wd.c revision 1.460.2.1 1 /* $NetBSD: wd.c,v 1.460.2.1 2020/04/20 11:29:03 bouyer Exp $ */
2
3 /*
4 * Copyright (c) 1998, 2001 Manuel Bouyer. All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25 */
26
27 /*-
28 * Copyright (c) 1998, 2003, 2004 The NetBSD Foundation, Inc.
29 * All rights reserved.
30 *
31 * This code is derived from software contributed to The NetBSD Foundation
32 * by Charles M. Hannum and by Onno van der Linden.
33 *
34 * Redistribution and use in source and binary forms, with or without
35 * modification, are permitted provided that the following conditions
36 * are met:
37 * 1. Redistributions of source code must retain the above copyright
38 * notice, this list of conditions and the following disclaimer.
39 * 2. Redistributions in binary form must reproduce the above copyright
40 * notice, this list of conditions and the following disclaimer in the
41 * documentation and/or other materials provided with the distribution.
42 *
43 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
44 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
45 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
46 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
47 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
48 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
49 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
50 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
51 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
52 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
53 * POSSIBILITY OF SUCH DAMAGE.
54 */
55
56 #include <sys/cdefs.h>
57 __KERNEL_RCSID(0, "$NetBSD: wd.c,v 1.460.2.1 2020/04/20 11:29:03 bouyer 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/device.h>
73 #include <sys/disklabel.h>
74 #include <sys/disk.h>
75 #include <sys/syslog.h>
76 #include <sys/proc.h>
77 #include <sys/reboot.h>
78 #include <sys/vnode.h>
79 #include <sys/rndsource.h>
80
81 #include <sys/intr.h>
82 #include <sys/bus.h>
83
84 #include <dev/ata/atareg.h>
85 #include <dev/ata/atavar.h>
86 #include <dev/ata/wdvar.h>
87 #include <dev/ic/wdcreg.h>
88 #include <sys/ataio.h>
89 #include "locators.h"
90
91 #include <prop/proplib.h>
92
93 #define WDIORETRIES_SINGLE 4 /* number of retries for single-sector */
94 #define WDIORETRIES 5 /* number of retries before giving up */
95 #define RECOVERYTIME hz/2 /* time to wait before retrying a cmd */
96
97 #define WDUNIT(dev) DISKUNIT(dev)
98 #define WDPART(dev) DISKPART(dev)
99 #define WDMINOR(unit, part) DISKMINOR(unit, part)
100 #define MAKEWDDEV(maj, unit, part) MAKEDISKDEV(maj, unit, part)
101
102 #define WDLABELDEV(dev) (MAKEWDDEV(major(dev), WDUNIT(dev), RAW_PART))
103
104 #define DEBUG_FUNCS 0x08
105 #define DEBUG_PROBE 0x10
106 #define DEBUG_DETACH 0x20
107 #define DEBUG_XFERS 0x40
108 #ifdef ATADEBUG
109 #ifndef ATADEBUG_WD_MASK
110 #define ATADEBUG_WD_MASK 0x0
111 #endif
112 int wdcdebug_wd_mask = ATADEBUG_WD_MASK;
113 #define ATADEBUG_PRINT(args, level) \
114 if (wdcdebug_wd_mask & (level)) \
115 printf args
116 #else
117 #define ATADEBUG_PRINT(args, level)
118 #endif
119
120 static int wdprobe(device_t, cfdata_t, void *);
121 static void wdattach(device_t, device_t, void *);
122 static int wddetach(device_t, int);
123 static void wdperror(const struct wd_softc *, struct ata_xfer *);
124
125 static void wdminphys(struct buf *);
126
127 static int wd_firstopen(device_t, dev_t, int, int);
128 static int wd_lastclose(device_t);
129 static bool wd_suspend(device_t, const pmf_qual_t *);
130 static int wd_standby(struct wd_softc *, int);
131
132 CFATTACH_DECL3_NEW(wd, sizeof(struct wd_softc),
133 wdprobe, wdattach, wddetach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
134
135 extern struct cfdriver wd_cd;
136
137 static dev_type_open(wdopen);
138 static dev_type_close(wdclose);
139 static dev_type_read(wdread);
140 static dev_type_write(wdwrite);
141 static dev_type_ioctl(wdioctl);
142 static dev_type_strategy(wdstrategy);
143 static dev_type_dump(wddump);
144 static dev_type_size(wdsize);
145 static dev_type_discard(wddiscard);
146
147 const struct bdevsw wd_bdevsw = {
148 .d_open = wdopen,
149 .d_close = wdclose,
150 .d_strategy = wdstrategy,
151 .d_ioctl = wdioctl,
152 .d_dump = wddump,
153 .d_psize = wdsize,
154 .d_discard = wddiscard,
155 .d_flag = D_DISK
156 };
157
158 const struct cdevsw wd_cdevsw = {
159 .d_open = wdopen,
160 .d_close = wdclose,
161 .d_read = wdread,
162 .d_write = wdwrite,
163 .d_ioctl = wdioctl,
164 .d_stop = nostop,
165 .d_tty = notty,
166 .d_poll = nopoll,
167 .d_mmap = nommap,
168 .d_kqfilter = nokqfilter,
169 .d_discard = wddiscard,
170 .d_flag = D_DISK
171 };
172
173 /* #define WD_DUMP_NOT_TRUSTED if you just want to watch */
174 static int wddoingadump = 0;
175 static int wddumprecalibrated = 0;
176
177 /*
178 * Glue necessary to hook WDCIOCCOMMAND into physio
179 */
180
181 struct wd_ioctl {
182 LIST_ENTRY(wd_ioctl) wi_list;
183 struct buf wi_bp;
184 struct uio wi_uio;
185 struct iovec wi_iov;
186 atareq_t wi_atareq;
187 struct wd_softc *wi_softc;
188 };
189
190 static struct wd_ioctl *wi_find(struct buf *);
191 static void wi_free(struct wd_ioctl *);
192 static struct wd_ioctl *wi_get(struct wd_softc *);
193 static void wdioctlstrategy(struct buf *);
194
195 static void wdrestart(void *);
196 static void wdstart1(struct wd_softc *, struct buf *, struct ata_xfer *);
197 static int wd_diskstart(device_t, struct buf *);
198 static int wd_dumpblocks(device_t, void *, daddr_t, int);
199 static void wd_iosize(device_t, int *);
200 static int wd_discard(device_t, off_t, off_t);
201 static void wdbioretry(void *);
202 static void wdbiorequeue(void *);
203 static void wddone(device_t, struct ata_xfer *);
204 static int wd_get_params(struct wd_softc *, 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 static const 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"
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 (at) 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", WD_QUIRK_SPLIT_MOD15_WRITE },
263 { "ST380023AS", WD_QUIRK_SPLIT_MOD15_WRITE },
264 { "ST360015AS", WD_QUIRK_SPLIT_MOD15_WRITE },
265 { NULL,
266 0 }
267 };
268
269 static const struct wd_quirk *
270 wd_lookup_quirks(const char *name)
271 {
272 const struct wd_quirk *wdq;
273 const char *estr;
274
275 for (wdq = wd_quirk_table; wdq->wdq_match != NULL; wdq++) {
276 /*
277 * We only want exact matches (which include matches
278 * against globbing characters).
279 */
280 if (pmatch(name, wdq->wdq_match, &estr) == 2)
281 return (wdq);
282 }
283 return (NULL);
284 }
285
286 static int
287 wdprobe(device_t parent, cfdata_t match, void *aux)
288 {
289 struct ata_device *adev = aux;
290
291 if (adev == NULL)
292 return 0;
293 if (adev->adev_bustype->bustype_type != SCSIPI_BUSTYPE_ATA)
294 return 0;
295
296 if (match->cf_loc[ATA_HLCF_DRIVE] != ATA_HLCF_DRIVE_DEFAULT &&
297 match->cf_loc[ATA_HLCF_DRIVE] != adev->adev_drv_data->drive)
298 return 0;
299 return 1;
300 }
301
302 static void
303 wdattach(device_t parent, device_t self, void *aux)
304 {
305 struct wd_softc *wd = device_private(self);
306 struct dk_softc *dksc = &wd->sc_dksc;
307 struct ata_device *adev= aux;
308 int i, blank;
309 char tbuf[41],pbuf[9], c, *p, *q;
310 const struct wd_quirk *wdq;
311 int dtype = DKTYPE_UNKNOWN;
312
313 dksc->sc_dev = self;
314
315 ATADEBUG_PRINT(("wdattach\n"), DEBUG_FUNCS | DEBUG_PROBE);
316 mutex_init(&wd->sc_lock, MUTEX_DEFAULT, IPL_BIO);
317 #ifdef WD_SOFTBADSECT
318 SLIST_INIT(&wd->sc_bslist);
319 #endif
320 wd->atabus = adev->adev_bustype;
321 wd->inflight = 0;
322 wd->drvp = adev->adev_drv_data;
323
324 wd->drvp->drv_openings = 1;
325 wd->drvp->drv_done = wddone;
326 wd->drvp->drv_softc = dksc->sc_dev; /* done in atabusconfig_thread()
327 but too late */
328
329 SLIST_INIT(&wd->sc_retry_list);
330 SLIST_INIT(&wd->sc_requeue_list);
331 callout_init(&wd->sc_retry_callout, 0); /* XXX MPSAFE */
332 callout_init(&wd->sc_requeue_callout, 0); /* XXX MPSAFE */
333 callout_init(&wd->sc_restart_diskqueue, 0); /* XXX MPSAFE */
334
335 aprint_naive("\n");
336 aprint_normal("\n");
337
338 /* read our drive info */
339 if (wd_get_params(wd, &wd->sc_params) != 0) {
340 aprint_error_dev(self, "IDENTIFY failed\n");
341 goto out;
342 }
343
344 for (blank = 0, p = wd->sc_params.atap_model, q = tbuf, i = 0;
345 i < sizeof(wd->sc_params.atap_model); i++) {
346 c = *p++;
347 if (c == '\0')
348 break;
349 if (c != ' ') {
350 if (blank) {
351 *q++ = ' ';
352 blank = 0;
353 }
354 *q++ = c;
355 } else
356 blank = 1;
357 }
358 *q++ = '\0';
359
360 wd->sc_typename = kmem_asprintf("%s", tbuf);
361 aprint_normal_dev(self, "<%s>\n", wd->sc_typename);
362
363 wdq = wd_lookup_quirks(tbuf);
364 if (wdq != NULL)
365 wd->sc_quirks = wdq->wdq_quirks;
366
367 if (wd->sc_quirks != 0) {
368 char sbuf[sizeof(WD_QUIRK_FMT) + 64];
369 snprintb(sbuf, sizeof(sbuf), WD_QUIRK_FMT, wd->sc_quirks);
370 aprint_normal_dev(self, "quirks %s\n", sbuf);
371
372 if (wd->sc_quirks & WD_QUIRK_SPLIT_MOD15_WRITE) {
373 aprint_error_dev(self, "drive corrupts write transfers with certain controllers, consider replacing\n");
374 }
375 }
376
377 if ((wd->sc_params.atap_multi & 0xff) > 1) {
378 wd->drvp->multi = wd->sc_params.atap_multi & 0xff;
379 } else {
380 wd->drvp->multi = 1;
381 }
382
383 aprint_verbose_dev(self, "drive supports %d-sector PIO transfers,",
384 wd->drvp->multi);
385
386 /* 48-bit LBA addressing */
387 if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0)
388 wd->sc_flags |= WDF_LBA48;
389
390 /* Prior to ATA-4, LBA was optional. */
391 if ((wd->sc_params.atap_capabilities1 & WDC_CAP_LBA) != 0)
392 wd->sc_flags |= WDF_LBA;
393 #if 0
394 /* ATA-4 requires LBA. */
395 if (wd->sc_params.atap_ataversion != 0xffff &&
396 wd->sc_params.atap_ataversion >= WDC_VER_ATA4)
397 wd->sc_flags |= WDF_LBA;
398 #endif
399
400 if ((wd->sc_flags & WDF_LBA48) != 0) {
401 aprint_verbose(" LBA48 addressing\n");
402 wd->sc_capacity =
403 ((uint64_t) wd->sc_params.atap_max_lba[3] << 48) |
404 ((uint64_t) wd->sc_params.atap_max_lba[2] << 32) |
405 ((uint64_t) wd->sc_params.atap_max_lba[1] << 16) |
406 ((uint64_t) wd->sc_params.atap_max_lba[0] << 0);
407 wd->sc_capacity28 =
408 (wd->sc_params.atap_capacity[1] << 16) |
409 wd->sc_params.atap_capacity[0];
410 } else if ((wd->sc_flags & WDF_LBA) != 0) {
411 aprint_verbose(" LBA addressing\n");
412 wd->sc_capacity28 = wd->sc_capacity =
413 (wd->sc_params.atap_capacity[1] << 16) |
414 wd->sc_params.atap_capacity[0];
415 } else {
416 aprint_verbose(" chs addressing\n");
417 wd->sc_capacity28 = wd->sc_capacity =
418 wd->sc_params.atap_cylinders *
419 wd->sc_params.atap_heads *
420 wd->sc_params.atap_sectors;
421 }
422 if ((wd->sc_params.atap_secsz & ATA_SECSZ_VALID_MASK) == ATA_SECSZ_VALID
423 && ((wd->sc_params.atap_secsz & ATA_SECSZ_LLS) != 0)) {
424 wd->sc_blksize = 2ULL *
425 ((uint32_t)((wd->sc_params.atap_lls_secsz[1] << 16) |
426 wd->sc_params.atap_lls_secsz[0]));
427 } else {
428 wd->sc_blksize = 512;
429 }
430 wd->sc_sectoralign.dsa_firstaligned = 0;
431 wd->sc_sectoralign.dsa_alignment = 1;
432 if ((wd->sc_params.atap_secsz & ATA_SECSZ_VALID_MASK) == ATA_SECSZ_VALID
433 && ((wd->sc_params.atap_secsz & ATA_SECSZ_LPS) != 0)) {
434 wd->sc_sectoralign.dsa_alignment = 1 <<
435 (wd->sc_params.atap_secsz & ATA_SECSZ_LPS_SZMSK);
436 if ((wd->sc_params.atap_logical_align & ATA_LA_VALID_MASK) ==
437 ATA_LA_VALID) {
438 wd->sc_sectoralign.dsa_firstaligned =
439 (wd->sc_sectoralign.dsa_alignment -
440 (wd->sc_params.atap_logical_align &
441 ATA_LA_MASK));
442 }
443 }
444 wd->sc_capacity512 = (wd->sc_capacity * wd->sc_blksize) / DEV_BSIZE;
445 format_bytes(pbuf, sizeof(pbuf), wd->sc_capacity * wd->sc_blksize);
446 aprint_normal_dev(self, "%s, %d cyl, %d head, %d sec, "
447 "%d bytes/sect x %llu sectors",
448 pbuf,
449 (wd->sc_flags & WDF_LBA) ? (int)(wd->sc_capacity /
450 (wd->sc_params.atap_heads * wd->sc_params.atap_sectors)) :
451 wd->sc_params.atap_cylinders,
452 wd->sc_params.atap_heads, wd->sc_params.atap_sectors,
453 wd->sc_blksize, (unsigned long long)wd->sc_capacity);
454 if (wd->sc_sectoralign.dsa_alignment != 1) {
455 aprint_normal(" (%d bytes/physsect",
456 wd->sc_sectoralign.dsa_alignment & wd->sc_blksize);
457 if (wd->sc_sectoralign.dsa_firstaligned != 0) {
458 aprint_normal("; first aligned sector: %jd",
459 (intmax_t)wd->sc_sectoralign.dsa_firstaligned);
460 }
461 aprint_normal(")");
462 }
463 aprint_normal("\n");
464
465 ATADEBUG_PRINT(("%s: atap_dmatiming_mimi=%d, atap_dmatiming_recom=%d\n",
466 device_xname(self), wd->sc_params.atap_dmatiming_mimi,
467 wd->sc_params.atap_dmatiming_recom), DEBUG_PROBE);
468
469 if (wd->sc_blksize <= 0 || !powerof2(wd->sc_blksize) ||
470 wd->sc_blksize < DEV_BSIZE || wd->sc_blksize > MAXPHYS) {
471 aprint_normal_dev(self, "WARNING: block size %u "
472 "might not actually work\n", wd->sc_blksize);
473 }
474
475 if (strcmp(wd->sc_params.atap_model, "ST506") == 0)
476 dtype = DKTYPE_ST506;
477 else
478 dtype = DKTYPE_ESDI;
479
480 out:
481 /*
482 * Initialize and attach the disk structure.
483 */
484 dk_init(dksc, self, dtype);
485 disk_init(&dksc->sc_dkdev, dksc->sc_xname, &wddkdriver);
486
487 /* Attach dk and disk subsystems */
488 dk_attach(dksc);
489 disk_attach(&dksc->sc_dkdev);
490 wd_set_geometry(wd);
491
492 bufq_alloc(&dksc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
493
494 /* reference to label structure, used by ata code */
495 wd->drvp->lp = dksc->sc_dkdev.dk_label;
496
497 /* Discover wedges on this disk. */
498 dkwedge_discover(&dksc->sc_dkdev);
499
500 if (!pmf_device_register1(self, wd_suspend, NULL, wd_shutdown))
501 aprint_error_dev(self, "couldn't establish power handler\n");
502
503 wd_sysctl_attach(wd);
504 }
505
506 static bool
507 wd_suspend(device_t dv, const pmf_qual_t *qual)
508 {
509 struct wd_softc *sc = device_private(dv);
510
511 /* the adapter needs to be enabled */
512 if (sc->atabus->ata_addref(sc->drvp))
513 return true; /* no need to complain */
514
515 wd_flushcache(sc, AT_WAIT, false);
516 wd_standby(sc, AT_WAIT);
517
518 sc->atabus->ata_delref(sc->drvp);
519 return true;
520 }
521
522 static int
523 wddetach(device_t self, int flags)
524 {
525 struct wd_softc *wd = device_private(self);
526 struct dk_softc *dksc = &wd->sc_dksc;
527 int bmaj, cmaj, i, mn, rc;
528
529 if ((rc = disk_begindetach(&dksc->sc_dkdev, wd_lastclose, self, flags)) != 0)
530 return rc;
531
532 /* locate the major number */
533 bmaj = bdevsw_lookup_major(&wd_bdevsw);
534 cmaj = cdevsw_lookup_major(&wd_cdevsw);
535
536 /* Nuke the vnodes for any open instances. */
537 for (i = 0; i < MAXPARTITIONS; i++) {
538 mn = WDMINOR(device_unit(self), i);
539 vdevgone(bmaj, mn, mn, VBLK);
540 vdevgone(cmaj, mn, mn, VCHR);
541 }
542
543 dk_drain(dksc);
544
545 /* Kill off any pending commands. */
546 mutex_enter(&wd->sc_lock);
547 wd->atabus->ata_killpending(wd->drvp);
548
549 callout_halt(&wd->sc_retry_callout, &wd->sc_lock);
550 callout_destroy(&wd->sc_retry_callout);
551 callout_halt(&wd->sc_requeue_callout, &wd->sc_lock);
552 callout_destroy(&wd->sc_requeue_callout);
553 callout_halt(&wd->sc_restart_diskqueue, &wd->sc_lock);
554 callout_destroy(&wd->sc_restart_diskqueue);
555
556 mutex_exit(&wd->sc_lock);
557
558 bufq_free(dksc->sc_bufq);
559
560 /* Delete all of our wedges. */
561 dkwedge_delall(&dksc->sc_dkdev);
562
563 if (flags & DETACH_POWEROFF)
564 wd_standby(wd, AT_POLL);
565
566 /* Detach from the disk list. */
567 disk_detach(&dksc->sc_dkdev);
568 disk_destroy(&dksc->sc_dkdev);
569
570 dk_detach(dksc);
571
572 #ifdef WD_SOFTBADSECT
573 /* Clean out the bad sector list */
574 while (!SLIST_EMPTY(&wd->sc_bslist)) {
575 struct disk_badsectors *dbs = SLIST_FIRST(&wd->sc_bslist);
576 SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
577 kmem_free(dbs, sizeof(*dbs));
578 }
579 wd->sc_bscount = 0;
580 #endif
581 if (wd->sc_typename != NULL) {
582 kmem_free(wd->sc_typename, strlen(wd->sc_typename) + 1);
583 wd->sc_typename = NULL;
584 }
585
586 pmf_device_deregister(self);
587
588 wd_sysctl_detach(wd);
589
590 mutex_destroy(&wd->sc_lock);
591
592 wd->drvp->drive_type = ATA_DRIVET_NONE; /* no drive any more here */
593 wd->drvp->drive_flags = 0;
594
595 return (0);
596 }
597
598 /*
599 * Read/write routine for a buffer. Validates the arguments and schedules the
600 * transfer. Does not wait for the transfer to complete.
601 */
602 static void
603 wdstrategy(struct buf *bp)
604 {
605 struct wd_softc *wd =
606 device_lookup_private(&wd_cd, WDUNIT(bp->b_dev));
607 struct dk_softc *dksc = &wd->sc_dksc;
608
609 ATADEBUG_PRINT(("wdstrategy (%s)\n", dksc->sc_xname),
610 DEBUG_XFERS);
611
612 /* If device invalidated (e.g. media change, door open,
613 * device detachment), then error.
614 */
615 if ((wd->sc_flags & WDF_LOADED) == 0 ||
616 !device_is_enabled(dksc->sc_dev))
617 goto err;
618
619 #ifdef WD_SOFTBADSECT
620 /*
621 * If the transfer about to be attempted contains only a block that
622 * is known to be bad then return an error for the transfer without
623 * even attempting to start a transfer up under the premis that we
624 * will just end up doing more retries for a transfer that will end
625 * up failing again.
626 */
627 if (__predict_false(!SLIST_EMPTY(&wd->sc_bslist))) {
628 struct disklabel *lp = dksc->sc_dkdev.dk_label;
629 struct disk_badsectors *dbs;
630 daddr_t blkno, maxblk;
631
632 /* convert the block number to absolute */
633 if (lp->d_secsize >= DEV_BSIZE)
634 blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
635 else
636 blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
637 if (WDPART(bp->b_dev) != RAW_PART)
638 blkno += lp->d_partitions[WDPART(bp->b_dev)].p_offset;
639 maxblk = blkno + (bp->b_bcount / wd->sc_blksize) - 1;
640
641 mutex_enter(&wd->sc_lock);
642 SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next)
643 if ((dbs->dbs_min <= bp->b_rawblkno &&
644 bp->b_rawblkno <= dbs->dbs_max) ||
645 (dbs->dbs_min <= maxblk && maxblk <= dbs->dbs_max)){
646 mutex_exit(&wd->sc_lock);
647 goto err;
648 }
649 mutex_exit(&wd->sc_lock);
650 }
651 #endif
652
653 dk_strategy(dksc, bp);
654 return;
655
656 err:
657 bp->b_error = EIO;
658 bp->b_resid = bp->b_bcount;
659 biodone(bp);
660 }
661
662 static void
663 wdstart1(struct wd_softc *wd, struct buf *bp, struct ata_xfer *xfer)
664 {
665 struct dk_softc *dksc = &wd->sc_dksc;
666
667 KASSERT(bp == xfer->c_bio.bp || xfer->c_bio.bp == NULL);
668 KASSERT((xfer->c_flags & (C_WAITACT|C_FREE)) == 0);
669 KASSERT(mutex_owned(&wd->sc_lock));
670
671 /* Reset state, so that retries don't use stale info */
672 if (__predict_false(xfer->c_retries > 0)) {
673 xfer->c_flags = 0;
674 memset(&xfer->c_bio, 0, sizeof(xfer->c_bio));
675 }
676
677 xfer->c_bio.blkno = bp->b_rawblkno;
678 xfer->c_bio.bcount = bp->b_bcount;
679 xfer->c_bio.databuf = bp->b_data;
680 xfer->c_bio.blkdone = 0;
681 xfer->c_bio.bp = bp;
682
683 #ifdef WD_CHAOS_MONKEY
684 /*
685 * Override blkno to be over device capacity to trigger error,
686 * but only if it's read, to avoid trashing disk contents should
687 * the command be clipped, or otherwise misinterpreted, by the
688 * driver or controller.
689 */
690 if (BUF_ISREAD(bp) && xfer->c_retries == 0 && wd->drv_chaos_freq > 0 &&
691 (++wd->drv_chaos_cnt % wd->drv_chaos_freq) == 0) {
692 device_printf(dksc->sc_dev, "%s: chaos xfer %"PRIxPTR"\n",
693 __func__, (intptr_t)xfer & PAGE_MASK);
694 xfer->c_bio.blkno = 7777777 + wd->sc_capacity;
695 xfer->c_flags |= C_CHAOS;
696 }
697 #endif
698
699 /*
700 * If we're retrying, retry in single-sector mode. This will give us
701 * the sector number of the problem, and will eventually allow the
702 * transfer to succeed. If FUA is requested, we can't actually
703 * do this, as ATA_SINGLE is usually executed as PIO transfer by drivers
704 * which support it, and that isn't compatible with NCQ/FUA.
705 */
706 if (xfer->c_retries >= WDIORETRIES_SINGLE &&
707 (bp->b_flags & B_MEDIA_FUA) == 0)
708 xfer->c_bio.flags = ATA_SINGLE;
709 else
710 xfer->c_bio.flags = 0;
711
712 /*
713 * request LBA48 transfers when supported by the controller
714 * and needed by transfer offset or size.
715 */
716 if (wd->sc_flags & WDF_LBA48 &&
717 (((xfer->c_bio.blkno +
718 xfer->c_bio.bcount / dksc->sc_dkdev.dk_geom.dg_secsize) >
719 wd->sc_capacity28) ||
720 ((xfer->c_bio.bcount / dksc->sc_dkdev.dk_geom.dg_secsize) > 128)))
721 xfer->c_bio.flags |= ATA_LBA48;
722
723 /*
724 * If NCQ was negotiated, always use it for the first several attempts.
725 * Since device cancels all outstanding requests on error, downgrade
726 * to non-NCQ on retry, so that the retried transfer would not cause
727 * cascade failure for the other transfers if it fails again.
728 * If FUA was requested, we can't downgrade, as that would violate
729 * the semantics - FUA would not be honored. In that case, continue
730 * retrying with NCQ.
731 */
732 if (WD_USE_NCQ(wd) && (xfer->c_retries < WDIORETRIES_SINGLE ||
733 (bp->b_flags & B_MEDIA_FUA) != 0)) {
734 xfer->c_bio.flags |= ATA_LBA48;
735 xfer->c_flags |= C_NCQ;
736
737 if (WD_USE_NCQ_PRIO(wd) &&
738 BIO_GETPRIO(bp) == BPRIO_TIMECRITICAL)
739 xfer->c_bio.flags |= ATA_PRIO_HIGH;
740 }
741
742 if (wd->sc_flags & WDF_LBA)
743 xfer->c_bio.flags |= ATA_LBA;
744 if (bp->b_flags & B_READ) {
745 xfer->c_bio.flags |= ATA_READ;
746 } else {
747 /* it's a write */
748 wd->sc_flags |= WDF_DIRTY;
749 }
750 if (bp->b_flags & B_MEDIA_FUA) {
751 /* If not using NCQ, the command WRITE DMA FUA EXT is LBA48 */
752 KASSERT((wd->sc_flags & WDF_LBA48) != 0);
753 if ((xfer->c_flags & C_NCQ) == 0)
754 xfer->c_bio.flags |= ATA_LBA48;
755
756 xfer->c_bio.flags |= ATA_FUA;
757 }
758
759 if (xfer->c_retries == 0)
760 wd->inflight++;
761 mutex_exit(&wd->sc_lock);
762
763 /* Queue the xfer */
764 wd->atabus->ata_bio(wd->drvp, xfer);
765
766 mutex_enter(&wd->sc_lock);
767 }
768
769 static int
770 wd_diskstart(device_t dev, struct buf *bp)
771 {
772 struct wd_softc *wd = device_private(dev);
773 #ifdef ATADEBUG
774 struct dk_softc *dksc = &wd->sc_dksc;
775 #endif
776 struct ata_xfer *xfer;
777 struct ata_channel *chp;
778 unsigned openings;
779 int ticks;
780
781 mutex_enter(&wd->sc_lock);
782
783 chp = wd->drvp->chnl_softc;
784
785 ata_channel_lock(chp);
786 openings = ata_queue_openings(chp);
787 ata_channel_unlock(chp);
788
789 openings = uimin(openings, wd->drvp->drv_openings);
790
791 if (wd->inflight >= openings) {
792 /*
793 * pretend we run out of memory when the queue is full,
794 * so that the operation is retried after a minimal
795 * delay.
796 */
797 xfer = NULL;
798 ticks = 1;
799 } else {
800 /*
801 * If there is no available memory, retry later. This
802 * happens very rarely and only under memory pressure,
803 * so wait relatively long before retry.
804 */
805 xfer = ata_get_xfer(chp, false);
806 ticks = hz/2;
807 }
808
809 if (xfer == NULL) {
810 ATADEBUG_PRINT(("wd_diskstart %s no xfer\n",
811 dksc->sc_xname), DEBUG_XFERS);
812
813 /*
814 * The disk queue is pushed automatically when an I/O
815 * operation finishes or another one is queued. We
816 * need this extra timeout because an ATA channel
817 * might be shared by more than one disk queue and
818 * all queues need to be restarted when another slot
819 * becomes available.
820 */
821 if (!callout_pending(&wd->sc_restart_diskqueue)) {
822 callout_reset(&wd->sc_restart_diskqueue, ticks,
823 wdrestart, dev);
824 }
825
826 mutex_exit(&wd->sc_lock);
827 return EAGAIN;
828 }
829
830 wdstart1(wd, bp, xfer);
831
832 mutex_exit(&wd->sc_lock);
833
834 return 0;
835 }
836
837 /*
838 * Queue a drive for I/O.
839 */
840 static void
841 wdrestart(void *x)
842 {
843 device_t self = x;
844 struct wd_softc *wd = device_private(self);
845 struct dk_softc *dksc = &wd->sc_dksc;
846
847 ATADEBUG_PRINT(("wdstart %s\n", dksc->sc_xname),
848 DEBUG_XFERS);
849
850 if (!device_is_active(dksc->sc_dev))
851 return;
852
853 dk_start(dksc, NULL);
854 }
855
856 static void
857 wddone(device_t self, struct ata_xfer *xfer)
858 {
859 struct wd_softc *wd = device_private(self);
860 struct dk_softc *dksc = &wd->sc_dksc;
861 const char *errmsg;
862 int do_perror = 0;
863 struct buf *bp;
864
865 ATADEBUG_PRINT(("wddone %s\n", dksc->sc_xname),
866 DEBUG_XFERS);
867
868 if (__predict_false(wddoingadump)) {
869 /* just drop it to the floor */
870 ata_free_xfer(wd->drvp->chnl_softc, xfer);
871 return;
872 }
873
874 bp = xfer->c_bio.bp;
875 KASSERT(bp != NULL);
876
877 bp->b_resid = xfer->c_bio.bcount;
878 switch (xfer->c_bio.error) {
879 case ERR_DMA:
880 errmsg = "DMA error";
881 goto retry;
882 case ERR_DF:
883 errmsg = "device fault";
884 goto retry;
885 case TIMEOUT:
886 errmsg = "device timeout";
887 goto retry;
888 case REQUEUE:
889 errmsg = "requeue";
890 goto retry2;
891 case ERR_RESET:
892 errmsg = "channel reset";
893 goto retry2;
894 case ERROR:
895 /* Don't care about media change bits */
896 if (xfer->c_bio.r_error != 0 &&
897 (xfer->c_bio.r_error & ~(WDCE_MC | WDCE_MCR)) == 0)
898 goto noerror;
899 errmsg = "error";
900 do_perror = 1;
901 retry: /* Just reset and retry. Can we do more ? */
902 if ((xfer->c_flags & C_RECOVERED) == 0) {
903 int wflags = (xfer->c_flags & C_POLL) ? AT_POLL : 0;
904 ata_channel_lock(wd->drvp->chnl_softc);
905 ata_thread_run(wd->drvp->chnl_softc, wflags,
906 ATACH_TH_DRIVE_RESET, wd->drvp->drive);
907 ata_channel_unlock(wd->drvp->chnl_softc);
908 }
909 retry2:
910 mutex_enter(&wd->sc_lock);
911
912 diskerr(bp, "wd", errmsg, LOG_PRINTF,
913 xfer->c_bio.blkdone, dksc->sc_dkdev.dk_label);
914 if (xfer->c_retries < WDIORETRIES)
915 printf(", xfer %"PRIxPTR", retry %d",
916 (intptr_t)xfer & PAGE_MASK,
917 xfer->c_retries);
918 printf("\n");
919 if (do_perror)
920 wdperror(wd, xfer);
921
922 if (xfer->c_retries < WDIORETRIES) {
923 xfer->c_retries++;
924
925 /* Rerun ASAP if just requeued */
926 if (xfer->c_bio.error == REQUEUE) {
927 SLIST_INSERT_HEAD(&wd->sc_requeue_list, xfer,
928 c_retrychain);
929 callout_reset(&wd->sc_requeue_callout,
930 1, wdbiorequeue, wd);
931 } else {
932 SLIST_INSERT_HEAD(&wd->sc_retry_list, xfer,
933 c_retrychain);
934 callout_reset(&wd->sc_retry_callout,
935 RECOVERYTIME, wdbioretry, wd);
936 }
937
938 mutex_exit(&wd->sc_lock);
939 return;
940 }
941
942 mutex_exit(&wd->sc_lock);
943
944 #ifdef WD_SOFTBADSECT
945 /*
946 * Not all errors indicate a failed block but those that do,
947 * put the block on the bad-block list for the device. Only
948 * do this for reads because the drive should do it for writes,
949 * itself, according to Manuel.
950 */
951 if ((bp->b_flags & B_READ) &&
952 ((wd->drvp->ata_vers >= 4 && xfer->c_bio.r_error & 64) ||
953 (wd->drvp->ata_vers < 4 && xfer->c_bio.r_error & 192))) {
954 struct disk_badsectors *dbs;
955
956 dbs = kmem_zalloc(sizeof *dbs, KM_NOSLEEP);
957 if (dbs == NULL) {
958 aprint_error_dev(dksc->sc_dev,
959 "failed to add bad block to list\n");
960 goto out;
961 }
962
963 dbs->dbs_min = bp->b_rawblkno;
964 dbs->dbs_max = dbs->dbs_min +
965 (bp->b_bcount /wd->sc_blksize) - 1;
966 microtime(&dbs->dbs_failedat);
967
968 mutex_enter(&wd->sc_lock);
969 SLIST_INSERT_HEAD(&wd->sc_bslist, dbs, dbs_next);
970 wd->sc_bscount++;
971 mutex_exit(&wd->sc_lock);
972 }
973 out:
974 #endif
975 bp->b_error = EIO;
976 break;
977 case NOERROR:
978 #ifdef WD_CHAOS_MONKEY
979 /*
980 * For example Parallels AHCI emulation doesn't actually
981 * return error for the invalid I/O, so just re-run
982 * the request and do not panic.
983 */
984 if (__predict_false(xfer->c_flags & C_CHAOS)) {
985 xfer->c_bio.error = REQUEUE;
986 errmsg = "chaos noerror";
987 goto retry2;
988 }
989 #endif
990
991 noerror: if ((xfer->c_bio.flags & ATA_CORR) || xfer->c_retries > 0)
992 device_printf(dksc->sc_dev,
993 "soft error (corrected) xfer %"PRIxPTR"\n",
994 (intptr_t)xfer & PAGE_MASK);
995 break;
996 case ERR_NODEV:
997 bp->b_error = EIO;
998 break;
999 }
1000 if (__predict_false(bp->b_error != 0) && bp->b_resid == 0) {
1001 /*
1002 * the disk or controller sometimes report a complete
1003 * xfer, when there has been an error. This is wrong,
1004 * assume nothing got transferred in this case
1005 */
1006 bp->b_resid = bp->b_bcount;
1007 }
1008
1009 ata_free_xfer(wd->drvp->chnl_softc, xfer);
1010
1011 mutex_enter(&wd->sc_lock);
1012 wd->inflight--;
1013 mutex_exit(&wd->sc_lock);
1014 dk_done(dksc, bp);
1015 dk_start(dksc, NULL);
1016 }
1017
1018 static void
1019 wdbioretry(void *v)
1020 {
1021 struct wd_softc *wd = v;
1022 struct ata_xfer *xfer;
1023
1024 ATADEBUG_PRINT(("%s %s\n", __func__, wd->sc_dksc.sc_xname),
1025 DEBUG_XFERS);
1026
1027 mutex_enter(&wd->sc_lock);
1028 while ((xfer = SLIST_FIRST(&wd->sc_retry_list))) {
1029 SLIST_REMOVE_HEAD(&wd->sc_retry_list, c_retrychain);
1030 wdstart1(wd, xfer->c_bio.bp, xfer);
1031 }
1032 mutex_exit(&wd->sc_lock);
1033 }
1034
1035 static void
1036 wdbiorequeue(void *v)
1037 {
1038 struct wd_softc *wd = v;
1039 struct ata_xfer *xfer;
1040
1041 ATADEBUG_PRINT(("%s %s\n", __func__, wd->sc_dksc.sc_xname),
1042 DEBUG_XFERS);
1043
1044 mutex_enter(&wd->sc_lock);
1045 while ((xfer = SLIST_FIRST(&wd->sc_requeue_list))) {
1046 SLIST_REMOVE_HEAD(&wd->sc_requeue_list, c_retrychain);
1047 wdstart1(wd, xfer->c_bio.bp, xfer);
1048 }
1049 mutex_exit(&wd->sc_lock);
1050 }
1051
1052 static void
1053 wdminphys(struct buf *bp)
1054 {
1055 const struct wd_softc * const wd =
1056 device_lookup_private(&wd_cd, WDUNIT(bp->b_dev));
1057 int maxsectors;
1058
1059 /*
1060 * The limit is actually 65536 for LBA48 and 256 for non-LBA48,
1061 * but that requires to set the count for the ATA command
1062 * to 0, which is somewhat error prone, so better stay safe.
1063 */
1064 if (wd->sc_flags & WDF_LBA48)
1065 maxsectors = 65535;
1066 else
1067 maxsectors = 128;
1068
1069 if (bp->b_bcount > (wd->sc_blksize * maxsectors))
1070 bp->b_bcount = (wd->sc_blksize * maxsectors);
1071
1072 minphys(bp);
1073 }
1074
1075 static void
1076 wd_iosize(device_t dev, int *count)
1077 {
1078 struct buf B;
1079 int bmaj;
1080
1081 bmaj = bdevsw_lookup_major(&wd_bdevsw);
1082 B.b_dev = MAKEWDDEV(bmaj,device_unit(dev),RAW_PART);
1083 B.b_bcount = *count;
1084
1085 wdminphys(&B);
1086
1087 *count = B.b_bcount;
1088 }
1089
1090 static int
1091 wdread(dev_t dev, struct uio *uio, int flags)
1092 {
1093
1094 ATADEBUG_PRINT(("wdread\n"), DEBUG_XFERS);
1095 return (physio(wdstrategy, NULL, dev, B_READ, wdminphys, uio));
1096 }
1097
1098 static int
1099 wdwrite(dev_t dev, struct uio *uio, int flags)
1100 {
1101
1102 ATADEBUG_PRINT(("wdwrite\n"), DEBUG_XFERS);
1103 return (physio(wdstrategy, NULL, dev, B_WRITE, wdminphys, uio));
1104 }
1105
1106 static int
1107 wdopen(dev_t dev, int flag, int fmt, struct lwp *l)
1108 {
1109 struct wd_softc *wd;
1110 struct dk_softc *dksc;
1111 int unit, part, error;
1112
1113 ATADEBUG_PRINT(("wdopen\n"), DEBUG_FUNCS);
1114 unit = WDUNIT(dev);
1115 wd = device_lookup_private(&wd_cd, unit);
1116 if (wd == NULL)
1117 return (ENXIO);
1118 dksc = &wd->sc_dksc;
1119
1120 if (! device_is_active(dksc->sc_dev))
1121 return (ENODEV);
1122
1123 part = WDPART(dev);
1124
1125 if (wd->sc_capacity == 0)
1126 return (ENODEV);
1127
1128 /*
1129 * If any partition is open, but the disk has been invalidated,
1130 * disallow further opens.
1131 */
1132 if ((wd->sc_flags & (WDF_OPEN | WDF_LOADED)) == WDF_OPEN) {
1133 if (part != RAW_PART || fmt != S_IFCHR)
1134 return EIO;
1135 }
1136
1137 error = dk_open(dksc, dev, flag, fmt, l);
1138
1139 return error;
1140 }
1141
1142 /*
1143 * Serialized by caller
1144 */
1145 static int
1146 wd_firstopen(device_t self, dev_t dev, int flag, int fmt)
1147 {
1148 struct wd_softc *wd = device_private(self);
1149 struct dk_softc *dksc = &wd->sc_dksc;
1150 int error;
1151
1152 error = wd->atabus->ata_addref(wd->drvp);
1153 if (error)
1154 return error;
1155
1156 if ((wd->sc_flags & WDF_LOADED) == 0) {
1157 int param_error;
1158
1159 /* Load the physical device parameters. */
1160 param_error = wd_get_params(wd, &wd->sc_params);
1161 if (param_error != 0) {
1162 aprint_error_dev(dksc->sc_dev, "IDENTIFY failed\n");
1163 error = EIO;
1164 goto bad;
1165 }
1166 wd_set_geometry(wd);
1167 wd->sc_flags |= WDF_LOADED;
1168 }
1169
1170 wd->sc_flags |= WDF_OPEN;
1171 return 0;
1172
1173 bad:
1174 wd->atabus->ata_delref(wd->drvp);
1175 return error;
1176 }
1177
1178 /*
1179 * Caller must hold wd->sc_dk.dk_openlock.
1180 */
1181 static int
1182 wd_lastclose(device_t self)
1183 {
1184 struct wd_softc *wd = device_private(self);
1185
1186 KASSERTMSG(bufq_peek(wd->sc_dksc.sc_bufq) == NULL, "bufq not empty");
1187
1188 if (wd->sc_flags & WDF_DIRTY)
1189 wd_flushcache(wd, AT_WAIT, false);
1190
1191 wd->atabus->ata_delref(wd->drvp);
1192 wd->sc_flags &= ~WDF_OPEN;
1193
1194 return 0;
1195 }
1196
1197 static int
1198 wdclose(dev_t dev, int flag, int fmt, struct lwp *l)
1199 {
1200 struct wd_softc *wd;
1201 struct dk_softc *dksc;
1202 int unit;
1203
1204 unit = WDUNIT(dev);
1205 wd = device_lookup_private(&wd_cd, unit);
1206 dksc = &wd->sc_dksc;
1207
1208 return dk_close(dksc, dev, flag, fmt, l);
1209 }
1210
1211 void
1212 wdperror(const struct wd_softc *wd, struct ata_xfer *xfer)
1213 {
1214 static const char *const errstr0_3[] = {"address mark not found",
1215 "track 0 not found", "aborted command", "media change requested",
1216 "id not found", "media changed", "uncorrectable data error",
1217 "bad block detected"};
1218 static const char *const errstr4_5[] = {
1219 "obsolete (address mark not found)",
1220 "no media/write protected", "aborted command",
1221 "media change requested", "id not found", "media changed",
1222 "uncorrectable data error", "interface CRC error"};
1223 const char *const *errstr;
1224 int i;
1225 const char *sep = "";
1226
1227 const struct dk_softc *dksc = &wd->sc_dksc;
1228 const char *devname = dksc->sc_xname;
1229 struct ata_drive_datas *drvp = wd->drvp;
1230 int errno = xfer->c_bio.r_error;
1231
1232 if (drvp->ata_vers >= 4)
1233 errstr = errstr4_5;
1234 else
1235 errstr = errstr0_3;
1236
1237 printf("%s: (", devname);
1238
1239 if (errno == 0)
1240 printf("error not notified");
1241
1242 for (i = 0; i < 8; i++) {
1243 if (errno & (1 << i)) {
1244 printf("%s%s", sep, errstr[i]);
1245 sep = ", ";
1246 }
1247 }
1248 printf(")\n");
1249 }
1250
1251 int
1252 wdioctl(dev_t dev, u_long cmd, void *addr, int flag, struct lwp *l)
1253 {
1254 struct wd_softc *wd =
1255 device_lookup_private(&wd_cd, WDUNIT(dev));
1256 struct dk_softc *dksc = &wd->sc_dksc;
1257
1258 ATADEBUG_PRINT(("wdioctl\n"), DEBUG_FUNCS);
1259
1260 if ((wd->sc_flags & WDF_LOADED) == 0)
1261 return EIO;
1262
1263 switch (cmd) {
1264 #ifdef HAS_BAD144_HANDLING
1265 case DIOCSBAD:
1266 if ((flag & FWRITE) == 0)
1267 return EBADF;
1268 dksc->sc_dkdev.dk_cpulabel->bad = *(struct dkbad *)addr;
1269 dksc->sc_dkdev.dk_label->d_flags |= D_BADSECT;
1270 bad144intern(wd);
1271 return 0;
1272 #endif
1273 #ifdef WD_SOFTBADSECT
1274 case DIOCBSLIST :
1275 {
1276 uint32_t count, missing, skip;
1277 struct disk_badsecinfo dbsi;
1278 struct disk_badsectors *dbs;
1279 size_t available;
1280 uint8_t *laddr;
1281
1282 dbsi = *(struct disk_badsecinfo *)addr;
1283 missing = wd->sc_bscount;
1284 count = 0;
1285 available = dbsi.dbsi_bufsize;
1286 skip = dbsi.dbsi_skip;
1287 laddr = (uint8_t *)dbsi.dbsi_buffer;
1288
1289 /*
1290 * We start this loop with the expectation that all of the
1291 * entries will be missed and decrement this counter each
1292 * time we either skip over one (already copied out) or
1293 * we actually copy it back to user space. The structs
1294 * holding the bad sector information are copied directly
1295 * back to user space whilst the summary is returned via
1296 * the struct passed in via the ioctl.
1297 */
1298 mutex_enter(&wd->sc_lock);
1299 SLIST_FOREACH(dbs, &wd->sc_bslist, dbs_next) {
1300 if (skip > 0) {
1301 missing--;
1302 skip--;
1303 continue;
1304 }
1305 if (available < sizeof(*dbs))
1306 break;
1307 available -= sizeof(*dbs);
1308 copyout(dbs, laddr, sizeof(*dbs));
1309 laddr += sizeof(*dbs);
1310 missing--;
1311 count++;
1312 }
1313 mutex_exit(&wd->sc_lock);
1314 dbsi.dbsi_left = missing;
1315 dbsi.dbsi_copied = count;
1316 *(struct disk_badsecinfo *)addr = dbsi;
1317 return 0;
1318 }
1319
1320 case DIOCBSFLUSH :
1321 /* Clean out the bad sector list */
1322 mutex_enter(&wd->sc_lock);
1323 while (!SLIST_EMPTY(&wd->sc_bslist)) {
1324 struct disk_badsectors *dbs =
1325 SLIST_FIRST(&wd->sc_bslist);
1326 SLIST_REMOVE_HEAD(&wd->sc_bslist, dbs_next);
1327 kmem_free(dbs, sizeof(*dbs));
1328 }
1329 mutex_exit(&wd->sc_lock);
1330 wd->sc_bscount = 0;
1331 return 0;
1332 #endif
1333
1334 #ifdef notyet
1335 case DIOCWFORMAT:
1336 if ((flag & FWRITE) == 0)
1337 return EBADF;
1338 {
1339 register struct format_op *fop;
1340 struct iovec aiov;
1341 struct uio auio;
1342 int error1;
1343
1344 fop = (struct format_op *)addr;
1345 aiov.iov_base = fop->df_buf;
1346 aiov.iov_len = fop->df_count;
1347 auio.uio_iov = &aiov;
1348 auio.uio_iovcnt = 1;
1349 auio.uio_resid = fop->df_count;
1350 auio.uio_offset =
1351 fop->df_startblk * wd->sc_dk.dk_label->d_secsize;
1352 auio.uio_vmspace = l->l_proc->p_vmspace;
1353 error1 = physio(wdformat, NULL, dev, B_WRITE, wdminphys,
1354 &auio);
1355 fop->df_count -= auio.uio_resid;
1356 fop->df_reg[0] = wdc->sc_status;
1357 fop->df_reg[1] = wdc->sc_error;
1358 return error1;
1359 }
1360 #endif
1361 case DIOCGCACHE:
1362 return wd_getcache(wd, (int *)addr);
1363
1364 case DIOCSCACHE:
1365 return wd_setcache(wd, *(int *)addr);
1366
1367 case DIOCCACHESYNC:
1368 return wd_flushcache(wd, AT_WAIT, true);
1369
1370 case ATAIOCCOMMAND:
1371 /*
1372 * Make sure this command is (relatively) safe first
1373 */
1374 if ((((atareq_t *) addr)->flags & ATACMD_READ) == 0 &&
1375 (flag & FWRITE) == 0)
1376 return (EBADF);
1377 {
1378 struct wd_ioctl *wi;
1379 atareq_t *atareq = (atareq_t *) addr;
1380 int error1;
1381
1382 wi = wi_get(wd);
1383 wi->wi_atareq = *atareq;
1384
1385 if (atareq->datalen && atareq->flags &
1386 (ATACMD_READ | ATACMD_WRITE)) {
1387 void *tbuf;
1388 if (atareq->datalen < DEV_BSIZE
1389 && atareq->command == WDCC_IDENTIFY) {
1390 tbuf = kmem_zalloc(DEV_BSIZE, KM_SLEEP);
1391 wi->wi_iov.iov_base = tbuf;
1392 wi->wi_iov.iov_len = DEV_BSIZE;
1393 UIO_SETUP_SYSSPACE(&wi->wi_uio);
1394 } else {
1395 tbuf = NULL;
1396 wi->wi_iov.iov_base = atareq->databuf;
1397 wi->wi_iov.iov_len = atareq->datalen;
1398 wi->wi_uio.uio_vmspace = l->l_proc->p_vmspace;
1399 }
1400 wi->wi_uio.uio_iov = &wi->wi_iov;
1401 wi->wi_uio.uio_iovcnt = 1;
1402 wi->wi_uio.uio_resid = atareq->datalen;
1403 wi->wi_uio.uio_offset = 0;
1404 wi->wi_uio.uio_rw =
1405 (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE;
1406 error1 = physio(wdioctlstrategy, &wi->wi_bp, dev,
1407 (atareq->flags & ATACMD_READ) ? B_READ : B_WRITE,
1408 wdminphys, &wi->wi_uio);
1409 if (tbuf != NULL && error1 == 0) {
1410 error1 = copyout(tbuf, atareq->databuf,
1411 atareq->datalen);
1412 kmem_free(tbuf, DEV_BSIZE);
1413 }
1414 } else {
1415 /* No need to call physio if we don't have any
1416 user data */
1417 wi->wi_bp.b_flags = 0;
1418 wi->wi_bp.b_data = 0;
1419 wi->wi_bp.b_bcount = 0;
1420 wi->wi_bp.b_dev = dev;
1421 wi->wi_bp.b_proc = l->l_proc;
1422 wdioctlstrategy(&wi->wi_bp);
1423 error1 = wi->wi_bp.b_error;
1424 }
1425 *atareq = wi->wi_atareq;
1426 wi_free(wi);
1427 return(error1);
1428 }
1429
1430 case DIOCGSECTORALIGN: {
1431 struct disk_sectoralign *dsa = addr;
1432 int part = WDPART(dev);
1433
1434 *dsa = wd->sc_sectoralign;
1435 if (part != RAW_PART) {
1436 struct disklabel *lp = dksc->sc_dkdev.dk_label;
1437 daddr_t offset = lp->d_partitions[part].p_offset;
1438 uint32_t r = offset % dsa->dsa_alignment;
1439
1440 if (r < dsa->dsa_firstaligned)
1441 dsa->dsa_firstaligned = dsa->dsa_firstaligned
1442 - r;
1443 else
1444 dsa->dsa_firstaligned = (dsa->dsa_firstaligned
1445 + dsa->dsa_alignment) - r;
1446 }
1447
1448 return 0;
1449 }
1450
1451 default:
1452 return dk_ioctl(dksc, dev, cmd, addr, flag, l);
1453 }
1454
1455 #ifdef DIAGNOSTIC
1456 panic("wdioctl: impossible");
1457 #endif
1458 }
1459
1460 static int
1461 wd_discard(device_t dev, off_t pos, off_t len)
1462 {
1463 struct wd_softc *wd = device_private(dev);
1464 daddr_t bno;
1465 long size, done;
1466 long maxatonce, amount;
1467 int result;
1468
1469 if (!(wd->sc_params.atap_ata_major & WDC_VER_ATA7)
1470 || !(wd->sc_params.support_dsm & ATA_SUPPORT_DSM_TRIM)) {
1471 /* not supported; ignore request */
1472 ATADEBUG_PRINT(("wddiscard (unsupported)\n"), DEBUG_FUNCS);
1473 return 0;
1474 }
1475 maxatonce = 0xffff; /*wd->sc_params.max_dsm_blocks*/
1476
1477 ATADEBUG_PRINT(("wddiscard\n"), DEBUG_FUNCS);
1478
1479 if ((wd->sc_flags & WDF_LOADED) == 0)
1480 return EIO;
1481
1482 /* round the start up and the end down */
1483 bno = (pos + wd->sc_blksize - 1) / wd->sc_blksize;
1484 size = ((pos + len) / wd->sc_blksize) - bno;
1485
1486 done = 0;
1487 while (done < size) {
1488 amount = size - done;
1489 if (amount > maxatonce) {
1490 amount = maxatonce;
1491 }
1492 result = wd_trim(wd, bno + done, amount);
1493 if (result) {
1494 return result;
1495 }
1496 done += amount;
1497 }
1498 return 0;
1499 }
1500
1501 static int
1502 wddiscard(dev_t dev, off_t pos, off_t len)
1503 {
1504 struct wd_softc *wd;
1505 struct dk_softc *dksc;
1506 int unit;
1507
1508 unit = WDUNIT(dev);
1509 wd = device_lookup_private(&wd_cd, unit);
1510 dksc = &wd->sc_dksc;
1511
1512 return dk_discard(dksc, dev, pos, len);
1513 }
1514
1515 #ifdef B_FORMAT
1516 int
1517 wdformat(struct buf *bp)
1518 {
1519
1520 bp->b_flags |= B_FORMAT;
1521 return wdstrategy(bp);
1522 }
1523 #endif
1524
1525 int
1526 wdsize(dev_t dev)
1527 {
1528 struct wd_softc *wd;
1529 struct dk_softc *dksc;
1530 int unit;
1531
1532 ATADEBUG_PRINT(("wdsize\n"), DEBUG_FUNCS);
1533
1534 unit = WDUNIT(dev);
1535 wd = device_lookup_private(&wd_cd, unit);
1536 if (wd == NULL)
1537 return (-1);
1538 dksc = &wd->sc_dksc;
1539
1540 if (!device_is_active(dksc->sc_dev))
1541 return (-1);
1542
1543 return dk_size(dksc, dev);
1544 }
1545
1546 /*
1547 * Dump core after a system crash.
1548 */
1549 static int
1550 wddump(dev_t dev, daddr_t blkno, void *va, size_t size)
1551 {
1552 struct wd_softc *wd;
1553 struct dk_softc *dksc;
1554 int unit;
1555
1556 /* Check if recursive dump; if so, punt. */
1557 if (wddoingadump)
1558 return EFAULT;
1559 wddoingadump = 1;
1560
1561 unit = WDUNIT(dev);
1562 wd = device_lookup_private(&wd_cd, unit);
1563 if (wd == NULL)
1564 return (ENXIO);
1565 dksc = &wd->sc_dksc;
1566
1567 return dk_dump(dksc, dev, blkno, va, size, 0);
1568 }
1569
1570 static int
1571 wd_dumpblocks(device_t dev, void *va, daddr_t blkno, int nblk)
1572 {
1573 struct wd_softc *wd = device_private(dev);
1574 struct dk_softc *dksc = &wd->sc_dksc;
1575 struct disk_geom *dg = &dksc->sc_dkdev.dk_geom;
1576 struct ata_xfer *xfer = &wd->dump_xfer;
1577 int err;
1578
1579 /* Recalibrate, if first dump transfer. */
1580 if (wddumprecalibrated == 0) {
1581 wddumprecalibrated = 1;
1582 ata_channel_lock(wd->drvp->chnl_softc);
1583 /* This will directly execute the reset due to AT_POLL */
1584 ata_thread_run(wd->drvp->chnl_softc, AT_POLL,
1585 ATACH_TH_DRIVE_RESET, wd->drvp->drive);
1586
1587 wd->drvp->state = RESET;
1588 ata_channel_unlock(wd->drvp->chnl_softc);
1589 }
1590
1591 memset(xfer, 0, sizeof(*xfer));
1592 xfer->c_flags |= C_PRIVATE_ALLOC | C_SKIP_QUEUE;
1593
1594 xfer->c_bio.blkno = blkno;
1595 xfer->c_bio.flags = ATA_POLL;
1596 if (wd->sc_flags & WDF_LBA48 &&
1597 (xfer->c_bio.blkno + nblk) > wd->sc_capacity28)
1598 xfer->c_bio.flags |= ATA_LBA48;
1599 if (wd->sc_flags & WDF_LBA)
1600 xfer->c_bio.flags |= ATA_LBA;
1601 xfer->c_bio.bcount = nblk * dg->dg_secsize;
1602 xfer->c_bio.databuf = va;
1603 #ifndef WD_DUMP_NOT_TRUSTED
1604 /* This will poll until the bio is complete */
1605 wd->atabus->ata_bio(wd->drvp, xfer);
1606
1607 switch(err = xfer->c_bio.error) {
1608 case TIMEOUT:
1609 printf("wddump: device timed out");
1610 err = EIO;
1611 break;
1612 case ERR_DF:
1613 printf("wddump: drive fault");
1614 err = EIO;
1615 break;
1616 case ERR_DMA:
1617 printf("wddump: DMA error");
1618 err = EIO;
1619 break;
1620 case ERROR:
1621 printf("wddump: ");
1622 wdperror(wd, xfer);
1623 err = EIO;
1624 break;
1625 case NOERROR:
1626 err = 0;
1627 break;
1628 default:
1629 panic("wddump: unknown error type %x", err);
1630 }
1631
1632 if (err != 0) {
1633 printf("\n");
1634 return err;
1635 }
1636 #else /* WD_DUMP_NOT_TRUSTED */
1637 /* Let's just talk about this first... */
1638 printf("wd%d: dump addr 0x%x, cylin %d, head %d, sector %d\n",
1639 unit, va, cylin, head, sector);
1640 delay(500 * 1000); /* half a second */
1641 #endif
1642
1643 wddoingadump = 0;
1644 return 0;
1645 }
1646
1647 #ifdef HAS_BAD144_HANDLING
1648 /*
1649 * Internalize the bad sector table.
1650 */
1651 void
1652 bad144intern(struct wd_softc *wd)
1653 {
1654 struct dk_softc *dksc = &wd->sc_dksc;
1655 struct dkbad *bt = &dksc->sc_dkdev.dk_cpulabel->bad;
1656 struct disklabel *lp = dksc->sc_dkdev.dk_label;
1657 int i = 0;
1658
1659 ATADEBUG_PRINT(("bad144intern\n"), DEBUG_XFERS);
1660
1661 for (; i < NBT_BAD; i++) {
1662 if (bt->bt_bad[i].bt_cyl == 0xffff)
1663 break;
1664 wd->drvp->badsect[i] =
1665 bt->bt_bad[i].bt_cyl * lp->d_secpercyl +
1666 (bt->bt_bad[i].bt_trksec >> 8) * lp->d_nsectors +
1667 (bt->bt_bad[i].bt_trksec & 0xff);
1668 }
1669 for (; i < NBT_BAD+1; i++)
1670 wd->drvp->badsect[i] = -1;
1671 }
1672 #endif
1673
1674 static void
1675 wd_set_geometry(struct wd_softc *wd)
1676 {
1677 struct dk_softc *dksc = &wd->sc_dksc;
1678 struct disk_geom *dg = &dksc->sc_dkdev.dk_geom;
1679
1680 memset(dg, 0, sizeof(*dg));
1681
1682 dg->dg_secperunit = wd->sc_capacity;
1683 dg->dg_secsize = wd->sc_blksize;
1684 dg->dg_nsectors = wd->sc_params.atap_sectors;
1685 dg->dg_ntracks = wd->sc_params.atap_heads;
1686 if ((wd->sc_flags & WDF_LBA) == 0)
1687 dg->dg_ncylinders = wd->sc_params.atap_cylinders;
1688
1689 disk_set_info(dksc->sc_dev, &dksc->sc_dkdev, wd->sc_typename);
1690 }
1691
1692 int
1693 wd_get_params(struct wd_softc *wd, struct ataparams *params)
1694 {
1695 int retry = 0;
1696 struct ata_channel *chp = wd->drvp->chnl_softc;
1697 const int flags = AT_WAIT;
1698
1699 again:
1700 switch (wd->atabus->ata_get_params(wd->drvp, flags, params)) {
1701 case CMD_AGAIN:
1702 return 1;
1703 case CMD_ERR:
1704 if (retry == 0) {
1705 retry++;
1706 ata_channel_lock(chp);
1707 (*wd->atabus->ata_reset_drive)(wd->drvp, flags, NULL);
1708 ata_channel_unlock(chp);
1709 goto again;
1710 }
1711
1712 if (wd->drvp->drive_type != ATA_DRIVET_OLD)
1713 return 1;
1714 /*
1715 * We `know' there's a drive here; just assume it's old.
1716 * This geometry is only used to read the MBR and print a
1717 * (false) attach message.
1718 */
1719 strncpy(params->atap_model, "ST506",
1720 sizeof params->atap_model);
1721 params->atap_config = ATA_CFG_FIXED;
1722 params->atap_cylinders = 1024;
1723 params->atap_heads = 8;
1724 params->atap_sectors = 17;
1725 params->atap_multi = 1;
1726 params->atap_capabilities1 = params->atap_capabilities2 = 0;
1727 wd->drvp->ata_vers = -1; /* Mark it as pre-ATA */
1728 /* FALLTHROUGH */
1729 case CMD_OK:
1730 return 0;
1731 default:
1732 panic("wd_get_params: bad return code from ata_get_params");
1733 /* NOTREACHED */
1734 }
1735 }
1736
1737 int
1738 wd_getcache(struct wd_softc *wd, int *bitsp)
1739 {
1740 struct ataparams params;
1741
1742 if (wd_get_params(wd, ¶ms) != 0)
1743 return EIO;
1744 if (params.atap_cmd_set1 == 0x0000 ||
1745 params.atap_cmd_set1 == 0xffff ||
1746 (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0) {
1747 *bitsp = 0;
1748 return 0;
1749 }
1750 *bitsp = DKCACHE_WCHANGE | DKCACHE_READ;
1751 if (params.atap_cmd1_en & WDC_CMD1_CACHE)
1752 *bitsp |= DKCACHE_WRITE;
1753
1754 if (WD_USE_NCQ(wd) || (wd->drvp->drive_flags & ATA_DRIVE_WFUA))
1755 *bitsp |= DKCACHE_FUA;
1756
1757 return 0;
1758 }
1759
1760 const char at_errbits[] = "\20\10ERROR\11TIMEOU\12DF";
1761
1762 int
1763 wd_setcache(struct wd_softc *wd, int bits)
1764 {
1765 struct dk_softc *dksc = &wd->sc_dksc;
1766 struct ataparams params;
1767 struct ata_xfer *xfer;
1768 int error;
1769
1770 if (wd_get_params(wd, ¶ms) != 0)
1771 return EIO;
1772
1773 if (params.atap_cmd_set1 == 0x0000 ||
1774 params.atap_cmd_set1 == 0xffff ||
1775 (params.atap_cmd_set1 & WDC_CMD1_CACHE) == 0)
1776 return EOPNOTSUPP;
1777
1778 if ((bits & DKCACHE_READ) == 0 ||
1779 (bits & DKCACHE_SAVE) != 0)
1780 return EOPNOTSUPP;
1781
1782 xfer = ata_get_xfer(wd->drvp->chnl_softc, true);
1783
1784 xfer->c_ata_c.r_command = SET_FEATURES;
1785 xfer->c_ata_c.r_st_bmask = 0;
1786 xfer->c_ata_c.r_st_pmask = 0;
1787 xfer->c_ata_c.timeout = 30000; /* 30s timeout */
1788 xfer->c_ata_c.flags = AT_WAIT;
1789 if (bits & DKCACHE_WRITE)
1790 xfer->c_ata_c.r_features = WDSF_WRITE_CACHE_EN;
1791 else
1792 xfer->c_ata_c.r_features = WDSF_WRITE_CACHE_DS;
1793
1794 wd->atabus->ata_exec_command(wd->drvp, xfer);
1795 ata_wait_cmd(wd->drvp->chnl_softc, xfer);
1796
1797 if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1798 char sbuf[sizeof(at_errbits) + 64];
1799 snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
1800 aprint_error_dev(dksc->sc_dev, "wd_setcache: status=%s\n", sbuf);
1801 error = EIO;
1802 goto out;
1803 }
1804
1805 error = 0;
1806
1807 out:
1808 ata_free_xfer(wd->drvp->chnl_softc, xfer);
1809 return error;
1810 }
1811
1812 static int
1813 wd_standby(struct wd_softc *wd, int flags)
1814 {
1815 struct dk_softc *dksc = &wd->sc_dksc;
1816 struct ata_xfer *xfer;
1817 int error;
1818
1819 aprint_debug_dev(dksc->sc_dev, "standby immediate\n");
1820 xfer = ata_get_xfer(wd->drvp->chnl_softc, true);
1821
1822 xfer->c_ata_c.r_command = WDCC_STANDBY_IMMED;
1823 xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
1824 xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
1825 xfer->c_ata_c.flags = flags;
1826 xfer->c_ata_c.timeout = 30000; /* 30s timeout */
1827
1828 wd->atabus->ata_exec_command(wd->drvp, xfer);
1829 ata_wait_cmd(wd->drvp->chnl_softc, xfer);
1830
1831 if (xfer->c_ata_c.flags & AT_ERROR) {
1832 if (xfer->c_ata_c.r_error == WDCE_ABRT) {
1833 /* command not supported */
1834 aprint_debug_dev(dksc->sc_dev,
1835 "standby immediate not supported\n");
1836 error = ENODEV;
1837 goto out;
1838 }
1839 }
1840 if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1841 char sbuf[sizeof(at_errbits) + 64];
1842 snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
1843 aprint_error_dev(dksc->sc_dev, "wd_standby: status=%s\n", sbuf);
1844 error = EIO;
1845 goto out;
1846 }
1847 error = 0;
1848
1849 out:
1850 ata_free_xfer(wd->drvp->chnl_softc, xfer);
1851 return error;
1852 }
1853
1854 int
1855 wd_flushcache(struct wd_softc *wd, int flags, bool start_self)
1856 {
1857 struct dk_softc *dksc = &wd->sc_dksc;
1858 struct ata_xfer *xfer;
1859 int error;
1860
1861 /*
1862 * WDCC_FLUSHCACHE is here since ATA-4, but some drives report
1863 * only ATA-2 and still support it.
1864 */
1865 if (wd->drvp->ata_vers < 4 &&
1866 ((wd->sc_params.atap_cmd_set2 & WDC_CMD2_FC) == 0 ||
1867 wd->sc_params.atap_cmd_set2 == 0xffff))
1868 return ENODEV;
1869
1870 xfer = ata_get_xfer(wd->drvp->chnl_softc, true);
1871
1872 if ((wd->sc_params.atap_cmd2_en & ATA_CMD2_LBA48) != 0 &&
1873 (wd->sc_params.atap_cmd2_en & ATA_CMD2_FCE) != 0) {
1874 xfer->c_ata_c.r_command = WDCC_FLUSHCACHE_EXT;
1875 flags |= AT_LBA48;
1876 } else
1877 xfer->c_ata_c.r_command = WDCC_FLUSHCACHE;
1878 xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
1879 xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
1880 xfer->c_ata_c.flags = flags | AT_READREG;
1881 xfer->c_ata_c.timeout = 300000; /* 5m timeout */
1882
1883 wd->atabus->ata_exec_command(wd->drvp, xfer);
1884 ata_wait_cmd(wd->drvp->chnl_softc, xfer);
1885
1886 if (xfer->c_ata_c.flags & AT_ERROR) {
1887 if (xfer->c_ata_c.r_error == WDCE_ABRT) {
1888 /* command not supported */
1889 error = ENODEV;
1890 goto out_xfer;
1891 }
1892 }
1893 if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1894 char sbuf[sizeof(at_errbits) + 64];
1895 snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
1896 aprint_error_dev(dksc->sc_dev, "wd_flushcache: status=%s\n",
1897 sbuf);
1898 error = EIO;
1899 goto out_xfer;
1900 }
1901 wd->sc_flags &= ~WDF_DIRTY;
1902 error = 0;
1903
1904 out_xfer:
1905 ata_free_xfer(wd->drvp->chnl_softc, xfer);
1906 return error;
1907 }
1908
1909 /*
1910 * Execute TRIM command, assumes sleep context.
1911 */
1912 static int
1913 wd_trim(struct wd_softc *wd, daddr_t bno, long size)
1914 {
1915 struct dk_softc *dksc = &wd->sc_dksc;
1916 struct ata_xfer *xfer;
1917 int error;
1918 unsigned char *req;
1919
1920 xfer = ata_get_xfer(wd->drvp->chnl_softc, true);
1921
1922 req = kmem_zalloc(512, KM_SLEEP);
1923 req[0] = bno & 0xff;
1924 req[1] = (bno >> 8) & 0xff;
1925 req[2] = (bno >> 16) & 0xff;
1926 req[3] = (bno >> 24) & 0xff;
1927 req[4] = (bno >> 32) & 0xff;
1928 req[5] = (bno >> 40) & 0xff;
1929 req[6] = size & 0xff;
1930 req[7] = (size >> 8) & 0xff;
1931
1932 /*
1933 * XXX We could possibly use NCQ TRIM, which supports executing
1934 * this command concurrently. It would need some investigation, some
1935 * early or not so early disk firmware caused data loss with NCQ TRIM.
1936 * atastart() et.al would need to be adjusted to allow and support
1937 * running several non-I/O ATA commands in parallel.
1938 */
1939
1940 xfer->c_ata_c.r_command = ATA_DATA_SET_MANAGEMENT;
1941 xfer->c_ata_c.r_count = 1;
1942 xfer->c_ata_c.r_features = ATA_SUPPORT_DSM_TRIM;
1943 xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
1944 xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
1945 xfer->c_ata_c.timeout = 30000; /* 30s timeout */
1946 xfer->c_ata_c.data = req;
1947 xfer->c_ata_c.bcount = 512;
1948 xfer->c_ata_c.flags |= AT_WRITE | AT_WAIT;
1949
1950 wd->atabus->ata_exec_command(wd->drvp, xfer);
1951 ata_wait_cmd(wd->drvp->chnl_softc, xfer);
1952
1953 kmem_free(req, 512);
1954 if (xfer->c_ata_c.flags & AT_ERROR) {
1955 if (xfer->c_ata_c.r_error == WDCE_ABRT) {
1956 /* command not supported */
1957 error = ENODEV;
1958 goto out;
1959 }
1960 }
1961 if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
1962 char sbuf[sizeof(at_errbits) + 64];
1963 snprintb(sbuf, sizeof(sbuf), at_errbits, xfer->c_ata_c.flags);
1964 aprint_error_dev(dksc->sc_dev, "wd_trim: status=%s\n",
1965 sbuf);
1966 error = EIO;
1967 goto out;
1968 }
1969 error = 0;
1970
1971 out:
1972 ata_free_xfer(wd->drvp->chnl_softc, xfer);
1973 return error;
1974 }
1975
1976 bool
1977 wd_shutdown(device_t dev, int how)
1978 {
1979 struct wd_softc *wd = device_private(dev);
1980
1981 /* the adapter needs to be enabled */
1982 if (wd->atabus->ata_addref(wd->drvp))
1983 return true; /* no need to complain */
1984
1985 wd_flushcache(wd, AT_POLL, false);
1986 if ((how & RB_POWERDOWN) == RB_POWERDOWN)
1987 wd_standby(wd, AT_POLL);
1988 return true;
1989 }
1990
1991 /*
1992 * Allocate space for a ioctl queue structure. Mostly taken from
1993 * scsipi_ioctl.c
1994 */
1995 struct wd_ioctl *
1996 wi_get(struct wd_softc *wd)
1997 {
1998 struct wd_ioctl *wi;
1999
2000 wi = kmem_zalloc(sizeof(struct wd_ioctl), KM_SLEEP);
2001 wi->wi_softc = wd;
2002 buf_init(&wi->wi_bp);
2003
2004 return (wi);
2005 }
2006
2007 /*
2008 * Free an ioctl structure and remove it from our list
2009 */
2010
2011 void
2012 wi_free(struct wd_ioctl *wi)
2013 {
2014 buf_destroy(&wi->wi_bp);
2015 kmem_free(wi, sizeof(*wi));
2016 }
2017
2018 /*
2019 * Find a wd_ioctl structure based on the struct buf.
2020 */
2021
2022 struct wd_ioctl *
2023 wi_find(struct buf *bp)
2024 {
2025 return container_of(bp, struct wd_ioctl, wi_bp);
2026 }
2027
2028 static uint
2029 wi_sector_size(const struct wd_ioctl * const wi)
2030 {
2031 switch (wi->wi_atareq.command) {
2032 case WDCC_READ:
2033 case WDCC_WRITE:
2034 case WDCC_READMULTI:
2035 case WDCC_WRITEMULTI:
2036 case WDCC_READDMA:
2037 case WDCC_WRITEDMA:
2038 case WDCC_READ_EXT:
2039 case WDCC_WRITE_EXT:
2040 case WDCC_READMULTI_EXT:
2041 case WDCC_WRITEMULTI_EXT:
2042 case WDCC_READDMA_EXT:
2043 case WDCC_WRITEDMA_EXT:
2044 case WDCC_READ_FPDMA_QUEUED:
2045 case WDCC_WRITE_FPDMA_QUEUED:
2046 return wi->wi_softc->sc_blksize;
2047 default:
2048 return 512;
2049 }
2050 }
2051
2052 /*
2053 * Ioctl pseudo strategy routine
2054 *
2055 * This is mostly stolen from scsipi_ioctl.c:scsistrategy(). What
2056 * happens here is:
2057 *
2058 * - wdioctl() queues a wd_ioctl structure.
2059 *
2060 * - wdioctl() calls physio/wdioctlstrategy based on whether or not
2061 * user space I/O is required. If physio() is called, physio() eventually
2062 * calls wdioctlstrategy().
2063 *
2064 * - In either case, wdioctlstrategy() calls wd->atabus->ata_exec_command()
2065 * to perform the actual command
2066 *
2067 * The reason for the use of the pseudo strategy routine is because
2068 * when doing I/O to/from user space, physio _really_ wants to be in
2069 * the loop. We could put the entire buffer into the ioctl request
2070 * structure, but that won't scale if we want to do things like download
2071 * microcode.
2072 */
2073
2074 void
2075 wdioctlstrategy(struct buf *bp)
2076 {
2077 struct wd_ioctl *wi;
2078 struct ata_xfer *xfer;
2079 int error = 0;
2080
2081 wi = wi_find(bp);
2082 if (wi == NULL) {
2083 printf("wdioctlstrategy: "
2084 "No matching ioctl request found in queue\n");
2085 error = EINVAL;
2086 goto out2;
2087 }
2088
2089 xfer = ata_get_xfer(wi->wi_softc->drvp->chnl_softc, true);
2090
2091 /*
2092 * Abort if physio broke up the transfer
2093 */
2094
2095 if (bp->b_bcount != wi->wi_atareq.datalen) {
2096 printf("physio split wd ioctl request... cannot proceed\n");
2097 error = EIO;
2098 goto out;
2099 }
2100
2101 /*
2102 * Abort if we didn't get a buffer size that was a multiple of
2103 * our sector size (or overflows CHS/LBA28 sector count)
2104 */
2105
2106 if ((bp->b_bcount % wi_sector_size(wi)) != 0 ||
2107 (bp->b_bcount / wi_sector_size(wi)) >=
2108 (1 << NBBY)) {
2109 error = EINVAL;
2110 goto out;
2111 }
2112
2113 /*
2114 * Make sure a timeout was supplied in the ioctl request
2115 */
2116
2117 if (wi->wi_atareq.timeout == 0) {
2118 error = EINVAL;
2119 goto out;
2120 }
2121
2122 if (wi->wi_atareq.flags & ATACMD_READ)
2123 xfer->c_ata_c.flags |= AT_READ;
2124 else if (wi->wi_atareq.flags & ATACMD_WRITE)
2125 xfer->c_ata_c.flags |= AT_WRITE;
2126
2127 if (wi->wi_atareq.flags & ATACMD_READREG)
2128 xfer->c_ata_c.flags |= AT_READREG;
2129
2130 if ((wi->wi_atareq.flags & ATACMD_LBA) != 0)
2131 xfer->c_ata_c.flags |= AT_LBA;
2132
2133 xfer->c_ata_c.flags |= AT_WAIT;
2134
2135 xfer->c_ata_c.timeout = wi->wi_atareq.timeout;
2136 xfer->c_ata_c.r_command = wi->wi_atareq.command;
2137 xfer->c_ata_c.r_lba = ((wi->wi_atareq.head & 0x0f) << 24) |
2138 (wi->wi_atareq.cylinder << 8) |
2139 wi->wi_atareq.sec_num;
2140 xfer->c_ata_c.r_count = wi->wi_atareq.sec_count;
2141 xfer->c_ata_c.r_features = wi->wi_atareq.features;
2142 xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
2143 xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
2144 xfer->c_ata_c.data = wi->wi_bp.b_data;
2145 xfer->c_ata_c.bcount = wi->wi_bp.b_bcount;
2146
2147 wi->wi_softc->atabus->ata_exec_command(wi->wi_softc->drvp, xfer);
2148 ata_wait_cmd(wi->wi_softc->drvp->chnl_softc, xfer);
2149
2150 if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
2151 if (xfer->c_ata_c.flags & AT_ERROR) {
2152 wi->wi_atareq.retsts = ATACMD_ERROR;
2153 wi->wi_atareq.error = xfer->c_ata_c.r_error;
2154 } else if (xfer->c_ata_c.flags & AT_DF)
2155 wi->wi_atareq.retsts = ATACMD_DF;
2156 else
2157 wi->wi_atareq.retsts = ATACMD_TIMEOUT;
2158 } else {
2159 wi->wi_atareq.retsts = ATACMD_OK;
2160 if (wi->wi_atareq.flags & ATACMD_READREG) {
2161 wi->wi_atareq.command = xfer->c_ata_c.r_status;
2162 wi->wi_atareq.features = xfer->c_ata_c.r_error;
2163 wi->wi_atareq.sec_count = xfer->c_ata_c.r_count;
2164 wi->wi_atareq.sec_num = xfer->c_ata_c.r_lba & 0xff;
2165 wi->wi_atareq.head = (xfer->c_ata_c.r_device & 0xf0) |
2166 ((xfer->c_ata_c.r_lba >> 24) & 0x0f);
2167 wi->wi_atareq.cylinder =
2168 (xfer->c_ata_c.r_lba >> 8) & 0xffff;
2169 wi->wi_atareq.error = xfer->c_ata_c.r_error;
2170 }
2171 }
2172
2173 out:
2174 ata_free_xfer(wi->wi_softc->drvp->chnl_softc, xfer);
2175 out2:
2176 bp->b_error = error;
2177 if (error)
2178 bp->b_resid = bp->b_bcount;
2179 biodone(bp);
2180 }
2181
2182 static void
2183 wd_sysctl_attach(struct wd_softc *wd)
2184 {
2185 struct dk_softc *dksc = &wd->sc_dksc;
2186 const struct sysctlnode *node;
2187 int error;
2188
2189 /* sysctl set-up */
2190 if (sysctl_createv(&wd->nodelog, 0, NULL, &node,
2191 0, CTLTYPE_NODE, dksc->sc_xname,
2192 SYSCTL_DESCR("wd driver settings"),
2193 NULL, 0, NULL, 0,
2194 CTL_HW, CTL_CREATE, CTL_EOL) != 0) {
2195 aprint_error_dev(dksc->sc_dev,
2196 "could not create %s.%s sysctl node\n",
2197 "hw", dksc->sc_xname);
2198 return;
2199 }
2200
2201 wd->drv_ncq = true;
2202 if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
2203 CTLFLAG_READWRITE, CTLTYPE_BOOL, "use_ncq",
2204 SYSCTL_DESCR("use NCQ if supported"),
2205 NULL, 0, &wd->drv_ncq, 0,
2206 CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
2207 != 0) {
2208 aprint_error_dev(dksc->sc_dev,
2209 "could not create %s.%s.use_ncq sysctl - error %d\n",
2210 "hw", dksc->sc_xname, error);
2211 return;
2212 }
2213
2214 wd->drv_ncq_prio = false;
2215 if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
2216 CTLFLAG_READWRITE, CTLTYPE_BOOL, "use_ncq_prio",
2217 SYSCTL_DESCR("use NCQ PRIORITY if supported"),
2218 NULL, 0, &wd->drv_ncq_prio, 0,
2219 CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
2220 != 0) {
2221 aprint_error_dev(dksc->sc_dev,
2222 "could not create %s.%s.use_ncq_prio sysctl - error %d\n",
2223 "hw", dksc->sc_xname, error);
2224 return;
2225 }
2226
2227 #ifdef WD_CHAOS_MONKEY
2228 wd->drv_chaos_freq = 0;
2229 if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
2230 CTLFLAG_READWRITE, CTLTYPE_INT, "chaos_freq",
2231 SYSCTL_DESCR("simulated bio read error rate"),
2232 NULL, 0, &wd->drv_chaos_freq, 0,
2233 CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
2234 != 0) {
2235 aprint_error_dev(dksc->sc_dev,
2236 "could not create %s.%s.chaos_freq sysctl - error %d\n",
2237 "hw", dksc->sc_xname, error);
2238 return;
2239 }
2240
2241 wd->drv_chaos_cnt = 0;
2242 if ((error = sysctl_createv(&wd->nodelog, 0, NULL, NULL,
2243 CTLFLAG_READONLY, CTLTYPE_INT, "chaos_cnt",
2244 SYSCTL_DESCR("number of processed bio reads"),
2245 NULL, 0, &wd->drv_chaos_cnt, 0,
2246 CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL))
2247 != 0) {
2248 aprint_error_dev(dksc->sc_dev,
2249 "could not create %s.%s.chaos_cnt sysctl - error %d\n",
2250 "hw", dksc->sc_xname, error);
2251 return;
2252 }
2253 #endif
2254
2255 }
2256
2257 static void
2258 wd_sysctl_detach(struct wd_softc *wd)
2259 {
2260 sysctl_teardown(&wd->nodelog);
2261 }
2262
2263 #ifdef ATADEBUG
2264 int wddebug(void);
2265
2266 int
2267 wddebug(void)
2268 {
2269 struct wd_softc *wd;
2270 struct dk_softc *dksc;
2271 int unit;
2272
2273 for (unit = 0; unit <= 3; unit++) {
2274 wd = device_lookup_private(&wd_cd, unit);
2275 if (wd == NULL)
2276 continue;
2277 dksc = &wd->sc_dksc;
2278 printf("%s fl %x bufq %p:\n",
2279 dksc->sc_xname, wd->sc_flags, bufq_peek(dksc->sc_bufq));
2280
2281 atachannel_debug(wd->drvp->chnl_softc);
2282 }
2283 return 0;
2284 }
2285 #endif /* ATADEBUG */
2286