wd.c revision 1.431 1 /* $NetBSD: wd.c,v 1.431 2017/10/14 13:15:14 jdolecek 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 1.431 2017/10/14 13:15:14 jdolecek 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_INTR 0x01
106 #define DEBUG_XFERS 0x02
107 #define DEBUG_STATUS 0x04
108 #define DEBUG_FUNCS 0x08
109 #define DEBUG_PROBE 0x10
110 #ifdef ATADEBUG
111 int wdcdebug_wd_mask = 0x0;
112 #define ATADEBUG_PRINT(args, level) \
113 if (wdcdebug_wd_mask & (level)) \
114 printf args
115 #else
116 #define ATADEBUG_PRINT(args, level)
117 #endif
118
119 int wdprobe(device_t, cfdata_t, void *);
120 void wdattach(device_t, device_t, void *);
121 int wddetach(device_t, int);
122 int wdprint(void *, char *);
123 void wdperror(const struct wd_softc *, struct ata_xfer *);
124
125 static void wdminphys(struct buf *);
126
127 static int wdlastclose(device_t);
128 static bool wd_suspend(device_t, const pmf_qual_t *);
129 static int wd_standby(struct wd_softc *, int);
130
131 CFATTACH_DECL3_NEW(wd, sizeof(struct wd_softc),
132 wdprobe, wdattach, wddetach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
133
134 extern struct cfdriver wd_cd;
135
136 dev_type_open(wdopen);
137 dev_type_close(wdclose);
138 dev_type_read(wdread);
139 dev_type_write(wdwrite);
140 dev_type_ioctl(wdioctl);
141 dev_type_strategy(wdstrategy);
142 dev_type_dump(wddump);
143 dev_type_size(wdsize);
144 static dev_type_discard(wddiscard);
145
146 const struct bdevsw wd_bdevsw = {
147 .d_open = wdopen,
148 .d_close = wdclose,
149 .d_strategy = wdstrategy,
150 .d_ioctl = wdioctl,
151 .d_dump = wddump,
152 .d_psize = wdsize,
153 .d_discard = wddiscard,
154 .d_flag = D_DISK
155 };
156
157 const struct cdevsw wd_cdevsw = {
158 .d_open = wdopen,
159 .d_close = wdclose,
160 .d_read = wdread,
161 .d_write = wdwrite,
162 .d_ioctl = wdioctl,
163 .d_stop = nostop,
164 .d_tty = notty,
165 .d_poll = nopoll,
166 .d_mmap = nommap,
167 .d_kqfilter = nokqfilter,
168 .d_discard = wddiscard,
169 .d_flag = D_DISK
170 };
171
172 /* #define WD_DUMP_NOT_TRUSTED if you just want to watch */
173 static int wddoingadump = 0;
174 static int wddumprecalibrated = 0;
175
176 /*
177 * Glue necessary to hook WDCIOCCOMMAND into physio
178 */
179
180 struct wd_ioctl {
181 LIST_ENTRY(wd_ioctl) wi_list;
182 struct buf wi_bp;
183 struct uio wi_uio;
184 struct iovec wi_iov;
185 atareq_t wi_atareq;
186 struct wd_softc *wi_softc;
187 };
188
189 struct wd_ioctl *wi_find(struct buf *);
190 void wi_free(struct wd_ioctl *);
191 struct wd_ioctl *wi_get(struct wd_softc *);
192 void wdioctlstrategy(struct buf *);
193
194 void wdgetdefaultlabel(struct wd_softc *, struct disklabel *);
195 void wdgetdisklabel(struct wd_softc *);
196 void wdstart(device_t);
197 void wdstart1(struct wd_softc *, struct buf *, struct ata_xfer *);
198 static void wdbiorestart(void *);
199 void wddone(device_t, struct ata_xfer *);
200 static void wd_params_to_properties(struct wd_softc *);
201 int wd_get_params(struct wd_softc *, uint8_t, struct ataparams *);
202 int wd_flushcache(struct wd_softc *, int, bool);
203 int wd_trim(struct wd_softc *, int, daddr_t, long);
204 bool wd_shutdown(device_t, int);
205
206 int wd_getcache(struct wd_softc *, int *);
207 int wd_setcache(struct wd_softc *, int);
208
209 static void wd_sysctl_attach(struct wd_softc *);
210 static void wd_sysctl_detach(struct wd_softc *);
211
212 struct dkdriver wddkdriver = {
213 .d_strategy = wdstrategy,
214 .d_minphys = wdminphys
215 };
216
217 #ifdef HAS_BAD144_HANDLING
218 static void bad144intern(struct wd_softc *);
219 #endif
220
221 #define WD_QUIRK_SPLIT_MOD15_WRITE 0x0001 /* must split certain writes */
222
223 #define WD_QUIRK_FMT "\20\1SPLIT_MOD15_WRITE\2FORCE_LBA48"
224
225 /*
226 * Quirk table for IDE drives. Put more-specific matches first, since
227 * a simple globing routine is used for matching.
228 */
229 static const struct wd_quirk {
230 const char *wdq_match; /* inquiry pattern to match */
231 int wdq_quirks; /* drive quirks */
232 } wd_quirk_table[] = {
233 /*
234 * Some Seagate S-ATA drives have a PHY which can get confused
235 * with the way data is packetized by some S-ATA controllers.
236 *
237 * The work-around is to split in two any write transfer whose
238 * sector count % 15 == 1 (assuming 512 byte sectors).
239 *
240 * XXX This is an incomplete list. There are at least a couple
241 * XXX more model numbers. If you have trouble with such transfers
242 * XXX (8K is the most common) on Seagate S-ATA drives, please
243 * XXX notify thorpej (at) NetBSD.org.
244 *
245 * The ST360015AS has not yet been confirmed to have this
246 * issue, however, it is the only other drive in the
247 * Seagate Barracuda Serial ATA V family.
248 *
249 */
250 { "ST3120023AS",
251 WD_QUIRK_SPLIT_MOD15_WRITE },
252 { "ST380023AS",
253 WD_QUIRK_SPLIT_MOD15_WRITE },
254 { "ST360015AS",
255 WD_QUIRK_SPLIT_MOD15_WRITE },
256 { NULL,
257 0 }
258 };
259
260 static const struct wd_quirk *
261 wd_lookup_quirks(const char *name)
262 {
263 const struct wd_quirk *wdq;
264 const char *estr;
265
266 for (wdq = wd_quirk_table; wdq->wdq_match != NULL; wdq++) {
267 /*
268 * We only want exact matches (which include matches
269 * against globbing characters).
270 */
271 if (pmatch(name, wdq->wdq_match, &estr) == 2)
272 return (wdq);
273 }
274 return (NULL);
275 }
276
277 int
278 wdprobe(device_t parent, cfdata_t match, void *aux)
279 {
280 struct ata_device *adev = aux;
281
282 if (adev == NULL)
283 return 0;
284 if (adev->adev_bustype->bustype_type != SCSIPI_BUSTYPE_ATA)
285 return 0;
286
287 if (match->cf_loc[ATA_HLCF_DRIVE] != ATA_HLCF_DRIVE_DEFAULT &&
288 match->cf_loc[ATA_HLCF_DRIVE] != adev->adev_drv_data->drive)
289 return 0;
290 return 1;
291 }
292
293 void
294 wdattach(device_t parent, device_t self, void *aux)
295 {
296 struct wd_softc *wd = device_private(self);
297 struct ata_device *adev= aux;
298 int i, blank;
299 char tbuf[41], pbuf[9], c, *p, *q;
300 const struct wd_quirk *wdq;
301
302 wd->sc_dev = self;
303
304 ATADEBUG_PRINT(("wdattach\n"), DEBUG_FUNCS | DEBUG_PROBE);
305 mutex_init(&wd->sc_lock, MUTEX_DEFAULT, IPL_BIO);
306 bufq_alloc(&wd->sc_q, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
307 #ifdef WD_SOFTBADSECT
308 SLIST_INIT(&wd->sc_bslist);
309 #endif
310 wd->atabus = adev->adev_bustype;
311 wd->drvp = adev->adev_drv_data;
312
313 wd->drvp->drv_openings = 1;
314 wd->drvp->drv_start = wdstart;
315 wd->drvp->drv_done = wddone;
316 wd->drvp->drv_softc = wd->sc_dev; /* done in atabusconfig_thread()
317 but too late */
318
319 aprint_naive("\n");
320 aprint_normal("\n");
321
322 /* read our drive info */
323 if (wd_get_params(wd, AT_WAIT, &wd->sc_params) != 0) {
324 aprint_error_dev(self, "IDENTIFY failed\n");
325 goto out;
326 }
327
328 for (blank = 0, p = wd->sc_params.atap_model, q = tbuf, i = 0;
329 i < sizeof(wd->sc_params.atap_model); i++) {
330 c = *p++;
331 if (c == '\0')
332 break;
333 if (c != ' ') {
334 if (blank) {
335 *q++ = ' ';
336 blank = 0;
337 }
338 *q++ = c;
339 } else
340 blank = 1;
341 }
342 *q++ = '\0';
343
344 aprint_normal_dev(self, "<%s>\n", tbuf);
345
346 wdq = wd_lookup_quirks(tbuf);
347 if (wdq != NULL)
348 wd->sc_quirks = wdq->wdq_quirks;
349
350 if (wd->sc_quirks != 0) {
351 char sbuf[sizeof(WD_QUIRK_FMT) + 64];
352 snprintb(sbuf, sizeof(sbuf), WD_QUIRK_FMT, wd->sc_quirks);
353 aprint_normal_dev(self, "quirks %s\n", sbuf);
354
355 if (wd->sc_quirks & WD_QUIRK_SPLIT_MOD15_WRITE) {
356 aprint_error_dev(self, "drive corrupts write transfers with certain controllers, consider replacing\n");
357 }
358 }
359
360 if ((wd->sc_params.atap_multi & 0xff) > 1) {
361 wd->drvp->multi = wd->sc_params.atap_multi & 0xff;
362 } else {
363 wd->drvp->multi = 1;
364 }
365
366 aprint_verbose_dev(self, "drive supports %d-sector PIO transfers,",
367 wd->drvp->multi);
368
369 /* 48-bit LBA addressing */
370 if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0)
371 wd->sc_flags |= WDF_LBA48;
372
373 /* Prior to ATA-4, LBA was optional. */
374 if ((wd->sc_params.atap_capabilities1 & WDC_CAP_LBA) != 0)
375 wd->sc_flags |= WDF_LBA;
376 #if 0
377 /* ATA-4 requires LBA. */
378 if (wd->sc_params.atap_ataversion != 0xffff &&
379 wd->sc_params.atap_ataversion >= WDC_VER_ATA4)
380 wd->sc_flags |= WDF_LBA;
381 #endif
382
383 if ((wd->sc_flags & WDF_LBA48) != 0) {
384 aprint_verbose(" LBA48 addressing\n");
385 wd->sc_capacity =
386 ((uint64_t) wd->sc_params.atap_max_lba[3] << 48) |
387 ((uint64_t) wd->sc_params.atap_max_lba[2] << 32) |
388 ((uint64_t) wd->sc_params.atap_max_lba[1] << 16) |
389 ((uint64_t) wd->sc_params.atap_max_lba[0] << 0);
390 wd->sc_capacity28 =
391 (wd->sc_params.atap_capacity[1] << 16) |
392 wd->sc_params.atap_capacity[0];
393 } else if ((wd->sc_flags & WDF_LBA) != 0) {
394 aprint_verbose(" LBA addressing\n");
395 wd->sc_capacity28 = wd->sc_capacity =
396 (wd->sc_params.atap_capacity[1] << 16) |
397 wd->sc_params.atap_capacity[0];
398 } else {
399 aprint_verbose(" chs addressing\n");
400 wd->sc_capacity28 = wd->sc_capacity =
401 wd->sc_params.atap_cylinders *
402 wd->sc_params.atap_heads *
403 wd->sc_params.atap_sectors;
404 }
405 if ((wd->sc_params.atap_secsz & ATA_SECSZ_VALID_MASK) == ATA_SECSZ_VALID
406 && ((wd->sc_params.atap_secsz & ATA_SECSZ_LLS) != 0)) {
407 wd->sc_blksize = 2ULL *
408 ((uint32_t)((wd->sc_params.atap_lls_secsz[1] << 16) |
409 wd->sc_params.atap_lls_secsz[0]));
410 } else {
411 wd->sc_blksize = 512;
412 }
413 wd->sc_capacity512 = (wd->sc_capacity * wd->sc_blksize) / DEV_BSIZE;
414 format_bytes(pbuf, sizeof(pbuf), wd->sc_capacity * wd->sc_blksize);
415 aprint_normal_dev(self, "%s, %d cyl, %d head, %d sec, "
416 "%d bytes/sect x %llu sectors\n",
417 pbuf,
418 (wd->sc_flags & WDF_LBA) ? (int)(wd->sc_capacity /
419 (wd->sc_params.atap_heads * wd->sc_params.atap_sectors)) :
420 wd->sc_params.atap_cylinders,
421 wd->sc_params.atap_heads, wd->sc_params.atap_sectors,
422 wd->sc_blksize, (unsigned long long)wd->sc_capacity);
423
424 ATADEBUG_PRINT(("%s: atap_dmatiming_mimi=%d, atap_dmatiming_recom=%d\n",
425 device_xname(self), wd->sc_params.atap_dmatiming_mimi,
426 wd->sc_params.atap_dmatiming_recom), DEBUG_PROBE);
427
428 if (wd->sc_blksize <= 0 || !powerof2(wd->sc_blksize) ||
429 wd->sc_blksize < DEV_BSIZE || wd->sc_blksize > MAXPHYS) {
430 aprint_normal_dev(self, "WARNING: block size %u "
431 "might not actually work\n", wd->sc_blksize);
432 }
433
434 out:
435 /*
436 * Initialize and attach the disk structure.
437 */
438 /* we fill in dk_info later */
439 disk_init(&wd->sc_dk, device_xname(wd->sc_dev), &wddkdriver);
440 disk_attach(&wd->sc_dk);
441 wd->drvp->lp = wd->sc_dk.dk_label;
442 wd_params_to_properties(wd);
443 rnd_attach_source(&wd->rnd_source, device_xname(wd->sc_dev),
444 RND_TYPE_DISK, RND_FLAG_DEFAULT);
445
446 /* Discover wedges on this disk. */
447 dkwedge_discover(&wd->sc_dk);
448
449 if (!pmf_device_register1(self, wd_suspend, NULL, wd_shutdown))
450 aprint_error_dev(self, "couldn't establish power handler\n");
451
452 wd_sysctl_attach(wd);
453 }
454
455 static bool
456 wd_suspend(device_t dv, const pmf_qual_t *qual)
457 {
458 struct wd_softc *sc = device_private(dv);
459
460 /* the adapter needs to be enabled */
461 if (sc->atabus->ata_addref(sc->drvp))
462 return true; /* no need to complain */
463
464 wd_flushcache(sc, AT_WAIT, false);
465 wd_standby(sc, AT_WAIT);
466
467 sc->atabus->ata_delref(sc->drvp);
468 return true;
469 }
470
471 int
472 wddetach(device_t self, int flags)
473 {
474 struct wd_softc *sc = device_private(self);
475 int bmaj, cmaj, i, mn, rc;
476
477 if ((rc = disk_begindetach(&sc->sc_dk, wdlastclose, self, flags)) != 0)
478 return rc;
479
480 /* locate the major number */
481 bmaj = bdevsw_lookup_major(&wd_bdevsw);
482 cmaj = cdevsw_lookup_major(&wd_cdevsw);
483
484 /* Nuke the vnodes for any open instances. */
485 for (i = 0; i < MAXPARTITIONS; i++) {
486 mn = WDMINOR(device_unit(self), i);
487 vdevgone(bmaj, mn, mn, VBLK);
488 vdevgone(cmaj, mn, mn, VCHR);
489 }
490
491 /* Delete all of our wedges. */
492 dkwedge_delall(&sc->sc_dk);
493
494 mutex_enter(&sc->sc_lock);
495
496 /* Kill off any queued buffers. */
497 bufq_drain(sc->sc_q);
498
499 sc->atabus->ata_killpending(sc->drvp);
500 mutex_exit(&sc->sc_lock);
501
502 if (flags & DETACH_POWEROFF)
503 wd_standby(sc, AT_POLL);
504
505 bufq_free(sc->sc_q);
506
507 /* Detach disk. */
508 disk_detach(&sc->sc_dk);
509 disk_destroy(&sc->sc_dk);
510
511 #ifdef WD_SOFTBADSECT
512 /* Clean out the bad sector list */
513 while (!SLIST_EMPTY(&sc->sc_bslist)) {
514 void *head = SLIST_FIRST(&sc->sc_bslist);
515 SLIST_REMOVE_HEAD(&sc->sc_bslist, dbs_next);
516 free(head, M_TEMP);
517 }
518 sc->sc_bscount = 0;
519 #endif
520
521 pmf_device_deregister(self);
522
523 wd_sysctl_detach(sc);
524
525 /* Unhook the entropy source. */
526 rnd_detach_source(&sc->rnd_source);
527
528 mutex_destroy(&sc->sc_lock);
529
530 sc->drvp->drive_type = ATA_DRIVET_NONE; /* no drive any more here */
531 sc->drvp->drive_flags = 0;
532
533 return (0);
534 }
535
536 /*
537 * Read/write routine for a buffer. Validates the arguments and schedules the
538 * transfer. Does not wait for the transfer to complete.
539 */
540 void
541 wdstrategy(struct buf *bp)
542 {
543 struct wd_softc *wd =
544 device_lookup_private(&wd_cd, WDUNIT(bp->b_dev));
545 struct disklabel *lp = wd->sc_dk.dk_label;
546 daddr_t blkno;
547
548 ATADEBUG_PRINT(("wdstrategy (%s)\n", device_xname(wd->sc_dev)),
549 DEBUG_XFERS);
550
551 /* Valid request? */
552 if (bp->b_blkno < 0 ||
553 (bp->b_bcount % lp->d_secsize) != 0 ||
554 (bp->b_bcount / lp->d_secsize) >= (1 << NBBY)) {
555 bp->b_error = EINVAL;
556 goto done;
557 }
558
559 /* If device invalidated (e.g. media change, door open,
560 * device detachment), then error.
561 */
562 if ((wd->sc_flags & WDF_LOADED) == 0 ||
563 !device_is_enabled(wd->sc_dev)) {
564 bp->b_error = EIO;
565 goto done;
566 }
567
568 /* If it's a null transfer, return immediately. */
569 if (bp->b_bcount == 0)
570 goto done;
571
572 /*
573 * Do bounds checking, adjust transfer. if error, process.
574 * If end of partition, just return.
575 */
576 if (WDPART(bp->b_dev) == RAW_PART) {
577 if (bounds_check_with_mediasize(bp, DEV_BSIZE,
578 wd->sc_capacity512) <= 0)
579 goto done;
580 } else {
581 if (bounds_check_with_label(&wd->sc_dk, bp,
582 (wd->sc_flags & (WDF_WLABEL|WDF_LABELLING)) != 0) <= 0)
583 goto done;
584 }
585
586 /*
587 * Now convert the block number to absolute and put it in
588 * terms of the device's logical block size.
589 */
590 if (lp->d_secsize >= DEV_BSIZE)
591 blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
592 else
593 blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
594
595 if (WDPART(bp->b_dev) != RAW_PART)
596 blkno += lp->d_partitions[WDPART(bp->b_dev)].p_offset;
597
598 bp->b_rawblkno = blkno;
599
600 #ifdef WD_SOFTBADSECT
601 /*
602 * If the transfer about to be attempted contains only a block that
603 * is known to be bad then return an error for the transfer without
604 * even attempting to start a transfer up under the premis that we
605 * will just end up doing more retries for a transfer that will end
606 * up failing again.
607 */
608 if (__predict_false(!SLIST_EMPTY(&wd->sc_bslist))) {
609 struct disk_badsectors *dbs;
610 daddr_t maxblk = blkno + (bp->b_bcount / wd->sc_blksize) - 1;
611
612 mutex_enter(&wd->sc_lock);
613 SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next)
614 if ((dbs->dbs_min <= blkno && blkno <= dbs->dbs_max) ||
615 (dbs->dbs_min <= maxblk && maxblk <= dbs->dbs_max)){
616 bp->b_error = EIO;
617 mutex_exit(&wd->sc_lock);
618 goto done;
619 }
620 mutex_exit(&wd->sc_lock);
621 }
622 #endif
623
624 /* Queue transfer on drive, activate drive and controller if idle. */
625 mutex_enter(&wd->sc_lock);
626 disk_wait(&wd->sc_dk);
627 bufq_put(wd->sc_q, bp);
628 mutex_exit(&wd->sc_lock);
629
630 /* Try to queue on the current drive only */
631 wdstart(wd->sc_dev);
632 return;
633 done:
634 /* Toss transfer; we're done early. */
635 bp->b_resid = bp->b_bcount;
636 biodone(bp);
637 }
638
639 /*
640 * Queue a drive for I/O.
641 */
642 void
643 wdstart(device_t self)
644 {
645 struct wd_softc *wd = device_private(self);
646 struct buf *bp;
647 struct ata_xfer *xfer;
648
649 ATADEBUG_PRINT(("wdstart %s\n", device_xname(wd->sc_dev)),
650 DEBUG_XFERS);
651
652 if (!device_is_active(wd->sc_dev))
653 return;
654
655 mutex_enter(&wd->sc_lock);
656
657 /*
658 * Do not queue any transfers until flush is finished, so that
659 * once flush is pending, it will get handled as soon as xfer
660 * is available.
661 */
662 if (ISSET(wd->sc_flags, WDF_FLUSH_PEND))
663 goto out;
664
665 while (bufq_peek(wd->sc_q) != NULL) {
666 /* First try to get xfer. Limit to drive openings iff NCQ. */
667 xfer = ata_get_xfer_ext(wd->drvp->chnl_softc, 0,
668 WD_USE_NCQ(wd) ? WD_MAX_OPENINGS(wd) : 0);
669 if (xfer == NULL)
670 break;
671
672 /* There is got to be a buf for us */
673 bp = bufq_get(wd->sc_q);
674 KASSERT(bp != NULL);
675
676 xfer->c_retries = 0;
677 wdstart1(wd, bp, xfer);
678 }
679
680 out:
681 mutex_exit(&wd->sc_lock);
682 }
683
684 void
685 wdstart1(struct wd_softc *wd, struct buf *bp, struct ata_xfer *xfer)
686 {
687 /* must be locked on entry */
688 KASSERT(mutex_owned(&wd->sc_lock));
689
690 KASSERT(bp == xfer->c_bio.bp || xfer->c_bio.bp == NULL);
691 KASSERT((xfer->c_flags & (C_WAITACT|C_FREE)) == 0);
692
693 /* Reset state, so that retries don't use stale info */
694 if (__predict_false(xfer->c_retries > 0)) {
695 xfer->c_flags = 0;
696 memset(&xfer->c_bio, 0, sizeof(xfer->c_bio));
697 }
698
699 xfer->c_bio.blkno = bp->b_rawblkno;
700 xfer->c_bio.bcount = bp->b_bcount;
701 xfer->c_bio.databuf = bp->b_data;
702 xfer->c_bio.blkdone = 0;
703 xfer->c_bio.bp = bp;
704
705 #ifdef WD_CHAOS_MONKEY
706 /*
707 * Override blkno to be over device capacity to trigger error,
708 * but only if it's read, to avoid trashing disk contents should
709 * the command be clipped, or otherwise misinterpreted, by the
710 * driver or controller.
711 */
712 if (BUF_ISREAD(bp) && xfer->c_retries == 0 && wd->drv_chaos_freq > 0 &&
713 (++wd->drv_chaos_cnt % wd->drv_chaos_freq) == 0) {
714 aprint_normal_dev(wd->sc_dev, "%s: chaos xfer %d\n",
715 __func__, xfer->c_slot);
716 xfer->c_bio.blkno = 7777777 + wd->sc_capacity;
717 xfer->c_flags |= C_CHAOS;
718 }
719 #endif
720
721 /*
722 * If we're retrying, retry in single-sector mode. This will give us
723 * the sector number of the problem, and will eventually allow the
724 * transfer to succeed. If FUA is requested, we can't actually
725 * do this, as ATA_SINGLE is usually executed as PIO transfer by drivers
726 * which support it, and that isn't compatible with NCQ/FUA.
727 */
728 if (xfer->c_retries >= WDIORETRIES_SINGLE &&
729 (bp->b_flags & B_MEDIA_FUA) == 0)
730 xfer->c_bio.flags = ATA_SINGLE;
731 else
732 xfer->c_bio.flags = 0;
733 if (wd->sc_flags & WDF_LBA48 &&
734 (((xfer->c_bio.blkno +
735 xfer->c_bio.bcount / wd->sc_dk.dk_label->d_secsize) >
736 wd->sc_capacity28) ||
737 ((xfer->c_bio.bcount / wd->sc_dk.dk_label->d_secsize) > 128)))
738 xfer->c_bio.flags |= ATA_LBA48;
739
740 /*
741 * If NCQ was negotiated, always use it for the first several attempts.
742 * Since device cancels all outstanding requests on error, downgrade
743 * to non-NCQ on retry, so that the retried transfer would not cause
744 * cascade failure for the other transfers if it fails again.
745 * If FUA was requested, we can't downgrade, as that would violate
746 * the semantics - FUA would not be honored. In that case, continue
747 * retrying with NCQ.
748 */
749 if (WD_USE_NCQ(wd) && (xfer->c_retries < WDIORETRIES_SINGLE ||
750 (bp->b_flags & B_MEDIA_FUA) != 0)) {
751 xfer->c_bio.flags |= ATA_LBA48;
752 xfer->c_flags |= C_NCQ;
753
754 if (WD_USE_NCQ_PRIO(wd) &&
755 BIO_GETPRIO(bp) == BPRIO_TIMECRITICAL)
756 xfer->c_bio.flags |= ATA_PRIO_HIGH;
757 }
758
759 if (wd->sc_flags & WDF_LBA)
760 xfer->c_bio.flags |= ATA_LBA;
761 if (bp->b_flags & B_READ)
762 xfer->c_bio.flags |= ATA_READ;
763 if (bp->b_flags & B_MEDIA_FUA) {
764 /* If not using NCQ, the command WRITE DMA FUA EXT is LBA48 */
765 KASSERT((wd->sc_flags & WDF_LBA48) != 0);
766 if ((xfer->c_flags & C_NCQ) == 0)
767 xfer->c_bio.flags |= ATA_LBA48;
768
769 xfer->c_bio.flags |= ATA_FUA;
770 }
771
772 /* Instrumentation. */
773 if (xfer->c_retries == 0)
774 disk_busy(&wd->sc_dk);
775 switch (wd->atabus->ata_bio(wd->drvp, xfer)) {
776 case ATACMD_TRY_AGAIN:
777 panic("wdstart1: try again");
778 break;
779 case ATACMD_QUEUED:
780 case ATACMD_COMPLETE:
781 break;
782 default:
783 panic("wdstart1: bad return code from ata_bio()");
784 }
785 }
786
787 void
788 wddone(device_t self, struct ata_xfer *xfer)
789 {
790 struct wd_softc *wd = device_private(self);
791 const char *errmsg;
792 int do_perror = 0;
793 struct buf *bp;
794
795 ATADEBUG_PRINT(("wddone %s\n", device_xname(wd->sc_dev)),
796 DEBUG_XFERS);
797
798 if (__predict_false(wddoingadump)) {
799 /* just drop it to the floor */
800 ata_free_xfer(wd->drvp->chnl_softc, xfer);
801 return;
802 }
803
804 bp = xfer->c_bio.bp;
805 KASSERT(bp != NULL);
806
807 bp->b_resid = xfer->c_bio.bcount;
808 switch (xfer->c_bio.error) {
809 case ERR_DMA:
810 errmsg = "DMA error";
811 goto retry;
812 case ERR_DF:
813 errmsg = "device fault";
814 goto retry;
815 case TIMEOUT:
816 errmsg = "device timeout";
817 goto retry;
818 case REQUEUE:
819 errmsg = "requeue";
820 goto retry2;
821 case ERR_RESET:
822 errmsg = "channel reset";
823 goto retry2;
824 case ERROR:
825 /* Don't care about media change bits */
826 if (xfer->c_bio.r_error != 0 &&
827 (xfer->c_bio.r_error & ~(WDCE_MC | WDCE_MCR)) == 0)
828 goto noerror;
829 errmsg = "error";
830 do_perror = 1;
831 retry: /* Just reset and retry. Can we do more ? */
832 if ((xfer->c_flags & C_RECOVERED) == 0) {
833 int wflags = (xfer->c_flags & C_POLL) ? AT_POLL : 0;
834 (*wd->atabus->ata_reset_drive)(wd->drvp, wflags, NULL);
835 }
836 retry2:
837 mutex_enter(&wd->sc_lock);
838
839 diskerr(bp, "wd", errmsg, LOG_PRINTF,
840 xfer->c_bio.blkdone, wd->sc_dk.dk_label);
841 if (xfer->c_retries < WDIORETRIES)
842 printf(", slot %d, retry %d", xfer->c_slot,
843 xfer->c_retries + 1);
844 printf("\n");
845 if (do_perror)
846 wdperror(wd, xfer);
847
848 if (xfer->c_retries < WDIORETRIES) {
849 xfer->c_retries++;
850
851 /* Rerun ASAP if just requeued */
852 callout_reset(&xfer->c_retry_callout,
853 (xfer->c_bio.error == REQUEUE) ? 1 : RECOVERYTIME,
854 wdbiorestart, xfer);
855
856 mutex_exit(&wd->sc_lock);
857 return;
858 }
859
860 mutex_exit(&wd->sc_lock);
861
862 #ifdef WD_SOFTBADSECT
863 /*
864 * Not all errors indicate a failed block but those that do,
865 * put the block on the bad-block list for the device. Only
866 * do this for reads because the drive should do it for writes,
867 * itself, according to Manuel.
868 */
869 if ((bp->b_flags & B_READ) &&
870 ((wd->drvp->ata_vers >= 4 && xfer->c_bio.r_error & 64) ||
871 (wd->drvp->ata_vers < 4 && xfer->c_bio.r_error & 192))) {
872 struct disk_badsectors *dbs;
873
874 dbs = malloc(sizeof *dbs, M_TEMP, M_NOWAIT);
875 if (dbs == NULL) {
876 aprint_error_dev(wd->sc_dev,
877 "failed to add bad block to list\n");
878 goto out;
879 }
880
881 dbs->dbs_min = bp->b_rawblkno;
882 dbs->dbs_max = dbs->dbs_min +
883 (bp->b_bcount /wd->sc_blksize) - 1;
884 microtime(&dbs->dbs_failedat);
885
886 mutex_enter(&wd->sc_lock);
887 SLIST_INSERT_HEAD(&wd->sc_bslist, dbs, dbs_next);
888 wd->sc_bscount++;
889 mutex_exit(&wd->sc_lock);
890 }
891 out:
892 #endif
893 bp->b_error = EIO;
894 break;
895 case NOERROR:
896 noerror: if ((xfer->c_bio.flags & ATA_CORR) || xfer->c_retries > 0)
897 aprint_error_dev(wd->sc_dev,
898 "soft error (corrected) slot %d\n", xfer->c_slot);
899 #ifdef WD_CHAOS_MONKEY
900 KASSERT((xfer->c_flags & C_CHAOS) == 0);
901 #endif
902 break;
903 case ERR_NODEV:
904 bp->b_error = EIO;
905 break;
906 }
907 if (__predict_false(bp->b_error != 0) && bp->b_resid == 0) {
908 /*
909 * the disk or controller sometimes report a complete
910 * xfer, when there has been an error. This is wrong,
911 * assume nothing got transfered in this case
912 */
913 bp->b_resid = bp->b_bcount;
914 }
915 disk_unbusy(&wd->sc_dk, (bp->b_bcount - bp->b_resid),
916 (bp->b_flags & B_READ));
917 rnd_add_uint32(&wd->rnd_source, bp->b_blkno);
918 ata_free_xfer(wd->drvp->chnl_softc, xfer);
919 biodone(bp);
920 ata_channel_start(wd->drvp->chnl_softc, wd->drvp->drive);
921 }
922
923 static void
924 wdbiorestart(void *v)
925 {
926 struct ata_xfer *xfer = v;
927 struct buf *bp = xfer->c_bio.bp;
928 struct wd_softc *wd = device_lookup_private(&wd_cd, WDUNIT(bp->b_dev));
929
930 ATADEBUG_PRINT(("wdrestart %s\n", device_xname(wd->sc_dev)),
931 DEBUG_XFERS);
932
933 mutex_enter(&wd->sc_lock);
934 wdstart1(wd, bp, xfer);
935 mutex_exit(&wd->sc_lock);
936 }
937
938 static void
939 wdminphys(struct buf *bp)
940 {
941 const struct wd_softc * const wd =
942 device_lookup_private(&wd_cd, WDUNIT(bp->b_dev));
943 uint32_t maxsectors;
944
945 /*
946 * The limit is actually 65536 for LBA48 and 256 for non-LBA48,
947 * but that requires to set the count for the ATA command
948 * to 0, which is somewhat error prone, so better stay safe.
949 */
950 if (wd->sc_flags & WDF_LBA48)
951 maxsectors = 65535;
952 else
953 maxsectors = 128;
954
955 if (bp->b_bcount > (wd->sc_blksize * maxsectors))
956 bp->b_bcount = (wd->sc_blksize * maxsectors);
957
958 minphys(bp);
959 }
960
961 int
962 wdread(dev_t dev, struct uio *uio, int flags)
963 {
964
965 ATADEBUG_PRINT(("wdread\n"), DEBUG_XFERS);
966 return (physio(wdstrategy, NULL, dev, B_READ, wdminphys, uio));
967 }
968
969 int
970 wdwrite(dev_t dev, struct uio *uio, int flags)
971 {
972
973 ATADEBUG_PRINT(("wdwrite\n"), DEBUG_XFERS);
974 return (physio(wdstrategy, NULL, dev, B_WRITE, wdminphys, uio));
975 }
976
977 int
978 wdopen(dev_t dev, int flag, int fmt, struct lwp *l)
979 {
980 struct wd_softc *wd;
981 int part, error;
982
983 ATADEBUG_PRINT(("wdopen\n"), DEBUG_FUNCS);
984 wd = device_lookup_private(&wd_cd, WDUNIT(dev));
985 if (wd == NULL)
986 return (ENXIO);
987
988 if (! device_is_active(wd->sc_dev))
989 return (ENODEV);
990
991 if (wd->sc_capacity == 0)
992 return (ENODEV);
993
994 part = WDPART(dev);
995
996 mutex_enter(&wd->sc_dk.dk_openlock);
997
998 /*
999 * If there are wedges, and this is not RAW_PART, then we
1000 * need to fail.
1001 */
1002 if (wd->sc_dk.dk_nwedges != 0 && part != RAW_PART) {
1003 error = EBUSY;
1004 goto bad1;
1005 }
1006
1007 /*
1008 * If this is the first open of this device, add a reference
1009 * to the adapter.
1010 */
1011 if (wd->sc_dk.dk_openmask == 0 &&
1012 (error = wd->atabus->ata_addref(wd->drvp)) != 0)
1013 goto bad1;
1014
1015 if (wd->sc_dk.dk_openmask != 0) {
1016 /*
1017 * If any partition is open, but the disk has been invalidated,
1018 * disallow further opens.
1019 */
1020 if ((wd->sc_flags & WDF_LOADED) == 0) {
1021 error = EIO;
1022 goto bad2;
1023 }
1024 } else {
1025 if ((wd->sc_flags & WDF_LOADED) == 0) {
1026
1027 /* Load the physical device parameters. */
1028 if (wd_get_params(wd, AT_WAIT, &wd->sc_params) != 0) {
1029 aprint_error_dev(wd->sc_dev,
1030 "IDENTIFY failed\n");
1031 error = EIO;
1032 goto bad2;
1033 }
1034 wd->sc_flags |= WDF_LOADED;
1035 /* Load the partition info if not already loaded. */
1036 wdgetdisklabel(wd);
1037 }
1038 }
1039
1040 /* Check that the partition exists. */
1041 if (part != RAW_PART &&
1042 (part >= wd->sc_dk.dk_label->d_npartitions ||
1043 wd->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
1044 error = ENXIO;
1045 goto bad2;
1046 }
1047
1048 /* Insure only one open at a time. */
1049 switch (fmt) {
1050 case S_IFCHR:
1051 wd->sc_dk.dk_copenmask |= (1 << part);
1052 break;
1053 case S_IFBLK:
1054 wd->sc_dk.dk_bopenmask |= (1 << part);
1055 break;
1056 }
1057 wd->sc_dk.dk_openmask =
1058 wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
1059
1060 mutex_exit(&wd->sc_dk.dk_openlock);
1061 return 0;
1062
1063 bad2:
1064 if (wd->sc_dk.dk_openmask == 0)
1065 wd->atabus->ata_delref(wd->drvp);
1066 bad1:
1067 mutex_exit(&wd->sc_dk.dk_openlock);
1068 return error;
1069 }
1070
1071 /*
1072 * Caller must hold wd->sc_dk.dk_openlock.
1073 */
1074 static int
1075 wdlastclose(device_t self)
1076 {
1077 struct wd_softc *wd = device_private(self);
1078
1079 wd_flushcache(wd, AT_WAIT, false);
1080
1081 if (! (wd->sc_flags & WDF_KLABEL))
1082 wd->sc_flags &= ~WDF_LOADED;
1083
1084 wd->atabus->ata_delref(wd->drvp);
1085
1086 return 0;
1087 }
1088
1089 int
1090 wdclose(dev_t dev, int flag, int fmt, struct lwp *l)
1091 {
1092 struct wd_softc *wd =
1093 device_lookup_private(&wd_cd, WDUNIT(dev));
1094 int part = WDPART(dev);
1095
1096 ATADEBUG_PRINT(("wdclose\n"), DEBUG_FUNCS);
1097
1098 mutex_enter(&wd->sc_dk.dk_openlock);
1099
1100 switch (fmt) {
1101 case S_IFCHR:
1102 wd->sc_dk.dk_copenmask &= ~(1 << part);
1103 break;
1104 case S_IFBLK:
1105 wd->sc_dk.dk_bopenmask &= ~(1 << part);
1106 break;
1107 }
1108 wd->sc_dk.dk_openmask =
1109 wd->sc_dk.dk_copenmask | wd->sc_dk.dk_bopenmask;
1110
1111 if (wd->sc_dk.dk_openmask == 0)
1112 wdlastclose(wd->sc_dev);
1113
1114 mutex_exit(&wd->sc_dk.dk_openlock);
1115 return 0;
1116 }
1117
1118 void
1119 wdgetdefaultlabel(struct wd_softc *wd, struct disklabel *lp)
1120 {
1121
1122 ATADEBUG_PRINT(("wdgetdefaultlabel\n"), DEBUG_FUNCS);
1123 memset(lp, 0, sizeof(struct disklabel));
1124
1125 lp->d_secsize = wd->sc_blksize;
1126 lp->d_ntracks = wd->sc_params.atap_heads;
1127 lp->d_nsectors = wd->sc_params.atap_sectors;
1128 lp->d_ncylinders = (wd->sc_flags & WDF_LBA) ? wd->sc_capacity /
1129 (wd->sc_params.atap_heads * wd->sc_params.atap_sectors) :
1130 wd->sc_params.atap_cylinders;
1131 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1132
1133 if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
1134 lp->d_type = DKTYPE_ST506;
1135 else
1136 lp->d_type = DKTYPE_ESDI;
1137
1138 strncpy(lp->d_typename, wd->sc_params.atap_model, 16);
1139 strncpy(lp->d_packname, "fictitious", 16);
1140 if (wd->sc_capacity > UINT32_MAX)
1141 lp->d_secperunit = UINT32_MAX;
1142 else
1143 lp->d_secperunit = wd->sc_capacity;
1144 lp->d_rpm = 3600;
1145 lp->d_interleave = 1;
1146 lp->d_flags = 0;
1147
1148 lp->d_partitions[RAW_PART].p_offset = 0;
1149 lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
1150 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1151 lp->d_npartitions = RAW_PART + 1;
1152
1153 lp->d_magic = DISKMAGIC;
1154 lp->d_magic2 = DISKMAGIC;
1155 lp->d_checksum = dkcksum(lp);
1156 }
1157
1158 /*
1159 * Fabricate a default disk label, and try to read the correct one.
1160 */
1161 void
1162 wdgetdisklabel(struct wd_softc *wd)
1163 {
1164 struct disklabel *lp = wd->sc_dk.dk_label;
1165 const char *errstring;
1166
1167 ATADEBUG_PRINT(("wdgetdisklabel\n"), DEBUG_FUNCS);
1168
1169 memset(wd->sc_dk.dk_cpulabel, 0, sizeof(struct cpu_disklabel));
1170
1171 wdgetdefaultlabel(wd, lp);
1172
1173 wd->drvp->badsect[0] = -1;
1174
1175 if (wd->drvp->state > RESET) {
1176 mutex_enter(&wd->sc_lock);
1177 wd->drvp->drive_flags |= ATA_DRIVE_RESET;
1178 mutex_exit(&wd->sc_lock);
1179 }
1180 errstring = readdisklabel(MAKEWDDEV(0, device_unit(wd->sc_dev),
1181 RAW_PART), wdstrategy, lp,
1182 wd->sc_dk.dk_cpulabel);
1183 if (errstring) {
1184 /*
1185 * This probably happened because the drive's default
1186 * geometry doesn't match the DOS geometry. We
1187 * assume the DOS geometry is now in the label and try
1188 * again. XXX This is a kluge.
1189 */
1190 if (wd->drvp->state > RESET) {
1191 mutex_enter(&wd->sc_lock);
1192 wd->drvp->drive_flags |= ATA_DRIVE_RESET;
1193 mutex_exit(&wd->sc_lock);
1194 }
1195 errstring = readdisklabel(MAKEWDDEV(0, device_unit(wd->sc_dev),
1196 RAW_PART), wdstrategy, lp, wd->sc_dk.dk_cpulabel);
1197 }
1198 if (errstring) {
1199 aprint_error_dev(wd->sc_dev, "%s\n", errstring);
1200 return;
1201 }
1202
1203 if (wd->drvp->state > RESET) {
1204 mutex_enter(&wd->sc_lock);
1205 wd->drvp->drive_flags |= ATA_DRIVE_RESET;
1206 mutex_exit(&wd->sc_lock);
1207 }
1208 #ifdef HAS_BAD144_HANDLING
1209 if ((lp->d_flags & D_BADSECT) != 0)
1210 bad144intern(wd);
1211 #endif
1212 }
1213
1214 void
1215 wdperror(const struct wd_softc *wd, struct ata_xfer *xfer)
1216 {
1217 static const char *const errstr0_3[] = {"address mark not found",
1218 "track 0 not found", "aborted command", "media change requested",
1219 "id not found", "media changed", "uncorrectable data error",
1220 "bad block detected"};
1221 static const char *const errstr4_5[] = {
1222 "obsolete (address mark not found)",
1223 "no media/write protected", "aborted command",
1224 "media change requested", "id not found", "media changed",
1225 "uncorrectable data error", "interface CRC error"};
1226 const char *const *errstr;
1227 int i;
1228 const char *sep = "";
1229
1230 const char *devname = device_xname(wd->sc_dev);
1231 struct ata_drive_datas *drvp = wd->drvp;
1232 int errno = xfer->c_bio.r_error;
1233
1234 if (drvp->ata_vers >= 4)
1235 errstr = errstr4_5;
1236 else
1237 errstr = errstr0_3;
1238
1239 printf("%s: (", devname);
1240
1241 if (errno == 0)
1242 printf("error not notified");
1243
1244 for (i = 0; i < 8; i++) {
1245 if (errno & (1 << i)) {
1246 printf("%s%s", sep, errstr[i]);
1247 sep = ", ";
1248 }
1249 }
1250 printf(")\n");
1251 }
1252
1253 int
1254 wdioctl(dev_t dev, u_long xfer, void *addr, int flag, struct lwp *l)
1255 {
1256 struct wd_softc *wd =
1257 device_lookup_private(&wd_cd, WDUNIT(dev));
1258 int error;
1259 #ifdef __HAVE_OLD_DISKLABEL
1260 struct disklabel *newlabel = NULL;
1261 #endif
1262
1263 ATADEBUG_PRINT(("wdioctl\n"), DEBUG_FUNCS);
1264
1265 if ((wd->sc_flags & WDF_LOADED) == 0)
1266 return EIO;
1267
1268 error = disk_ioctl(&wd->sc_dk, dev, xfer, addr, flag, l);
1269 if (error != EPASSTHROUGH)
1270 return error;
1271
1272 error = 0;
1273 switch (xfer) {
1274 #ifdef HAS_BAD144_HANDLING
1275 case DIOCSBAD:
1276 if ((flag & FWRITE) == 0)
1277 return EBADF;
1278 wd->sc_dk.dk_cpulabel->bad = *(struct dkbad *)addr;
1279 wd->sc_dk.dk_label->d_flags |= D_BADSECT;
1280 bad144intern(wd);
1281 return 0;
1282 #endif
1283 #ifdef WD_SOFTBADSECT
1284 case DIOCBSLIST :
1285 {
1286 uint32_t count, missing, skip;
1287 struct disk_badsecinfo dbsi;
1288 struct disk_badsectors *dbs;
1289 size_t available;
1290 uint8_t *laddr;
1291
1292 dbsi = *(struct disk_badsecinfo *)addr;
1293 missing = wd->sc_bscount;
1294 count = 0;
1295 available = dbsi.dbsi_bufsize;
1296 skip = dbsi.dbsi_skip;
1297 laddr = (uint8_t *)dbsi.dbsi_buffer;
1298
1299 /*
1300 * We start this loop with the expectation that all of the
1301 * entries will be missed and decrement this counter each
1302 * time we either skip over one (already copied out) or
1303 * we actually copy it back to user space. The structs
1304 * holding the bad sector information are copied directly
1305 * back to user space whilst the summary is returned via
1306 * the struct passed in via the ioctl.
1307 */
1308 SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next) {
1309 if (skip > 0) {
1310 missing--;
1311 skip--;
1312 continue;
1313 }
1314 if (available < sizeof(*dbs))
1315 break;
1316 available -= sizeof(*dbs);
1317 copyout(dbs, laddr, sizeof(*dbs));
1318 laddr += sizeof(*dbs);
1319 missing--;
1320 count++;
1321 }
1322 dbsi.dbsi_left = missing;
1323 dbsi.dbsi_copied = count;
1324 *(struct disk_badsecinfo *)addr = dbsi;
1325 return 0;
1326 }
1327
1328 case DIOCBSFLUSH :
1329 /* Clean out the bad sector list */
1330 while (!SLIST_EMPTY(&wd->sc_bslist)) {
1331 void *head = SLIST_FIRST(&wd->sc_bslist);
1332 SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
1333 free(head, M_TEMP);
1334 }
1335 wd->sc_bscount = 0;
1336 return 0;
1337 #endif
1338
1339 case DIOCWDINFO:
1340 case DIOCSDINFO:
1341 #ifdef __HAVE_OLD_DISKLABEL
1342 case ODIOCWDINFO:
1343 case ODIOCSDINFO:
1344 #endif
1345 {
1346 struct disklabel *lp;
1347
1348 if ((flag & FWRITE) == 0)
1349 return EBADF;
1350
1351 #ifdef __HAVE_OLD_DISKLABEL
1352 if (xfer == ODIOCSDINFO || xfer == ODIOCWDINFO) {
1353 newlabel = malloc(sizeof *newlabel, M_TEMP,
1354 M_WAITOK | M_ZERO);
1355 if (newlabel == NULL)
1356 return EIO;
1357 memcpy(newlabel, addr, sizeof (struct olddisklabel));
1358 lp = newlabel;
1359 } else
1360 #endif
1361 lp = (struct disklabel *)addr;
1362
1363 mutex_enter(&wd->sc_dk.dk_openlock);
1364 wd->sc_flags |= WDF_LABELLING;
1365
1366 error = setdisklabel(wd->sc_dk.dk_label,
1367 lp, /*wd->sc_dk.dk_openmask : */0,
1368 wd->sc_dk.dk_cpulabel);
1369 if (error == 0) {
1370 if (wd->drvp->state > RESET) {
1371 mutex_enter(&wd->sc_lock);
1372 wd->drvp->drive_flags |= ATA_DRIVE_RESET;
1373 mutex_exit(&wd->sc_lock);
1374 }
1375 if (xfer == DIOCWDINFO
1376 #ifdef __HAVE_OLD_DISKLABEL
1377 || xfer == ODIOCWDINFO
1378 #endif
1379 )
1380 error = writedisklabel(WDLABELDEV(dev),
1381 wdstrategy, wd->sc_dk.dk_label,
1382 wd->sc_dk.dk_cpulabel);
1383 }
1384
1385 wd->sc_flags &= ~WDF_LABELLING;
1386 mutex_exit(&wd->sc_dk.dk_openlock);
1387 #ifdef __HAVE_OLD_DISKLABEL
1388 if (newlabel != NULL)
1389 free(newlabel, M_TEMP);
1390 #endif
1391 return error;
1392 }
1393
1394 case DIOCKLABEL:
1395 if (*(int *)addr)
1396 wd->sc_flags |= WDF_KLABEL;
1397 else
1398 wd->sc_flags &= ~WDF_KLABEL;
1399 return 0;
1400
1401 case DIOCWLABEL:
1402 if ((flag & FWRITE) == 0)
1403 return EBADF;
1404 if (*(int *)addr)
1405 wd->sc_flags |= WDF_WLABEL;
1406 else
1407 wd->sc_flags &= ~WDF_WLABEL;
1408 return 0;
1409
1410 case DIOCGDEFLABEL:
1411 wdgetdefaultlabel(wd, (struct disklabel *)addr);
1412 return 0;
1413 #ifdef __HAVE_OLD_DISKLABEL
1414 case ODIOCGDEFLABEL:
1415 newlabel = malloc(sizeof *newlabel, M_TEMP, M_WAITOK);
1416 if (newlabel == NULL)
1417 return EIO;
1418 wdgetdefaultlabel(wd, newlabel);
1419 if (newlabel->d_npartitions <= OLDMAXPARTITIONS)
1420 memcpy(addr, &newlabel, sizeof (struct olddisklabel));
1421 else
1422 error = ENOTTY;
1423 free(newlabel, M_TEMP);
1424 return error;
1425 #endif
1426
1427 #ifdef notyet
1428 case DIOCWFORMAT:
1429 if ((flag & FWRITE) == 0)
1430 return EBADF;
1431 {
1432 register struct format_op *fop;
1433 struct iovec aiov;
1434 struct uio auio;
1435
1436 fop = (struct format_op *)addr;
1437 aiov.iov_base = fop->df_buf;
1438 aiov.iov_len = fop->df_count;
1439 auio.uio_iov = &aiov;
1440 auio.uio_iovcnt = 1;
1441 auio.uio_resid = fop->df_count;
1442 auio.uio_offset =
1443 fop->df_startblk * wd->sc_dk.dk_label->d_secsize;
1444 auio.uio_vmspace = l->l_proc->p_vmspace;
1445 error = physio(wdformat, NULL, dev, B_WRITE, wdminphys,
1446 &auio);
1447 fop->df_count -= auio.uio_resid;
1448 fop->df_reg[0] = wdc->sc_status;
1449 fop->df_reg[1] = wdc->sc_error;
1450 return error;
1451 }
1452 #endif
1453 case DIOCGCACHE:
1454 return wd_getcache(wd, (int *)addr);
1455
1456 case DIOCSCACHE:
1457 return wd_setcache(wd, *(int *)addr);
1458
1459 case DIOCCACHESYNC:
1460 return wd_flushcache(wd, AT_WAIT, true);
1461
1462 case ATAIOCCOMMAND:
1463 /*
1464 * Make sure this command is (relatively) safe first
1465 */
1466 if ((((atareq_t *) addr)->flags & ATACMD_READ) == 0 &&
1467 (flag & FWRITE) == 0)
1468 return (EBADF);
1469 {
1470 struct wd_ioctl *wi;
1471 atareq_t *atareq = (atareq_t *) addr;
1472 int error1;
1473
1474 wi = wi_get(wd);
1475 wi->wi_atareq = *atareq;
1476
1477 if (atareq->datalen && atareq->flags &
1478 (ATACMD_READ | ATACMD_WRITE)) {
1479 void *tbuf;
1480 if (atareq->datalen < DEV_BSIZE
1481 && atareq->command == WDCC_IDENTIFY) {
1482 tbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1483 wi->wi_iov.iov_base = tbuf;
1484 wi->wi_iov.iov_len = DEV_BSIZE;
1485 UIO_SETUP_SYSSPACE(&wi->wi_uio);
1486 } else {
1487 tbuf = NULL;
1488 wi->wi_iov.iov_base = atareq->databuf;
1489 wi->wi_iov.iov_len = atareq->datalen;
1490 wi->wi_uio.uio_vmspace = l->l_proc->p_vmspace;
1491 }
1492 wi->wi_uio.uio_iov = &wi->wi_iov;
1493 wi->wi_uio.uio_iovcnt = 1;
1494 wi->wi_uio.uio_resid = atareq->datalen;
1495 wi->wi_uio.uio_offset = 0;
1496 wi->wi_uio.uio_rw =
1497 (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE;
1498 error1 = physio(wdioctlstrategy, &wi->wi_bp, dev,
1499 (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE,
1500 wdminphys, &wi->wi_uio);
1501 if (tbuf != NULL && error1 == 0) {
1502 error1 = copyout(tbuf, atareq->databuf,
1503 atareq->datalen);
1504 free(tbuf, M_TEMP);
1505 }
1506 } else {
1507 /* No need to call physio if we don't have any
1508 user data */
1509 wi->wi_bp.b_flags = 0;
1510 wi->wi_bp.b_data = 0;
1511 wi->wi_bp.b_bcount = 0;
1512 wi->wi_bp.b_dev = dev;
1513 wi->wi_bp.b_proc = l->l_proc;
1514 wdioctlstrategy(&wi->wi_bp);
1515 error1 = wi->wi_bp.b_error;
1516 }
1517 *atareq = wi->wi_atareq;
1518 wi_free(wi);
1519 return(error1);
1520 }
1521
1522 case DIOCGSTRATEGY:
1523 {
1524 struct disk_strategy *dks = (void *)addr;
1525
1526 mutex_enter(&wd->sc_lock);
1527 strlcpy(dks->dks_name, bufq_getstrategyname(wd->sc_q),
1528 sizeof(dks->dks_name));
1529 mutex_exit(&wd->sc_lock);
1530 dks->dks_paramlen = 0;
1531
1532 return 0;
1533 }
1534
1535 case DIOCSSTRATEGY:
1536 {
1537 struct disk_strategy *dks = (void *)addr;
1538 struct bufq_state *new;
1539 struct bufq_state *old;
1540
1541 if ((flag & FWRITE) == 0) {
1542 return EBADF;
1543 }
1544 if (dks->dks_param != NULL) {
1545 return EINVAL;
1546 }
1547 dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
1548 error = bufq_alloc(&new, dks->dks_name,
1549 BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
1550 if (error) {
1551 return error;
1552 }
1553 mutex_enter(&wd->sc_lock);
1554 old = wd->sc_q;
1555 bufq_move(new, old);
1556 wd->sc_q = new;
1557 mutex_exit(&wd->sc_lock);
1558 bufq_free(old);
1559
1560 return 0;
1561 }
1562
1563 default:
1564 return ENOTTY;
1565 }
1566
1567 #ifdef DIAGNOSTIC
1568 panic("wdioctl: impossible");
1569 #endif
1570 }
1571
1572 static int
1573 wddiscard(dev_t dev, off_t pos, off_t len)
1574 {
1575 struct wd_softc *wd = device_lookup_private(&wd_cd, WDUNIT(dev));
1576 daddr_t bno;
1577 long size, done;
1578 long maxatonce, amount;
1579 int result;
1580
1581 if (!(wd->sc_params.atap_ata_major & WDC_VER_ATA7)
1582 || !(wd->sc_params.support_dsm & ATA_SUPPORT_DSM_TRIM)) {
1583 /* not supported; ignore request */
1584 ATADEBUG_PRINT(("wddiscard (unsupported)\n"), DEBUG_FUNCS);
1585 return 0;
1586 }
1587 maxatonce = 0xffff; /*wd->sc_params.max_dsm_blocks*/
1588
1589 ATADEBUG_PRINT(("wddiscard\n"), DEBUG_FUNCS);
1590
1591 if ((wd->sc_flags & WDF_LOADED) == 0)
1592 return EIO;
1593
1594 /* round the start up and the end down */
1595 bno = (pos + wd->sc_blksize - 1) / wd->sc_blksize;
1596 size = ((pos + len) / wd->sc_blksize) - bno;
1597
1598 done = 0;
1599 while (done < size) {
1600 amount = size - done;
1601 if (amount > maxatonce) {
1602 amount = maxatonce;
1603 }
1604 result = wd_trim(wd, WDPART(dev), bno + done, amount);
1605 if (result) {
1606 return result;
1607 }
1608 done += amount;
1609 }
1610 return 0;
1611 }
1612
1613 #ifdef B_FORMAT
1614 int
1615 wdformat(struct buf *bp)
1616 {
1617
1618 bp->b_flags |= B_FORMAT;
1619 return wdstrategy(bp);
1620 }
1621 #endif
1622
1623 int
1624 wdsize(dev_t dev)
1625 {
1626 struct wd_softc *wd;
1627 int part, omask;
1628 int size;
1629
1630 ATADEBUG_PRINT(("wdsize\n"), DEBUG_FUNCS);
1631
1632 wd = device_lookup_private(&wd_cd, WDUNIT(dev));
1633 if (wd == NULL)
1634 return (-1);
1635
1636 part = WDPART(dev);
1637 omask = wd->sc_dk.dk_openmask & (1 << part);
1638
1639 if (omask == 0 && wdopen(dev, 0, S_IFBLK, NULL) != 0)
1640 return (-1);
1641 if (wd->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
1642 size = -1;
1643 else
1644 size = wd->sc_dk.dk_label->d_partitions[part].p_size *
1645 (wd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
1646 if (omask == 0 && wdclose(dev, 0, S_IFBLK, NULL) != 0)
1647 return (-1);
1648 return (size);
1649 }
1650
1651 /*
1652 * Dump core after a system crash.
1653 */
1654 int
1655 wddump(dev_t dev, daddr_t blkno, void *va, size_t size)
1656 {
1657 struct wd_softc *wd; /* disk unit to do the I/O */
1658 struct disklabel *lp; /* disk's disklabel */
1659 int part, err;
1660 int nblks; /* total number of sectors left to write */
1661 struct ata_xfer *xfer;
1662
1663 /* Check if recursive dump; if so, punt. */
1664 if (wddoingadump)
1665 return EFAULT;
1666 wddoingadump = 1;
1667
1668 wd = device_lookup_private(&wd_cd, WDUNIT(dev));
1669 if (wd == NULL)
1670 return (ENXIO);
1671
1672 part = WDPART(dev);
1673
1674 /* Convert to disk sectors. Request must be a multiple of size. */
1675 lp = wd->sc_dk.dk_label;
1676 if ((size % lp->d_secsize) != 0)
1677 return EFAULT;
1678 nblks = size / lp->d_secsize;
1679 blkno = blkno / (lp->d_secsize / DEV_BSIZE);
1680
1681 /* Check transfer bounds against partition size. */
1682 if ((blkno < 0) || ((blkno + nblks) > lp->d_partitions[part].p_size))
1683 return EINVAL;
1684
1685 /* Offset block number to start of partition. */
1686 blkno += lp->d_partitions[part].p_offset;
1687
1688 /* Recalibrate, if first dump transfer. */
1689 if (wddumprecalibrated == 0) {
1690 wddumprecalibrated = 1;
1691 (*wd->atabus->ata_reset_drive)(wd->drvp,
1692 AT_POLL | AT_RST_EMERG, NULL);
1693 wd->drvp->state = RESET;
1694 }
1695
1696 xfer = ata_get_xfer_ext(wd->drvp->chnl_softc, 0, 0);
1697 if (xfer == NULL) {
1698 printf("%s: no xfer\n", __func__);
1699 return EAGAIN;
1700 }
1701
1702 xfer->c_bio.blkno = blkno;
1703 xfer->c_bio.flags = ATA_POLL;
1704 if (wd->sc_flags & WDF_LBA48 &&
1705 (xfer->c_bio.blkno + nblks) > wd->sc_capacity28)
1706 xfer->c_bio.flags |= ATA_LBA48;
1707 if (wd->sc_flags & WDF_LBA)
1708 xfer->c_bio.flags |= ATA_LBA;
1709 xfer->c_bio.bcount = nblks * lp->d_secsize;
1710 xfer->c_bio.databuf = va;
1711 #ifndef WD_DUMP_NOT_TRUSTED
1712 switch (err = wd->atabus->ata_bio(wd->drvp, xfer)) {
1713 case ATACMD_TRY_AGAIN:
1714 panic("wddump: try again");
1715 break;
1716 case ATACMD_QUEUED:
1717 panic("wddump: polled command has been queued");
1718 break;
1719 case ATACMD_COMPLETE:
1720 break;
1721 default:
1722 panic("wddump: unknown atacmd code %d", err);
1723 }
1724 switch(err = xfer->c_bio.error) {
1725 case TIMEOUT:
1726 printf("wddump: device timed out");
1727 err = EIO;
1728 break;
1729 case ERR_DF:
1730 printf("wddump: drive fault");
1731 err = EIO;
1732 break;
1733 case ERR_DMA:
1734 printf("wddump: DMA error");
1735 err = EIO;
1736 break;
1737 case ERROR:
1738 printf("wddump: ");
1739 wdperror(wd, xfer);
1740 err = EIO;
1741 break;
1742 case NOERROR:
1743 err = 0;
1744 break;
1745 default:
1746 panic("wddump: unknown error type %d", err);
1747 }
1748
1749 if (err != 0) {
1750 printf("\n");
1751 return err;
1752 }
1753 #else /* WD_DUMP_NOT_TRUSTED */
1754 /* Let's just talk about this first... */
1755 printf("wd%d: dump addr 0x%x, cylin %d, head %d, sector %d\n",
1756 unit, va, cylin, head, sector);
1757 delay(500 * 1000); /* half a second */
1758 #endif
1759
1760 wddoingadump = 0;
1761 return 0;
1762 }
1763
1764 #ifdef HAS_BAD144_HANDLING
1765 /*
1766 * Internalize the bad sector table.
1767 */
1768 void
1769 bad144intern(struct wd_softc *wd)
1770 {
1771 struct dkbad *bt = &wd->sc_dk.dk_cpulabel->bad;
1772 struct disklabel *lp = wd->sc_dk.dk_label;
1773 int i = 0;
1774
1775 ATADEBUG_PRINT(("bad144intern\n"), DEBUG_XFERS);
1776
1777 for (; i < NBT_BAD; i++) {
1778 if (bt->bt_bad[i].bt_cyl == 0xffff)
1779 break;
1780 wd->drvp->badsect[i] =
1781 bt->bt_bad[i].bt_cyl * lp->d_secpercyl +
1782 (bt->bt_bad[i].bt_trksec >> 8) * lp->d_nsectors +
1783 (bt->bt_bad[i].bt_trksec & 0xff);
1784 }
1785 for (; i < NBT_BAD+1; i++)
1786 wd->drvp->badsect[i] = -1;
1787 }
1788 #endif
1789
1790 static void
1791 wd_params_to_properties(struct wd_softc *wd)
1792 {
1793 struct disk_geom *dg = &wd->sc_dk.dk_geom;
1794
1795 memset(dg, 0, sizeof(*dg));
1796
1797 dg->dg_secperunit = wd->sc_capacity;
1798 dg->dg_secsize = wd->sc_blksize;
1799 dg->dg_nsectors = wd->sc_params.atap_sectors;
1800 dg->dg_ntracks = wd->sc_params.atap_heads;
1801 if ((wd->sc_flags & WDF_LBA) == 0)
1802 dg->dg_ncylinders = wd->sc_params.atap_cylinders;
1803
1804 /* XXX Should have a case for ATA here, too. */
1805 const char *cp = strcmp(wd->sc_params.atap_model, "ST506") ?
1806 "ST506" : "ESDI";
1807
1808 disk_set_info(wd->sc_dev, &wd->sc_dk, cp);
1809 }
1810
1811 int
1812 wd_get_params(struct wd_softc *wd, uint8_t flags, struct ataparams *params)
1813 {
1814
1815 switch (wd->atabus->ata_get_params(wd->drvp, flags, params)) {
1816 case CMD_AGAIN:
1817 return 1;
1818 case CMD_ERR:
1819 if (wd->drvp->drive_type != ATA_DRIVET_OLD)
1820 return 1;
1821 /*
1822 * We `know' there's a drive here; just assume it's old.
1823 * This geometry is only used to read the MBR and print a
1824 * (false) attach message.
1825 */
1826 strncpy(params->atap_model, "ST506",
1827 sizeof params->atap_model);
1828 params->atap_config = ATA_CFG_FIXED;
1829 params->atap_cylinders = 1024;
1830 params->atap_heads = 8;
1831 params->atap_sectors = 17;
1832 params->atap_multi = 1;
1833 params->atap_capabilities1 = params->atap_capabilities2 = 0;
1834 wd->drvp->ata_vers = -1; /* Mark it as pre-ATA */
1835 /* FALLTHROUGH */
1836 case CMD_OK:
1837 return 0;
1838 default:
1839 panic("wd_get_params: bad return code from ata_get_params");
1840 /* NOTREACHED */
1841 }
1842 }
1843
1844 int
1845 wd_getcache(struct wd_softc *wd, int *bitsp)
1846 {
1847 struct ataparams params;
1848
1849 if (wd_get_params(wd, AT_WAIT, ¶ms) != 0)
1850 return EIO;
1851 if (params.atap_cmd_set1 == 0x0000 ||
1852 params.atap_cmd_set1 == 0xffff ||
1853 (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0) {
1854 *bitsp = 0;
1855 return 0;
1856 }
1857 *bitsp = DKCACHE_WCHANGE | DKCACHE_READ;
1858 if (params.atap_cmd1_en & WDC_CMD1_CACHE)
1859 *bitsp |= DKCACHE_WRITE;
1860
1861 if (WD_USE_NCQ(wd) || (wd->drvp->drive_flags & ATA_DRIVE_WFUA))
1862 *bitsp |= DKCACHE_FUA;
1863
1864 return 0;
1865 }
1866
1867 const char at_errbits[] = "\20\10ERROR\11TIMEOU\12DF";
1868
1869 int
1870 wd_setcache(struct wd_softc *wd, int bits)
1871 {
1872 struct ataparams params;
1873 struct ata_xfer *xfer;
1874 int error;
1875
1876 if (wd_get_params(wd, AT_WAIT, ¶ms) != 0)
1877 return EIO;
1878
1879 if (params.atap_cmd_set1 == 0x0000 ||
1880 params.atap_cmd_set1 == 0xffff ||
1881 (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0)
1882 return EOPNOTSUPP;
1883
1884 if ((bits & DKCACHE_READ) == 0 ||
1885 (bits & DKCACHE_SAVE) != 0)
1886 return EOPNOTSUPP;
1887
1888 xfer = ata_get_xfer(wd->drvp->chnl_softc);
1889 if (xfer == NULL)
1890 return EINTR;
1891
1892 xfer->c_ata_c.r_command = SET_FEATURES;
1893 xfer->c_ata_c.r_st_bmask = 0;
1894 xfer->c_ata_c.r_st_pmask = 0;
1895 xfer->c_ata_c.timeout = 30000; /* 30s timeout */
1896 xfer->c_ata_c.flags = AT_WAIT;
1897 if (bits & DKCACHE_WRITE)
1898 xfer->c_ata_c.r_features = WDSF_WRITE_CACHE_EN;
1899 else
1900 xfer->c_ata_c.r_features = WDSF_WRITE_CACHE_DS;
1901 if (wd->atabus->ata_exec_command(wd->drvp, xfer) != ATACMD_COMPLETE) {
1902 aprint_error_dev(wd->sc_dev,
1903 "wd_setcache command not complete\n");
1904 error = EIO;
1905 goto out;
1906 }
1907
1908 if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1909 char sbuf[sizeof(at_errbits) + 64];
1910 snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
1911 aprint_error_dev(wd->sc_dev, "wd_setcache: status=%s\n", sbuf);
1912 error = EIO;
1913 goto out;
1914 }
1915
1916 error = 0;
1917
1918 out:
1919 ata_free_xfer(wd->drvp->chnl_softc, xfer);
1920 ata_channel_start(wd->drvp->chnl_softc, wd->drvp->drive);
1921 return error;
1922 }
1923
1924 static int
1925 wd_standby(struct wd_softc *wd, int flags)
1926 {
1927 struct ata_xfer *xfer;
1928 int error;
1929
1930 xfer = ata_get_xfer(wd->drvp->chnl_softc);
1931 if (xfer == NULL)
1932 return EINTR;
1933
1934 xfer->c_ata_c.r_command = WDCC_STANDBY_IMMED;
1935 xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
1936 xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
1937 xfer->c_ata_c.flags = flags;
1938 xfer->c_ata_c.timeout = 30000; /* 30s timeout */
1939 if (wd->atabus->ata_exec_command(wd->drvp, xfer) != ATACMD_COMPLETE) {
1940 aprint_error_dev(wd->sc_dev,
1941 "standby immediate command didn't complete\n");
1942 error = EIO;
1943 goto out;
1944 }
1945 if (xfer->c_ata_c.flags & AT_ERROR) {
1946 if (xfer->c_ata_c.r_error == WDCE_ABRT) {
1947 /* command not supported */
1948 error = ENODEV;
1949 goto out;
1950 }
1951 }
1952 if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1953 char sbuf[sizeof(at_errbits) + 64];
1954 snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
1955 aprint_error_dev(wd->sc_dev, "wd_standby: status=%s\n", sbuf);
1956 error = EIO;
1957 goto out;
1958 }
1959 error = 0;
1960
1961 out:
1962 ata_free_xfer(wd->drvp->chnl_softc, xfer);
1963 /* drive is supposed to go idle, do not call ata_channel_start() */
1964 return error;
1965 }
1966
1967 int
1968 wd_flushcache(struct wd_softc *wd, int flags, bool start)
1969 {
1970 struct ata_xfer *xfer;
1971 int error;
1972
1973 /*
1974 * WDCC_FLUSHCACHE is here since ATA-4, but some drives report
1975 * only ATA-2 and still support it.
1976 */
1977 if (wd->drvp->ata_vers < 4 &&
1978 ((wd->sc_params.atap_cmd_set2 & WDC_CMD2_FC) == 0 ||
1979 wd->sc_params.atap_cmd_set2 == 0xffff))
1980 return ENODEV;
1981
1982 mutex_enter(&wd->sc_lock);
1983 SET(wd->sc_flags, WDF_FLUSH_PEND);
1984 mutex_exit(&wd->sc_lock);
1985
1986 xfer = ata_get_xfer(wd->drvp->chnl_softc);
1987
1988 mutex_enter(&wd->sc_lock);
1989 CLR(wd->sc_flags, WDF_FLUSH_PEND);
1990 mutex_exit(&wd->sc_lock);
1991
1992 if (xfer == NULL) {
1993 error = EINTR;
1994 goto out;
1995 }
1996
1997 if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0 &&
1998 (wd->sc_params.atap_cmd2_en & ATA_CMD2_FCE) != 0) {
1999 xfer->c_ata_c.r_command = WDCC_FLUSHCACHE_EXT;
2000 flags |= AT_LBA48;
2001 } else
2002 xfer->c_ata_c.r_command = WDCC_FLUSHCACHE;
2003 xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
2004 xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
2005 xfer->c_ata_c.flags = flags | AT_READREG;
2006 xfer->c_ata_c.timeout = 300000; /* 5m timeout */
2007 if (wd->atabus->ata_exec_command(wd->drvp, xfer) != ATACMD_COMPLETE) {
2008 aprint_error_dev(wd->sc_dev,
2009 "flush cache command didn't complete\n");
2010 error = EIO;
2011 goto out_xfer;
2012 }
2013 if (xfer->c_ata_c.flags & AT_ERROR) {
2014 if (xfer->c_ata_c.r_error == WDCE_ABRT) {
2015 /* command not supported */
2016 error = ENODEV;
2017 goto out_xfer;
2018 }
2019 }
2020 if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
2021 char sbuf[sizeof(at_errbits) + 64];
2022 snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
2023 aprint_error_dev(wd->sc_dev, "wd_flushcache: status=%s\n",
2024 sbuf);
2025 error = EIO;
2026 goto out_xfer;
2027 }
2028 error = 0;
2029
2030 out_xfer:
2031 ata_free_xfer(wd->drvp->chnl_softc, xfer);
2032
2033 out:
2034 /* kick queue processing blocked while waiting for flush xfer */
2035 if (start)
2036 ata_channel_start(wd->drvp->chnl_softc, wd->drvp->drive);
2037
2038 return error;
2039 }
2040
2041 int
2042 wd_trim(struct wd_softc *wd, int part, daddr_t bno, long size)
2043 {
2044 struct ata_xfer *xfer;
2045 int error;
2046 unsigned char *req;
2047
2048 if (part != RAW_PART)
2049 bno += wd->sc_dk.dk_label->d_partitions[part].p_offset;;
2050
2051 xfer = ata_get_xfer(wd->drvp->chnl_softc);
2052 if (xfer == NULL)
2053 return EINTR;
2054
2055 req = kmem_zalloc(512, KM_SLEEP);
2056 req[0] = bno & 0xff;
2057 req[1] = (bno >> 8) & 0xff;
2058 req[2] = (bno >> 16) & 0xff;
2059 req[3] = (bno >> 24) & 0xff;
2060 req[4] = (bno >> 32) & 0xff;
2061 req[5] = (bno >> 40) & 0xff;
2062 req[6] = size & 0xff;
2063 req[7] = (size >> 8) & 0xff;
2064
2065 xfer->c_ata_c.r_command = ATA_DATA_SET_MANAGEMENT;
2066 xfer->c_ata_c.r_count = 1;
2067 xfer->c_ata_c.r_features = ATA_SUPPORT_DSM_TRIM;
2068 xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
2069 xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
2070 xfer->c_ata_c.timeout = 30000; /* 30s timeout */
2071 xfer->c_ata_c.data = req;
2072 xfer->c_ata_c.bcount = 512;
2073 xfer->c_ata_c.flags |= AT_WRITE | AT_WAIT;
2074 if (wd->atabus->ata_exec_command(wd->drvp, xfer) != ATACMD_COMPLETE) {
2075 aprint_error_dev(wd->sc_dev,
2076 "trim command didn't complete\n");
2077 kmem_free(req, 512);
2078 error = EIO;
2079 goto out;
2080 }
2081 kmem_free(req, 512);
2082 if (xfer->c_ata_c.flags & AT_ERROR) {
2083 if (xfer->c_ata_c.r_error == WDCE_ABRT) {
2084 /* command not supported */
2085 error = ENODEV;
2086 goto out;
2087 }
2088 }
2089 if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
2090 char sbuf[sizeof(at_errbits) + 64];
2091 snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
2092 aprint_error_dev(wd->sc_dev, "wd_trim: status=%s\n",
2093 sbuf);
2094 error = EIO;
2095 goto out;
2096 }
2097 error = 0;
2098
2099 out:
2100 ata_free_xfer(wd->drvp->chnl_softc, xfer);
2101 ata_channel_start(wd->drvp->chnl_softc, wd->drvp->drive);
2102 return error;
2103 }
2104
2105 bool
2106 wd_shutdown(device_t dev, int how)
2107 {
2108 struct wd_softc *wd = device_private(dev);
2109
2110 /* the adapter needs to be enabled */
2111 if (wd->atabus->ata_addref(wd->drvp))
2112 return true; /* no need to complain */
2113
2114 wd_flushcache(wd, AT_POLL, false);
2115 if ((how & RB_POWERDOWN) == RB_POWERDOWN)
2116 wd_standby(wd, AT_POLL);
2117 return true;
2118 }
2119
2120 /*
2121 * Allocate space for a ioctl queue structure. Mostly taken from
2122 * scsipi_ioctl.c
2123 */
2124 struct wd_ioctl *
2125 wi_get(struct wd_softc *wd)
2126 {
2127 struct wd_ioctl *wi;
2128
2129 wi = malloc(sizeof(struct wd_ioctl), M_TEMP, M_WAITOK|M_ZERO);
2130 wi->wi_softc = wd;
2131 buf_init(&wi->wi_bp);
2132
2133 return (wi);
2134 }
2135
2136 /*
2137 * Free an ioctl structure and remove it from our list
2138 */
2139
2140 void
2141 wi_free(struct wd_ioctl *wi)
2142 {
2143 buf_destroy(&wi->wi_bp);
2144 free(wi, M_TEMP);
2145 }
2146
2147 /*
2148 * Find a wd_ioctl structure based on the struct buf.
2149 */
2150
2151 struct wd_ioctl *
2152 wi_find(struct buf *bp)
2153 {
2154 return container_of(bp, struct wd_ioctl, wi_bp);
2155 }
2156
2157 static uint
2158 wi_sector_size(const struct wd_ioctl * const wi)
2159 {
2160 switch (wi->wi_atareq.command) {
2161 case WDCC_READ:
2162 case WDCC_WRITE:
2163 case WDCC_READMULTI:
2164 case WDCC_WRITEMULTI:
2165 case WDCC_READDMA:
2166 case WDCC_WRITEDMA:
2167 case WDCC_READ_EXT:
2168 case WDCC_WRITE_EXT:
2169 case WDCC_READMULTI_EXT:
2170 case WDCC_WRITEMULTI_EXT:
2171 case WDCC_READDMA_EXT:
2172 case WDCC_WRITEDMA_EXT:
2173 case WDCC_READ_FPDMA_QUEUED:
2174 case WDCC_WRITE_FPDMA_QUEUED:
2175 return wi->wi_softc->sc_blksize;
2176 default:
2177 return 512;
2178 }
2179 }
2180
2181 /*
2182 * Ioctl pseudo strategy routine
2183 *
2184 * This is mostly stolen from scsipi_ioctl.c:scsistrategy(). What
2185 * happens here is:
2186 *
2187 * - wdioctl() queues a wd_ioctl structure.
2188 *
2189 * - wdioctl() calls physio/wdioctlstrategy based on whether or not
2190 * user space I/O is required. If physio() is called, physio() eventually
2191 * calls wdioctlstrategy().
2192 *
2193 * - In either case, wdioctlstrategy() calls wd->atabus->ata_exec_command()
2194 * to perform the actual command
2195 *
2196 * The reason for the use of the pseudo strategy routine is because
2197 * when doing I/O to/from user space, physio _really_ wants to be in
2198 * the loop. We could put the entire buffer into the ioctl request
2199 * structure, but that won't scale if we want to do things like download
2200 * microcode.
2201 */
2202
2203 void
2204 wdioctlstrategy(struct buf *bp)
2205 {
2206 struct wd_ioctl *wi;
2207 struct ata_xfer *xfer;
2208 int error = 0;
2209
2210 wi = wi_find(bp);
2211 if (wi == NULL) {
2212 printf("wdioctlstrategy: "
2213 "No matching ioctl request found in queue\n");
2214 error = EINVAL;
2215 goto out2;
2216 }
2217
2218 xfer = ata_get_xfer(wi->wi_softc->drvp->chnl_softc);
2219 if (xfer == NULL) {
2220 error = EINTR;
2221 goto out2;
2222 }
2223
2224 /*
2225 * Abort if physio broke up the transfer
2226 */
2227
2228 if (bp->b_bcount != wi->wi_atareq.datalen) {
2229 printf("physio split wd ioctl request... cannot proceed\n");
2230 error = EIO;
2231 goto out;
2232 }
2233
2234 /*
2235 * Abort if we didn't get a buffer size that was a multiple of
2236 * our sector size (or overflows CHS/LBA28 sector count)
2237 */
2238
2239 if ((bp->b_bcount % wi_sector_size(wi)) != 0 ||
2240 (bp->b_bcount / wi_sector_size(wi)) >=
2241 (1 << NBBY)) {
2242 error = EINVAL;
2243 goto out;
2244 }
2245
2246 /*
2247 * Make sure a timeout was supplied in the ioctl request
2248 */
2249
2250 if (wi->wi_atareq.timeout == 0) {
2251 error = EINVAL;
2252 goto out;
2253 }
2254
2255 if (wi->wi_atareq.flags & ATACMD_READ)
2256 xfer->c_ata_c.flags |= AT_READ;
2257 else if (wi->wi_atareq.flags & ATACMD_WRITE)
2258 xfer->c_ata_c.flags |= AT_WRITE;
2259
2260 if (wi->wi_atareq.flags & ATACMD_READREG)
2261 xfer->c_ata_c.flags |= AT_READREG;
2262
2263 if ((wi->wi_atareq.flags & ATACMD_LBA) != 0)
2264 xfer->c_ata_c.flags |= AT_LBA;
2265
2266 xfer->c_ata_c.flags |= AT_WAIT;
2267
2268 xfer->c_ata_c.timeout = wi->wi_atareq.timeout;
2269 xfer->c_ata_c.r_command = wi->wi_atareq.command;
2270 xfer->c_ata_c.r_lba = ((wi->wi_atareq.head & 0x0f) << 24) |
2271 (wi->wi_atareq.cylinder << 8) |
2272 wi->wi_atareq.sec_num;
2273 xfer->c_ata_c.r_count = wi->wi_atareq.sec_count;
2274 xfer->c_ata_c.r_features = wi->wi_atareq.features;
2275 xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
2276 xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
2277 xfer->c_ata_c.data = wi->wi_bp.b_data;
2278 xfer->c_ata_c.bcount = wi->wi_bp.b_bcount;
2279
2280 if (wi->wi_softc->atabus->ata_exec_command(wi->wi_softc->drvp, xfer)
2281 != ATACMD_COMPLETE) {
2282 wi->wi_atareq.retsts = ATACMD_ERROR;
2283 error = EIO;
2284 goto out;
2285 }
2286
2287 if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
2288 if (xfer->c_ata_c.flags & AT_ERROR) {
2289 wi->wi_atareq.retsts = ATACMD_ERROR;
2290 wi->wi_atareq.error = xfer->c_ata_c.r_error;
2291 } else if (xfer->c_ata_c.flags & AT_DF)
2292 wi->wi_atareq.retsts = ATACMD_DF;
2293 else
2294 wi->wi_atareq.retsts = ATACMD_TIMEOUT;
2295 } else {
2296 wi->wi_atareq.retsts = ATACMD_OK;
2297 if (wi->wi_atareq.flags & ATACMD_READREG) {
2298 wi->wi_atareq.command = xfer->c_ata_c.r_status;
2299 wi->wi_atareq.features = xfer->c_ata_c.r_error;
2300 wi->wi_atareq.sec_count = xfer->c_ata_c.r_count;
2301 wi->wi_atareq.sec_num = xfer->c_ata_c.r_lba & 0xff;
2302 wi->wi_atareq.head = (xfer->c_ata_c.r_device & 0xf0) |
2303 ((xfer->c_ata_c.r_lba >> 24) & 0x0f);
2304 wi->wi_atareq.cylinder =
2305 (xfer->c_ata_c.r_lba >> 8) & 0xffff;
2306 wi->wi_atareq.error = xfer->c_ata_c.r_error;
2307 }
2308 }
2309
2310 out:
2311 ata_free_xfer(wi->wi_softc->drvp->chnl_softc, xfer);
2312 ata_channel_start(wi->wi_softc->drvp->chnl_softc,
2313 wi->wi_softc->drvp->drive);
2314 out2:
2315 bp->b_error = error;
2316 if (error)
2317 bp->b_resid = bp->b_bcount;
2318 biodone(bp);
2319 }
2320
2321 static void
2322 wd_sysctl_attach(struct wd_softc *wd)
2323 {
2324 const struct sysctlnode *node;
2325 int error;
2326
2327 /* sysctl set-up */
2328 if (sysctl_createv(&wd->nodelog, 0, NULL, &node,
2329 0, CTLTYPE_NODE, device_xname(wd->sc_dev),
2330 SYSCTL_DESCR("wd driver settings"),
2331 NULL, 0, NULL, 0,
2332 CTL_HW, CTL_CREATE, CTL_EOL) != 0) {
2333 aprint_error_dev(wd->sc_dev,
2334 "could not create %s.%s sysctl node\n",
2335 "hw", device_xname(wd->sc_dev));
2336 return;
2337 }
2338
2339 wd->drv_max_tags = ATA_MAX_OPENINGS;
2340 if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
2341 CTLFLAG_READWRITE, CTLTYPE_INT, "max_tags",
2342 SYSCTL_DESCR("max number of NCQ tags to use"),
2343 NULL, 0, &wd->drv_max_tags, 0,
2344 CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
2345 != 0) {
2346 aprint_error_dev(wd->sc_dev,
2347 "could not create %s.%s.max_tags sysctl - error %d\n",
2348 "hw", device_xname(wd->sc_dev), error);
2349 return;
2350 }
2351
2352 wd->drv_ncq = true;
2353 if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
2354 CTLFLAG_READWRITE, CTLTYPE_BOOL, "use_ncq",
2355 SYSCTL_DESCR("use NCQ if supported"),
2356 NULL, 0, &wd->drv_ncq, 0,
2357 CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
2358 != 0) {
2359 aprint_error_dev(wd->sc_dev,
2360 "could not create %s.%s.use_ncq sysctl - error %d\n",
2361 "hw", device_xname(wd->sc_dev), error);
2362 return;
2363 }
2364
2365 wd->drv_ncq_prio = true;
2366 if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
2367 CTLFLAG_READWRITE, CTLTYPE_BOOL, "use_ncq_prio",
2368 SYSCTL_DESCR("use NCQ PRIORITY if supported"),
2369 NULL, 0, &wd->drv_ncq_prio, 0,
2370 CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
2371 != 0) {
2372 aprint_error_dev(wd->sc_dev,
2373 "could not create %s.%s.use_ncq_prio sysctl - error %d\n",
2374 "hw", device_xname(wd->sc_dev), error);
2375 return;
2376 }
2377
2378 #ifdef WD_CHAOS_MONKEY
2379 wd->drv_chaos_freq = 0;
2380 if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
2381 CTLFLAG_READWRITE, CTLTYPE_INT, "chaos_freq",
2382 SYSCTL_DESCR("simulated bio read error rate"),
2383 NULL, 0, &wd->drv_chaos_freq, 0,
2384 CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
2385 != 0) {
2386 aprint_error_dev(wd->sc_dev,
2387 "could not create %s.%s.chaos_freq sysctl - error %d\n",
2388 "hw", device_xname(wd->sc_dev), error);
2389 return;
2390 }
2391
2392 wd->drv_chaos_cnt = 0;
2393 if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
2394 CTLFLAG_READONLY, CTLTYPE_INT, "chaos_cnt",
2395 SYSCTL_DESCR("number of processed bio reads"),
2396 NULL, 0, &wd->drv_chaos_cnt, 0,
2397 CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
2398 != 0) {
2399 aprint_error_dev(wd->sc_dev,
2400 "could not create %s.%s.chaos_cnt sysctl - error %d\n",
2401 "hw", device_xname(wd->sc_dev), error);
2402 return;
2403 }
2404 #endif
2405
2406 }
2407
2408 static void
2409 wd_sysctl_detach(struct wd_softc *wd)
2410 {
2411 sysctl_teardown(&wd->nodelog);
2412 }
2413
2414