xref: /netbsd-src/sys/dev/ata/wd.c (revision 7f21db1c0118155e0dd40b75182e30c589d9f63e)
1 /*	$NetBSD: wd.c,v 1.383 2010/01/23 18:54:53 bouyer 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 1.383 2010/01/23 18:54:53 bouyer 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, 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, 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 suspension), then error.
493 	 */
494 	if ((wd->sc_flags & WDF_LOADED) == 0 || !device_is_active(wd->sc_dev)) {
495 		bp->b_error = EIO;
496 		goto done;
497 	}
498 
499 	/* If it's a null transfer, return immediately. */
500 	if (bp->b_bcount == 0)
501 		goto done;
502 
503 	/*
504 	 * Do bounds checking, adjust transfer. if error, process.
505 	 * If end of partition, just return.
506 	 */
507 	if (WDPART(bp->b_dev) == RAW_PART) {
508 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
509 		    wd->sc_capacity) <= 0)
510 			goto done;
511 	} else {
512 		if (bounds_check_with_label(&wd->sc_dk, bp,
513 		    (wd->sc_flags & (WDF_WLABEL|WDF_LABELLING)) != 0) <= 0)
514 			goto done;
515 	}
516 
517 	/*
518 	 * Now convert the block number to absolute and put it in
519 	 * terms of the device's logical block size.
520 	 */
521 	if (lp->d_secsize >= DEV_BSIZE)
522 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
523 	else
524 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
525 
526 	if (WDPART(bp->b_dev) != RAW_PART)
527 		blkno += lp->d_partitions[WDPART(bp->b_dev)].p_offset;
528 
529 	bp->b_rawblkno = blkno;
530 
531 #ifdef WD_SOFTBADSECT
532 	/*
533 	 * If the transfer about to be attempted contains only a block that
534 	 * is known to be bad then return an error for the transfer without
535 	 * even attempting to start a transfer up under the premis that we
536 	 * will just end up doing more retries for a transfer that will end
537 	 * up failing again.
538 	 * XXX:SMP - mutex required to protect with DIOCBSFLUSH
539 	 */
540 	if (__predict_false(!SLIST_EMPTY(&wd->sc_bslist))) {
541 		struct disk_badsectors *dbs;
542 		daddr_t maxblk = blkno + (bp->b_bcount >> DEV_BSHIFT) - 1;
543 
544 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next)
545 			if ((dbs->dbs_min <= blkno && blkno <= dbs->dbs_max) ||
546 			    (dbs->dbs_min <= maxblk && maxblk <= dbs->dbs_max)){
547 				bp->b_error = EIO;
548 				goto done;
549 			}
550 	}
551 #endif
552 
553 	/* Queue transfer on drive, activate drive and controller if idle. */
554 	s = splbio();
555 	bufq_put(wd->sc_q, bp);
556 	wdstart(wd);
557 	splx(s);
558 	return;
559 done:
560 	/* Toss transfer; we're done early. */
561 	bp->b_resid = bp->b_bcount;
562 	biodone(bp);
563 }
564 
565 /*
566  * Queue a drive for I/O.
567  */
568 void
569 wdstart(void *arg)
570 {
571 	struct wd_softc *wd = arg;
572 	struct buf *bp = NULL;
573 
574 	ATADEBUG_PRINT(("wdstart %s\n", device_xname(wd->sc_dev)),
575 	    DEBUG_XFERS);
576 	while (wd->openings > 0) {
577 
578 		/* Is there a buf for us ? */
579 		if ((bp = bufq_get(wd->sc_q)) == NULL)
580 			return;
581 
582 		/*
583 		 * Make the command. First lock the device
584 		 */
585 		wd->openings--;
586 
587 		wd->retries = 0;
588 		wdstart1(wd, bp);
589 	}
590 }
591 
592 static void
593 wd_split_mod15_write(struct buf *bp)
594 {
595 	struct buf *obp = bp->b_private;
596 	struct wd_softc *sc =
597 	    device_lookup_private(&wd_cd, DISKUNIT(obp->b_dev));
598 	int s;
599 
600 	if (__predict_false(bp->b_error != 0)) {
601 		/*
602 		 * Propagate the error.  If this was the first half of
603 		 * the original transfer, make sure to account for that
604 		 * in the residual.
605 		 */
606 		if (bp->b_data == obp->b_data)
607 			bp->b_resid += bp->b_bcount;
608 		goto done;
609 	}
610 
611 	/*
612 	 * If this was the second half of the transfer, we're all done!
613 	 */
614 	if (bp->b_data != obp->b_data)
615 		goto done;
616 
617 	/*
618 	 * Advance the pointer to the second half and issue that command
619 	 * using the same opening.
620 	 */
621 	bp->b_flags = obp->b_flags;
622 	bp->b_oflags = obp->b_oflags;
623 	bp->b_cflags = obp->b_cflags;
624 	bp->b_data = (char *)bp->b_data + bp->b_bcount;
625 	bp->b_blkno += (bp->b_bcount / 512);
626 	bp->b_rawblkno += (bp->b_bcount / 512);
627 	s = splbio();
628 	wdstart1(sc, bp);
629 	splx(s);
630 	return;
631 
632  done:
633 	obp->b_error = bp->b_error;
634 	obp->b_resid = bp->b_resid;
635 	s = splbio();
636 	putiobuf(bp);
637 	biodone(obp);
638 	sc->openings++;
639 	splx(s);
640 	/* wddone() will call wdstart() */
641 }
642 
643 void
644 wdstart1(struct wd_softc *wd, struct buf *bp)
645 {
646 
647 	/*
648 	 * Deal with the "split mod15 write" quirk.  We just divide the
649 	 * transfer in two, doing the first half and then then second half
650 	 * with the same command opening.
651 	 *
652 	 * Note we MUST do this here, because we can't let insertion
653 	 * into the bufq cause the transfers to be re-merged.
654 	 */
655 	if (__predict_false((wd->sc_quirks & WD_QUIRK_SPLIT_MOD15_WRITE) != 0 &&
656 			    (bp->b_flags & B_READ) == 0 &&
657 			    bp->b_bcount > 512 &&
658 			    ((bp->b_bcount / 512) % 15) == 1)) {
659 		struct buf *nbp;
660 
661 		/* already at splbio */
662 		nbp = getiobuf(NULL, false);
663 		if (__predict_false(nbp == NULL)) {
664 			/* No memory -- fail the iop. */
665 			bp->b_error = ENOMEM;
666 			bp->b_resid = bp->b_bcount;
667 			biodone(bp);
668 			wd->openings++;
669 			return;
670 		}
671 
672 		nbp->b_error = 0;
673 		nbp->b_proc = bp->b_proc;
674 		nbp->b_dev = bp->b_dev;
675 
676 		nbp->b_bcount = bp->b_bcount / 2;
677 		nbp->b_bufsize = bp->b_bcount / 2;
678 		nbp->b_data = bp->b_data;
679 
680 		nbp->b_blkno = bp->b_blkno;
681 		nbp->b_rawblkno = bp->b_rawblkno;
682 
683 		nbp->b_flags = bp->b_flags;
684 		nbp->b_oflags = bp->b_oflags;
685 		nbp->b_cflags = bp->b_cflags;
686 		nbp->b_iodone = wd_split_mod15_write;
687 
688 		/* Put ptr to orig buf in b_private and use new buf */
689 		nbp->b_private = bp;
690 
691 		BIO_COPYPRIO(nbp, bp);
692 
693 		bp = nbp;
694 	}
695 
696 	wd->sc_wdc_bio.blkno = bp->b_rawblkno;
697 	wd->sc_wdc_bio.bcount = bp->b_bcount;
698 	wd->sc_wdc_bio.databuf = bp->b_data;
699 	wd->sc_wdc_bio.blkdone =0;
700 	wd->sc_bp = bp;
701 	/*
702 	 * If we're retrying, retry in single-sector mode. This will give us
703 	 * the sector number of the problem, and will eventually allow the
704 	 * transfer to succeed.
705 	 */
706 	if (wd->retries >= WDIORETRIES_SINGLE)
707 		wd->sc_wdc_bio.flags = ATA_SINGLE;
708 	else
709 		wd->sc_wdc_bio.flags = 0;
710 	if (wd->sc_flags & WDF_LBA48 &&
711 	    (wd->sc_wdc_bio.blkno +
712 	     wd->sc_wdc_bio.bcount / wd->sc_dk.dk_label->d_secsize) >
713 	    wd->sc_capacity28)
714 		wd->sc_wdc_bio.flags |= ATA_LBA48;
715 	if (wd->sc_flags & WDF_LBA)
716 		wd->sc_wdc_bio.flags |= ATA_LBA;
717 	if (bp->b_flags & B_READ)
718 		wd->sc_wdc_bio.flags |= ATA_READ;
719 	/* Instrumentation. */
720 	disk_busy(&wd->sc_dk);
721 	switch (wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
722 	case ATACMD_TRY_AGAIN:
723 		callout_reset(&wd->sc_restart_ch, hz, wdrestart, wd);
724 		break;
725 	case ATACMD_QUEUED:
726 	case ATACMD_COMPLETE:
727 		break;
728 	default:
729 		panic("wdstart1: bad return code from ata_bio()");
730 	}
731 }
732 
733 void
734 wddone(void *v)
735 {
736 	struct wd_softc *wd = device_private(v);
737 	struct buf *bp = wd->sc_bp;
738 	const char *errmsg;
739 	int do_perror = 0;
740 
741 	ATADEBUG_PRINT(("wddone %s\n", device_xname(wd->sc_dev)),
742 	    DEBUG_XFERS);
743 	if (bp == NULL)
744 		return;
745 	bp->b_resid = wd->sc_wdc_bio.bcount;
746 	switch (wd->sc_wdc_bio.error) {
747 	case ERR_DMA:
748 		errmsg = "DMA error";
749 		goto retry;
750 	case ERR_DF:
751 		errmsg = "device fault";
752 		goto retry;
753 	case TIMEOUT:
754 		errmsg = "device timeout";
755 		goto retry;
756 	case ERR_RESET:
757 		errmsg = "channel reset";
758 		goto retry2;
759 	case ERROR:
760 		/* Don't care about media change bits */
761 		if (wd->sc_wdc_bio.r_error != 0 &&
762 		    (wd->sc_wdc_bio.r_error & ~(WDCE_MC | WDCE_MCR)) == 0)
763 			goto noerror;
764 		errmsg = "error";
765 		do_perror = 1;
766 retry:		/* Just reset and retry. Can we do more ? */
767 		(*wd->atabus->ata_reset_drive)(wd->drvp, AT_RST_NOCMD);
768 retry2:
769 		diskerr(bp, "wd", errmsg, LOG_PRINTF,
770 		    wd->sc_wdc_bio.blkdone, wd->sc_dk.dk_label);
771 		if (wd->retries < WDIORETRIES)
772 			printf(", retrying");
773 		printf("\n");
774 		if (do_perror)
775 			wdperror(wd);
776 		if (wd->retries < WDIORETRIES) {
777 			wd->retries++;
778 			callout_reset(&wd->sc_restart_ch, RECOVERYTIME,
779 			    wdrestart, wd);
780 			return;
781 		}
782 
783 #ifdef WD_SOFTBADSECT
784 		/*
785 		 * Not all errors indicate a failed block but those that do,
786 		 * put the block on the bad-block list for the device.  Only
787 		 * do this for reads because the drive should do it for writes,
788 		 * itself, according to Manuel.
789 		 */
790 		if ((bp->b_flags & B_READ) &&
791 		    ((wd->drvp->ata_vers >= 4 && wd->sc_wdc_bio.r_error & 64) ||
792 	     	     (wd->drvp->ata_vers < 4 && wd->sc_wdc_bio.r_error & 192))) {
793 			struct disk_badsectors *dbs;
794 
795 			dbs = malloc(sizeof *dbs, M_TEMP, M_WAITOK);
796 			dbs->dbs_min = bp->b_rawblkno;
797 			dbs->dbs_max = dbs->dbs_min + (bp->b_bcount >> DEV_BSHIFT) - 1;
798 			microtime(&dbs->dbs_failedat);
799 			SLIST_INSERT_HEAD(&wd->sc_bslist, dbs, dbs_next);
800 			wd->sc_bscount++;
801 		}
802 #endif
803 		bp->b_error = EIO;
804 		break;
805 	case NOERROR:
806 noerror:	if ((wd->sc_wdc_bio.flags & ATA_CORR) || wd->retries > 0)
807 			aprint_error_dev(wd->sc_dev,
808 			    "soft error (corrected)\n");
809 		break;
810 	case ERR_NODEV:
811 		bp->b_error = EIO;
812 		break;
813 	}
814 	disk_unbusy(&wd->sc_dk, (bp->b_bcount - bp->b_resid),
815 	    (bp->b_flags & B_READ));
816 #if NRND > 0
817 	rnd_add_uint32(&wd->rnd_source, bp->b_blkno);
818 #endif
819 	/* XXX Yuck, but we don't want to increment openings in this case */
820 	if (__predict_false(bp->b_iodone == wd_split_mod15_write))
821 		biodone(bp);
822 	else {
823 		biodone(bp);
824 		wd->openings++;
825 	}
826 	wdstart(wd);
827 }
828 
829 void
830 wdrestart(void *v)
831 {
832 	struct wd_softc *wd = v;
833 	struct buf *bp = wd->sc_bp;
834 	int s;
835 
836 	ATADEBUG_PRINT(("wdrestart %s\n", device_xname(wd->sc_dev)),
837 	    DEBUG_XFERS);
838 	s = splbio();
839 	wdstart1(v, bp);
840 	splx(s);
841 }
842 
843 int
844 wdread(dev_t dev, struct uio *uio, int flags)
845 {
846 
847 	ATADEBUG_PRINT(("wdread\n"), DEBUG_XFERS);
848 	return (physio(wdstrategy, NULL, dev, B_READ, minphys, uio));
849 }
850 
851 int
852 wdwrite(dev_t dev, struct uio *uio, int flags)
853 {
854 
855 	ATADEBUG_PRINT(("wdwrite\n"), DEBUG_XFERS);
856 	return (physio(wdstrategy, NULL, dev, B_WRITE, minphys, uio));
857 }
858 
859 int
860 wdopen(dev_t dev, int flag, int fmt, struct lwp *l)
861 {
862 	struct wd_softc *wd;
863 	int part, error;
864 
865 	ATADEBUG_PRINT(("wdopen\n"), DEBUG_FUNCS);
866 	wd = device_lookup_private(&wd_cd, WDUNIT(dev));
867 	if (wd == NULL)
868 		return (ENXIO);
869 
870 	if (! device_is_active(wd->sc_dev))
871 		return (ENODEV);
872 
873 	part = WDPART(dev);
874 
875 	mutex_enter(&wd->sc_dk.dk_openlock);
876 
877 	/*
878 	 * If there are wedges, and this is not RAW_PART, then we
879 	 * need to fail.
880 	 */
881 	if (wd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
882 		error = EBUSY;
883 		goto bad1;
884 	}
885 
886 	/*
887 	 * If this is the first open of this device, add a reference
888 	 * to the adapter.
889 	 */
890 	if (wd->sc_dk.dk_openmask == 0 &&
891 	    (error = wd->atabus->ata_addref(wd->drvp)) != 0)
892 		goto bad1;
893 
894 	if (wd->sc_dk.dk_openmask != 0) {
895 		/*
896 		 * If any partition is open, but the disk has been invalidated,
897 		 * disallow further opens.
898 		 */
899 		if ((wd->sc_flags & WDF_LOADED) == 0) {
900 			error = EIO;
901 			goto bad2;
902 		}
903 	} else {
904 		if ((wd->sc_flags & WDF_LOADED) == 0) {
905 			wd->sc_flags |= WDF_LOADED;
906 
907 			/* Load the physical device parameters. */
908 			wd_get_params(wd, AT_WAIT, &wd->sc_params);
909 
910 			/* Load the partition info if not already loaded. */
911 			wdgetdisklabel(wd);
912 		}
913 	}
914 
915 	/* Check that the partition exists. */
916 	if (part != RAW_PART &&
917 	    (part >= wd->sc_dk.dk_label->d_npartitions ||
918 	     wd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
919 		error = ENXIO;
920 		goto bad2;
921 	}
922 
923 	/* Insure only one open at a time. */
924 	switch (fmt) {
925 	case S_IFCHR:
926 		wd->sc_dk.dk_copenmask |= (1 << part);
927 		break;
928 	case S_IFBLK:
929 		wd->sc_dk.dk_bopenmask |= (1 << part);
930 		break;
931 	}
932 	wd->sc_dk.dk_openmask =
933 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
934 
935 	mutex_exit(&wd->sc_dk.dk_openlock);
936 	return 0;
937 
938  bad2:
939 	if (wd->sc_dk.dk_openmask == 0)
940 		wd->atabus->ata_delref(wd->drvp);
941  bad1:
942 	mutex_exit(&wd->sc_dk.dk_openlock);
943 	return error;
944 }
945 
946 /*
947  * Caller must hold wd->sc_dk.dk_openlock.
948  */
949 static int
950 wdlastclose(device_t self)
951 {
952 	struct wd_softc *wd = device_private(self);
953 
954 	wd_flushcache(wd, AT_WAIT);
955 
956 	if (! (wd->sc_flags & WDF_KLABEL))
957 		wd->sc_flags &= ~WDF_LOADED;
958 
959 	wd->atabus->ata_delref(wd->drvp);
960 
961 	return 0;
962 }
963 
964 int
965 wdclose(dev_t dev, int flag, int fmt, struct lwp *l)
966 {
967 	struct wd_softc *wd =
968 	    device_lookup_private(&wd_cd, WDUNIT(dev));
969 	int part = WDPART(dev);
970 
971 	ATADEBUG_PRINT(("wdclose\n"), DEBUG_FUNCS);
972 
973 	mutex_enter(&wd->sc_dk.dk_openlock);
974 
975 	switch (fmt) {
976 	case S_IFCHR:
977 		wd->sc_dk.dk_copenmask &= ~(1 << part);
978 		break;
979 	case S_IFBLK:
980 		wd->sc_dk.dk_bopenmask &= ~(1 << part);
981 		break;
982 	}
983 	wd->sc_dk.dk_openmask =
984 	    wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
985 
986 	if (wd->sc_dk.dk_openmask == 0)
987 		wdlastclose(wd->sc_dev);
988 
989 	mutex_exit(&wd->sc_dk.dk_openlock);
990 	return 0;
991 }
992 
993 void
994 wdgetdefaultlabel(struct wd_softc *wd, struct disklabel *lp)
995 {
996 
997 	ATADEBUG_PRINT(("wdgetdefaultlabel\n"), DEBUG_FUNCS);
998 	memset(lp, 0, sizeof(struct disklabel));
999 
1000 	lp->d_secsize = DEV_BSIZE;
1001 	lp->d_ntracks = wd->sc_params.atap_heads;
1002 	lp->d_nsectors = wd->sc_params.atap_sectors;
1003 	lp->d_ncylinders = (wd->sc_flags & WDF_LBA) ? wd->sc_capacity /
1004 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors) :
1005 		wd->sc_params.atap_cylinders;
1006 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1007 
1008 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
1009 		lp->d_type = DTYPE_ST506;
1010 	else
1011 		lp->d_type = DTYPE_ESDI;
1012 
1013 	strncpy(lp->d_typename, wd->sc_params.atap_model, 16);
1014 	strncpy(lp->d_packname, "fictitious", 16);
1015 	if (wd->sc_capacity > UINT32_MAX)
1016 		lp->d_secperunit = UINT32_MAX;
1017 	else
1018 		lp->d_secperunit = wd->sc_capacity;
1019 	lp->d_rpm = 3600;
1020 	lp->d_interleave = 1;
1021 	lp->d_flags = 0;
1022 
1023 	lp->d_partitions[RAW_PART].p_offset = 0;
1024 	lp->d_partitions[RAW_PART].p_size =
1025 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
1026 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1027 	lp->d_npartitions = RAW_PART + 1;
1028 
1029 	lp->d_magic = DISKMAGIC;
1030 	lp->d_magic2 = DISKMAGIC;
1031 	lp->d_checksum = dkcksum(lp);
1032 }
1033 
1034 /*
1035  * Fabricate a default disk label, and try to read the correct one.
1036  */
1037 void
1038 wdgetdisklabel(struct wd_softc *wd)
1039 {
1040 	struct disklabel *lp = wd->sc_dk.dk_label;
1041 	const char *errstring;
1042 	int s;
1043 
1044 	ATADEBUG_PRINT(("wdgetdisklabel\n"), DEBUG_FUNCS);
1045 
1046 	memset(wd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1047 
1048 	wdgetdefaultlabel(wd, lp);
1049 
1050 	wd->sc_badsect[0] = -1;
1051 
1052 	if (wd->drvp->state > RESET) {
1053 		s = splbio();
1054 		wd->drvp->drive_flags |= DRIVE_RESET;
1055 		splx(s);
1056 	}
1057 	errstring = readdisklabel(MAKEWDDEV(0, device_unit(wd->sc_dev),
1058 				  RAW_PART), wdstrategy, lp,
1059 				  wd->sc_dk.dk_cpulabel);
1060 	if (errstring) {
1061 		/*
1062 		 * This probably happened because the drive's default
1063 		 * geometry doesn't match the DOS geometry.  We
1064 		 * assume the DOS geometry is now in the label and try
1065 		 * again.  XXX This is a kluge.
1066 		 */
1067 		if (wd->drvp->state > RESET) {
1068 			s = splbio();
1069 			wd->drvp->drive_flags |= DRIVE_RESET;
1070 			splx(s);
1071 		}
1072 		errstring = readdisklabel(MAKEWDDEV(0, device_unit(wd->sc_dev),
1073 		    RAW_PART), wdstrategy, lp, wd->sc_dk.dk_cpulabel);
1074 	}
1075 	if (errstring) {
1076 		aprint_error_dev(wd->sc_dev, "%s\n", errstring);
1077 		return;
1078 	}
1079 
1080 	if (wd->drvp->state > RESET) {
1081 		s = splbio();
1082 		wd->drvp->drive_flags |= DRIVE_RESET;
1083 		splx(s);
1084 	}
1085 #ifdef HAS_BAD144_HANDLING
1086 	if ((lp->d_flags & D_BADSECT) != 0)
1087 		bad144intern(wd);
1088 #endif
1089 }
1090 
1091 void
1092 wdperror(const struct wd_softc *wd)
1093 {
1094 	static const char *const errstr0_3[] = {"address mark not found",
1095 	    "track 0 not found", "aborted command", "media change requested",
1096 	    "id not found", "media changed", "uncorrectable data error",
1097 	    "bad block detected"};
1098 	static const char *const errstr4_5[] = {
1099 	    "obsolete (address mark not found)",
1100 	    "no media/write protected", "aborted command",
1101 	    "media change requested", "id not found", "media changed",
1102 	    "uncorrectable data error", "interface CRC error"};
1103 	const char *const *errstr;
1104 	int i;
1105 	const char *sep = "";
1106 
1107 	const char *devname = device_xname(wd->sc_dev);
1108 	struct ata_drive_datas *drvp = wd->drvp;
1109 	int errno = wd->sc_wdc_bio.r_error;
1110 
1111 	if (drvp->ata_vers >= 4)
1112 		errstr = errstr4_5;
1113 	else
1114 		errstr = errstr0_3;
1115 
1116 	printf("%s: (", devname);
1117 
1118 	if (errno == 0)
1119 		printf("error not notified");
1120 
1121 	for (i = 0; i < 8; i++) {
1122 		if (errno & (1 << i)) {
1123 			printf("%s%s", sep, errstr[i]);
1124 			sep = ", ";
1125 		}
1126 	}
1127 	printf(")\n");
1128 }
1129 
1130 int
1131 wdioctl(dev_t dev, u_long xfer, void *addr, int flag, struct lwp *l)
1132 {
1133 	struct wd_softc *wd =
1134 	    device_lookup_private(&wd_cd, WDUNIT(dev));
1135 	int error = 0, s;
1136 #ifdef __HAVE_OLD_DISKLABEL
1137 	struct disklabel *newlabel = NULL;
1138 #endif
1139 
1140 	ATADEBUG_PRINT(("wdioctl\n"), DEBUG_FUNCS);
1141 
1142 	if ((wd->sc_flags & WDF_LOADED) == 0)
1143 		return EIO;
1144 
1145 	error = disk_ioctl(&wd->sc_dk, xfer, addr, flag, l);
1146 	if (error != EPASSTHROUGH)
1147 		return (error);
1148 
1149 	switch (xfer) {
1150 #ifdef HAS_BAD144_HANDLING
1151 	case DIOCSBAD:
1152 		if ((flag & FWRITE) == 0)
1153 			return EBADF;
1154 		wd->sc_dk.dk_cpulabel->bad = *(struct dkbad *)addr;
1155 		wd->sc_dk.dk_label->d_flags |= D_BADSECT;
1156 		bad144intern(wd);
1157 		return 0;
1158 #endif
1159 #ifdef WD_SOFTBADSECT
1160 	case DIOCBSLIST :
1161 	{
1162 		u_int32_t count, missing, skip;
1163 		struct disk_badsecinfo dbsi;
1164 		struct disk_badsectors *dbs;
1165 		size_t available;
1166 		uint8_t *laddr;
1167 
1168 		dbsi = *(struct disk_badsecinfo *)addr;
1169 		missing = wd->sc_bscount;
1170 		count = 0;
1171 		available = dbsi.dbsi_bufsize;
1172 		skip = dbsi.dbsi_skip;
1173 		laddr = (uint8_t *)dbsi.dbsi_buffer;
1174 
1175 		/*
1176 		 * We start this loop with the expectation that all of the
1177 		 * entries will be missed and decrement this counter each
1178 		 * time we either skip over one (already copied out) or
1179 		 * we actually copy it back to user space.  The structs
1180 		 * holding the bad sector information are copied directly
1181 		 * back to user space whilst the summary is returned via
1182 		 * the struct passed in via the ioctl.
1183 		 */
1184 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next) {
1185 			if (skip > 0) {
1186 				missing--;
1187 				skip--;
1188 				continue;
1189 			}
1190 			if (available < sizeof(*dbs))
1191 				break;
1192 			available -= sizeof(*dbs);
1193 			copyout(dbs, laddr, sizeof(*dbs));
1194 			laddr += sizeof(*dbs);
1195 			missing--;
1196 			count++;
1197 		}
1198 		dbsi.dbsi_left = missing;
1199 		dbsi.dbsi_copied = count;
1200 		*(struct disk_badsecinfo *)addr = dbsi;
1201 		return 0;
1202 	}
1203 
1204 	case DIOCBSFLUSH :
1205 		/* Clean out the bad sector list */
1206 		while (!SLIST_EMPTY(&wd->sc_bslist)) {
1207 			void *head = SLIST_FIRST(&wd->sc_bslist);
1208 			SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
1209 			free(head, M_TEMP);
1210 		}
1211 		wd->sc_bscount = 0;
1212 		return 0;
1213 #endif
1214 	case DIOCGDINFO:
1215 		*(struct disklabel *)addr = *(wd->sc_dk.dk_label);
1216 		return 0;
1217 #ifdef __HAVE_OLD_DISKLABEL
1218 	case ODIOCGDINFO:
1219 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1220 		if (newlabel == NULL)
1221 			return EIO;
1222 		*newlabel = *(wd->sc_dk.dk_label);
1223 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1224 			memcpy(addr, newlabel, sizeof (struct olddisklabel));
1225 		else
1226 			error = ENOTTY;
1227 		free(newlabel, M_TEMP);
1228 		return error;
1229 #endif
1230 
1231 	case DIOCGPART:
1232 		((struct partinfo *)addr)->disklab = wd->sc_dk.dk_label;
1233 		((struct partinfo *)addr)->part =
1234 		    &wd->sc_dk.dk_label->d_partitions[WDPART(dev)];
1235 		return 0;
1236 
1237 	case DIOCWDINFO:
1238 	case DIOCSDINFO:
1239 #ifdef __HAVE_OLD_DISKLABEL
1240 	case ODIOCWDINFO:
1241 	case ODIOCSDINFO:
1242 #endif
1243 	{
1244 		struct disklabel *lp;
1245 
1246 		if ((flag & FWRITE) == 0)
1247 			return EBADF;
1248 
1249 #ifdef __HAVE_OLD_DISKLABEL
1250 		if (xfer == ODIOCSDINFO || xfer == ODIOCWDINFO) {
1251 			newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1252 			if (newlabel == NULL)
1253 				return EIO;
1254 			memset(newlabel, 0, sizeof newlabel);
1255 			memcpy(newlabel, addr, sizeof (struct olddisklabel));
1256 			lp = newlabel;
1257 		} else
1258 #endif
1259 		lp = (struct disklabel *)addr;
1260 
1261 		mutex_enter(&wd->sc_dk.dk_openlock);
1262 		wd->sc_flags |= WDF_LABELLING;
1263 
1264 		error = setdisklabel(wd->sc_dk.dk_label,
1265 		    lp, /*wd->sc_dk.dk_openmask : */0,
1266 		    wd->sc_dk.dk_cpulabel);
1267 		if (error == 0) {
1268 			if (wd->drvp->state > RESET) {
1269 				s = splbio();
1270 				wd->drvp->drive_flags |= DRIVE_RESET;
1271 				splx(s);
1272 			}
1273 			if (xfer == DIOCWDINFO
1274 #ifdef __HAVE_OLD_DISKLABEL
1275 			    || xfer == ODIOCWDINFO
1276 #endif
1277 			    )
1278 				error = writedisklabel(WDLABELDEV(dev),
1279 				    wdstrategy, wd->sc_dk.dk_label,
1280 				    wd->sc_dk.dk_cpulabel);
1281 		}
1282 
1283 		wd->sc_flags &= ~WDF_LABELLING;
1284 		mutex_exit(&wd->sc_dk.dk_openlock);
1285 #ifdef __HAVE_OLD_DISKLABEL
1286 		if (newlabel != NULL)
1287 			free(newlabel, M_TEMP);
1288 #endif
1289 		return error;
1290 	}
1291 
1292 	case DIOCKLABEL:
1293 		if (*(int *)addr)
1294 			wd->sc_flags |= WDF_KLABEL;
1295 		else
1296 			wd->sc_flags &= ~WDF_KLABEL;
1297 		return 0;
1298 
1299 	case DIOCWLABEL:
1300 		if ((flag & FWRITE) == 0)
1301 			return EBADF;
1302 		if (*(int *)addr)
1303 			wd->sc_flags |= WDF_WLABEL;
1304 		else
1305 			wd->sc_flags &= ~WDF_WLABEL;
1306 		return 0;
1307 
1308 	case DIOCGDEFLABEL:
1309 		wdgetdefaultlabel(wd, (struct disklabel *)addr);
1310 		return 0;
1311 #ifdef __HAVE_OLD_DISKLABEL
1312 	case ODIOCGDEFLABEL:
1313 		newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1314 		if (newlabel == NULL)
1315 			return EIO;
1316 		wdgetdefaultlabel(wd, newlabel);
1317 		if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1318 			memcpy(addr, &newlabel, sizeof (struct olddisklabel));
1319 		else
1320 			error = ENOTTY;
1321 		free(newlabel, M_TEMP);
1322 		return error;
1323 #endif
1324 
1325 #ifdef notyet
1326 	case DIOCWFORMAT:
1327 		if ((flag & FWRITE) == 0)
1328 			return EBADF;
1329 		{
1330 		register struct format_op *fop;
1331 		struct iovec aiov;
1332 		struct uio auio;
1333 
1334 		fop = (struct format_op *)addr;
1335 		aiov.iov_base = fop->df_buf;
1336 		aiov.iov_len = fop->df_count;
1337 		auio.uio_iov = &aiov;
1338 		auio.uio_iovcnt = 1;
1339 		auio.uio_resid = fop->df_count;
1340 		auio.uio_offset =
1341 			fop->df_startblk * wd->sc_dk.dk_label->d_secsize;
1342 		auio.uio_vmspace = l->l_proc->p_vmspace;
1343 		error = physio(wdformat, NULL, dev, B_WRITE, minphys,
1344 		    &auio);
1345 		fop->df_count -= auio.uio_resid;
1346 		fop->df_reg[0] = wdc->sc_status;
1347 		fop->df_reg[1] = wdc->sc_error;
1348 		return error;
1349 		}
1350 #endif
1351 	case DIOCGCACHE:
1352 		return wd_getcache(wd, (int *)addr);
1353 
1354 	case DIOCSCACHE:
1355 		return wd_setcache(wd, *(int *)addr);
1356 
1357 	case DIOCCACHESYNC:
1358 		return wd_flushcache(wd, AT_WAIT);
1359 
1360 	case ATAIOCCOMMAND:
1361 		/*
1362 		 * Make sure this command is (relatively) safe first
1363 		 */
1364 		if ((((atareq_t *) addr)->flags & ATACMD_READ) == 0 &&
1365 		    (flag & FWRITE) == 0)
1366 			return (EBADF);
1367 		{
1368 		struct wd_ioctl *wi;
1369 		atareq_t *atareq = (atareq_t *) addr;
1370 		int error1;
1371 
1372 		wi = wi_get();
1373 		wi->wi_softc = wd;
1374 		wi->wi_atareq = *atareq;
1375 
1376 		if (atareq->datalen && atareq->flags &
1377 		    (ATACMD_READ | ATACMD_WRITE)) {
1378 			void *tbuf;
1379 			if (atareq->datalen < DEV_BSIZE
1380 			    && atareq->command == WDCC_IDENTIFY) {
1381 				tbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1382 				wi->wi_iov.iov_base = tbuf;
1383 				wi->wi_iov.iov_len = DEV_BSIZE;
1384 				UIO_SETUP_SYSSPACE(&wi->wi_uio);
1385 			} else {
1386 				tbuf = NULL;
1387 				wi->wi_iov.iov_base = atareq->databuf;
1388 				wi->wi_iov.iov_len = atareq->datalen;
1389 				wi->wi_uio.uio_vmspace = l->l_proc->p_vmspace;
1390 			}
1391 			wi->wi_uio.uio_iov = &wi->wi_iov;
1392 			wi->wi_uio.uio_iovcnt = 1;
1393 			wi->wi_uio.uio_resid = atareq->datalen;
1394 			wi->wi_uio.uio_offset = 0;
1395 			wi->wi_uio.uio_rw =
1396 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE;
1397 			error1 = physio(wdioctlstrategy, &wi->wi_bp, dev,
1398 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE,
1399 			    minphys, &wi->wi_uio);
1400 			if (tbuf != NULL && error1 == 0) {
1401 				error1 = copyout(tbuf, atareq->databuf,
1402 				    atareq->datalen);
1403 				free(tbuf, M_TEMP);
1404 			}
1405 		} else {
1406 			/* No need to call physio if we don't have any
1407 			   user data */
1408 			wi->wi_bp.b_flags = 0;
1409 			wi->wi_bp.b_data = 0;
1410 			wi->wi_bp.b_bcount = 0;
1411 			wi->wi_bp.b_dev = 0;
1412 			wi->wi_bp.b_proc = l->l_proc;
1413 			wdioctlstrategy(&wi->wi_bp);
1414 			error1 = wi->wi_bp.b_error;
1415 		}
1416 		*atareq = wi->wi_atareq;
1417 		wi_free(wi);
1418 		return(error1);
1419 		}
1420 
1421 	case DIOCAWEDGE:
1422 	    {
1423 	    	struct dkwedge_info *dkw = (void *) addr;
1424 
1425 		if ((flag & FWRITE) == 0)
1426 			return (EBADF);
1427 
1428 		/* If the ioctl happens here, the parent is us. */
1429 		strcpy(dkw->dkw_parent, device_xname(wd->sc_dev));
1430 		return (dkwedge_add(dkw));
1431 	    }
1432 
1433 	case DIOCDWEDGE:
1434 	    {
1435 	    	struct dkwedge_info *dkw = (void *) addr;
1436 
1437 		if ((flag & FWRITE) == 0)
1438 			return (EBADF);
1439 
1440 		/* If the ioctl happens here, the parent is us. */
1441 		strcpy(dkw->dkw_parent, device_xname(wd->sc_dev));
1442 		return (dkwedge_del(dkw));
1443 	    }
1444 
1445 	case DIOCLWEDGES:
1446 	    {
1447 	    	struct dkwedge_list *dkwl = (void *) addr;
1448 
1449 		return (dkwedge_list(&wd->sc_dk, dkwl, l));
1450 	    }
1451 
1452 	case DIOCGSTRATEGY:
1453 	    {
1454 		struct disk_strategy *dks = (void *)addr;
1455 
1456 		s = splbio();
1457 		strlcpy(dks->dks_name, bufq_getstrategyname(wd->sc_q),
1458 		    sizeof(dks->dks_name));
1459 		splx(s);
1460 		dks->dks_paramlen = 0;
1461 
1462 		return 0;
1463 	    }
1464 
1465 	case DIOCSSTRATEGY:
1466 	    {
1467 		struct disk_strategy *dks = (void *)addr;
1468 		struct bufq_state *new;
1469 		struct bufq_state *old;
1470 
1471 		if ((flag & FWRITE) == 0) {
1472 			return EBADF;
1473 		}
1474 		if (dks->dks_param != NULL) {
1475 			return EINVAL;
1476 		}
1477 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
1478 		error = bufq_alloc(&new, dks->dks_name,
1479 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
1480 		if (error) {
1481 			return error;
1482 		}
1483 		s = splbio();
1484 		old = wd->sc_q;
1485 		bufq_move(new, old);
1486 		wd->sc_q = new;
1487 		splx(s);
1488 		bufq_free(old);
1489 
1490 		return 0;
1491 	    }
1492 
1493 	default:
1494 		return ENOTTY;
1495 	}
1496 
1497 #ifdef DIAGNOSTIC
1498 	panic("wdioctl: impossible");
1499 #endif
1500 }
1501 
1502 #ifdef B_FORMAT
1503 int
1504 wdformat(struct buf *bp)
1505 {
1506 
1507 	bp->b_flags |= B_FORMAT;
1508 	return wdstrategy(bp);
1509 }
1510 #endif
1511 
1512 int
1513 wdsize(dev_t dev)
1514 {
1515 	struct wd_softc *wd;
1516 	int part, omask;
1517 	int size;
1518 
1519 	ATADEBUG_PRINT(("wdsize\n"), DEBUG_FUNCS);
1520 
1521 	wd = device_lookup_private(&wd_cd, WDUNIT(dev));
1522 	if (wd == NULL)
1523 		return (-1);
1524 
1525 	part = WDPART(dev);
1526 	omask = wd->sc_dk.dk_openmask & (1 << part);
1527 
1528 	if (omask == 0 && wdopen(dev, 0, S_IFBLK, NULL) != 0)
1529 		return (-1);
1530 	if (wd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1531 		size = -1;
1532 	else
1533 		size = wd->sc_dk.dk_label->d_partitions[part].p_size *
1534 		    (wd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1535 	if (omask == 0 && wdclose(dev, 0, S_IFBLK, NULL) != 0)
1536 		return (-1);
1537 	return (size);
1538 }
1539 
1540 /* #define WD_DUMP_NOT_TRUSTED if you just want to watch */
1541 static int wddoingadump = 0;
1542 static int wddumprecalibrated = 0;
1543 
1544 /*
1545  * Dump core after a system crash.
1546  */
1547 int
1548 wddump(dev_t dev, daddr_t blkno, void *va, size_t size)
1549 {
1550 	struct wd_softc *wd;	/* disk unit to do the I/O */
1551 	struct disklabel *lp;   /* disk's disklabel */
1552 	int part, err;
1553 	int nblks;	/* total number of sectors left to write */
1554 
1555 	/* Check if recursive dump; if so, punt. */
1556 	if (wddoingadump)
1557 		return EFAULT;
1558 	wddoingadump = 1;
1559 
1560 	wd = device_lookup_private(&wd_cd, WDUNIT(dev));
1561 	if (wd == NULL)
1562 		return (ENXIO);
1563 
1564 	part = WDPART(dev);
1565 
1566 	/* Convert to disk sectors.  Request must be a multiple of size. */
1567 	lp = wd->sc_dk.dk_label;
1568 	if ((size % lp->d_secsize) != 0)
1569 		return EFAULT;
1570 	nblks = size / lp->d_secsize;
1571 	blkno = blkno / (lp->d_secsize / DEV_BSIZE);
1572 
1573 	/* Check transfer bounds against partition size. */
1574 	if ((blkno < 0) || ((blkno + nblks) > lp->d_partitions[part].p_size))
1575 		return EINVAL;
1576 
1577 	/* Offset block number to start of partition. */
1578 	blkno += lp->d_partitions[part].p_offset;
1579 
1580 	/* Recalibrate, if first dump transfer. */
1581 	if (wddumprecalibrated == 0) {
1582 		wddumprecalibrated = 1;
1583 		(*wd->atabus->ata_reset_drive)(wd->drvp,
1584 					       AT_POLL | AT_RST_EMERG);
1585 		wd->drvp->state = RESET;
1586 	}
1587 
1588 	wd->sc_bp = NULL;
1589 	wd->sc_wdc_bio.blkno = blkno;
1590 	wd->sc_wdc_bio.flags = ATA_POLL;
1591 	if (wd->sc_flags & WDF_LBA48 &&
1592 	    (wd->sc_wdc_bio.blkno + nblks) > wd->sc_capacity28)
1593 		wd->sc_wdc_bio.flags |= ATA_LBA48;
1594 	if (wd->sc_flags & WDF_LBA)
1595 		wd->sc_wdc_bio.flags |= ATA_LBA;
1596 	wd->sc_wdc_bio.bcount = nblks * lp->d_secsize;
1597 	wd->sc_wdc_bio.databuf = va;
1598 #ifndef WD_DUMP_NOT_TRUSTED
1599 	switch (err = wd->atabus->ata_bio(wd->drvp, &wd->sc_wdc_bio)) {
1600 	case ATACMD_TRY_AGAIN:
1601 		panic("wddump: try again");
1602 		break;
1603 	case ATACMD_QUEUED:
1604 		panic("wddump: polled command has been queued");
1605 		break;
1606 	case ATACMD_COMPLETE:
1607 		break;
1608 	default:
1609 		panic("wddump: unknown atacmd code %d", err);
1610 	}
1611 	switch(err = wd->sc_wdc_bio.error) {
1612 	case TIMEOUT:
1613 		printf("wddump: device timed out");
1614 		err = EIO;
1615 		break;
1616 	case ERR_DF:
1617 		printf("wddump: drive fault");
1618 		err = EIO;
1619 		break;
1620 	case ERR_DMA:
1621 		printf("wddump: DMA error");
1622 		err = EIO;
1623 		break;
1624 	case ERROR:
1625 		printf("wddump: ");
1626 		wdperror(wd);
1627 		err = EIO;
1628 		break;
1629 	case NOERROR:
1630 		err = 0;
1631 		break;
1632 	default:
1633 		panic("wddump: unknown error type %d", err);
1634 	}
1635 	if (err != 0) {
1636 		printf("\n");
1637 		return err;
1638 	}
1639 #else	/* WD_DUMP_NOT_TRUSTED */
1640 	/* Let's just talk about this first... */
1641 	printf("wd%d: dump addr 0x%x, cylin %d, head %d, sector %d\n",
1642 	    unit, va, cylin, head, sector);
1643 	delay(500 * 1000);	/* half a second */
1644 #endif
1645 
1646 	wddoingadump = 0;
1647 	return 0;
1648 }
1649 
1650 #ifdef HAS_BAD144_HANDLING
1651 /*
1652  * Internalize the bad sector table.
1653  */
1654 void
1655 bad144intern(struct wd_softc *wd)
1656 {
1657 	struct dkbad *bt = &wd->sc_dk.dk_cpulabel->bad;
1658 	struct disklabel *lp = wd->sc_dk.dk_label;
1659 	int i = 0;
1660 
1661 	ATADEBUG_PRINT(("bad144intern\n"), DEBUG_XFERS);
1662 
1663 	for (; i < NBT_BAD; i++) {
1664 		if (bt->bt_bad[i].bt_cyl == 0xffff)
1665 			break;
1666 		wd->sc_badsect[i] =
1667 		    bt->bt_bad[i].bt_cyl * lp->d_secpercyl +
1668 		    (bt->bt_bad[i].bt_trksec >> 8) * lp->d_nsectors +
1669 		    (bt->bt_bad[i].bt_trksec & 0xff);
1670 	}
1671 	for (; i < NBT_BAD+1; i++)
1672 		wd->sc_badsect[i] = -1;
1673 }
1674 #endif
1675 
1676 static void
1677 wd_params_to_properties(struct wd_softc *wd, struct ataparams *params)
1678 {
1679 	prop_dictionary_t disk_info, odisk_info, geom;
1680 	const char *cp;
1681 
1682 	disk_info = prop_dictionary_create();
1683 
1684 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
1685 		cp = "ST506";
1686 	else {
1687 		/* XXX Should have a case for ATA here, too. */
1688 		cp = "ESDI";
1689 	}
1690 	prop_dictionary_set_cstring_nocopy(disk_info, "type", cp);
1691 
1692 	geom = prop_dictionary_create();
1693 
1694 	prop_dictionary_set_uint64(geom, "sectors-per-unit", wd->sc_capacity);
1695 
1696 	prop_dictionary_set_uint32(geom, "sector-size",
1697 				   DEV_BSIZE /* XXX 512? */);
1698 
1699 	prop_dictionary_set_uint16(geom, "sectors-per-track",
1700 				   wd->sc_params.atap_sectors);
1701 
1702 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
1703 				   wd->sc_params.atap_heads);
1704 
1705 	if (wd->sc_flags & WDF_LBA)
1706 		prop_dictionary_set_uint64(geom, "cylinders-per-unit",
1707 					   wd->sc_capacity /
1708 					       (wd->sc_params.atap_heads *
1709 					        wd->sc_params.atap_sectors));
1710 	else
1711 		prop_dictionary_set_uint16(geom, "cylinders-per-unit",
1712 					   wd->sc_params.atap_cylinders);
1713 
1714 	prop_dictionary_set(disk_info, "geometry", geom);
1715 	prop_object_release(geom);
1716 
1717 	prop_dictionary_set(device_properties(wd->sc_dev),
1718 			    "disk-info", disk_info);
1719 
1720 	/*
1721 	 * Don't release disk_info here; we keep a reference to it.
1722 	 * disk_detach() will release it when we go away.
1723 	 */
1724 
1725 	odisk_info = wd->sc_dk.dk_info;
1726 	wd->sc_dk.dk_info = disk_info;
1727 	if (odisk_info)
1728 		prop_object_release(odisk_info);
1729 }
1730 
1731 int
1732 wd_get_params(struct wd_softc *wd, u_int8_t flags, struct ataparams *params)
1733 {
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 		aprint_error_dev(wd->sc_dev,
1815 		    "wd_setcache command not complete\n");
1816 		return EIO;
1817 	}
1818 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1819 		char sbuf[sizeof(at_errbits) + 64];
1820 		snprintb(sbuf, sizeof(sbuf), at_errbits, ata_c.flags);
1821 		aprint_error_dev(wd->sc_dev, "wd_setcache: status=%s\n", sbuf);
1822 		return EIO;
1823 	}
1824 	return 0;
1825 }
1826 
1827 static int
1828 wd_standby(struct wd_softc *wd, int flags)
1829 {
1830 	struct ata_command ata_c;
1831 
1832 	memset(&ata_c, 0, sizeof(struct ata_command));
1833 	ata_c.r_command = WDCC_STANDBY_IMMED;
1834 	ata_c.r_st_bmask = WDCS_DRDY;
1835 	ata_c.r_st_pmask = WDCS_DRDY;
1836 	ata_c.flags = flags;
1837 	ata_c.timeout = 30000; /* 30s timeout */
1838 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1839 		aprint_error_dev(wd->sc_dev,
1840 		    "standby immediate command didn't complete\n");
1841 		return EIO;
1842 	}
1843 	if (ata_c.flags & AT_ERROR) {
1844 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
1845 			return ENODEV;
1846 	}
1847 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1848 		char sbuf[sizeof(at_errbits) + 64];
1849 		snprintb(sbuf, sizeof(sbuf), at_errbits, ata_c.flags);
1850 		aprint_error_dev(wd->sc_dev, "wd_standby: status=%s\n", sbuf);
1851 		return EIO;
1852 	}
1853 	return 0;
1854 }
1855 
1856 int
1857 wd_flushcache(struct wd_softc *wd, int flags)
1858 {
1859 	struct ata_command ata_c;
1860 
1861 	/*
1862 	 * WDCC_FLUSHCACHE is here since ATA-4, but some drives report
1863 	 * only ATA-2 and still support it.
1864 	 */
1865 	if (wd->drvp->ata_vers < 4 &&
1866 	    ((wd->sc_params.atap_cmd_set2 & WDC_CMD2_FC) == 0 ||
1867 	    wd->sc_params.atap_cmd_set2 == 0xffff))
1868 		return ENODEV;
1869 	memset(&ata_c, 0, sizeof(struct ata_command));
1870 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0 &&
1871 	    (wd->sc_params.atap_cmd2_en & ATA_CMD2_FCE) != 0)
1872 		ata_c.r_command = WDCC_FLUSHCACHE_EXT;
1873 	else
1874 		ata_c.r_command = WDCC_FLUSHCACHE;
1875 	ata_c.r_st_bmask = WDCS_DRDY;
1876 	ata_c.r_st_pmask = WDCS_DRDY;
1877 	ata_c.flags = flags;
1878 	ata_c.timeout = 30000; /* 30s timeout */
1879 	if (wd->atabus->ata_exec_command(wd->drvp, &ata_c) != ATACMD_COMPLETE) {
1880 		aprint_error_dev(wd->sc_dev,
1881 		    "flush cache command didn't complete\n");
1882 		return EIO;
1883 	}
1884 	if (ata_c.flags & AT_ERROR) {
1885 		if (ata_c.r_error == WDCE_ABRT) /* command not supported */
1886 			return ENODEV;
1887 	}
1888 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1889 		char sbuf[sizeof(at_errbits) + 64];
1890 		snprintb(sbuf, sizeof(sbuf), at_errbits, ata_c.flags);
1891 		aprint_error_dev(wd->sc_dev, "wd_flushcache: status=%s\n",
1892 		    sbuf);
1893 		return EIO;
1894 	}
1895 	return 0;
1896 }
1897 
1898 bool
1899 wd_shutdown(device_t dev, int how)
1900 {
1901 	struct wd_softc *wd = device_private(dev);
1902 
1903 	/* the adapter needs to be enabled */
1904 	if (wd->atabus->ata_addref(wd->drvp))
1905 		return true; /* no need to complain */
1906 
1907 	wd_flushcache(wd, AT_POLL);
1908 	if ((how & RB_POWERDOWN) == RB_POWERDOWN)
1909 		wd_standby(wd, AT_POLL);
1910 	return true;
1911 }
1912 
1913 /*
1914  * Allocate space for a ioctl queue structure.  Mostly taken from
1915  * scsipi_ioctl.c
1916  */
1917 struct wd_ioctl *
1918 wi_get(void)
1919 {
1920 	struct wd_ioctl *wi;
1921 	int s;
1922 
1923 	wi = malloc(sizeof(struct wd_ioctl), M_TEMP, M_WAITOK|M_ZERO);
1924 	buf_init(&wi->wi_bp);
1925 	s = splbio();
1926 	LIST_INSERT_HEAD(&wi_head, wi, wi_list);
1927 	splx(s);
1928 	return (wi);
1929 }
1930 
1931 /*
1932  * Free an ioctl structure and remove it from our list
1933  */
1934 
1935 void
1936 wi_free(struct wd_ioctl *wi)
1937 {
1938 	int s;
1939 
1940 	s = splbio();
1941 	LIST_REMOVE(wi, wi_list);
1942 	splx(s);
1943 	buf_destroy(&wi->wi_bp);
1944 	free(wi, M_TEMP);
1945 }
1946 
1947 /*
1948  * Find a wd_ioctl structure based on the struct buf.
1949  */
1950 
1951 struct wd_ioctl *
1952 wi_find(struct buf *bp)
1953 {
1954 	struct wd_ioctl *wi;
1955 	int s;
1956 
1957 	s = splbio();
1958 	for (wi = wi_head.lh_first; wi != 0; wi = wi->wi_list.le_next)
1959 		if (bp == &wi->wi_bp)
1960 			break;
1961 	splx(s);
1962 	return (wi);
1963 }
1964 
1965 /*
1966  * Ioctl pseudo strategy routine
1967  *
1968  * This is mostly stolen from scsipi_ioctl.c:scsistrategy().  What
1969  * happens here is:
1970  *
1971  * - wdioctl() queues a wd_ioctl structure.
1972  *
1973  * - wdioctl() calls physio/wdioctlstrategy based on whether or not
1974  *   user space I/O is required.  If physio() is called, physio() eventually
1975  *   calls wdioctlstrategy().
1976  *
1977  * - In either case, wdioctlstrategy() calls wd->atabus->ata_exec_command()
1978  *   to perform the actual command
1979  *
1980  * The reason for the use of the pseudo strategy routine is because
1981  * when doing I/O to/from user space, physio _really_ wants to be in
1982  * the loop.  We could put the entire buffer into the ioctl request
1983  * structure, but that won't scale if we want to do things like download
1984  * microcode.
1985  */
1986 
1987 void
1988 wdioctlstrategy(struct buf *bp)
1989 {
1990 	struct wd_ioctl *wi;
1991 	struct ata_command ata_c;
1992 	int error = 0;
1993 
1994 	wi = wi_find(bp);
1995 	if (wi == NULL) {
1996 		printf("wdioctlstrategy: "
1997 		    "No matching ioctl request found in queue\n");
1998 		error = EINVAL;
1999 		goto bad;
2000 	}
2001 
2002 	memset(&ata_c, 0, sizeof(ata_c));
2003 
2004 	/*
2005 	 * Abort if physio broke up the transfer
2006 	 */
2007 
2008 	if (bp->b_bcount != wi->wi_atareq.datalen) {
2009 		printf("physio split wd ioctl request... cannot proceed\n");
2010 		error = EIO;
2011 		goto bad;
2012 	}
2013 
2014 	/*
2015 	 * Abort if we didn't get a buffer size that was a multiple of
2016 	 * our sector size (or was larger than NBBY)
2017 	 */
2018 
2019 	if ((bp->b_bcount % wi->wi_softc->sc_dk.dk_label->d_secsize) != 0 ||
2020 	    (bp->b_bcount / wi->wi_softc->sc_dk.dk_label->d_secsize) >=
2021 	     (1 << NBBY)) {
2022 		error = EINVAL;
2023 		goto bad;
2024 	}
2025 
2026 	/*
2027 	 * Make sure a timeout was supplied in the ioctl request
2028 	 */
2029 
2030 	if (wi->wi_atareq.timeout == 0) {
2031 		error = EINVAL;
2032 		goto bad;
2033 	}
2034 
2035 	if (wi->wi_atareq.flags & ATACMD_READ)
2036 		ata_c.flags |= AT_READ;
2037 	else if (wi->wi_atareq.flags & ATACMD_WRITE)
2038 		ata_c.flags |= AT_WRITE;
2039 
2040 	if (wi->wi_atareq.flags & ATACMD_READREG)
2041 		ata_c.flags |= AT_READREG;
2042 
2043 	ata_c.flags |= AT_WAIT;
2044 
2045 	ata_c.timeout = wi->wi_atareq.timeout;
2046 	ata_c.r_command = wi->wi_atareq.command;
2047 	ata_c.r_head = wi->wi_atareq.head & 0x0f;
2048 	ata_c.r_cyl = wi->wi_atareq.cylinder;
2049 	ata_c.r_sector = wi->wi_atareq.sec_num;
2050 	ata_c.r_count = wi->wi_atareq.sec_count;
2051 	ata_c.r_features = wi->wi_atareq.features;
2052 	ata_c.r_st_bmask = WDCS_DRDY;
2053 	ata_c.r_st_pmask = WDCS_DRDY;
2054 	ata_c.data = wi->wi_bp.b_data;
2055 	ata_c.bcount = wi->wi_bp.b_bcount;
2056 
2057 	if (wi->wi_softc->atabus->ata_exec_command(wi->wi_softc->drvp, &ata_c)
2058 	    != ATACMD_COMPLETE) {
2059 		wi->wi_atareq.retsts = ATACMD_ERROR;
2060 		goto bad;
2061 	}
2062 
2063 	if (ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
2064 		if (ata_c.flags & AT_ERROR) {
2065 			wi->wi_atareq.retsts = ATACMD_ERROR;
2066 			wi->wi_atareq.error = ata_c.r_error;
2067 		} else if (ata_c.flags & AT_DF)
2068 			wi->wi_atareq.retsts = ATACMD_DF;
2069 		else
2070 			wi->wi_atareq.retsts = ATACMD_TIMEOUT;
2071 	} else {
2072 		wi->wi_atareq.retsts = ATACMD_OK;
2073 		if (wi->wi_atareq.flags & ATACMD_READREG) {
2074 			wi->wi_atareq.head = ata_c.r_head ;
2075 			wi->wi_atareq.cylinder = ata_c.r_cyl;
2076 			wi->wi_atareq.sec_num = ata_c.r_sector;
2077 			wi->wi_atareq.sec_count = ata_c.r_count;
2078 			wi->wi_atareq.features = ata_c.r_features;
2079 			wi->wi_atareq.error = ata_c.r_error;
2080 		}
2081 	}
2082 
2083 	bp->b_error = 0;
2084 	biodone(bp);
2085 	return;
2086 bad:
2087 	bp->b_error = error;
2088 	biodone(bp);
2089 }
2090