xref: /netbsd-src/sys/dev/ld.c (revision f21b7d7f2cbdd5c14b3882c4e8a3d43580d460a6)
1 /*	$NetBSD: ld.c,v 1.97 2016/09/27 03:33:32 pgoyette Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 2000 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Doran and Charles M. Hannum.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Disk driver for use by RAID controllers.
34  */
35 
36 #include <sys/cdefs.h>
37 __KERNEL_RCSID(0, "$NetBSD: ld.c,v 1.97 2016/09/27 03:33:32 pgoyette Exp $");
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #include <sys/device.h>
43 #include <sys/queue.h>
44 #include <sys/proc.h>
45 #include <sys/buf.h>
46 #include <sys/bufq.h>
47 #include <sys/endian.h>
48 #include <sys/disklabel.h>
49 #include <sys/disk.h>
50 #include <sys/dkio.h>
51 #include <sys/stat.h>
52 #include <sys/conf.h>
53 #include <sys/fcntl.h>
54 #include <sys/vnode.h>
55 #include <sys/syslog.h>
56 #include <sys/mutex.h>
57 #include <sys/module.h>
58 
59 #include <dev/ldvar.h>
60 
61 static void	ldminphys(struct buf *bp);
62 static bool	ld_suspend(device_t, const pmf_qual_t *);
63 static bool	ld_shutdown(device_t, int);
64 static int	ld_diskstart(device_t, struct buf *bp);
65 static void	ld_iosize(device_t, int *);
66 static int	ld_dumpblocks(device_t, void *, daddr_t, int);
67 static void	ld_fake_geometry(struct ld_softc *);
68 static void	ld_set_geometry(struct ld_softc *);
69 static void	ld_config_interrupts (device_t);
70 static int	ld_lastclose(device_t);
71 static int	ld_discard(device_t, off_t, off_t);
72 
73 extern struct	cfdriver ld_cd;
74 
75 static dev_type_open(ldopen);
76 static dev_type_close(ldclose);
77 static dev_type_read(ldread);
78 static dev_type_write(ldwrite);
79 static dev_type_ioctl(ldioctl);
80 static dev_type_strategy(ldstrategy);
81 static dev_type_dump(lddump);
82 static dev_type_size(ldsize);
83 static dev_type_discard(lddiscard);
84 
85 const struct bdevsw ld_bdevsw = {
86 	.d_open = ldopen,
87 	.d_close = ldclose,
88 	.d_strategy = ldstrategy,
89 	.d_ioctl = ldioctl,
90 	.d_dump = lddump,
91 	.d_psize = ldsize,
92 	.d_discard = lddiscard,
93 	.d_flag = D_DISK | D_MPSAFE
94 };
95 
96 const struct cdevsw ld_cdevsw = {
97 	.d_open = ldopen,
98 	.d_close = ldclose,
99 	.d_read = ldread,
100 	.d_write = ldwrite,
101 	.d_ioctl = ldioctl,
102 	.d_stop = nostop,
103 	.d_tty = notty,
104 	.d_poll = nopoll,
105 	.d_mmap = nommap,
106 	.d_kqfilter = nokqfilter,
107 	.d_discard = lddiscard,
108 	.d_flag = D_DISK | D_MPSAFE
109 };
110 
111 static struct	dkdriver lddkdriver = {
112 	.d_open = ldopen,
113 	.d_close = ldclose,
114 	.d_strategy = ldstrategy,
115 	.d_iosize = ld_iosize,
116 	.d_minphys  = ldminphys,
117 	.d_diskstart = ld_diskstart,
118 	.d_dumpblocks = ld_dumpblocks,
119 	.d_lastclose = ld_lastclose,
120 	.d_discard = ld_discard
121 };
122 
123 void
124 ldattach(struct ld_softc *sc, const char *default_strategy)
125 {
126 	device_t self = sc->sc_dv;
127 	struct dk_softc *dksc = &sc->sc_dksc;
128 
129 	mutex_init(&sc->sc_mutex, MUTEX_DEFAULT, IPL_VM);
130 	cv_init(&sc->sc_drain, "lddrain");
131 
132 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
133 		return;
134 	}
135 
136 	/* Initialise dk and disk structure. */
137 	dk_init(dksc, self, DKTYPE_LD);
138 	disk_init(&dksc->sc_dkdev, dksc->sc_xname, &lddkdriver);
139 
140 	if (sc->sc_maxxfer > MAXPHYS)
141 		sc->sc_maxxfer = MAXPHYS;
142 
143 	/* Build synthetic geometry if necessary. */
144 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
145 	    sc->sc_ncylinders == 0)
146 	    ld_fake_geometry(sc);
147 
148 	sc->sc_disksize512 = sc->sc_secperunit * sc->sc_secsize / DEV_BSIZE;
149 
150 	/* Attach dk and disk subsystems */
151 	dk_attach(dksc);
152 	disk_attach(&dksc->sc_dkdev);
153 	ld_set_geometry(sc);
154 
155 	bufq_alloc(&dksc->sc_bufq, default_strategy, BUFQ_SORT_RAWBLOCK);
156 
157 	/* Register with PMF */
158 	if (!pmf_device_register1(dksc->sc_dev, ld_suspend, NULL, ld_shutdown))
159 		aprint_error_dev(dksc->sc_dev,
160 		    "couldn't establish power handler\n");
161 
162 	/* Discover wedges on this disk. */
163 	config_interrupts(sc->sc_dv, ld_config_interrupts);
164 }
165 
166 int
167 ldadjqparam(struct ld_softc *sc, int xmax)
168 {
169 
170 	mutex_enter(&sc->sc_mutex);
171 	sc->sc_maxqueuecnt = xmax;
172 	mutex_exit(&sc->sc_mutex);
173 
174 	return (0);
175 }
176 
177 int
178 ldbegindetach(struct ld_softc *sc, int flags)
179 {
180 	struct dk_softc *dksc = &sc->sc_dksc;
181 	int rv = 0;
182 
183 	if ((sc->sc_flags & LDF_ENABLED) == 0)
184 		return (0);
185 
186 	rv = disk_begindetach(&dksc->sc_dkdev, ld_lastclose, dksc->sc_dev, flags);
187 
188 	if (rv != 0)
189 		return rv;
190 
191 	mutex_enter(&sc->sc_mutex);
192 	sc->sc_maxqueuecnt = 0;
193 
194 	while (sc->sc_queuecnt > 0) {
195 		sc->sc_flags |= LDF_DRAIN;
196 		cv_wait(&sc->sc_drain, &sc->sc_mutex);
197 	}
198 	mutex_exit(&sc->sc_mutex);
199 
200 	return (rv);
201 }
202 
203 void
204 ldenddetach(struct ld_softc *sc)
205 {
206 	struct dk_softc *dksc = &sc->sc_dksc;
207 	int bmaj, cmaj, i, mn;
208 
209 	if ((sc->sc_flags & LDF_ENABLED) == 0)
210 		return;
211 
212 	mutex_enter(&sc->sc_mutex);
213 
214 	/* Wait for commands queued with the hardware to complete. */
215 	if (sc->sc_queuecnt != 0) {
216 		if (cv_timedwait(&sc->sc_drain, &sc->sc_mutex, 30 * hz))
217 			printf("%s: not drained\n", dksc->sc_xname);
218 	}
219 	mutex_exit(&sc->sc_mutex);
220 
221 	/* Kill off any queued buffers. */
222 	dk_drain(dksc);
223 	bufq_free(dksc->sc_bufq);
224 
225 	/* Locate the major numbers. */
226 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
227 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
228 
229 	/* Nuke the vnodes for any open instances. */
230 	for (i = 0; i < MAXPARTITIONS; i++) {
231 		mn = DISKMINOR(device_unit(dksc->sc_dev), i);
232 		vdevgone(bmaj, mn, mn, VBLK);
233 		vdevgone(cmaj, mn, mn, VCHR);
234 	}
235 
236 	/* Delete all of our wedges. */
237 	dkwedge_delall(&dksc->sc_dkdev);
238 
239 	/* Detach from the disk list. */
240 	disk_detach(&dksc->sc_dkdev);
241 	disk_destroy(&dksc->sc_dkdev);
242 
243 	dk_detach(dksc);
244 
245 	/* Deregister with PMF */
246 	pmf_device_deregister(dksc->sc_dev);
247 
248 	/*
249 	 * XXX We can't really flush the cache here, beceause the
250 	 * XXX device may already be non-existent from the controller's
251 	 * XXX perspective.
252 	 */
253 #if 0
254 	/* Flush the device's cache. */
255 	if (sc->sc_flush != NULL)
256 		if ((*sc->sc_flush)(sc, 0) != 0)
257 			device_printf(dksc->sc_dev, "unable to flush cache\n");
258 #endif
259 	cv_destroy(&sc->sc_drain);
260 	mutex_destroy(&sc->sc_mutex);
261 }
262 
263 /* ARGSUSED */
264 static bool
265 ld_suspend(device_t dev, const pmf_qual_t *qual)
266 {
267 	return ld_shutdown(dev, 0);
268 }
269 
270 /* ARGSUSED */
271 static bool
272 ld_shutdown(device_t dev, int flags)
273 {
274 	struct ld_softc *sc = device_private(dev);
275 	struct dk_softc *dksc = &sc->sc_dksc;
276 
277 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, LDFL_POLL) != 0) {
278 		device_printf(dksc->sc_dev, "unable to flush cache\n");
279 		return false;
280 	}
281 
282 	return true;
283 }
284 
285 /* ARGSUSED */
286 static int
287 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
288 {
289 	struct ld_softc *sc;
290 	struct dk_softc *dksc;
291 	int unit;
292 
293 	unit = DISKUNIT(dev);
294 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
295 		return (ENXIO);
296 	dksc = &sc->sc_dksc;
297 
298 	return dk_open(dksc, dev, flags, fmt, l);
299 }
300 
301 static int
302 ld_lastclose(device_t self)
303 {
304 	struct ld_softc *sc = device_private(self);
305 
306 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, 0) != 0)
307 		device_printf(self, "unable to flush cache\n");
308 
309 	return 0;
310 }
311 
312 /* ARGSUSED */
313 static int
314 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
315 {
316 	struct ld_softc *sc;
317 	struct dk_softc *dksc;
318 	int unit;
319 
320 	unit = DISKUNIT(dev);
321 	sc = device_lookup_private(&ld_cd, unit);
322 	dksc = &sc->sc_dksc;
323 
324 	return dk_close(dksc, dev, flags, fmt, l);
325 }
326 
327 /* ARGSUSED */
328 static int
329 ldread(dev_t dev, struct uio *uio, int ioflag)
330 {
331 
332 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
333 }
334 
335 /* ARGSUSED */
336 static int
337 ldwrite(dev_t dev, struct uio *uio, int ioflag)
338 {
339 
340 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
341 }
342 
343 /* ARGSUSED */
344 static int
345 ldioctl(dev_t dev, u_long cmd, void *addr, int32_t flag, struct lwp *l)
346 {
347 	struct ld_softc *sc;
348 	struct dk_softc *dksc;
349 	int unit, error;
350 
351 	unit = DISKUNIT(dev);
352 	sc = device_lookup_private(&ld_cd, unit);
353 	dksc = &sc->sc_dksc;
354 
355 	error = 0;
356 
357 	switch (cmd) {
358 	case DIOCCACHESYNC:
359 		/*
360 		 * XXX Do we really need to care about having a writable
361 		 * file descriptor here?
362 		 */
363 		if ((flag & FWRITE) == 0)
364 			error = EBADF;
365 		else if (sc->sc_flush)
366 			error = (*sc->sc_flush)(sc, 0);
367 		else
368 			error = 0;	/* XXX Error out instead? */
369 		break;
370 
371 	default:
372 		error = dk_ioctl(dksc, dev, cmd, addr, flag, l);
373 		break;
374 	}
375 
376 	return (error);
377 }
378 
379 static void
380 ldstrategy(struct buf *bp)
381 {
382 	struct ld_softc *sc;
383 	struct dk_softc *dksc;
384 	int unit;
385 
386 	unit = DISKUNIT(bp->b_dev);
387 	sc = device_lookup_private(&ld_cd, unit);
388 	dksc = &sc->sc_dksc;
389 
390 	dk_strategy(dksc, bp);
391 }
392 
393 static int
394 ld_diskstart(device_t dev, struct buf *bp)
395 {
396 	struct ld_softc *sc = device_private(dev);
397 	int error;
398 
399 	if (sc->sc_queuecnt >= sc->sc_maxqueuecnt)
400 		return EAGAIN;
401 
402 	mutex_enter(&sc->sc_mutex);
403 
404 	if (sc->sc_queuecnt >= sc->sc_maxqueuecnt)
405 		error = EAGAIN;
406 	else {
407 		error = (*sc->sc_start)(sc, bp);
408 		if (error == 0)
409 			sc->sc_queuecnt++;
410 	}
411 
412 	mutex_exit(&sc->sc_mutex);
413 
414 	return error;
415 }
416 
417 void
418 lddone(struct ld_softc *sc, struct buf *bp)
419 {
420 	struct dk_softc *dksc = &sc->sc_dksc;
421 
422 	dk_done(dksc, bp);
423 
424 	mutex_enter(&sc->sc_mutex);
425 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
426 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
427 			sc->sc_flags &= ~LDF_DRAIN;
428 			cv_broadcast(&sc->sc_drain);
429 		}
430 		mutex_exit(&sc->sc_mutex);
431 		dk_start(dksc, NULL);
432 	} else
433 		mutex_exit(&sc->sc_mutex);
434 }
435 
436 static int
437 ldsize(dev_t dev)
438 {
439 	struct ld_softc *sc;
440 	struct dk_softc *dksc;
441 	int unit;
442 
443 	unit = DISKUNIT(dev);
444 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
445 		return (ENODEV);
446 	dksc = &sc->sc_dksc;
447 
448 	if ((sc->sc_flags & LDF_ENABLED) == 0)
449 		return (ENODEV);
450 
451 	return dk_size(dksc, dev);
452 }
453 
454 /*
455  * Take a dump.
456  */
457 static int
458 lddump(dev_t dev, daddr_t blkno, void *va, size_t size)
459 {
460 	struct ld_softc *sc;
461 	struct dk_softc *dksc;
462 	int unit;
463 
464 	unit = DISKUNIT(dev);
465 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
466 		return (ENXIO);
467 	dksc = &sc->sc_dksc;
468 
469 	if ((sc->sc_flags & LDF_ENABLED) == 0)
470 		return (ENODEV);
471 
472 	return dk_dump(dksc, dev, blkno, va, size);
473 }
474 
475 static int
476 ld_dumpblocks(device_t dev, void *va, daddr_t blkno, int nblk)
477 {
478 	struct ld_softc *sc = device_private(dev);
479 
480 	if (sc->sc_dump == NULL)
481 		return (ENODEV);
482 
483 	return (*sc->sc_dump)(sc, va, blkno, nblk);
484 }
485 
486 /*
487  * Adjust the size of a transfer.
488  */
489 static void
490 ldminphys(struct buf *bp)
491 {
492 	int unit;
493 	struct ld_softc *sc;
494 
495 	unit = DISKUNIT(bp->b_dev);
496 	sc = device_lookup_private(&ld_cd, unit);
497 
498 	ld_iosize(sc->sc_dv, &bp->b_bcount);
499 	minphys(bp);
500 }
501 
502 static void
503 ld_iosize(device_t d, int *countp)
504 {
505 	struct ld_softc *sc = device_private(d);
506 
507 	if (*countp > sc->sc_maxxfer)
508 		*countp = sc->sc_maxxfer;
509 }
510 
511 static void
512 ld_fake_geometry(struct ld_softc *sc)
513 {
514 	uint64_t ncyl;
515 
516 	if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
517 		sc->sc_nheads = 16;
518 	else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
519 		sc->sc_nheads = 32;
520 	else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
521 		sc->sc_nheads = 64;
522 	else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
523 		sc->sc_nheads = 128;
524 	else
525 		sc->sc_nheads = 255;
526 
527 	sc->sc_nsectors = 63;
528 	sc->sc_ncylinders = INT_MAX;
529 	ncyl = sc->sc_secperunit /
530 	    (sc->sc_nheads * sc->sc_nsectors);
531 	if (ncyl < INT_MAX)
532 		sc->sc_ncylinders = (int)ncyl;
533 }
534 
535 static void
536 ld_set_geometry(struct ld_softc *sc)
537 {
538 	struct dk_softc *dksc = &sc->sc_dksc;
539 	struct disk_geom *dg = &dksc->sc_dkdev.dk_geom;
540 	char tbuf[9];
541 
542 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
543 	    sc->sc_secsize);
544 	aprint_normal_dev(dksc->sc_dev, "%s, %d cyl, %d head, %d sec, "
545 	    "%d bytes/sect x %"PRIu64" sectors\n",
546 	    tbuf, sc->sc_ncylinders, sc->sc_nheads,
547 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
548 
549 	memset(dg, 0, sizeof(*dg));
550 	dg->dg_secperunit = sc->sc_secperunit;
551 	dg->dg_secsize = sc->sc_secsize;
552 	dg->dg_nsectors = sc->sc_nsectors;
553 	dg->dg_ntracks = sc->sc_nheads;
554 	dg->dg_ncylinders = sc->sc_ncylinders;
555 
556 	disk_set_info(dksc->sc_dev, &dksc->sc_dkdev, NULL);
557 }
558 
559 static void
560 ld_config_interrupts(device_t d)
561 {
562 	struct ld_softc *sc = device_private(d);
563 	struct dk_softc *dksc = &sc->sc_dksc;
564 
565 	dkwedge_discover(&dksc->sc_dkdev);
566 }
567 
568 static int
569 ld_discard(device_t dev, off_t pos, off_t len)
570 {
571 	struct ld_softc *sc = device_private(dev);
572 
573 	if (sc->sc_discard == NULL)
574 		return (ENODEV);
575 
576 	return (*sc->sc_discard)(sc, pos, len);
577 }
578 
579 static int
580 lddiscard(dev_t dev, off_t pos, off_t len)
581 {
582 	struct ld_softc *sc;
583 	struct dk_softc *dksc;
584 	int unit;
585 
586 	unit = DISKUNIT(dev);
587 	sc = device_lookup_private(&ld_cd, unit);
588 	dksc = &sc->sc_dksc;
589 
590 	return dk_discard(dksc, dev, pos, len);
591 }
592 
593 MODULE(MODULE_CLASS_DRIVER, ld, "dk_subr");
594 
595 #ifdef _MODULE
596 CFDRIVER_DECL(ld, DV_DISK, NULL);
597 #endif
598 
599 static int
600 ld_modcmd(modcmd_t cmd, void *opaque)
601 {
602 #ifdef _MODULE
603 	devmajor_t bmajor, cmajor;
604 #endif
605 	int error = 0;
606 
607 #ifdef _MODULE
608 	switch (cmd) {
609 	case MODULE_CMD_INIT:
610 		bmajor = cmajor = -1;
611 		error = devsw_attach(ld_cd.cd_name, &ld_bdevsw, &bmajor,
612 		    &ld_cdevsw, &cmajor);
613 		if (error)
614 			break;
615 		error = config_cfdriver_attach(&ld_cd);
616 		break;
617 	case MODULE_CMD_FINI:
618 		error = config_cfdriver_detach(&ld_cd);
619 		if (error)
620 			break;
621 		devsw_detach(&ld_bdevsw, &ld_cdevsw);
622 		break;
623 	default:
624 		error = ENOTTY;
625 		break;
626 	}
627 #endif
628 
629 	return error;
630 }
631