xref: /netbsd-src/sys/dev/ata/wd.c (revision c38e7cc395b1472a774ff828e46123de44c628e9)
1 /*	$NetBSD: wd.c,v 1.438 2018/01/07 11:37:30 mlelstv 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 1.438 2018/01/07 11:37:30 mlelstv Exp $");
58 
59 #include "opt_ata.h"
60 #include "opt_wd.h"
61 
62 #include <sys/param.h>
63 #include <sys/systm.h>
64 #include <sys/kernel.h>
65 #include <sys/conf.h>
66 #include <sys/file.h>
67 #include <sys/stat.h>
68 #include <sys/ioctl.h>
69 #include <sys/buf.h>
70 #include <sys/bufq.h>
71 #include <sys/uio.h>
72 #include <sys/malloc.h>
73 #include <sys/device.h>
74 #include <sys/disklabel.h>
75 #include <sys/disk.h>
76 #include <sys/syslog.h>
77 #include <sys/proc.h>
78 #include <sys/reboot.h>
79 #include <sys/vnode.h>
80 #include <sys/rndsource.h>
81 
82 #include <sys/intr.h>
83 #include <sys/bus.h>
84 
85 #include <dev/ata/atareg.h>
86 #include <dev/ata/atavar.h>
87 #include <dev/ata/wdvar.h>
88 #include <dev/ic/wdcreg.h>
89 #include <sys/ataio.h>
90 #include "locators.h"
91 
92 #include <prop/proplib.h>
93 
94 #define	WDIORETRIES_SINGLE 4	/* number of retries for single-sector */
95 #define	WDIORETRIES	5	/* number of retries before giving up */
96 #define	RECOVERYTIME hz/2	/* time to wait before retrying a cmd */
97 
98 #define	WDUNIT(dev)		DISKUNIT(dev)
99 #define	WDPART(dev)		DISKPART(dev)
100 #define	WDMINOR(unit, part)	DISKMINOR(unit, part)
101 #define	MAKEWDDEV(maj, unit, part)	MAKEDISKDEV(maj, unit, part)
102 
103 #define	WDLABELDEV(dev)	(MAKEWDDEV(major(dev), WDUNIT(dev), RAW_PART))
104 
105 #define DEBUG_FUNCS  0x08
106 #define DEBUG_PROBE  0x10
107 #define DEBUG_DETACH 0x20
108 #define	DEBUG_XFERS  0x40
109 #ifdef ATADEBUG
110 #ifndef ATADEBUG_WD_MASK
111 #define ATADEBUG_WD_MASK 0x0
112 #endif
113 int wdcdebug_wd_mask = ATADEBUG_WD_MASK;
114 #define ATADEBUG_PRINT(args, level) \
115 	if (wdcdebug_wd_mask & (level)) \
116 		printf args
117 #else
118 #define ATADEBUG_PRINT(args, level)
119 #endif
120 
121 static int	wdprobe(device_t, cfdata_t, void *);
122 static void	wdattach(device_t, device_t, void *);
123 static int	wddetach(device_t, int);
124 static void	wdperror(const struct wd_softc *, struct ata_xfer *);
125 
126 static void	wdminphys(struct buf *);
127 
128 static int	wd_firstopen(device_t, dev_t, int, int);
129 static int	wd_lastclose(device_t);
130 static bool	wd_suspend(device_t, const pmf_qual_t *);
131 static int	wd_standby(struct wd_softc *, int);
132 
133 CFATTACH_DECL3_NEW(wd, sizeof(struct wd_softc),
134     wdprobe, wdattach, wddetach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
135 
136 extern struct cfdriver wd_cd;
137 
138 static dev_type_open(wdopen);
139 static dev_type_close(wdclose);
140 static dev_type_read(wdread);
141 static dev_type_write(wdwrite);
142 static dev_type_ioctl(wdioctl);
143 static dev_type_strategy(wdstrategy);
144 static dev_type_dump(wddump);
145 static dev_type_size(wdsize);
146 static dev_type_discard(wddiscard);
147 
148 const struct bdevsw wd_bdevsw = {
149 	.d_open = wdopen,
150 	.d_close = wdclose,
151 	.d_strategy = wdstrategy,
152 	.d_ioctl = wdioctl,
153 	.d_dump = wddump,
154 	.d_psize = wdsize,
155 	.d_discard = wddiscard,
156 	.d_flag = D_DISK
157 };
158 
159 const struct cdevsw wd_cdevsw = {
160 	.d_open = wdopen,
161 	.d_close = wdclose,
162 	.d_read = wdread,
163 	.d_write = wdwrite,
164 	.d_ioctl = wdioctl,
165 	.d_stop = nostop,
166 	.d_tty = notty,
167 	.d_poll = nopoll,
168 	.d_mmap = nommap,
169 	.d_kqfilter = nokqfilter,
170 	.d_discard = wddiscard,
171 	.d_flag = D_DISK
172 };
173 
174 /* #define WD_DUMP_NOT_TRUSTED if you just want to watch */
175 static int wddoingadump = 0;
176 static int wddumprecalibrated = 0;
177 
178 /*
179  * Glue necessary to hook WDCIOCCOMMAND into physio
180  */
181 
182 struct wd_ioctl {
183 	LIST_ENTRY(wd_ioctl) wi_list;
184 	struct buf wi_bp;
185 	struct uio wi_uio;
186 	struct iovec wi_iov;
187 	atareq_t wi_atareq;
188 	struct wd_softc *wi_softc;
189 };
190 
191 static struct	wd_ioctl *wi_find(struct buf *);
192 static void	wi_free(struct wd_ioctl *);
193 static struct	wd_ioctl *wi_get(struct wd_softc *);
194 static void	wdioctlstrategy(struct buf *);
195 
196 static void	wdstart(device_t);
197 static void	wdstart1(struct wd_softc *, struct buf *, struct ata_xfer *);
198 static int	wd_diskstart(device_t, struct buf *);
199 static int	wd_dumpblocks(device_t, void *, daddr_t, int);
200 static void	wd_iosize(device_t, int *);
201 static int	wd_discard(device_t, off_t, off_t);
202 static void	wdbiorestart(void *);
203 static void	wddone(device_t, struct ata_xfer *);
204 static int	wd_get_params(struct wd_softc *, uint8_t, struct ataparams *);
205 static void	wd_set_geometry(struct wd_softc *);
206 static int	wd_flushcache(struct wd_softc *, int, bool);
207 static int	wd_trim(struct wd_softc *, daddr_t, long);
208 static bool	wd_shutdown(device_t, int);
209 
210 static int wd_getcache(struct wd_softc *, int *);
211 static int wd_setcache(struct wd_softc *, int);
212 
213 static void wd_sysctl_attach(struct wd_softc *);
214 static void wd_sysctl_detach(struct wd_softc *);
215 
216 struct dkdriver wddkdriver = {
217 	.d_open = wdopen,
218 	.d_close = wdclose,
219 	.d_strategy = wdstrategy,
220 	.d_minphys = wdminphys,
221 	.d_diskstart = wd_diskstart,
222 	.d_dumpblocks = wd_dumpblocks,
223 	.d_iosize = wd_iosize,
224 	.d_firstopen = wd_firstopen,
225 	.d_lastclose = wd_lastclose,
226 	.d_discard = wd_discard
227 };
228 
229 #ifdef HAS_BAD144_HANDLING
230 static void bad144intern(struct wd_softc *);
231 #endif
232 
233 #define	WD_QUIRK_SPLIT_MOD15_WRITE	0x0001	/* must split certain writes */
234 
235 #define	WD_QUIRK_FMT "\20\1SPLIT_MOD15_WRITE\2FORCE_LBA48"
236 
237 /*
238  * Quirk table for IDE drives.  Put more-specific matches first, since
239  * a simple globing routine is used for matching.
240  */
241 static const struct wd_quirk {
242 	const char *wdq_match;		/* inquiry pattern to match */
243 	int wdq_quirks;			/* drive quirks */
244 } wd_quirk_table[] = {
245 	/*
246 	 * Some Seagate S-ATA drives have a PHY which can get confused
247 	 * with the way data is packetized by some S-ATA controllers.
248 	 *
249 	 * The work-around is to split in two any write transfer whose
250 	 * sector count % 15 == 1 (assuming 512 byte sectors).
251 	 *
252 	 * XXX This is an incomplete list.  There are at least a couple
253 	 * XXX more model numbers.  If you have trouble with such transfers
254 	 * XXX (8K is the most common) on Seagate S-ATA drives, please
255 	 * XXX notify thorpej@NetBSD.org.
256 	 *
257 	 * The ST360015AS has not yet been confirmed to have this
258 	 * issue, however, it is the only other drive in the
259 	 * Seagate Barracuda Serial ATA V family.
260 	 *
261 	 */
262 	{ "ST3120023AS",
263 	  WD_QUIRK_SPLIT_MOD15_WRITE },
264 	{ "ST380023AS",
265 	  WD_QUIRK_SPLIT_MOD15_WRITE },
266 	{ "ST360015AS",
267 	  WD_QUIRK_SPLIT_MOD15_WRITE },
268 	{ NULL,
269 	  0 }
270 };
271 
272 static const struct wd_quirk *
273 wd_lookup_quirks(const char *name)
274 {
275 	const struct wd_quirk *wdq;
276 	const char *estr;
277 
278 	for (wdq = wd_quirk_table; wdq->wdq_match != NULL; wdq++) {
279 		/*
280 		 * We only want exact matches (which include matches
281 		 * against globbing characters).
282 		 */
283 		if (pmatch(name, wdq->wdq_match, &estr) == 2)
284 			return (wdq);
285 	}
286 	return (NULL);
287 }
288 
289 static int
290 wdprobe(device_t parent, cfdata_t match, void *aux)
291 {
292 	struct ata_device *adev = aux;
293 
294 	if (adev == NULL)
295 		return 0;
296 	if (adev->adev_bustype->bustype_type != SCSIPI_BUSTYPE_ATA)
297 		return 0;
298 
299 	if (match->cf_loc[ATA_HLCF_DRIVE] != ATA_HLCF_DRIVE_DEFAULT &&
300 	    match->cf_loc[ATA_HLCF_DRIVE] != adev->adev_drv_data->drive)
301 		return 0;
302 	return 1;
303 }
304 
305 static void
306 wdattach(device_t parent, device_t self, void *aux)
307 {
308 	struct wd_softc *wd = device_private(self);
309 	struct dk_softc *dksc = &wd->sc_dksc;
310 	struct ata_device *adev= aux;
311 	int i, blank;
312 	char tbuf[41], pbuf[9], c, *p, *q;
313 	const struct wd_quirk *wdq;
314 	int dtype = DKTYPE_UNKNOWN;
315 
316 	dksc->sc_dev = self;
317 
318 	ATADEBUG_PRINT(("wdattach\n"), DEBUG_FUNCS | DEBUG_PROBE);
319 	mutex_init(&wd->sc_lock, MUTEX_DEFAULT, IPL_BIO);
320 #ifdef WD_SOFTBADSECT
321 	SLIST_INIT(&wd->sc_bslist);
322 #endif
323 	wd->atabus = adev->adev_bustype;
324 	wd->drvp = adev->adev_drv_data;
325 
326 	wd->drvp->drv_openings = 1;
327 	wd->drvp->drv_start = wdstart;
328 	wd->drvp->drv_done = wddone;
329 	wd->drvp->drv_softc = dksc->sc_dev; /* done in atabusconfig_thread()
330 					     but too late */
331 
332 	aprint_naive("\n");
333 	aprint_normal("\n");
334 
335 	/* read our drive info */
336 	if (wd_get_params(wd, AT_WAIT, &wd->sc_params) != 0) {
337 		aprint_error_dev(self, "IDENTIFY failed\n");
338 		goto out;
339 	}
340 
341 	for (blank = 0, p = wd->sc_params.atap_model, q = tbuf, i = 0;
342 	    i < sizeof(wd->sc_params.atap_model); i++) {
343 		c = *p++;
344 		if (c == '\0')
345 			break;
346 		if (c != ' ') {
347 			if (blank) {
348 				*q++ = ' ';
349 				blank = 0;
350 			}
351 			*q++ = c;
352 		} else
353 			blank = 1;
354 	}
355 	*q++ = '\0';
356 
357 	aprint_normal_dev(self, "<%s>\n", tbuf);
358 
359 	wdq = wd_lookup_quirks(tbuf);
360 	if (wdq != NULL)
361 		wd->sc_quirks = wdq->wdq_quirks;
362 
363 	if (wd->sc_quirks != 0) {
364 		char sbuf[sizeof(WD_QUIRK_FMT) + 64];
365 		snprintb(sbuf, sizeof(sbuf), WD_QUIRK_FMT, wd->sc_quirks);
366 		aprint_normal_dev(self, "quirks %s\n", sbuf);
367 
368 		if (wd->sc_quirks & WD_QUIRK_SPLIT_MOD15_WRITE) {
369 			aprint_error_dev(self, "drive corrupts write transfers with certain controllers, consider replacing\n");
370 		}
371 	}
372 
373 	if ((wd->sc_params.atap_multi & 0xff) > 1) {
374 		wd->drvp->multi = wd->sc_params.atap_multi & 0xff;
375 	} else {
376 		wd->drvp->multi = 1;
377 	}
378 
379 	aprint_verbose_dev(self, "drive supports %d-sector PIO transfers,",
380 	    wd->drvp->multi);
381 
382 	/* 48-bit LBA addressing */
383 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0)
384 		wd->sc_flags |= WDF_LBA48;
385 
386 	/* Prior to ATA-4, LBA was optional. */
387 	if ((wd->sc_params.atap_capabilities1 & WDC_CAP_LBA) != 0)
388 		wd->sc_flags |= WDF_LBA;
389 #if 0
390 	/* ATA-4 requires LBA. */
391 	if (wd->sc_params.atap_ataversion != 0xffff &&
392 	    wd->sc_params.atap_ataversion >= WDC_VER_ATA4)
393 		wd->sc_flags |= WDF_LBA;
394 #endif
395 
396 	if ((wd->sc_flags & WDF_LBA48) != 0) {
397 		aprint_verbose(" LBA48 addressing\n");
398 		wd->sc_capacity =
399 		    ((uint64_t) wd->sc_params.atap_max_lba[3] << 48) |
400 		    ((uint64_t) wd->sc_params.atap_max_lba[2] << 32) |
401 		    ((uint64_t) wd->sc_params.atap_max_lba[1] << 16) |
402 		    ((uint64_t) wd->sc_params.atap_max_lba[0] <<  0);
403 		wd->sc_capacity28 =
404 		    (wd->sc_params.atap_capacity[1] << 16) |
405 		    wd->sc_params.atap_capacity[0];
406 	} else if ((wd->sc_flags & WDF_LBA) != 0) {
407 		aprint_verbose(" LBA addressing\n");
408 		wd->sc_capacity28 = wd->sc_capacity =
409 		    (wd->sc_params.atap_capacity[1] << 16) |
410 		    wd->sc_params.atap_capacity[0];
411 	} else {
412 		aprint_verbose(" chs addressing\n");
413 		wd->sc_capacity28 = wd->sc_capacity =
414 		    wd->sc_params.atap_cylinders *
415 		    wd->sc_params.atap_heads *
416 		    wd->sc_params.atap_sectors;
417 	}
418 	if ((wd->sc_params.atap_secsz & ATA_SECSZ_VALID_MASK) == ATA_SECSZ_VALID
419 	    && ((wd->sc_params.atap_secsz & ATA_SECSZ_LLS) != 0)) {
420 		wd->sc_blksize = 2ULL *
421 		    ((uint32_t)((wd->sc_params.atap_lls_secsz[1] << 16) |
422 		    wd->sc_params.atap_lls_secsz[0]));
423 	} else {
424 		wd->sc_blksize = 512;
425 	}
426 	wd->sc_capacity512 = (wd->sc_capacity * wd->sc_blksize) / DEV_BSIZE;
427 	format_bytes(pbuf, sizeof(pbuf), wd->sc_capacity * wd->sc_blksize);
428 	aprint_normal_dev(self, "%s, %d cyl, %d head, %d sec, "
429 	    "%d bytes/sect x %llu sectors\n",
430 	    pbuf,
431 	    (wd->sc_flags & WDF_LBA) ? (int)(wd->sc_capacity /
432 		(wd->sc_params.atap_heads * wd->sc_params.atap_sectors)) :
433 		wd->sc_params.atap_cylinders,
434 	    wd->sc_params.atap_heads, wd->sc_params.atap_sectors,
435 	    wd->sc_blksize, (unsigned long long)wd->sc_capacity);
436 
437 	ATADEBUG_PRINT(("%s: atap_dmatiming_mimi=%d, atap_dmatiming_recom=%d\n",
438 	    device_xname(self), wd->sc_params.atap_dmatiming_mimi,
439 	    wd->sc_params.atap_dmatiming_recom), DEBUG_PROBE);
440 
441 	if (wd->sc_blksize <= 0 || !powerof2(wd->sc_blksize) ||
442 	    wd->sc_blksize < DEV_BSIZE || wd->sc_blksize > MAXPHYS) {
443 		aprint_normal_dev(self, "WARNING: block size %u "
444 		    "might not actually work\n", wd->sc_blksize);
445 	}
446 
447 	if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
448 		dtype = DKTYPE_ST506;
449 	else
450 		dtype = DKTYPE_ESDI;
451 
452 out:
453 	/*
454 	 * Initialize and attach the disk structure.
455 	 */
456 	dk_init(dksc, self, dtype);
457 	disk_init(&dksc->sc_dkdev, dksc->sc_xname, &wddkdriver);
458 
459 	/* Attach dk and disk subsystems */
460 	dk_attach(dksc);
461 	disk_attach(&dksc->sc_dkdev);
462 	wd_set_geometry(wd);
463 
464 	bufq_alloc(&dksc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
465 
466 	/* reference to label structure, used by ata code */
467 	wd->drvp->lp = dksc->sc_dkdev.dk_label;
468 
469 	/* Discover wedges on this disk. */
470 	dkwedge_discover(&dksc->sc_dkdev);
471 
472 	if (!pmf_device_register1(self, wd_suspend, NULL, wd_shutdown))
473 		aprint_error_dev(self, "couldn't establish power handler\n");
474 
475 	wd_sysctl_attach(wd);
476 }
477 
478 static bool
479 wd_suspend(device_t dv, const pmf_qual_t *qual)
480 {
481 	struct wd_softc *sc = device_private(dv);
482 
483 	/* the adapter needs to be enabled */
484 	if (sc->atabus->ata_addref(sc->drvp))
485 		return true; /* no need to complain */
486 
487 	wd_flushcache(sc, AT_WAIT, false);
488 	wd_standby(sc, AT_WAIT);
489 
490 	sc->atabus->ata_delref(sc->drvp);
491 	return true;
492 }
493 
494 static int
495 wddetach(device_t self, int flags)
496 {
497 	struct wd_softc *wd = device_private(self);
498 	struct dk_softc *dksc = &wd->sc_dksc;
499 	int bmaj, cmaj, i, mn, rc;
500 
501 	if ((rc = disk_begindetach(&dksc->sc_dkdev, wd_lastclose, self, flags)) != 0)
502 		return rc;
503 
504 	/* locate the major number */
505 	bmaj = bdevsw_lookup_major(&wd_bdevsw);
506 	cmaj = cdevsw_lookup_major(&wd_cdevsw);
507 
508 	/* Nuke the vnodes for any open instances. */
509 	for (i = 0; i < MAXPARTITIONS; i++) {
510 		mn = WDMINOR(device_unit(self), i);
511 		vdevgone(bmaj, mn, mn, VBLK);
512 		vdevgone(cmaj, mn, mn, VCHR);
513 	}
514 
515 	dk_drain(dksc);
516 
517 	/* Kill off any pending commands. */
518 	mutex_enter(&wd->sc_lock);
519 	wd->atabus->ata_killpending(wd->drvp);
520 	mutex_exit(&wd->sc_lock);
521 
522 	bufq_free(dksc->sc_bufq);
523 
524 	if (flags & DETACH_POWEROFF)
525 		wd_standby(wd, AT_POLL);
526 
527 	/* Delete all of our wedges. */
528 	dkwedge_delall(&dksc->sc_dkdev);
529 
530 	/* Detach from the disk list. */
531 	disk_detach(&dksc->sc_dkdev);
532 	disk_destroy(&dksc->sc_dkdev);
533 
534 	dk_detach(dksc);
535 
536 #ifdef WD_SOFTBADSECT
537 	/* Clean out the bad sector list */
538 	while (!SLIST_EMPTY(&wd->sc_bslist)) {
539 		void *head = SLIST_FIRST(&wd->sc_bslist);
540 		SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
541 		free(head, M_TEMP);
542 	}
543 	wd->sc_bscount = 0;
544 #endif
545 
546 	pmf_device_deregister(self);
547 
548 	wd_sysctl_detach(wd);
549 
550 	mutex_destroy(&wd->sc_lock);
551 
552 	wd->drvp->drive_type = ATA_DRIVET_NONE; /* no drive any more here */
553 	wd->drvp->drive_flags = 0;
554 
555 	return (0);
556 }
557 
558 /*
559  * Read/write routine for a buffer.  Validates the arguments and schedules the
560  * transfer.  Does not wait for the transfer to complete.
561  */
562 static void
563 wdstrategy(struct buf *bp)
564 {
565 	struct wd_softc *wd =
566 	    device_lookup_private(&wd_cd, WDUNIT(bp->b_dev));
567 	struct dk_softc *dksc = &wd->sc_dksc;
568 
569 	ATADEBUG_PRINT(("wdstrategy (%s)\n", dksc->sc_xname),
570 	    DEBUG_XFERS);
571 
572 	/* If device invalidated (e.g. media change, door open,
573 	 * device detachment), then error.
574 	 */
575 	if ((wd->sc_flags & WDF_LOADED) == 0 ||
576 	    !device_is_enabled(dksc->sc_dev))
577 		goto err;
578 
579 #ifdef WD_SOFTBADSECT
580 	/*
581 	 * If the transfer about to be attempted contains only a block that
582 	 * is known to be bad then return an error for the transfer without
583 	 * even attempting to start a transfer up under the premis that we
584 	 * will just end up doing more retries for a transfer that will end
585 	 * up failing again.
586 	 */
587 	if (__predict_false(!SLIST_EMPTY(&wd->sc_bslist))) {
588 		struct disklabel *lp = dksc->sc_dkdev.dk_label;
589 		struct disk_badsectors *dbs;
590 		daddr_t blkno, maxblk;
591 
592 		/* convert the block number to absolute */
593 		if (lp->d_secsize >= DEV_BSIZE)
594 			blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
595 		else
596 			blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
597 		if (WDPART(bp->b_dev) != RAW_PART)
598 			blkno += lp->d_partitions[WDPART(bp->b_dev)].p_offset;
599 		maxblk = blkno + (bp->b_bcount / wd->sc_blksize) - 1;
600 
601 		mutex_enter(&wd->sc_lock);
602 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next)
603 			if ((dbs->dbs_min <= bp->b_rawblkno &&
604 			     bp->b_rawblkno <= dbs->dbs_max) ||
605 			    (dbs->dbs_min <= maxblk && maxblk <= dbs->dbs_max)){
606 				mutex_exit(&wd->sc_lock);
607 				goto err;
608 			}
609 		mutex_exit(&wd->sc_lock);
610 	}
611 #endif
612 
613 	dk_strategy(dksc, bp);
614 	return;
615 
616 err:
617 	bp->b_error = EIO;
618 	bp->b_resid = bp->b_bcount;
619 	biodone(bp);
620 }
621 
622 static void
623 wdstart1(struct wd_softc *wd, struct buf *bp, struct ata_xfer *xfer)
624 {
625 	struct dk_softc *dksc = &wd->sc_dksc;
626 
627 	KASSERT(bp == xfer->c_bio.bp || xfer->c_bio.bp == NULL);
628 	KASSERT((xfer->c_flags & (C_WAITACT|C_FREE)) == 0);
629 
630 	/* Reset state, so that retries don't use stale info */
631 	if (__predict_false(xfer->c_retries > 0)) {
632 		xfer->c_flags = 0;
633 		memset(&xfer->c_bio, 0, sizeof(xfer->c_bio));
634 	}
635 
636 	xfer->c_bio.blkno = bp->b_rawblkno;
637 	xfer->c_bio.bcount = bp->b_bcount;
638 	xfer->c_bio.databuf = bp->b_data;
639 	xfer->c_bio.blkdone = 0;
640 	xfer->c_bio.bp = bp;
641 
642 #ifdef WD_CHAOS_MONKEY
643 	/*
644 	 * Override blkno to be over device capacity to trigger error,
645 	 * but only if it's read, to avoid trashing disk contents should
646 	 * the command be clipped, or otherwise misinterpreted, by the
647 	 * driver or controller.
648 	 */
649 	if (BUF_ISREAD(bp) && xfer->c_retries == 0 && wd->drv_chaos_freq > 0 &&
650 	    (++wd->drv_chaos_cnt % wd->drv_chaos_freq) == 0) {
651 		aprint_normal_dev(dksc->sc_dev, "%s: chaos xfer %d\n",
652 		    __func__, xfer->c_slot);
653 		xfer->c_bio.blkno = 7777777 + wd->sc_capacity;
654 		xfer->c_flags |= C_CHAOS;
655 	}
656 #endif
657 
658 	/*
659 	 * If we're retrying, retry in single-sector mode. This will give us
660 	 * the sector number of the problem, and will eventually allow the
661 	 * transfer to succeed. If FUA is requested, we can't actually
662 	 * do this, as ATA_SINGLE is usually executed as PIO transfer by drivers
663 	 * which support it, and that isn't compatible with NCQ/FUA.
664 	 */
665 	if (xfer->c_retries >= WDIORETRIES_SINGLE &&
666 	    (bp->b_flags & B_MEDIA_FUA) == 0)
667 		xfer->c_bio.flags = ATA_SINGLE;
668 	else
669 		xfer->c_bio.flags = 0;
670 
671 	/*
672 	 * request LBA48 transfers when supported by the controller
673 	 * and needed by transfer offset or size.
674 	 */
675 	if (wd->sc_flags & WDF_LBA48 &&
676 	    (((xfer->c_bio.blkno +
677 	     xfer->c_bio.bcount / dksc->sc_dkdev.dk_geom.dg_secsize) >
678 	    wd->sc_capacity28) ||
679 	    ((xfer->c_bio.bcount / dksc->sc_dkdev.dk_geom.dg_secsize) > 128)))
680 		xfer->c_bio.flags |= ATA_LBA48;
681 
682 	/*
683 	 * If NCQ was negotiated, always use it for the first several attempts.
684 	 * Since device cancels all outstanding requests on error, downgrade
685 	 * to non-NCQ on retry, so that the retried transfer would not cause
686 	 * cascade failure for the other transfers if it fails again.
687 	 * If FUA was requested, we can't downgrade, as that would violate
688 	 * the semantics - FUA would not be honored. In that case, continue
689 	 * retrying with NCQ.
690 	 */
691 	if (WD_USE_NCQ(wd) && (xfer->c_retries < WDIORETRIES_SINGLE ||
692 	    (bp->b_flags & B_MEDIA_FUA) != 0)) {
693 		xfer->c_bio.flags |= ATA_LBA48;
694 		xfer->c_flags |= C_NCQ;
695 
696 		if (WD_USE_NCQ_PRIO(wd) &&
697 		    BIO_GETPRIO(bp) == BPRIO_TIMECRITICAL)
698 			xfer->c_bio.flags |= ATA_PRIO_HIGH;
699 	}
700 
701 	if (wd->sc_flags & WDF_LBA)
702 		xfer->c_bio.flags |= ATA_LBA;
703 	if (bp->b_flags & B_READ)
704 		xfer->c_bio.flags |= ATA_READ;
705 	if (bp->b_flags & B_MEDIA_FUA) {
706 		/* If not using NCQ, the command WRITE DMA FUA EXT is LBA48 */
707 		KASSERT((wd->sc_flags & WDF_LBA48) != 0);
708 		if ((xfer->c_flags & C_NCQ) == 0)
709 			xfer->c_bio.flags |= ATA_LBA48;
710 
711 		xfer->c_bio.flags |= ATA_FUA;
712 	}
713 
714 	switch (wd->atabus->ata_bio(wd->drvp, xfer)) {
715 	case ATACMD_TRY_AGAIN:
716 		panic("wdstart1: try again");
717 		break;
718 	case ATACMD_QUEUED:
719 	case ATACMD_COMPLETE:
720 		break;
721 	default:
722 		panic("wdstart1: bad return code from ata_bio()");
723 	}
724 }
725 
726 static int
727 wd_diskstart(device_t dev, struct buf *bp)
728 {
729 	struct wd_softc *wd = device_private(dev);
730 #ifdef ATADEBUG
731 	struct dk_softc *dksc = &wd->sc_dksc;
732 #endif
733 	struct ata_xfer *xfer;
734 
735 	mutex_enter(&wd->sc_lock);
736 
737 	xfer = ata_get_xfer_ext(wd->drvp->chnl_softc, 0,
738 	    WD_USE_NCQ(wd) ? WD_MAX_OPENINGS(wd) : 0);
739 	if (xfer == NULL) {
740 		ATADEBUG_PRINT(("wd_diskstart %s no xfer\n",
741 		    dksc->sc_xname), DEBUG_XFERS);
742 		mutex_exit(&wd->sc_lock);
743 		return EAGAIN;
744 	}
745 
746 	wdstart1(wd, bp, xfer);
747 
748 	mutex_exit(&wd->sc_lock);
749 
750 	return 0;
751 }
752 
753 /*
754  * Queue a drive for I/O.
755  */
756 static void
757 wdstart(device_t self)
758 {
759 	struct wd_softc *wd = device_private(self);
760 	struct dk_softc *dksc = &wd->sc_dksc;
761 
762 	ATADEBUG_PRINT(("wdstart %s\n", dksc->sc_xname),
763 	    DEBUG_XFERS);
764 
765 	if (!device_is_active(dksc->sc_dev))
766 		return;
767 
768 	/*
769 	 * Do not queue any transfers until flush is finished, so that
770 	 * once flush is pending, it will get handled as soon as xfer
771 	 * is available.
772 	 */
773 	if (ISSET(wd->sc_flags, WDF_FLUSH_PEND)) {
774 		ATADEBUG_PRINT(("wdstart %s flush pend\n",
775 		    dksc->sc_xname), DEBUG_XFERS);
776 		return;
777 	}
778 
779 	dk_start(dksc, NULL);
780 }
781 
782 static void
783 wddone(device_t self, struct ata_xfer *xfer)
784 {
785 	struct wd_softc *wd = device_private(self);
786 	struct dk_softc *dksc = &wd->sc_dksc;
787 	const char *errmsg;
788 	int do_perror = 0;
789 	struct buf *bp;
790 
791 	ATADEBUG_PRINT(("wddone %s\n", dksc->sc_xname),
792 	    DEBUG_XFERS);
793 
794 	if (__predict_false(wddoingadump)) {
795 		/* just drop it to the floor */
796 		ata_free_xfer(wd->drvp->chnl_softc, xfer);
797 		return;
798 	}
799 
800 	bp = xfer->c_bio.bp;
801 	KASSERT(bp != NULL);
802 
803 	bp->b_resid = xfer->c_bio.bcount;
804 	switch (xfer->c_bio.error) {
805 	case ERR_DMA:
806 		errmsg = "DMA error";
807 		goto retry;
808 	case ERR_DF:
809 		errmsg = "device fault";
810 		goto retry;
811 	case TIMEOUT:
812 		errmsg = "device timeout";
813 		goto retry;
814 	case REQUEUE:
815 		errmsg = "requeue";
816 		goto retry2;
817 	case ERR_RESET:
818 		errmsg = "channel reset";
819 		goto retry2;
820 	case ERROR:
821 		/* Don't care about media change bits */
822 		if (xfer->c_bio.r_error != 0 &&
823 		    (xfer->c_bio.r_error & ~(WDCE_MC | WDCE_MCR)) == 0)
824 			goto noerror;
825 		errmsg = "error";
826 		do_perror = 1;
827 retry:		/* Just reset and retry. Can we do more ? */
828 		if ((xfer->c_flags & C_RECOVERED) == 0) {
829 			int wflags = (xfer->c_flags & C_POLL) ? AT_POLL : 0;
830 			(*wd->atabus->ata_reset_drive)(wd->drvp, wflags, NULL);
831 		}
832 retry2:
833 		mutex_enter(&wd->sc_lock);
834 
835 		diskerr(bp, "wd", errmsg, LOG_PRINTF,
836 		    xfer->c_bio.blkdone, dksc->sc_dkdev.dk_label);
837 		if (xfer->c_retries < WDIORETRIES)
838 			printf(", slot %d, retry %d", xfer->c_slot,
839 			    xfer->c_retries + 1);
840 		printf("\n");
841 		if (do_perror)
842 			wdperror(wd, xfer);
843 
844 		if (xfer->c_retries < WDIORETRIES) {
845 			xfer->c_retries++;
846 
847 			/* Rerun ASAP if just requeued */
848 			callout_reset(&xfer->c_retry_callout,
849 			    (xfer->c_bio.error == REQUEUE) ? 1 : RECOVERYTIME,
850 			    wdbiorestart, xfer);
851 
852 			mutex_exit(&wd->sc_lock);
853 			return;
854 		}
855 
856 		mutex_exit(&wd->sc_lock);
857 
858 #ifdef WD_SOFTBADSECT
859 		/*
860 		 * Not all errors indicate a failed block but those that do,
861 		 * put the block on the bad-block list for the device.  Only
862 		 * do this for reads because the drive should do it for writes,
863 		 * itself, according to Manuel.
864 		 */
865 		if ((bp->b_flags & B_READ) &&
866 		    ((wd->drvp->ata_vers >= 4 && xfer->c_bio.r_error & 64) ||
867 		     (wd->drvp->ata_vers < 4 && xfer->c_bio.r_error & 192))) {
868 			struct disk_badsectors *dbs;
869 
870 			dbs = malloc(sizeof *dbs, M_TEMP, M_NOWAIT);
871 			if (dbs == NULL) {
872 				aprint_error_dev(dksc->sc_dev,
873 				    "failed to add bad block to list\n");
874 				goto out;
875 			}
876 
877 			dbs->dbs_min = bp->b_rawblkno;
878 			dbs->dbs_max = dbs->dbs_min +
879 			    (bp->b_bcount /wd->sc_blksize) - 1;
880 			microtime(&dbs->dbs_failedat);
881 
882 			mutex_enter(&wd->sc_lock);
883 			SLIST_INSERT_HEAD(&wd->sc_bslist, dbs, dbs_next);
884 			wd->sc_bscount++;
885 			mutex_exit(&wd->sc_lock);
886 		}
887 out:
888 #endif
889 		bp->b_error = EIO;
890 		break;
891 	case NOERROR:
892 noerror:	if ((xfer->c_bio.flags & ATA_CORR) || xfer->c_retries > 0)
893 			aprint_error_dev(dksc->sc_dev,
894 			    "soft error (corrected) slot %d\n", xfer->c_slot);
895 #ifdef WD_CHAOS_MONKEY
896 		KASSERT((xfer->c_flags & C_CHAOS) == 0);
897 #endif
898 		break;
899 	case ERR_NODEV:
900 		bp->b_error = EIO;
901 		break;
902 	}
903 	if (__predict_false(bp->b_error != 0) && bp->b_resid == 0) {
904 		/*
905 		 * the disk or controller sometimes report a complete
906 		 * xfer, when there has been an error. This is wrong,
907 		 * assume nothing got transfered in this case
908 		 */
909 		bp->b_resid = bp->b_bcount;
910 	}
911 
912 	ata_free_xfer(wd->drvp->chnl_softc, xfer);
913 
914 	dk_done(dksc, bp);
915 	ata_channel_start(wd->drvp->chnl_softc, wd->drvp->drive);
916 }
917 
918 static void
919 wdbiorestart(void *v)
920 {
921 	struct ata_xfer *xfer = v;
922 	struct buf *bp = xfer->c_bio.bp;
923 	struct wd_softc *wd = device_lookup_private(&wd_cd, WDUNIT(bp->b_dev));
924 #ifdef ATADEBUG
925 	struct dk_softc *dksc = &wd->sc_dksc;
926 #endif
927 
928 	ATADEBUG_PRINT(("wdbiorestart %s\n", dksc->sc_xname),
929 	    DEBUG_XFERS);
930 
931 	mutex_enter(&wd->sc_lock);
932 	wdstart1(wd, bp, xfer);
933 	mutex_exit(&wd->sc_lock);
934 }
935 
936 static void
937 wdminphys(struct buf *bp)
938 {
939 	const struct wd_softc * const wd =
940 	    device_lookup_private(&wd_cd, WDUNIT(bp->b_dev));
941 	int maxsectors;
942 
943 	/*
944 	 * The limit is actually 65536 for LBA48 and 256 for non-LBA48,
945 	 * but that requires to set the count for the ATA command
946 	 * to 0, which is somewhat error prone, so better stay safe.
947 	 */
948 	if (wd->sc_flags & WDF_LBA48)
949 		maxsectors = 65535;
950 	else
951 		maxsectors = 128;
952 
953 	if (bp->b_bcount > (wd->sc_blksize * maxsectors))
954 		bp->b_bcount = (wd->sc_blksize * maxsectors);
955 
956 	minphys(bp);
957 }
958 
959 static void
960 wd_iosize(device_t dev, int *count)
961 {
962 	struct buf B;
963 	int bmaj;
964 
965 	bmaj       = bdevsw_lookup_major(&wd_bdevsw);
966 	B.b_dev    = MAKEWDDEV(bmaj,device_unit(dev),RAW_PART);
967 	B.b_bcount = *count;
968 
969 	wdminphys(&B);
970 
971 	*count = B.b_bcount;
972 }
973 
974 static int
975 wdread(dev_t dev, struct uio *uio, int flags)
976 {
977 
978 	ATADEBUG_PRINT(("wdread\n"), DEBUG_XFERS);
979 	return (physio(wdstrategy, NULL, dev, B_READ, wdminphys, uio));
980 }
981 
982 static int
983 wdwrite(dev_t dev, struct uio *uio, int flags)
984 {
985 
986 	ATADEBUG_PRINT(("wdwrite\n"), DEBUG_XFERS);
987 	return (physio(wdstrategy, NULL, dev, B_WRITE, wdminphys, uio));
988 }
989 
990 static int
991 wdopen(dev_t dev, int flag, int fmt, struct lwp *l)
992 {
993 	struct wd_softc *wd;
994 	struct dk_softc *dksc;
995 	int unit, part, error;
996 
997 	ATADEBUG_PRINT(("wdopen\n"), DEBUG_FUNCS);
998 	unit = WDUNIT(dev);
999 	wd = device_lookup_private(&wd_cd, unit);
1000 	if (wd == NULL)
1001 		return (ENXIO);
1002 	dksc = &wd->sc_dksc;
1003 
1004 	if (! device_is_active(dksc->sc_dev))
1005 		return (ENODEV);
1006 
1007 	part = WDPART(dev);
1008 
1009 	if (wd->sc_capacity == 0)
1010 		return (ENODEV);
1011 
1012 	/*
1013 	 * If any partition is open, but the disk has been invalidated,
1014 	 * disallow further opens.
1015 	 */
1016 	if ((wd->sc_flags & (WDF_OPEN | WDF_LOADED)) == WDF_OPEN) {
1017 		if (part != RAW_PART || fmt != S_IFCHR)
1018 			return EIO;
1019 	}
1020 
1021 	error = dk_open(dksc, dev, flag, fmt, l);
1022 
1023 	return error;
1024 }
1025 
1026 /*
1027  * Serialized by caller
1028  */
1029 static int
1030 wd_firstopen(device_t self, dev_t dev, int flag, int fmt)
1031 {
1032 	struct wd_softc *wd = device_private(self);
1033 	struct dk_softc *dksc = &wd->sc_dksc;
1034 	int error;
1035 
1036 	error = wd->atabus->ata_addref(wd->drvp);
1037 	if (error)
1038 		return error;
1039 
1040 	if ((wd->sc_flags & WDF_LOADED) == 0) {
1041 		int param_error;
1042 
1043 		/* Load the physical device parameters. */
1044 		param_error = wd_get_params(wd, AT_WAIT, &wd->sc_params);
1045 		if (param_error != 0) {
1046 			aprint_error_dev(dksc->sc_dev, "IDENTIFY failed\n");
1047 			error = EIO;
1048 			goto bad;
1049 		}
1050 		wd_set_geometry(wd);
1051 		wd->sc_flags |= WDF_LOADED;
1052 	}
1053 
1054 	wd->sc_flags |= WDF_OPEN;
1055 	return 0;
1056 
1057 bad:
1058 	wd->atabus->ata_delref(wd->drvp);
1059 	return error;
1060 }
1061 
1062 /*
1063  * Caller must hold wd->sc_dk.dk_openlock.
1064  */
1065 static int
1066 wd_lastclose(device_t self)
1067 {
1068 	struct wd_softc *wd = device_private(self);
1069 
1070 	wd_flushcache(wd, AT_WAIT, false);
1071 
1072 	wd->atabus->ata_delref(wd->drvp);
1073 	wd->sc_flags &= ~WDF_OPEN;
1074 
1075 	return 0;
1076 }
1077 
1078 static int
1079 wdclose(dev_t dev, int flag, int fmt, struct lwp *l)
1080 {
1081 	struct wd_softc *wd;
1082 	struct dk_softc *dksc;
1083 	int unit;
1084 
1085 	unit = WDUNIT(dev);
1086 	wd = device_lookup_private(&wd_cd, unit);
1087 	dksc = &wd->sc_dksc;
1088 
1089 	return dk_close(dksc, dev, flag, fmt, l);
1090 }
1091 
1092 void
1093 wdperror(const struct wd_softc *wd, struct ata_xfer *xfer)
1094 {
1095 	static const char *const errstr0_3[] = {"address mark not found",
1096 	    "track 0 not found", "aborted command", "media change requested",
1097 	    "id not found", "media changed", "uncorrectable data error",
1098 	    "bad block detected"};
1099 	static const char *const errstr4_5[] = {
1100 	    "obsolete (address mark not found)",
1101 	    "no media/write protected", "aborted command",
1102 	    "media change requested", "id not found", "media changed",
1103 	    "uncorrectable data error", "interface CRC error"};
1104 	const char *const *errstr;
1105 	int i;
1106 	const char *sep = "";
1107 
1108 	const struct dk_softc *dksc = &wd->sc_dksc;
1109 	const char *devname = dksc->sc_xname;
1110 	struct ata_drive_datas *drvp = wd->drvp;
1111 	int errno = xfer->c_bio.r_error;
1112 
1113 	if (drvp->ata_vers >= 4)
1114 		errstr = errstr4_5;
1115 	else
1116 		errstr = errstr0_3;
1117 
1118 	printf("%s: (", devname);
1119 
1120 	if (errno == 0)
1121 		printf("error not notified");
1122 
1123 	for (i = 0; i < 8; i++) {
1124 		if (errno & (1 << i)) {
1125 			printf("%s%s", sep, errstr[i]);
1126 			sep = ", ";
1127 		}
1128 	}
1129 	printf(")\n");
1130 }
1131 
1132 int
1133 wdioctl(dev_t dev, u_long cmd, void *addr, int flag, struct lwp *l)
1134 {
1135 	struct wd_softc *wd =
1136 	    device_lookup_private(&wd_cd, WDUNIT(dev));
1137 	struct dk_softc *dksc = &wd->sc_dksc;
1138 
1139 	ATADEBUG_PRINT(("wdioctl\n"), DEBUG_FUNCS);
1140 
1141 	if ((wd->sc_flags & WDF_LOADED) == 0)
1142 		return EIO;
1143 
1144 	switch (cmd) {
1145 #ifdef HAS_BAD144_HANDLING
1146 	case DIOCSBAD:
1147 		if ((flag & FWRITE) == 0)
1148 			return EBADF;
1149 		dksc->sc_dkdev.dk_cpulabel->bad = *(struct dkbad *)addr;
1150 		dksc->sc_dkdev.dk_label->d_flags |= D_BADSECT;
1151 		bad144intern(wd);
1152 		return 0;
1153 #endif
1154 #ifdef WD_SOFTBADSECT
1155 	case DIOCBSLIST :
1156 	{
1157 		uint32_t count, missing, skip;
1158 		struct disk_badsecinfo dbsi;
1159 		struct disk_badsectors *dbs;
1160 		size_t available;
1161 		uint8_t *laddr;
1162 
1163 		dbsi = *(struct disk_badsecinfo *)addr;
1164 		missing = wd->sc_bscount;
1165 		count = 0;
1166 		available = dbsi.dbsi_bufsize;
1167 		skip = dbsi.dbsi_skip;
1168 		laddr = (uint8_t *)dbsi.dbsi_buffer;
1169 
1170 		/*
1171 		 * We start this loop with the expectation that all of the
1172 		 * entries will be missed and decrement this counter each
1173 		 * time we either skip over one (already copied out) or
1174 		 * we actually copy it back to user space.  The structs
1175 		 * holding the bad sector information are copied directly
1176 		 * back to user space whilst the summary is returned via
1177 		 * the struct passed in via the ioctl.
1178 		 */
1179 		SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next) {
1180 			if (skip > 0) {
1181 				missing--;
1182 				skip--;
1183 				continue;
1184 			}
1185 			if (available < sizeof(*dbs))
1186 				break;
1187 			available -= sizeof(*dbs);
1188 			copyout(dbs, laddr, sizeof(*dbs));
1189 			laddr += sizeof(*dbs);
1190 			missing--;
1191 			count++;
1192 		}
1193 		dbsi.dbsi_left = missing;
1194 		dbsi.dbsi_copied = count;
1195 		*(struct disk_badsecinfo *)addr = dbsi;
1196 		return 0;
1197 	}
1198 
1199 	case DIOCBSFLUSH :
1200 		/* Clean out the bad sector list */
1201 		while (!SLIST_EMPTY(&wd->sc_bslist)) {
1202 			void *head = SLIST_FIRST(&wd->sc_bslist);
1203 			SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
1204 			free(head, M_TEMP);
1205 		}
1206 		wd->sc_bscount = 0;
1207 		return 0;
1208 #endif
1209 
1210 #ifdef notyet
1211 	case DIOCWFORMAT:
1212 		if ((flag & FWRITE) == 0)
1213 			return EBADF;
1214 		{
1215 		register struct format_op *fop;
1216 		struct iovec aiov;
1217 		struct uio auio;
1218 		int error1;
1219 
1220 		fop = (struct format_op *)addr;
1221 		aiov.iov_base = fop->df_buf;
1222 		aiov.iov_len = fop->df_count;
1223 		auio.uio_iov = &aiov;
1224 		auio.uio_iovcnt = 1;
1225 		auio.uio_resid = fop->df_count;
1226 		auio.uio_offset =
1227 			fop->df_startblk * wd->sc_dk.dk_label->d_secsize;
1228 		auio.uio_vmspace = l->l_proc->p_vmspace;
1229 		error1 = physio(wdformat, NULL, dev, B_WRITE, wdminphys,
1230 		    &auio);
1231 		fop->df_count -= auio.uio_resid;
1232 		fop->df_reg[0] = wdc->sc_status;
1233 		fop->df_reg[1] = wdc->sc_error;
1234 		return error1;
1235 		}
1236 #endif
1237 	case DIOCGCACHE:
1238 		return wd_getcache(wd, (int *)addr);
1239 
1240 	case DIOCSCACHE:
1241 		return wd_setcache(wd, *(int *)addr);
1242 
1243 	case DIOCCACHESYNC:
1244 		return wd_flushcache(wd, AT_WAIT, true);
1245 
1246 	case ATAIOCCOMMAND:
1247 		/*
1248 		 * Make sure this command is (relatively) safe first
1249 		 */
1250 		if ((((atareq_t *) addr)->flags & ATACMD_READ) == 0 &&
1251 		    (flag & FWRITE) == 0)
1252 			return (EBADF);
1253 		{
1254 		struct wd_ioctl *wi;
1255 		atareq_t *atareq = (atareq_t *) addr;
1256 		int error1;
1257 
1258 		wi = wi_get(wd);
1259 		wi->wi_atareq = *atareq;
1260 
1261 		if (atareq->datalen && atareq->flags &
1262 		    (ATACMD_READ | ATACMD_WRITE)) {
1263 			void *tbuf;
1264 			if (atareq->datalen < DEV_BSIZE
1265 			    && atareq->command == WDCC_IDENTIFY) {
1266 				tbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1267 				wi->wi_iov.iov_base = tbuf;
1268 				wi->wi_iov.iov_len = DEV_BSIZE;
1269 				UIO_SETUP_SYSSPACE(&wi->wi_uio);
1270 			} else {
1271 				tbuf = NULL;
1272 				wi->wi_iov.iov_base = atareq->databuf;
1273 				wi->wi_iov.iov_len = atareq->datalen;
1274 				wi->wi_uio.uio_vmspace = l->l_proc->p_vmspace;
1275 			}
1276 			wi->wi_uio.uio_iov = &wi->wi_iov;
1277 			wi->wi_uio.uio_iovcnt = 1;
1278 			wi->wi_uio.uio_resid = atareq->datalen;
1279 			wi->wi_uio.uio_offset = 0;
1280 			wi->wi_uio.uio_rw =
1281 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE;
1282 			error1 = physio(wdioctlstrategy, &wi->wi_bp, dev,
1283 			    (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE,
1284 			    wdminphys, &wi->wi_uio);
1285 			if (tbuf != NULL && error1 == 0) {
1286 				error1 = copyout(tbuf, atareq->databuf,
1287 				    atareq->datalen);
1288 				free(tbuf, M_TEMP);
1289 			}
1290 		} else {
1291 			/* No need to call physio if we don't have any
1292 			   user data */
1293 			wi->wi_bp.b_flags = 0;
1294 			wi->wi_bp.b_data = 0;
1295 			wi->wi_bp.b_bcount = 0;
1296 			wi->wi_bp.b_dev = dev;
1297 			wi->wi_bp.b_proc = l->l_proc;
1298 			wdioctlstrategy(&wi->wi_bp);
1299 			error1 = wi->wi_bp.b_error;
1300 		}
1301 		*atareq = wi->wi_atareq;
1302 		wi_free(wi);
1303 		return(error1);
1304 		}
1305 
1306 	default:
1307 		return dk_ioctl(dksc, dev, cmd, addr, flag, l);
1308 	}
1309 
1310 #ifdef DIAGNOSTIC
1311 	panic("wdioctl: impossible");
1312 #endif
1313 }
1314 
1315 static int
1316 wd_discard(device_t dev, off_t pos, off_t len)
1317 {
1318 	struct wd_softc *wd = device_private(dev);
1319 	daddr_t bno;
1320 	long size, done;
1321 	long maxatonce, amount;
1322 	int result;
1323 
1324 	if (!(wd->sc_params.atap_ata_major & WDC_VER_ATA7)
1325 	    || !(wd->sc_params.support_dsm & ATA_SUPPORT_DSM_TRIM)) {
1326 		/* not supported; ignore request */
1327 		ATADEBUG_PRINT(("wddiscard (unsupported)\n"), DEBUG_FUNCS);
1328 		return 0;
1329 	}
1330 	maxatonce = 0xffff; /*wd->sc_params.max_dsm_blocks*/
1331 
1332 	ATADEBUG_PRINT(("wddiscard\n"), DEBUG_FUNCS);
1333 
1334 	if ((wd->sc_flags & WDF_LOADED) == 0)
1335 		return EIO;
1336 
1337 	/* round the start up and the end down */
1338 	bno = (pos + wd->sc_blksize - 1) / wd->sc_blksize;
1339 	size = ((pos + len) / wd->sc_blksize) - bno;
1340 
1341 	done = 0;
1342 	while (done < size) {
1343 	     amount = size - done;
1344 	     if (amount > maxatonce) {
1345 		     amount = maxatonce;
1346 	     }
1347 	     result = wd_trim(wd, bno + done, amount);
1348 	     if (result) {
1349 		     return result;
1350 	     }
1351 	     done += amount;
1352 	}
1353 	return 0;
1354 }
1355 
1356 static int
1357 wddiscard(dev_t dev, off_t pos, off_t len)
1358 {
1359 	struct wd_softc *wd;
1360 	struct dk_softc *dksc;
1361 	int unit;
1362 
1363 	unit = WDUNIT(dev);
1364 	wd = device_lookup_private(&wd_cd, unit);
1365 	dksc = &wd->sc_dksc;
1366 
1367 	return dk_discard(dksc, dev, pos, len);
1368 }
1369 
1370 #ifdef B_FORMAT
1371 int
1372 wdformat(struct buf *bp)
1373 {
1374 
1375 	bp->b_flags |= B_FORMAT;
1376 	return wdstrategy(bp);
1377 }
1378 #endif
1379 
1380 int
1381 wdsize(dev_t dev)
1382 {
1383 	struct wd_softc *wd;
1384 	struct dk_softc *dksc;
1385 	int unit;
1386 
1387 	ATADEBUG_PRINT(("wdsize\n"), DEBUG_FUNCS);
1388 
1389 	unit = WDUNIT(dev);
1390 	wd = device_lookup_private(&wd_cd, unit);
1391 	if (wd == NULL)
1392 		return (-1);
1393 	dksc = &wd->sc_dksc;
1394 
1395 	if (!device_is_active(dksc->sc_dev))
1396 		return (-1);
1397 
1398 	return dk_size(dksc, dev);
1399 }
1400 
1401 /*
1402  * Dump core after a system crash.
1403  */
1404 static int
1405 wddump(dev_t dev, daddr_t blkno, void *va, size_t size)
1406 {
1407 	struct wd_softc *wd;
1408 	struct dk_softc *dksc;
1409 	int unit;
1410 
1411 	/* Check if recursive dump; if so, punt. */
1412 	if (wddoingadump)
1413 		return EFAULT;
1414 	wddoingadump = 1;
1415 
1416 	unit = WDUNIT(dev);
1417 	wd = device_lookup_private(&wd_cd, unit);
1418 	if (wd == NULL)
1419 		return (ENXIO);
1420 	dksc = &wd->sc_dksc;
1421 
1422 	return dk_dump(dksc, dev, blkno, va, size);
1423 }
1424 
1425 static int
1426 wd_dumpblocks(device_t dev, void *va, daddr_t blkno, int nblk)
1427 {
1428 	struct wd_softc *wd = device_private(dev);
1429 	struct dk_softc *dksc = &wd->sc_dksc;
1430 	struct disk_geom *dg = &dksc->sc_dkdev.dk_geom;
1431 	struct ata_xfer *xfer;
1432 	int err;
1433 
1434 	/* Recalibrate, if first dump transfer. */
1435 	if (wddumprecalibrated == 0) {
1436 		wddumprecalibrated = 1;
1437 		(*wd->atabus->ata_reset_drive)(wd->drvp,
1438 					       AT_POLL | AT_RST_EMERG, NULL);
1439 		wd->drvp->state = RESET;
1440 	}
1441 
1442 	xfer = ata_get_xfer_ext(wd->drvp->chnl_softc, 0, 0);
1443 	if (xfer == NULL) {
1444 		printf("%s: no xfer\n", __func__);
1445 		return EAGAIN;
1446 	}
1447 
1448 	xfer->c_bio.blkno = blkno;
1449 	xfer->c_bio.flags = ATA_POLL;
1450 	if (wd->sc_flags & WDF_LBA48 &&
1451 	    (xfer->c_bio.blkno + nblk) > wd->sc_capacity28)
1452 		xfer->c_bio.flags |= ATA_LBA48;
1453 	if (wd->sc_flags & WDF_LBA)
1454 		xfer->c_bio.flags |= ATA_LBA;
1455 	xfer->c_bio.bcount = nblk * dg->dg_secsize;
1456 	xfer->c_bio.databuf = va;
1457 #ifndef WD_DUMP_NOT_TRUSTED
1458 	switch (err = wd->atabus->ata_bio(wd->drvp, xfer)) {
1459 	case ATACMD_TRY_AGAIN:
1460 		panic("wddump: try again");
1461 		break;
1462 	case ATACMD_QUEUED:
1463 		panic("wddump: polled command has been queued");
1464 		break;
1465 	case ATACMD_COMPLETE:
1466 		break;
1467 	default:
1468 		panic("wddump: unknown atacmd code %d", err);
1469 	}
1470 	switch(err = xfer->c_bio.error) {
1471 	case TIMEOUT:
1472 		printf("wddump: device timed out");
1473 		err = EIO;
1474 		break;
1475 	case ERR_DF:
1476 		printf("wddump: drive fault");
1477 		err = EIO;
1478 		break;
1479 	case ERR_DMA:
1480 		printf("wddump: DMA error");
1481 		err = EIO;
1482 		break;
1483 	case ERROR:
1484 		printf("wddump: ");
1485 		wdperror(wd, xfer);
1486 		err = EIO;
1487 		break;
1488 	case NOERROR:
1489 		err = 0;
1490 		break;
1491 	default:
1492 		panic("wddump: unknown error type %d", err);
1493 	}
1494 
1495 	if (err != 0) {
1496 		printf("\n");
1497 		return err;
1498 	}
1499 #else	/* WD_DUMP_NOT_TRUSTED */
1500 	/* Let's just talk about this first... */
1501 	printf("wd%d: dump addr 0x%x, cylin %d, head %d, sector %d\n",
1502 	    unit, va, cylin, head, sector);
1503 	delay(500 * 1000);	/* half a second */
1504 #endif
1505 
1506 	wddoingadump = 0;
1507 	return 0;
1508 }
1509 
1510 #ifdef HAS_BAD144_HANDLING
1511 /*
1512  * Internalize the bad sector table.
1513  */
1514 void
1515 bad144intern(struct wd_softc *wd)
1516 {
1517 	struct dk_softc *dksc = &wd->sc_dksc;
1518 	struct dkbad *bt = &dksc->sc_dkdev.dk_cpulabel->bad;
1519 	struct disklabel *lp = dksc->sc_dkdev.dk_label;
1520 	int i = 0;
1521 
1522 	ATADEBUG_PRINT(("bad144intern\n"), DEBUG_XFERS);
1523 
1524 	for (; i < NBT_BAD; i++) {
1525 		if (bt->bt_bad[i].bt_cyl == 0xffff)
1526 			break;
1527 		wd->drvp->badsect[i] =
1528 		    bt->bt_bad[i].bt_cyl * lp->d_secpercyl +
1529 		    (bt->bt_bad[i].bt_trksec >> 8) * lp->d_nsectors +
1530 		    (bt->bt_bad[i].bt_trksec & 0xff);
1531 	}
1532 	for (; i < NBT_BAD+1; i++)
1533 		wd->drvp->badsect[i] = -1;
1534 }
1535 #endif
1536 
1537 static void
1538 wd_set_geometry(struct wd_softc *wd)
1539 {
1540 	struct dk_softc *dksc = &wd->sc_dksc;
1541 	struct disk_geom *dg = &dksc->sc_dkdev.dk_geom;
1542 
1543 	memset(dg, 0, sizeof(*dg));
1544 
1545 	dg->dg_secperunit = wd->sc_capacity;
1546 	dg->dg_secsize = wd->sc_blksize;
1547 	dg->dg_nsectors = wd->sc_params.atap_sectors;
1548 	dg->dg_ntracks = wd->sc_params.atap_heads;
1549 	if ((wd->sc_flags & WDF_LBA) == 0)
1550 		dg->dg_ncylinders = wd->sc_params.atap_cylinders;
1551 
1552 	disk_set_info(dksc->sc_dev, &dksc->sc_dkdev, NULL);
1553 }
1554 
1555 int
1556 wd_get_params(struct wd_softc *wd, uint8_t flags, struct ataparams *params)
1557 {
1558 
1559 	switch (wd->atabus->ata_get_params(wd->drvp, flags, params)) {
1560 	case CMD_AGAIN:
1561 		return 1;
1562 	case CMD_ERR:
1563 		if (wd->drvp->drive_type != ATA_DRIVET_OLD)
1564 			return 1;
1565 		/*
1566 		 * We `know' there's a drive here; just assume it's old.
1567 		 * This geometry is only used to read the MBR and print a
1568 		 * (false) attach message.
1569 		 */
1570 		strncpy(params->atap_model, "ST506",
1571 		    sizeof params->atap_model);
1572 		params->atap_config = ATA_CFG_FIXED;
1573 		params->atap_cylinders = 1024;
1574 		params->atap_heads = 8;
1575 		params->atap_sectors = 17;
1576 		params->atap_multi = 1;
1577 		params->atap_capabilities1 = params->atap_capabilities2 = 0;
1578 		wd->drvp->ata_vers = -1; /* Mark it as pre-ATA */
1579 		/* FALLTHROUGH */
1580 	case CMD_OK:
1581 		return 0;
1582 	default:
1583 		panic("wd_get_params: bad return code from ata_get_params");
1584 		/* NOTREACHED */
1585 	}
1586 }
1587 
1588 int
1589 wd_getcache(struct wd_softc *wd, int *bitsp)
1590 {
1591 	struct ataparams params;
1592 
1593 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
1594 		return EIO;
1595 	if (params.atap_cmd_set1 == 0x0000 ||
1596 	    params.atap_cmd_set1 == 0xffff ||
1597 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0) {
1598 		*bitsp = 0;
1599 		return 0;
1600 	}
1601 	*bitsp = DKCACHE_WCHANGE | DKCACHE_READ;
1602 	if (params.atap_cmd1_en & WDC_CMD1_CACHE)
1603 		*bitsp |= DKCACHE_WRITE;
1604 
1605 	if (WD_USE_NCQ(wd) || (wd->drvp->drive_flags & ATA_DRIVE_WFUA))
1606 		*bitsp |= DKCACHE_FUA;
1607 
1608 	return 0;
1609 }
1610 
1611 const char at_errbits[] = "\20\10ERROR\11TIMEOU\12DF";
1612 
1613 int
1614 wd_setcache(struct wd_softc *wd, int bits)
1615 {
1616 	struct dk_softc *dksc = &wd->sc_dksc;
1617 	struct ataparams params;
1618 	struct ata_xfer *xfer;
1619 	int error;
1620 
1621 	if (wd_get_params(wd, AT_WAIT, &params) != 0)
1622 		return EIO;
1623 
1624 	if (params.atap_cmd_set1 == 0x0000 ||
1625 	    params.atap_cmd_set1 == 0xffff ||
1626 	    (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0)
1627 		return EOPNOTSUPP;
1628 
1629 	if ((bits & DKCACHE_READ) == 0 ||
1630 	    (bits & DKCACHE_SAVE) != 0)
1631 		return EOPNOTSUPP;
1632 
1633 	xfer = ata_get_xfer(wd->drvp->chnl_softc);
1634 	if (xfer == NULL)
1635 		return EINTR;
1636 
1637 	xfer->c_ata_c.r_command = SET_FEATURES;
1638 	xfer->c_ata_c.r_st_bmask = 0;
1639 	xfer->c_ata_c.r_st_pmask = 0;
1640 	xfer->c_ata_c.timeout = 30000; /* 30s timeout */
1641 	xfer->c_ata_c.flags = AT_WAIT;
1642 	if (bits & DKCACHE_WRITE)
1643 		xfer->c_ata_c.r_features = WDSF_WRITE_CACHE_EN;
1644 	else
1645 		xfer->c_ata_c.r_features = WDSF_WRITE_CACHE_DS;
1646 	if (wd->atabus->ata_exec_command(wd->drvp, xfer) != ATACMD_COMPLETE) {
1647 		aprint_error_dev(dksc->sc_dev,
1648 		    "wd_setcache command not complete\n");
1649 		error = EIO;
1650 		goto out;
1651 	}
1652 
1653 	if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1654 		char sbuf[sizeof(at_errbits) + 64];
1655 		snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
1656 		aprint_error_dev(dksc->sc_dev, "wd_setcache: status=%s\n", sbuf);
1657 		error = EIO;
1658 		goto out;
1659 	}
1660 
1661 	error = 0;
1662 
1663 out:
1664 	ata_free_xfer(wd->drvp->chnl_softc, xfer);
1665 	ata_channel_start(wd->drvp->chnl_softc, wd->drvp->drive);
1666 	return error;
1667 }
1668 
1669 static int
1670 wd_standby(struct wd_softc *wd, int flags)
1671 {
1672 	struct dk_softc *dksc = &wd->sc_dksc;
1673 	struct ata_xfer *xfer;
1674 	int error;
1675 
1676 	xfer = ata_get_xfer(wd->drvp->chnl_softc);
1677 	if (xfer == NULL)
1678 		return EINTR;
1679 
1680 	xfer->c_ata_c.r_command = WDCC_STANDBY_IMMED;
1681 	xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
1682 	xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
1683 	xfer->c_ata_c.flags = flags;
1684 	xfer->c_ata_c.timeout = 30000; /* 30s timeout */
1685 	if (wd->atabus->ata_exec_command(wd->drvp, xfer) != ATACMD_COMPLETE) {
1686 		aprint_error_dev(dksc->sc_dev,
1687 		    "standby immediate command didn't complete\n");
1688 		error = EIO;
1689 		goto out;
1690 	}
1691 	if (xfer->c_ata_c.flags & AT_ERROR) {
1692 		if (xfer->c_ata_c.r_error == WDCE_ABRT) {
1693 			/* command not supported */
1694 			error = ENODEV;
1695 			goto out;
1696 		}
1697 	}
1698 	if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1699 		char sbuf[sizeof(at_errbits) + 64];
1700 		snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
1701 		aprint_error_dev(dksc->sc_dev, "wd_standby: status=%s\n", sbuf);
1702 		error = EIO;
1703 		goto out;
1704 	}
1705 	error = 0;
1706 
1707 out:
1708 	ata_free_xfer(wd->drvp->chnl_softc, xfer);
1709 	/* drive is supposed to go idle, do not call ata_channel_start() */
1710 	return error;
1711 }
1712 
1713 int
1714 wd_flushcache(struct wd_softc *wd, int flags, bool start)
1715 {
1716 	struct dk_softc *dksc = &wd->sc_dksc;
1717 	struct ata_xfer *xfer;
1718 	int error;
1719 
1720 	/*
1721 	 * WDCC_FLUSHCACHE is here since ATA-4, but some drives report
1722 	 * only ATA-2 and still support it.
1723 	 */
1724 	if (wd->drvp->ata_vers < 4 &&
1725 	    ((wd->sc_params.atap_cmd_set2 & WDC_CMD2_FC) == 0 ||
1726 	    wd->sc_params.atap_cmd_set2 == 0xffff))
1727 		return ENODEV;
1728 
1729 	mutex_enter(&wd->sc_lock);
1730 	SET(wd->sc_flags, WDF_FLUSH_PEND);
1731 	mutex_exit(&wd->sc_lock);
1732 
1733 	xfer = ata_get_xfer(wd->drvp->chnl_softc);
1734 
1735 	mutex_enter(&wd->sc_lock);
1736 	CLR(wd->sc_flags, WDF_FLUSH_PEND);
1737 	mutex_exit(&wd->sc_lock);
1738 
1739 	if (xfer == NULL) {
1740 		error = EINTR;
1741 		goto out;
1742 	}
1743 
1744 	if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0 &&
1745 	    (wd->sc_params.atap_cmd2_en & ATA_CMD2_FCE) != 0) {
1746 		xfer->c_ata_c.r_command = WDCC_FLUSHCACHE_EXT;
1747 		flags |= AT_LBA48;
1748 	} else
1749 		xfer->c_ata_c.r_command = WDCC_FLUSHCACHE;
1750 	xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
1751 	xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
1752 	xfer->c_ata_c.flags = flags | AT_READREG;
1753 	xfer->c_ata_c.timeout = 300000; /* 5m timeout */
1754 	if (wd->atabus->ata_exec_command(wd->drvp, xfer) != ATACMD_COMPLETE) {
1755 		aprint_error_dev(dksc->sc_dev,
1756 		    "flush cache command didn't complete\n");
1757 		error = EIO;
1758 		goto out_xfer;
1759 	}
1760 	if (xfer->c_ata_c.flags & AT_ERROR) {
1761 		if (xfer->c_ata_c.r_error == WDCE_ABRT) {
1762 			/* command not supported */
1763 			error = ENODEV;
1764 			goto out_xfer;
1765 		}
1766 	}
1767 	if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1768 		char sbuf[sizeof(at_errbits) + 64];
1769 		snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
1770 		aprint_error_dev(dksc->sc_dev, "wd_flushcache: status=%s\n",
1771 		    sbuf);
1772 		error = EIO;
1773 		goto out_xfer;
1774 	}
1775 	error = 0;
1776 
1777 out_xfer:
1778 	ata_free_xfer(wd->drvp->chnl_softc, xfer);
1779 
1780 out:
1781 	/* kick queue processing blocked while waiting for flush xfer */
1782 	if (start)
1783 		ata_channel_start(wd->drvp->chnl_softc, wd->drvp->drive);
1784 
1785 	return error;
1786 }
1787 
1788 static int
1789 wd_trim(struct wd_softc *wd, daddr_t bno, long size)
1790 {
1791 	struct dk_softc *dksc = &wd->sc_dksc;
1792 	struct ata_xfer *xfer;
1793 	int error;
1794 	unsigned char *req;
1795 
1796 	xfer = ata_get_xfer(wd->drvp->chnl_softc);
1797 	if (xfer == NULL)
1798 		return EINTR;
1799 
1800 	req = kmem_zalloc(512, KM_SLEEP);
1801 	req[0] = bno & 0xff;
1802 	req[1] = (bno >> 8) & 0xff;
1803 	req[2] = (bno >> 16) & 0xff;
1804 	req[3] = (bno >> 24) & 0xff;
1805 	req[4] = (bno >> 32) & 0xff;
1806 	req[5] = (bno >> 40) & 0xff;
1807 	req[6] = size & 0xff;
1808 	req[7] = (size >> 8) & 0xff;
1809 
1810 	xfer->c_ata_c.r_command = ATA_DATA_SET_MANAGEMENT;
1811 	xfer->c_ata_c.r_count = 1;
1812 	xfer->c_ata_c.r_features = ATA_SUPPORT_DSM_TRIM;
1813 	xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
1814 	xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
1815 	xfer->c_ata_c.timeout = 30000; /* 30s timeout */
1816 	xfer->c_ata_c.data = req;
1817 	xfer->c_ata_c.bcount = 512;
1818 	xfer->c_ata_c.flags |= AT_WRITE | AT_WAIT;
1819 	if (wd->atabus->ata_exec_command(wd->drvp, xfer) != ATACMD_COMPLETE) {
1820 		aprint_error_dev(dksc->sc_dev,
1821 		    "trim command didn't complete\n");
1822 		kmem_free(req, 512);
1823 		error = EIO;
1824 		goto out;
1825 	}
1826 	kmem_free(req, 512);
1827 	if (xfer->c_ata_c.flags & AT_ERROR) {
1828 		if (xfer->c_ata_c.r_error == WDCE_ABRT) {
1829 			/* command not supported */
1830 			error = ENODEV;
1831 			goto out;
1832 		}
1833 	}
1834 	if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1835 		char sbuf[sizeof(at_errbits) + 64];
1836 		snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
1837 		aprint_error_dev(dksc->sc_dev, "wd_trim: status=%s\n",
1838 		    sbuf);
1839 		error = EIO;
1840 		goto out;
1841 	}
1842 	error = 0;
1843 
1844 out:
1845 	ata_free_xfer(wd->drvp->chnl_softc, xfer);
1846 	ata_channel_start(wd->drvp->chnl_softc, wd->drvp->drive);
1847 	return error;
1848 }
1849 
1850 bool
1851 wd_shutdown(device_t dev, int how)
1852 {
1853 	struct wd_softc *wd = device_private(dev);
1854 
1855 	/* the adapter needs to be enabled */
1856 	if (wd->atabus->ata_addref(wd->drvp))
1857 		return true; /* no need to complain */
1858 
1859 	wd_flushcache(wd, AT_POLL, false);
1860 	if ((how & RB_POWERDOWN) == RB_POWERDOWN)
1861 		wd_standby(wd, AT_POLL);
1862 	return true;
1863 }
1864 
1865 /*
1866  * Allocate space for a ioctl queue structure.  Mostly taken from
1867  * scsipi_ioctl.c
1868  */
1869 struct wd_ioctl *
1870 wi_get(struct wd_softc *wd)
1871 {
1872 	struct wd_ioctl *wi;
1873 
1874 	wi = malloc(sizeof(struct wd_ioctl), M_TEMP, M_WAITOK|M_ZERO);
1875 	wi->wi_softc = wd;
1876 	buf_init(&wi->wi_bp);
1877 
1878 	return (wi);
1879 }
1880 
1881 /*
1882  * Free an ioctl structure and remove it from our list
1883  */
1884 
1885 void
1886 wi_free(struct wd_ioctl *wi)
1887 {
1888 	buf_destroy(&wi->wi_bp);
1889 	free(wi, M_TEMP);
1890 }
1891 
1892 /*
1893  * Find a wd_ioctl structure based on the struct buf.
1894  */
1895 
1896 struct wd_ioctl *
1897 wi_find(struct buf *bp)
1898 {
1899 	return container_of(bp, struct wd_ioctl, wi_bp);
1900 }
1901 
1902 static uint
1903 wi_sector_size(const struct wd_ioctl * const wi)
1904 {
1905 	switch (wi->wi_atareq.command) {
1906 	case WDCC_READ:
1907 	case WDCC_WRITE:
1908 	case WDCC_READMULTI:
1909 	case WDCC_WRITEMULTI:
1910 	case WDCC_READDMA:
1911 	case WDCC_WRITEDMA:
1912 	case WDCC_READ_EXT:
1913 	case WDCC_WRITE_EXT:
1914 	case WDCC_READMULTI_EXT:
1915 	case WDCC_WRITEMULTI_EXT:
1916 	case WDCC_READDMA_EXT:
1917 	case WDCC_WRITEDMA_EXT:
1918 	case WDCC_READ_FPDMA_QUEUED:
1919 	case WDCC_WRITE_FPDMA_QUEUED:
1920 		return wi->wi_softc->sc_blksize;
1921 	default:
1922 		return 512;
1923 	}
1924 }
1925 
1926 /*
1927  * Ioctl pseudo strategy routine
1928  *
1929  * This is mostly stolen from scsipi_ioctl.c:scsistrategy().  What
1930  * happens here is:
1931  *
1932  * - wdioctl() queues a wd_ioctl structure.
1933  *
1934  * - wdioctl() calls physio/wdioctlstrategy based on whether or not
1935  *   user space I/O is required.  If physio() is called, physio() eventually
1936  *   calls wdioctlstrategy().
1937  *
1938  * - In either case, wdioctlstrategy() calls wd->atabus->ata_exec_command()
1939  *   to perform the actual command
1940  *
1941  * The reason for the use of the pseudo strategy routine is because
1942  * when doing I/O to/from user space, physio _really_ wants to be in
1943  * the loop.  We could put the entire buffer into the ioctl request
1944  * structure, but that won't scale if we want to do things like download
1945  * microcode.
1946  */
1947 
1948 void
1949 wdioctlstrategy(struct buf *bp)
1950 {
1951 	struct wd_ioctl *wi;
1952 	struct ata_xfer *xfer;
1953 	int error = 0;
1954 
1955 	wi = wi_find(bp);
1956 	if (wi == NULL) {
1957 		printf("wdioctlstrategy: "
1958 		    "No matching ioctl request found in queue\n");
1959 		error = EINVAL;
1960 		goto out2;
1961 	}
1962 
1963 	xfer = ata_get_xfer(wi->wi_softc->drvp->chnl_softc);
1964 	if (xfer == NULL) {
1965 		error = EINTR;
1966 		goto out2;
1967 	}
1968 
1969 	/*
1970 	 * Abort if physio broke up the transfer
1971 	 */
1972 
1973 	if (bp->b_bcount != wi->wi_atareq.datalen) {
1974 		printf("physio split wd ioctl request... cannot proceed\n");
1975 		error = EIO;
1976 		goto out;
1977 	}
1978 
1979 	/*
1980 	 * Abort if we didn't get a buffer size that was a multiple of
1981 	 * our sector size (or overflows CHS/LBA28 sector count)
1982 	 */
1983 
1984 	if ((bp->b_bcount % wi_sector_size(wi)) != 0 ||
1985 	    (bp->b_bcount / wi_sector_size(wi)) >=
1986 	     (1 << NBBY)) {
1987 		error = EINVAL;
1988 		goto out;
1989 	}
1990 
1991 	/*
1992 	 * Make sure a timeout was supplied in the ioctl request
1993 	 */
1994 
1995 	if (wi->wi_atareq.timeout == 0) {
1996 		error = EINVAL;
1997 		goto out;
1998 	}
1999 
2000 	if (wi->wi_atareq.flags & ATACMD_READ)
2001 		xfer->c_ata_c.flags |= AT_READ;
2002 	else if (wi->wi_atareq.flags & ATACMD_WRITE)
2003 		xfer->c_ata_c.flags |= AT_WRITE;
2004 
2005 	if (wi->wi_atareq.flags & ATACMD_READREG)
2006 		xfer->c_ata_c.flags |= AT_READREG;
2007 
2008 	if ((wi->wi_atareq.flags & ATACMD_LBA) != 0)
2009 		xfer->c_ata_c.flags |= AT_LBA;
2010 
2011 	xfer->c_ata_c.flags |= AT_WAIT;
2012 
2013 	xfer->c_ata_c.timeout = wi->wi_atareq.timeout;
2014 	xfer->c_ata_c.r_command = wi->wi_atareq.command;
2015 	xfer->c_ata_c.r_lba = ((wi->wi_atareq.head & 0x0f) << 24) |
2016 	    (wi->wi_atareq.cylinder << 8) |
2017 	    wi->wi_atareq.sec_num;
2018 	xfer->c_ata_c.r_count = wi->wi_atareq.sec_count;
2019 	xfer->c_ata_c.r_features = wi->wi_atareq.features;
2020 	xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
2021 	xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
2022 	xfer->c_ata_c.data = wi->wi_bp.b_data;
2023 	xfer->c_ata_c.bcount = wi->wi_bp.b_bcount;
2024 
2025 	if (wi->wi_softc->atabus->ata_exec_command(wi->wi_softc->drvp, xfer)
2026 	    != ATACMD_COMPLETE) {
2027 		wi->wi_atareq.retsts = ATACMD_ERROR;
2028 		error = EIO;
2029 		goto out;
2030 	}
2031 
2032 	if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
2033 		if (xfer->c_ata_c.flags & AT_ERROR) {
2034 			wi->wi_atareq.retsts = ATACMD_ERROR;
2035 			wi->wi_atareq.error = xfer->c_ata_c.r_error;
2036 		} else if (xfer->c_ata_c.flags & AT_DF)
2037 			wi->wi_atareq.retsts = ATACMD_DF;
2038 		else
2039 			wi->wi_atareq.retsts = ATACMD_TIMEOUT;
2040 	} else {
2041 		wi->wi_atareq.retsts = ATACMD_OK;
2042 		if (wi->wi_atareq.flags & ATACMD_READREG) {
2043 			wi->wi_atareq.command = xfer->c_ata_c.r_status;
2044 			wi->wi_atareq.features = xfer->c_ata_c.r_error;
2045 			wi->wi_atareq.sec_count = xfer->c_ata_c.r_count;
2046 			wi->wi_atareq.sec_num = xfer->c_ata_c.r_lba & 0xff;
2047 			wi->wi_atareq.head = (xfer->c_ata_c.r_device & 0xf0) |
2048 			    ((xfer->c_ata_c.r_lba >> 24) & 0x0f);
2049 			wi->wi_atareq.cylinder =
2050 			    (xfer->c_ata_c.r_lba >> 8) & 0xffff;
2051 			wi->wi_atareq.error = xfer->c_ata_c.r_error;
2052 		}
2053 	}
2054 
2055 out:
2056 	ata_free_xfer(wi->wi_softc->drvp->chnl_softc, xfer);
2057 	ata_channel_start(wi->wi_softc->drvp->chnl_softc,
2058 	    wi->wi_softc->drvp->drive);
2059 out2:
2060 	bp->b_error = error;
2061 	if (error)
2062 		bp->b_resid = bp->b_bcount;
2063 	biodone(bp);
2064 }
2065 
2066 static void
2067 wd_sysctl_attach(struct wd_softc *wd)
2068 {
2069 	struct dk_softc *dksc = &wd->sc_dksc;
2070 	const struct sysctlnode *node;
2071 	int error;
2072 
2073 	/* sysctl set-up */
2074 	if (sysctl_createv(&wd->nodelog, 0, NULL, &node,
2075 				0, CTLTYPE_NODE, dksc->sc_xname,
2076 				SYSCTL_DESCR("wd driver settings"),
2077 				NULL, 0, NULL, 0,
2078 				CTL_HW, CTL_CREATE, CTL_EOL) != 0) {
2079 		aprint_error_dev(dksc->sc_dev,
2080 		    "could not create %s.%s sysctl node\n",
2081 		    "hw", dksc->sc_xname);
2082 		return;
2083 	}
2084 
2085 	wd->drv_max_tags = ATA_MAX_OPENINGS;
2086 	if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
2087 				CTLFLAG_READWRITE, CTLTYPE_INT, "max_tags",
2088 				SYSCTL_DESCR("max number of NCQ tags to use"),
2089 				NULL, 0, &wd->drv_max_tags, 0,
2090 				CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
2091 				!= 0) {
2092 		aprint_error_dev(dksc->sc_dev,
2093 		    "could not create %s.%s.max_tags sysctl - error %d\n",
2094 		    "hw", dksc->sc_xname, error);
2095 		return;
2096 	}
2097 
2098 	wd->drv_ncq = true;
2099 	if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
2100 				CTLFLAG_READWRITE, CTLTYPE_BOOL, "use_ncq",
2101 				SYSCTL_DESCR("use NCQ if supported"),
2102 				NULL, 0, &wd->drv_ncq, 0,
2103 				CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
2104 				!= 0) {
2105 		aprint_error_dev(dksc->sc_dev,
2106 		    "could not create %s.%s.use_ncq sysctl - error %d\n",
2107 		    "hw", dksc->sc_xname, error);
2108 		return;
2109 	}
2110 
2111 	wd->drv_ncq_prio = false;
2112 	if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
2113 				CTLFLAG_READWRITE, CTLTYPE_BOOL, "use_ncq_prio",
2114 				SYSCTL_DESCR("use NCQ PRIORITY if supported"),
2115 				NULL, 0, &wd->drv_ncq_prio, 0,
2116 				CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
2117 				!= 0) {
2118 		aprint_error_dev(dksc->sc_dev,
2119 		    "could not create %s.%s.use_ncq_prio sysctl - error %d\n",
2120 		    "hw", dksc->sc_xname, error);
2121 		return;
2122 	}
2123 
2124 #ifdef WD_CHAOS_MONKEY
2125 	wd->drv_chaos_freq = 0;
2126 	if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
2127 				CTLFLAG_READWRITE, CTLTYPE_INT, "chaos_freq",
2128 				SYSCTL_DESCR("simulated bio read error rate"),
2129 				NULL, 0, &wd->drv_chaos_freq, 0,
2130 				CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
2131 				!= 0) {
2132 		aprint_error_dev(dksc->sc_dev,
2133 		    "could not create %s.%s.chaos_freq sysctl - error %d\n",
2134 		    "hw", dksc->sc_xname, error);
2135 		return;
2136 	}
2137 
2138 	wd->drv_chaos_cnt = 0;
2139 	if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
2140 				CTLFLAG_READONLY, CTLTYPE_INT, "chaos_cnt",
2141 				SYSCTL_DESCR("number of processed bio reads"),
2142 				NULL, 0, &wd->drv_chaos_cnt, 0,
2143 				CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
2144 				!= 0) {
2145 		aprint_error_dev(dksc->sc_dev,
2146 		    "could not create %s.%s.chaos_cnt sysctl - error %d\n",
2147 		    "hw", dksc->sc_xname, error);
2148 		return;
2149 	}
2150 #endif
2151 
2152 }
2153 
2154 static void
2155 wd_sysctl_detach(struct wd_softc *wd)
2156 {
2157 	sysctl_teardown(&wd->nodelog);
2158 }
2159 
2160