0;
1665 
1666 	ATADEBUG_PRINT(("bad144intern\n"), DEBUG_XFERS);
1667 
1668 	for (; i < NBT_BAD; i++) {
1669 		if (bt->bt_bad[i].bt_cyl == 0xffff)
1670 			break;
1671 		wd->sc_badsect[i] =
1672 		    bt->bt_bad[i].bt_cyl * lp->d_secpercyl +
1673 		    (bt->bt_bad[i].bt_trksec >> 8) * lp->d_nsectors +
1674 		    (bt->bt_bad[i].bt_trksec & 0xff);
1675 	}
1676 	for (; i < NBT_BAD+1; i++)
1677 		wd->sc_badsect[i] = -1;
1678 }
1679 #endif
1680 
1681 static void
1682 wd_params_to_properties(struct wd_softc *wd, struct ataparams *params)
1683 {
1684 	prop_dictionary_t disk_info, odisk_info, geom;
1685 	const char *cp;
1686 
1687 	disk_info = prop_dictionary_create();
1688 
1689 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
1690 		cp = "ST506";
1691 	else {
1692 		/* XXX Should have a case for ATA here, too. */
1693 		cp = "ESDI";
1694 	}
1695 	prop_dictionary_set_cstring_nocopy(disk_info, "type", cp);
1696 
1697 	geom = prop_dictionary_create();
1698 
1699 	prop_dictionary_set_uint64(geom, "sectors-per-unit", wd->sc_capacity);
1700 
1701 	prop_dictionary_set_uint32(geom, "sector-size",
1702 				   DEV_BSIZE /* XXX 512? */);
1703 
1704 	prop_dictionary_set_uint16(geom, "sectors-per-track",
1705 				   wd->sc_params.atap_sectors);
1706 
1707 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
1708 				   wd->sc_params.atap_heads);
1709 
1710 	if (wd->sc_flags & WDF_LBA)
1711 		prop_dictionary_set_uint64(geom, "cylinders-per-unit",
1712 					   wd->sc_capacity /
1713 					       (wd->sc_params.atap_heads *
1714 					        wd->sc_params.atap_sectors));
1715 	else
1716 		prop_dictionary_set_uint16(geom, "cylinders-per-unit",
1717 					   wd->sc_params.atap_cylinders);
1718 
1719 	prop_dictionary_set(disk_info, "geometry", geom);
1720 	prop_object_release(geom);
1721 
1722 	prop_dictionary_set(device_properties(wd->sc_dev),
1723 			    "disk-info", disk_info);
1724 
1725 	/*
1726 	 * Don't release disk_info here; we keep a reference to it.
1727 	 * disk_detach() will release it when we go away.
1728 	 */
1729 
1730 	odisk_info = wd->sc_dk.dk_info;
1731 	wd->sc_dk.dk_info = disk_info;
1732 	if (odisk_info)
1733 		prop_object_release(odisk_info);
1734 }
1735 
1736 int
1737 wd_get_params(struct wd_softc *wd, u_int8_t flags, struct ataparams *params)
1738 {
1739 
1740 	switch (wd->atabus->ata_get_params(wd->drvp, flags, params)) {
1741 	case CMD_AGAIN:
1742 		return 1;
1743 	case CMD_ERR:
1744 		/*
1745 		 * We `know' there's a drive here; just assume it's old.
1746 		 * This geometry is only used to read the MBR and print a
1747 		 * (false) attach message.
1748 		 */
1749 		strncpy(params->atap_model, "ST506",
1750 		    sizeof params->atap_model);
1751 		params->atap_config = ATA_CFG_FIXED;
1752 		params->atap_cylinders = 1024;
1753 		params->atap_heads = 8;
1754 		params->atap_sectors = 17;
1755 		params->atap_multi = 1;
1756 		params->atap_capabilities1 = params->atap_capabilities2 = 0;
1757 		wd->drvp->ata_vers = -1; /* Mark it as pre-ATA */
1758 		/* FALLTHROUGH */
1759 	case CMD_OK:
1760 		wd_params_to_properties(wd, params);
1761 		return 0;
1762 	default:
1763 		panic("wd_get_params: bad return code from ata_get_params");
1764 		/* NOTREACHED */
1765 	}
1766 }
1767 
1768 int
1769 wd_getcache(struct wd_softc *wd, int *bitsp)
1770 {
1771 	struct ataparams params;
1772 
1773 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
1774 		return EIO;
1775 	if (params.atap_cmd_set1 == 0x0000 ||
1776 	    params.atap_cmd_set1 == 0xffff ||
1777 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0) {
1778 		*bitsp = 0;
1779 		return 0;
1780 	}
1781 	*bitsp = DKCACHE_WCHANGE | DKCACHE_READ;
1782 	if (params.atap_cmd1_en & WDC_CMD1_CACHE)
1783 		*bitsp |= DKCACHE_WRITE;
1784 
1785 	return 0;
1786 }
1787 
1788 const char at_errbits[] = "\20\10ERROR\11TIMEOU\12DF";
1789 
1790 int
1791 wd_setcache(struct wd_softc *wd, int bits)
1792 {
1793 	struct ataparams params;
1794 	struct ata_command ata_c;
1795 
1796 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
1797 		return EIO;
1798 
1799 	if (params.atap_cmd_set1 == 0x0000 ||
1800 	    params.atap_cmd_set1 == 0xffff ||
1801 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0)
1802 		return EOPNOTSUPP;
1803 
1804 	if ((bits & DKCACHE_READ) == 0 ||
1805 	    (bits & DKCACHE_SAVE) != 0)
1806 		return EOPNOTSUPP;
1807 
1808 	memset(&ata_c, 0, sizeof(struct ata_command));
1809 	ata_c.r_command = SET_FEATURES;
1810 	ata_c.r_st_bmask = 0;
1811 	ata_c.r_st_pmask = 0;
1812 	ata_c.timeout = 30000; /* 30s timeout */
1813 	ata_c.flags = AT_WAIT;
1814 	if (bits & DKCACHE_WRITE)
1815 		ata_c.r_features = WDSF_WRITE_CACHE_EN;
1816 	else
1817 		ata_c.r_features = WDSF_WRITE_CACHE_DS;
1818 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1819 		aprint_error_dev(wd->sc_dev,
1820 		    "wd_setcache command not complete\n");
1821 		return EIO;
1822 	}
1823 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1824 		char sbuf[sizeof(at_errbits) + 64];
1825 		snprintb(sbuf, sizeof(sbuf), at_errbits, ata_c.flags);
1826 		aprint_error_dev(wd->sc_dev, "wd_setcache: status=%s\n", sbuf);
1827 		return EIO;
1828 	}
1829 	return 0;
1830 }
1831 
1832 static int
1833 wd_standby(struct wd_softc *wd, int flags)
1834 {
1835 	struct ata_command ata_c;
1836 
1837 	memset(&ata_c, 0, sizeof(struct ata_command));
1838 	ata_c.r_command = WDCC_STANDBY_IMMED;
1839 	ata_c.r_st_bmask = WDCS_DRDY;
1840 	ata_c.r_st_pmask = WDCS_DRDY;
1841 	ata_c.flags = flags;
1842 	ata_c.timeout = 30000; /* 30s timeout */
1843 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1844 		aprint_error_dev(wd->sc_dev,
1845 		    "standby immediate command didn't complete\n");
1846 		return EIO;
1847 	}
1848 	if (ata_c.flags & AT_ERROR) {
1849 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
1850 			return ENODEV;
1851 	}
1852 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1853 		char sbuf[sizeof(at_errbits) + 64];
1854 		snprintb(sbuf, sizeof(sbuf), at_errbits, ata_c.flags);
1855 		aprint_error_dev(wd->sc_dev, "wd_standby: status=%s\n", sbuf);
1856 		return EIO;
1857 	}
1858 	return 0;
1859 }
1860 
1861 int
1862 wd_flushcache(struct wd_softc *wd, int flags)
1863 {
1864 	struct ata_command ata_c;
1865 
1866 	/*
1867 	 * WDCC_FLUSHCACHE is here since ATA-4, but some drives report
1868 	 * only ATA-2 and still support it.
1869 	 */
1870 	if (wd->drvp->ata_vers < 4 &&
1871 	    ((wd->sc_params.atap_cmd_set2 & WDC_CMD2_FC) == 0 ||
1872 	    wd->sc_params.atap_cmd_set2 == 0xffff))
1873 		return ENODEV;
1874 	memset(&ata_c, 0, sizeof(struct ata_command));
1875 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0 &&
1876 	    (wd->sc_params.atap_cmd2_en & ATA_CMD2_FCE) != 0)
1877 		ata_c.r_command = WDCC_FLUSHCACHE_EXT;
1878 	else
1879 		ata_c.r_command = WDCC_FLUSHCACHE;
1880 	ata_c.r_st_bmask = WDCS_DRDY;
1881 	ata_c.r_st_pmask = WDCS_DRDY;
1882 	ata_c.flags = flags;
1883 	ata_c.timeout = 30000; /* 30s timeout */
1884 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1885 		aprint_error_dev(wd->sc_dev,
1886 		    "flush cache command didn't complete\n");
1887 		return EIO;
1888 	}
1889 	if (ata_c.flags & AT_ERROR) {
1890 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
1891 			return ENODEV;
1892 	}
1893 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1894 		char sbuf[sizeof(at_errbits) + 64];
1895 		snprintb(sbuf, sizeof(sbuf), at_errbits, ata_c.flags);
1896 		aprint_error_dev(wd->sc_dev, "wd_flushcache: status=%s\n",
1897 		    sbuf);
1898 		return EIO;
1899 	}
1900 	return 0;
1901 }
1902 
1903 bool
1904 wd_shutdown(device_t dev, int how)
1905 {
1906 	struct wd_softc *wd = device_private(dev);
1907 
1908 	/* the adapter needs to be enabled */
1909 	if (wd->atabus->ata_addref(wd->drvp))
1910 		return true; /* no need to complain */
1911 
1912 	wd_flushcache(wd, AT_POLL);
1913 	if ((how & RB_POWERDOWN) == RB_POWERDOWN)
1914 		wd_standby(wd, AT_POLL);
1915 	return true;
1916 }
1917 
1918 /*
1919  * Allocate space for a ioctl queue structure.  Mostly taken from
1920  * scsipi_ioctl.c
1921  */
1922 struct wd_ioctl *
1923 wi_get(void)
1924 {
1925 	struct wd_ioctl *wi;
1926 	int s;
1927 
1928 	wi = malloc(sizeof(struct wd_ioctl), M_TEMP, M_WAITOK|M_ZERO);
1929 	buf_init(&wi->wi_bp);
1930 	s = splbio();
1931 	LIST_INSERT_HEAD(&wi_head, wi, wi_list);
1932 	splx(s);
1933 	return (wi);
1934 }
1935 
1936 /*
1937  * Free an ioctl structure and remove it from our list
1938  */
1939 
1940 void
1941 wi_free(struct wd_ioctl *wi)
1942 {
1943 	int s;
1944 
1945 	s = splbio();
1946 	LIST_REMOVE(wi, wi_list);
1947 	splx(s);
1948 	buf_destroy(&wi->wi_bp);
1949 	free(wi, M_TEMP);
1950 }
1951 
1952 /*
1953  * Find a wd_ioctl structure based on the struct buf.
1954  */
1955 
1956 struct wd_ioctl *
1957 wi_find(struct buf *bp)
1958 {
1959 	struct wd_ioctl *wi;
1960 	int s;
1961 
1962 	s = splbio();
1963 	for (wi = wi_head.lh_first; wi != 0; wi = wi->wi_list.le_next)
1964 		if (bp == &wi->wi_bp)
1965 			break;
1966 	splx(s);
1967 	return (wi);
1968 }
1969 
1970 /*
1971  * Ioctl pseudo strategy routine
1972  *
1973  * This is mostly stolen from scsipi_ioctl.c:scsistrategy().  What
1974  * happens here is:
1975  *
1976  * - wdioctl() queues a wd_ioctl structure.
1977  *
1978  * - wdioctl() calls physio/wdioctlstrategy based on whether or not
1979  *   user space I/O is required.  If physio() is called, physio() eventually
1980  *   calls wdioctlstrategy().
1981  *
1982  * - In either case, wdioctlstrategy() calls wd->atabus->ata_exec_command()
1983  *   to perform the actual command
1984  *
1985  * The reason for the use of the pseudo strategy routine is because
1986  * when doing I/O to/from user space, physio _really_ wants to be in
1987  * the loop.  We could put the entire buffer into the ioctl request
1988  * structure, but that won't scale if we want to do things like download
1989  * microcode.
1990  */
1991 
1992 void
1993 wdioctlstrategy(struct buf *bp)
1994 {
1995 	struct wd_ioctl *wi;
1996 	struct ata_command ata_c;
1997 	int error = 0;
1998 
1999 	wi = wi_find(bp);
2000 	if (wi == NULL) {
2001 		printf("wdioctlstrategy: "
2002 		    "No matching ioctl request found in queue\n");
2003 		error = EINVAL;
2004 		goto bad;
2005 	}
2006 
2007 	memset(&ata_c, 0, sizeof(ata_c));
2008 
2009 	/*
2010 	 * Abort if physio broke up the transfer
2011 	 */
2012 
2013 	if (bp->b_bcount != wi->wi_atareq.datalen) {
2014 		printf("physio split wd ioctl request... cannot proceed\n");
2015 		error = EIO;
2016 		goto bad;
2017 	}
2018 
2019 	/*
2020 	 * Abort if we didn't get a buffer size that was a multiple of
2021 	 * our sector size (or was larger than NBBY)
2022 	 */
2023 
2024 	if ((bp->b_bcount % wi->wi_softc->sc_dk.dk_label->d_secsize) != 0 ||
2025 	    (bp->b_bcount / wi->wi_softc->sc_dk.dk_label->d_secsize) >=
2026 	     (1 << NBBY)) {
2027 		error = EINVAL;
2028 		goto bad;
2029 	}
2030 
2031 	/*
2032 	 * Make sure a timeout was supplied in the ioctl request
2033 	 */
2034 
2035 	if (wi->wi_atareq.timeout == 0) {
2036 		error = EINVAL;
2037 		goto bad;
2038 	}
2039 
2040 	if (wi->wi_atareq.flags & ATACMD_READ)
2041 		ata_c.flags |= AT_READ;
2042 	else if (wi->wi_atareq.flags & ATACMD_WRITE)
2043 		ata_c.flags |= AT_WRITE;
2044 
2045 	if (wi->wi_atareq.flags & ATACMD_READREG)
2046 		ata_c.flags |= AT_READREG;
2047 
2048 	ata_c.flags |= AT_WAIT;
2049 
2050 	ata_c.timeout = wi->wi_atareq.timeout;
2051 	ata_c.r_command = wi->wi_atareq.command;
2052 	ata_c.r_head = wi->wi_atareq.head & 0x0f;
2053 	ata_c.r_cyl = wi->wi_atareq.cylinder;
2054 	ata_c.r_sector = wi->wi_atareq.sec_num;
2055 	ata_c.r_count = wi->wi_atareq.sec_count;
2056 	ata_c.r_features = wi->wi_atareq.features;
2057 	ata_c.r_st_bmask = WDCS_DRDY;
2058 	ata_c.r_st_pmask = WDCS_DRDY;
2059 	ata_c.data = wi->wi_bp.b_data;
2060 	ata_c.bcount = wi->wi_bp.b_bcount;
2061 
2062 	if (wi->wi_softc->atabus->ata_exec_command(wi->wi_softc->drvp, &ata_c)
2063 	    != ATACMD_COMPLETE) {
2064 		wi->wi_atareq.retsts = ATACMD_ERROR;
2065 		goto bad;
2066 	}
2067 
2068 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
2069 		if (ata_c.flags & AT_ERROR) {
2070 			wi->wi_atareq.retsts = ATACMD_ERROR;
2071 			wi->wi_atareq.error = ata_c.r_error;
2072 		} else if (ata_c.flags & AT_DF)
2073 			wi->wi_atareq.retsts = ATACMD_DF;
2074 		else
2075 			wi->wi_atareq.retsts = ATACMD_TIMEOUT;
2076 	} else {
2077 		wi->wi_atareq.retsts = ATACMD_OK;
2078 		if (wi->wi_atareq.flags & ATACMD_READREG) {
2079 			wi->wi_atareq.head = ata_c.r_head ;
2080 			wi->wi_atareq.cylinder = ata_c.r_cyl;
2081 			wi->wi_atareq.sec_num = ata_c.r_sector;
2082 			wi->wi_atareq.sec_count = ata_c.r_count;
2083 			wi->wi_atareq.features = ata_c.r_features;
2084 			wi->wi_atareq.error = ata_c.r_error;
2085 		}
2086 	}
2087 
2088 	bp->b_error = 0;
2089 	biodone(bp);
2090 	return;
2091 bad:
2092 	bp->b_error = error;
2093 	biodone(bp);
2094 }
2095