xref: /dragonfly/sys/bus/cam/scsi/scsi_da.c (revision e8c03636)
1 /*
2  * Implementation of SCSI Direct Access Peripheral driver for CAM.
3  *
4  * Copyright (c) 1997 Justin T. Gibbs.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions, and the following disclaimer,
12  *    without modification, immediately at the beginning of the file.
13  * 2. The name of the author may not be used to endorse or promote products
14  *    derived from this software without specific prior written permission.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
20  * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  * $FreeBSD: src/sys/cam/scsi/scsi_da.c,v 1.42.2.46 2003/10/21 22:18:19 thomas Exp $
29  */
30 
31 #include <sys/param.h>
32 
33 #ifdef _KERNEL
34 
35 #include <sys/systm.h>
36 #include <sys/kernel.h>
37 #include <sys/buf.h>
38 #include <sys/sysctl.h>
39 #include <sys/taskqueue.h>
40 #include <sys/lock.h>
41 #include <sys/conf.h>
42 #include <sys/devicestat.h>
43 #include <sys/disk.h>
44 #include <sys/dtype.h>
45 #include <sys/eventhandler.h>
46 #include <sys/malloc.h>
47 #include <sys/cons.h>
48 #include <sys/proc.h>
49 #include <sys/ioctl_compat.h>
50 
51 #include <sys/buf2.h>
52 #include <sys/thread2.h>
53 #include <sys/mplock2.h>
54 
55 #endif /* _KERNEL */
56 
57 #ifdef _KERNEL
58 #include <vm/pmap.h>
59 #endif
60 
61 #ifndef _KERNEL
62 #include <stdio.h>
63 #include <string.h>
64 #endif /* _KERNEL */
65 
66 #include <sys/camlib.h>
67 #include "../cam.h"
68 #include "../cam_ccb.h"
69 #include "../cam_extend.h"
70 #include "../cam_periph.h"
71 #include "../cam_xpt_periph.h"
72 #include "../cam_sim.h"
73 
74 #include "scsi_message.h"
75 
76 #ifndef _KERNEL
77 #include "scsi_da.h"
78 #endif /* !_KERNEL */
79 
80 #ifdef _KERNEL
81 typedef enum {
82 	DA_STATE_PROBE,
83 	DA_STATE_PROBE2,
84 	DA_STATE_NORMAL
85 } da_state;
86 
87 typedef enum {
88 	DA_FLAG_PACK_INVALID	= 0x001,
89 	DA_FLAG_NEW_PACK	= 0x002,
90 	DA_FLAG_PACK_LOCKED	= 0x004,
91 	DA_FLAG_PACK_REMOVABLE	= 0x008,
92 	DA_FLAG_TAGGED_QUEUING	= 0x010,
93 	DA_FLAG_NEED_OTAG	= 0x020,
94 	DA_FLAG_WENT_IDLE	= 0x040,
95 	DA_FLAG_RETRY_UA	= 0x080,
96 	DA_FLAG_OPEN		= 0x100,
97 	DA_FLAG_SCTX_INIT	= 0x200,
98 	DA_FLAG_RD_LIMIT	= 0x400,
99 	DA_FLAG_WR_LIMIT	= 0x800,
100 	DA_FLAG_CAN_TRIM	= 0x1000
101 } da_flags;
102 
103 typedef enum {
104 	DA_Q_NONE		= 0x00,
105 	DA_Q_NO_SYNC_CACHE	= 0x01,
106 	DA_Q_NO_6_BYTE		= 0x02,
107 	DA_Q_NO_PREVENT		= 0x04
108 } da_quirks;
109 
110 typedef enum {
111 	DA_CCB_PROBE		= 0x01,
112 	DA_CCB_PROBE2		= 0x02,
113 	DA_CCB_BUFFER_IO	= 0x03,
114 	DA_CCB_WAITING		= 0x04,
115 	DA_CCB_DUMP		= 0x05,
116 	DA_CCB_TRIM		= 0x06,
117 	DA_CCB_TYPE_MASK	= 0x0F,
118 	DA_CCB_RETRY_UA		= 0x10
119 } da_ccb_state;
120 
121 /* Offsets into our private area for storing information */
122 #define ccb_state	ppriv_field0
123 #define ccb_bio		ppriv_ptr1
124 
125 struct disk_params {
126 	u_int8_t  heads;
127 	u_int32_t cylinders;
128 	u_int8_t  secs_per_track;
129 	u_int32_t secsize;	/* Number of bytes/sector */
130 	u_int64_t sectors;	/* total number sectors */
131 };
132 
133 #define TRIM_MAX_BLOCKS 8
134 #define TRIM_MAX_RANGES TRIM_MAX_BLOCKS * 64
135 struct trim_request {
136         uint8_t         data[TRIM_MAX_RANGES * 8];
137         struct bio      *bios[TRIM_MAX_RANGES];
138 };
139 
140 struct da_softc {
141 	struct	 bio_queue_head bio_queue_rd;
142 	struct	 bio_queue_head bio_queue_wr;
143 	struct	 bio_queue_head bio_queue_trim;
144 	struct	 devstat device_stats;
145 	SLIST_ENTRY(da_softc) links;
146 	LIST_HEAD(, ccb_hdr) pending_ccbs;
147 	da_state state;
148 	da_flags flags;
149 	da_quirks quirks;
150 	int	 minimum_cmd_size;
151 	int	 ordered_tag_count;
152 	int	 outstanding_cmds_rd;
153 	int	 outstanding_cmds_wr;
154 	int      trim_max_ranges;
155 	int      trim_running;
156 	int      trim_enabled;
157 	struct	 disk_params params;
158 	struct	 disk disk;
159 	union	 ccb saved_ccb;
160 	struct task		sysctl_task;
161 	struct sysctl_ctx_list	sysctl_ctx;
162 	struct sysctl_oid	*sysctl_tree;
163 	struct callout		sendordered_c;
164 	struct trim_request     trim_req;
165 };
166 
167 struct da_quirk_entry {
168 	struct scsi_inquiry_pattern inq_pat;
169 	da_quirks quirks;
170 };
171 
172 static const char quantum[] = "QUANTUM";
173 static const char microp[] = "MICROP";
174 
175 static struct da_quirk_entry da_quirk_table[] =
176 {
177 	/* SPI, FC devices */
178 	{
179 		/*
180 		 * Fujitsu M2513A MO drives.
181 		 * Tested devices: M2513A2 firmware versions 1200 & 1300.
182 		 * (dip switch selects whether T_DIRECT or T_OPTICAL device)
183 		 * Reported by: W.Scholten <whs@xs4all.nl>
184 		 */
185 		{T_DIRECT, SIP_MEDIA_REMOVABLE, "FUJITSU", "M2513A", "*"},
186 		/*quirks*/ DA_Q_NO_SYNC_CACHE
187 	},
188 	{
189 		/* See above. */
190 		{T_OPTICAL, SIP_MEDIA_REMOVABLE, "FUJITSU", "M2513A", "*"},
191 		/*quirks*/ DA_Q_NO_SYNC_CACHE
192 	},
193 	{
194 		/*
195 		 * This particular Fujitsu drive doesn't like the
196 		 * synchronize cache command.
197 		 * Reported by: Tom Jackson <toj@gorilla.net>
198 		 */
199 		{T_DIRECT, SIP_MEDIA_FIXED, "FUJITSU", "M2954*", "*"},
200 		/*quirks*/ DA_Q_NO_SYNC_CACHE
201 	},
202 	{
203 		/*
204 		 * This drive doesn't like the synchronize cache command
205 		 * either.  Reported by: Matthew Jacob <mjacob@feral.com>
206 		 * in NetBSD PR kern/6027, August 24, 1998.
207 		 */
208 		{T_DIRECT, SIP_MEDIA_FIXED, microp, "2217*", "*"},
209 		/*quirks*/ DA_Q_NO_SYNC_CACHE
210 	},
211 	{
212 		/*
213 		 * This drive doesn't like the synchronize cache command
214 		 * either.  Reported by: Hellmuth Michaelis (hm@kts.org)
215 		 * (PR 8882).
216 		 */
217 		{T_DIRECT, SIP_MEDIA_FIXED, microp, "2112*", "*"},
218 		/*quirks*/ DA_Q_NO_SYNC_CACHE
219 	},
220 	{
221 		/*
222 		 * Doesn't like the synchronize cache command.
223 		 * Reported by: Blaz Zupan <blaz@gold.amis.net>
224 		 */
225 		{T_DIRECT, SIP_MEDIA_FIXED, "NEC", "D3847*", "*"},
226 		/*quirks*/ DA_Q_NO_SYNC_CACHE
227 	},
228 	{
229 		/*
230 		 * Doesn't like the synchronize cache command.
231 		 * Reported by: Blaz Zupan <blaz@gold.amis.net>
232 		 */
233 		{T_DIRECT, SIP_MEDIA_FIXED, quantum, "MAVERICK 540S", "*"},
234 		/*quirks*/ DA_Q_NO_SYNC_CACHE
235 	},
236 	{
237 		/*
238 		 * Doesn't like the synchronize cache command.
239 		 */
240 		{T_DIRECT, SIP_MEDIA_FIXED, quantum, "LPS525S", "*"},
241 		/*quirks*/ DA_Q_NO_SYNC_CACHE
242 	},
243 	{
244 		/*
245 		 * Doesn't like the synchronize cache command.
246 		 * Reported by: walter@pelissero.de
247 		 */
248 		{T_DIRECT, SIP_MEDIA_FIXED, quantum, "LPS540S", "*"},
249 		/*quirks*/ DA_Q_NO_SYNC_CACHE
250 	},
251 	{
252 		/*
253 		 * Doesn't work correctly with 6 byte reads/writes.
254 		 * Returns illegal request, and points to byte 9 of the
255 		 * 6-byte CDB.
256 		 * Reported by:  Adam McDougall <bsdx@spawnet.com>
257 		 */
258 		{T_DIRECT, SIP_MEDIA_FIXED, quantum, "VIKING 4*", "*"},
259 		/*quirks*/ DA_Q_NO_6_BYTE
260 	},
261 	{
262 		/* See above. */
263 		{T_DIRECT, SIP_MEDIA_FIXED, quantum, "VIKING 2*", "*"},
264 		/*quirks*/ DA_Q_NO_6_BYTE
265 	},
266 	{
267 		/*
268 		 * Doesn't like the synchronize cache command.
269 		 * Reported by: walter@pelissero.de
270 		 */
271 		{T_DIRECT, SIP_MEDIA_FIXED, "CONNER", "CP3500*", "*"},
272 		/*quirks*/ DA_Q_NO_SYNC_CACHE
273 	},
274 	{
275 		/*
276 		 * The CISS RAID controllers do not support SYNC_CACHE
277 		 */
278 		{T_DIRECT, SIP_MEDIA_FIXED, "COMPAQ", "RAID*", "*"},
279 		/*quirks*/ DA_Q_NO_SYNC_CACHE
280 	},
281 	{
282 		/*
283 		 * The same goes for the mly(4) controllers
284 		 */
285 		{T_DIRECT, SIP_MEDIA_FIXED, "MLY*", "*", "MYLX"},
286 		/*quirks*/ DA_Q_NO_SYNC_CACHE
287 	},
288 	/*
289 	 * USB mass storage devices supported by umass(4)
290 	 *
291 	 * NOTE: USB attachments automatically set DA_Q_NO_SYNC_CACHE so
292 	 *	 it does not have to be specified here.
293 	 */
294  	{
295  		/*
296  		 * Creative Nomad MUVO mp3 player (USB)
297  		 * PR: kern/53094
298  		 */
299  		{T_DIRECT, SIP_MEDIA_REMOVABLE, "CREATIVE", "NOMAD_MUVO", "*"},
300 		/*quirks*/ DA_Q_NO_PREVENT
301  	},
302 	{
303 		/*
304 		 * Sigmatel USB Flash MP3 Player
305 		 * PR: kern/57046
306 		 */
307 		{T_DIRECT, SIP_MEDIA_REMOVABLE, "SigmaTel", "MSCN", "*"},
308 		/*quirks*/ DA_Q_NO_PREVENT
309 	},
310 	{
311 		/*
312 		 * SEAGRAND NP-900 MP3 Player
313 		 * PR: kern/64563
314 		 */
315 		{T_DIRECT, SIP_MEDIA_REMOVABLE, "SEAGRAND", "NP-900*", "*"},
316 		/*quirks*/ DA_Q_NO_PREVENT
317 	},
318 	{
319 		/*
320 		 * Creative MUVO Slim mp3 player (USB)
321 		 * PR: usb/86131
322 		 */
323 		{T_DIRECT, SIP_MEDIA_REMOVABLE, "CREATIVE", "MuVo Slim",
324 		"*"}, /*quirks*/ DA_Q_NO_PREVENT
325 	},
326 	{
327 		/*
328 		 * Philips USB Key Audio KEY013
329 		 * PR: usb/68412
330 		 */
331 		{T_DIRECT, SIP_MEDIA_REMOVABLE, "PHILIPS", "Key*", "*"},
332 		/*quirks*/ DA_Q_NO_PREVENT
333 	},
334 };
335 
336 static	d_open_t	daopen;
337 static	d_close_t	daclose;
338 static	d_strategy_t	dastrategy;
339 static	d_dump_t	dadump;
340 static	d_ioctl_t	daioctl;
341 static	periph_init_t	dainit;
342 static	void		daasync(void *callback_arg, u_int32_t code,
343 				struct cam_path *path, void *arg);
344 static	int		dacmdsizesysctl(SYSCTL_HANDLER_ARGS);
345 static	periph_ctor_t	daregister;
346 static	periph_dtor_t	dacleanup;
347 static	periph_start_t	dastart;
348 static	periph_oninv_t	daoninvalidate;
349 static	void		dadone(struct cam_periph *periph,
350 			       union ccb *done_ccb);
351 static  int		daerror(union ccb *ccb, u_int32_t cam_flags,
352 				u_int32_t sense_flags);
353 static void		daprevent(struct cam_periph *periph, int action);
354 static int		dagetcapacity(struct cam_periph *periph);
355 static int		dacheckmedia(struct cam_periph *periph);
356 static void		dasetgeom(struct cam_periph *periph, uint32_t block_len,
357 				  uint64_t maxsector);
358 static void		daflushbioq(struct bio_queue_head *bioq, int error);
359 static timeout_t	dasendorderedtag;
360 static void		dashutdown(void *arg, int howto);
361 
362 #ifndef DA_DEFAULT_TIMEOUT
363 #define DA_DEFAULT_TIMEOUT 60	/* Timeout in seconds */
364 #endif
365 
366 #ifndef	DA_DEFAULT_RETRY
367 #define	DA_DEFAULT_RETRY	4
368 #endif
369 
370 #ifndef	DA_DEFAULT_SEND_ORDERED
371 #define	DA_DEFAULT_SEND_ORDERED	1
372 #endif
373 
374 static int da_retry_count = DA_DEFAULT_RETRY;
375 static int da_default_timeout = DA_DEFAULT_TIMEOUT;
376 static int da_send_ordered = DA_DEFAULT_SEND_ORDERED;
377 static struct callout dasendorderedtag_ch;
378 
379 SYSCTL_NODE(_kern_cam, OID_AUTO, da, CTLFLAG_RD, 0,
380             "CAM Direct Access Disk driver");
381 SYSCTL_INT(_kern_cam_da, OID_AUTO, retry_count, CTLFLAG_RW,
382            &da_retry_count, 0, "Normal I/O retry count");
383 TUNABLE_INT("kern.cam.da.retry_count", &da_retry_count);
384 SYSCTL_INT(_kern_cam_da, OID_AUTO, default_timeout, CTLFLAG_RW,
385            &da_default_timeout, 0, "Normal I/O timeout (in seconds)");
386 TUNABLE_INT("kern.cam.da.default_timeout", &da_default_timeout);
387 SYSCTL_INT(_kern_cam_da, OID_AUTO, da_send_ordered, CTLFLAG_RW,
388            &da_send_ordered, 0, "Send Ordered Tags");
389 TUNABLE_INT("kern.cam.da.da_send_ordered", &da_send_ordered);
390 
391 /*
392  * DA_ORDEREDTAG_INTERVAL determines how often, relative
393  * to the default timeout, we check to see whether an ordered
394  * tagged transaction is appropriate to prevent simple tag
395  * starvation.  Since we'd like to ensure that there is at least
396  * 1/2 of the timeout length left for a starved transaction to
397  * complete after we've sent an ordered tag, we must poll at least
398  * four times in every timeout period.  This takes care of the worst
399  * case where a starved transaction starts during an interval that
400  * meets the requirement "don't send an ordered tag" test so it takes
401  * us two intervals to determine that a tag must be sent.
402  */
403 #ifndef DA_ORDEREDTAG_INTERVAL
404 #define DA_ORDEREDTAG_INTERVAL 4
405 #endif
406 
407 static struct periph_driver dadriver =
408 {
409 	dainit, "da",
410 	TAILQ_HEAD_INITIALIZER(dadriver.units), /* generation */ 0
411 };
412 
413 PERIPHDRIVER_DECLARE(da, dadriver);
414 
415 static struct dev_ops da_ops = {
416 	{ "da", 0, D_DISK | D_MPSAFE },
417 	.d_open =	daopen,
418 	.d_close =	daclose,
419 	.d_read =	physread,
420 	.d_write =	physwrite,
421 	.d_strategy =	dastrategy,
422 	.d_dump =	dadump,
423 	.d_ioctl =	daioctl
424 };
425 
426 static struct extend_array *daperiphs;
427 
428 MALLOC_DEFINE(M_SCSIDA, "scsi_da", "scsi_da buffers");
429 
430 static int
431 daioctl(struct dev_ioctl_args *ap)
432 {
433 	int unit;
434 	int error = 0;
435 	struct buf *bp;
436 	struct cam_periph *periph;
437 	int byte_count;
438 
439 	off_t *del_num = (off_t*)ap->a_data;
440 	off_t bytes_left;
441 	off_t bytes_start;
442 
443 	cdev_t dev = ap->a_head.a_dev;
444 
445 
446 	unit = dkunit(dev);
447 	periph = cam_extend_get(daperiphs, unit);
448 	if (periph == NULL)
449 		return(ENXIO);
450 
451 	switch (ap->a_cmd) {
452 	case IOCTLTRIM:
453 	{
454 
455 		bytes_left = del_num[1];
456 		bytes_start = del_num[0];
457 
458 		/* TRIM occurs on 512-byte sectors. */
459 		KKASSERT((bytes_left % 512) == 0);
460 		KKASSERT((bytes_start% 512) == 0);
461 
462 
463 		/* Break TRIM up into int-sized commands because of b_bcount */
464 		while(bytes_left) {
465 
466 			/*
467 			 * Rather than than squezing out more blocks in b_bcount
468 			 * and having to break up the TRIM request in da_start(),
469 			 * we ensure we can always TRIM this many bytes with one
470 			 * TRIM command (this happens if the device only
471 			 * supports one TRIM block).
472 			 *
473 			 * With min TRIM blksize of 1, TRIM command free
474 			 * 4194240 blks(64*65535): each LBA range can address
475 			 * 65535 blks and there 64 such ranges in a 512-byte
476 			 * block. And, 4194240 * 512 = 0x7FFF8000
477 			 *
478 			 */
479 			byte_count = MIN(bytes_left,0x7FFF8000);
480 			bp = getnewbuf(0,0,0,1);
481 
482 			bp->b_cmd = BUF_CMD_FREEBLKS;
483 			bp->b_bio1.bio_offset = bytes_start;
484 			bp->b_bcount = byte_count;
485 			bp->b_bio1.bio_flags |= BIO_SYNC;
486 			bp->b_bio1.bio_done = biodone_sync;
487 
488 			dev_dstrategy(ap->a_head.a_dev, &bp->b_bio1);
489 
490 			if (biowait(&bp->b_bio1, "TRIM")) {
491 				kprintf("Error:%d\n", bp->b_error);
492 				return(bp->b_error ? bp->b_error : EIO);
493 			}
494 			brelse(bp);
495 			bytes_left -= byte_count;
496 			bytes_start += byte_count;
497 		}
498 		break;
499 	}
500 	default:
501 		return(EINVAL);
502 	}
503 
504 	return(error);
505 }
506 
507 static int
508 daopen(struct dev_open_args *ap)
509 {
510 	cdev_t dev = ap->a_head.a_dev;
511 	struct cam_periph *periph;
512 	struct da_softc *softc;
513 	struct disk_info info;
514 	int unit;
515 	int error;
516 
517 	unit = dkunit(dev);
518 	periph = cam_extend_get(daperiphs, unit);
519 	if (periph == NULL) {
520 		return (ENXIO);
521 	}
522 
523 	if (cam_periph_acquire(periph) != CAM_REQ_CMP) {
524 		return(ENXIO);
525 	}
526 
527 	cam_periph_lock(periph);
528 	if ((error = cam_periph_hold(periph, PCATCH)) != 0) {
529 		cam_periph_unlock(periph);
530 		cam_periph_release(periph);
531 		return (error);
532 	}
533 
534 	unit = periph->unit_number;
535 	softc = (struct da_softc *)periph->softc;
536 
537 	CAM_DEBUG(periph->path, CAM_DEBUG_TRACE,
538 	    ("daopen: dev=%s (unit %d)\n", devtoname(dev),
539 	     unit));
540 
541 	if ((softc->flags & DA_FLAG_PACK_INVALID) != 0) {
542 		/* Invalidate our pack information. */
543 		disk_invalidate(&softc->disk);
544 		softc->flags &= ~DA_FLAG_PACK_INVALID;
545 	}
546 
547 	error = dacheckmedia(periph);
548 	softc->flags |= DA_FLAG_OPEN;
549 
550 	if (error == 0) {
551 		struct ccb_getdev cgd;
552 
553 		/* Build disk information structure */
554 		bzero(&info, sizeof(info));
555 		info.d_type = DTYPE_SCSI;
556 
557 		/*
558 		 * Grab the inquiry data to get the vendor and product names.
559 		 * Put them in the typename and packname for the label.
560 		 */
561 		xpt_setup_ccb(&cgd.ccb_h, periph->path, /*priority*/ 1);
562 		cgd.ccb_h.func_code = XPT_GDEV_TYPE;
563 		xpt_action((union ccb *)&cgd);
564 
565 		/*
566 		 * Check to see whether or not the blocksize is set yet.
567 		 * If it isn't, set it and then clear the blocksize
568 		 * unavailable flag for the device statistics.
569 		 */
570 		if ((softc->device_stats.flags & DEVSTAT_BS_UNAVAILABLE) != 0){
571 			softc->device_stats.block_size = softc->params.secsize;
572 			softc->device_stats.flags &= ~DEVSTAT_BS_UNAVAILABLE;
573 		}
574 	}
575 
576 	if (error == 0) {
577 		if ((softc->flags & DA_FLAG_PACK_REMOVABLE) != 0 &&
578 		    (softc->quirks & DA_Q_NO_PREVENT) == 0)
579 			daprevent(periph, PR_PREVENT);
580 	} else {
581 		softc->flags &= ~DA_FLAG_OPEN;
582 		cam_periph_release(periph);
583 	}
584 	cam_periph_unhold(periph, 1);
585 	return (error);
586 }
587 
588 static int
589 daclose(struct dev_close_args *ap)
590 {
591 	cdev_t dev = ap->a_head.a_dev;
592 	struct	cam_periph *periph;
593 	struct	da_softc *softc;
594 	int	unit;
595 	int	error;
596 
597 	unit = dkunit(dev);
598 	periph = cam_extend_get(daperiphs, unit);
599 	if (periph == NULL)
600 		return (ENXIO);
601 
602 	cam_periph_lock(periph);
603 	if ((error = cam_periph_hold(periph, 0)) != 0) {
604 		cam_periph_unlock(periph);
605 		cam_periph_release(periph);
606 		return (error);
607 	}
608 
609 	softc = (struct da_softc *)periph->softc;
610 
611 	if ((softc->quirks & DA_Q_NO_SYNC_CACHE) == 0) {
612 		union	ccb *ccb;
613 
614 		ccb = cam_periph_getccb(periph, /*priority*/1);
615 
616 		scsi_synchronize_cache(&ccb->csio,
617 				       /*retries*/1,
618 				       /*cbfcnp*/dadone,
619 				       MSG_SIMPLE_Q_TAG,
620 				       /*begin_lba*/0,/* Cover the whole disk */
621 				       /*lb_count*/0,
622 				       SSD_FULL_SIZE,
623 				       5 * 60 * 1000);
624 
625 		cam_periph_runccb(ccb, /*error_routine*/NULL, /*cam_flags*/0,
626 				  /*sense_flags*/SF_RETRY_UA,
627 				  &softc->device_stats);
628 
629 		if ((ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
630 			if ((ccb->ccb_h.status & CAM_STATUS_MASK) ==
631 			     CAM_SCSI_STATUS_ERROR) {
632 				int asc, ascq;
633 				int sense_key, error_code;
634 
635 				scsi_extract_sense(&ccb->csio.sense_data,
636 						   &error_code,
637 						   &sense_key,
638 						   &asc, &ascq);
639 				if (sense_key != SSD_KEY_ILLEGAL_REQUEST)
640 					scsi_sense_print(&ccb->csio);
641 			} else {
642 				xpt_print(periph->path, "Synchronize cache "
643 				    "failed, status == 0x%x, scsi status == "
644 				    "0x%x\n", ccb->csio.ccb_h.status,
645 				    ccb->csio.scsi_status);
646 			}
647 		}
648 
649 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) != 0)
650 			cam_release_devq(ccb->ccb_h.path,
651 					 /*relsim_flags*/0,
652 					 /*reduction*/0,
653 					 /*timeout*/0,
654 					 /*getcount_only*/0);
655 
656 		xpt_release_ccb(ccb);
657 
658 	}
659 
660 	if ((softc->flags & DA_FLAG_PACK_REMOVABLE) != 0) {
661 		if ((softc->quirks & DA_Q_NO_PREVENT) == 0)
662 			daprevent(periph, PR_ALLOW);
663 		/*
664 		 * If we've got removeable media, mark the blocksize as
665 		 * unavailable, since it could change when new media is
666 		 * inserted.
667 		 */
668 		softc->device_stats.flags |= DEVSTAT_BS_UNAVAILABLE;
669 	}
670 
671 	/*
672 	 * Don't compound any ref counting software bugs with more.
673 	 */
674 	if (softc->flags & DA_FLAG_OPEN) {
675 		softc->flags &= ~DA_FLAG_OPEN;
676 		cam_periph_release(periph);
677 	} else {
678 		xpt_print(periph->path,
679 			  "daclose() called on an already closed device!\n");
680 	}
681 	cam_periph_unhold(periph, 1);
682 	return (0);
683 }
684 
685 /*
686  * Actually translate the requested transfer into one the physical driver
687  * can understand.  The transfer is described by a buf and will include
688  * only one physical transfer.
689  */
690 static int
691 dastrategy(struct dev_strategy_args *ap)
692 {
693 	cdev_t dev = ap->a_head.a_dev;
694 	struct bio *bio = ap->a_bio;
695 	struct buf *bp = bio->bio_buf;
696 	struct cam_periph *periph;
697 	struct da_softc *softc;
698 	u_int  unit;
699 
700 	unit = dkunit(dev);
701 	periph = cam_extend_get(daperiphs, unit);
702 	if (periph == NULL) {
703 		bp->b_error = ENXIO;
704 		goto bad;
705 	}
706 	softc = (struct da_softc *)periph->softc;
707 
708 	cam_periph_lock(periph);
709 
710 #if 0
711 	/*
712 	 * check it's not too big a transfer for our adapter
713 	 */
714 	scsi_minphys(bp, &sd_switch);
715 #endif
716 
717 	/*
718 	 * Mask interrupts so that the pack cannot be invalidated until
719 	 * after we are in the queue.  Otherwise, we might not properly
720 	 * clean up one of the buffers.
721 	 */
722 
723 	/*
724 	 * If the device has been made invalid, error out
725 	 */
726 	if ((softc->flags & DA_FLAG_PACK_INVALID)) {
727 		cam_periph_unlock(periph);
728 		bp->b_error = ENXIO;
729 		goto bad;
730 	}
731 
732 	/*
733 	 * Place it in the queue of disk activities for this disk
734 	 */
735 	if (bp->b_cmd == BUF_CMD_WRITE || bp->b_cmd == BUF_CMD_FLUSH)
736 		bioqdisksort(&softc->bio_queue_wr, bio);
737 	else if (bp->b_cmd == BUF_CMD_FREEBLKS)
738 		bioqdisksort(&softc->bio_queue_trim, bio);
739 	else
740 		bioqdisksort(&softc->bio_queue_rd, bio);
741 
742 	/*
743 	 * Schedule ourselves for performing the work.
744 	 */
745 	xpt_schedule(periph, /* XXX priority */1);
746 	cam_periph_unlock(periph);
747 
748 	return(0);
749 bad:
750 	bp->b_flags |= B_ERROR;
751 
752 	/*
753 	 * Correctly set the buf to indicate a completed xfer
754 	 */
755 	bp->b_resid = bp->b_bcount;
756 	biodone(bio);
757 	return(0);
758 }
759 
760 static int
761 dadump(struct dev_dump_args *ap)
762 {
763 	cdev_t dev = ap->a_head.a_dev;
764 	struct	    cam_periph *periph;
765 	struct	    da_softc *softc;
766 	u_int	    unit;
767 	u_int32_t   secsize;
768 	struct	    ccb_scsiio csio;
769 
770 	unit = dkunit(dev);
771 	periph = cam_extend_get(daperiphs, unit);
772 	if (periph == NULL)
773 		return (ENXIO);
774 
775 	softc = (struct da_softc *)periph->softc;
776 	cam_periph_lock(periph);
777 	secsize = softc->params.secsize; /* XXX: or ap->a_secsize? */
778 
779 	if ((softc->flags & DA_FLAG_PACK_INVALID) != 0) {
780 		cam_periph_unlock(periph);
781 		return (ENXIO);
782 	}
783 
784 	/*
785 	 * because length == 0 means we are supposed to flush cache, we only
786 	 * try to write something if length > 0.
787 	 */
788 	if (ap->a_length > 0) {
789 		xpt_setup_ccb(&csio.ccb_h, periph->path, /*priority*/1);
790 		csio.ccb_h.flags |= CAM_POLLED;
791 		csio.ccb_h.ccb_state = DA_CCB_DUMP;
792 		scsi_read_write(&csio,
793 				/*retries*/1,
794 				dadone,
795 				MSG_ORDERED_Q_TAG,
796 				/*read*/FALSE,
797 				/*byte2*/0,
798 				/*minimum_cmd_size*/ softc->minimum_cmd_size,
799 				ap->a_offset / secsize,
800 				ap->a_length / secsize,
801 				/*data_ptr*/(u_int8_t *) ap->a_virtual,
802 				/*dxfer_len*/ap->a_length,
803 				/*sense_len*/SSD_FULL_SIZE,
804 				DA_DEFAULT_TIMEOUT * 1000);
805 		xpt_polled_action((union ccb *)&csio);
806 
807 		if ((csio.ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
808 			kprintf("Aborting dump due to I/O error.\n");
809 			if ((csio.ccb_h.status & CAM_STATUS_MASK) ==
810 			     CAM_SCSI_STATUS_ERROR)
811 				scsi_sense_print(&csio);
812 			else
813 				kprintf("status == 0x%x, scsi status == 0x%x\n",
814 				       csio.ccb_h.status, csio.scsi_status);
815 			return(EIO);
816 		}
817 		cam_periph_unlock(periph);
818 		return 0;
819 	}
820 
821 	/*
822 	 * Sync the disk cache contents to the physical media.
823 	 */
824 	if ((softc->quirks & DA_Q_NO_SYNC_CACHE) == 0) {
825 
826 		xpt_setup_ccb(&csio.ccb_h, periph->path, /*priority*/1);
827 		csio.ccb_h.ccb_state = DA_CCB_DUMP;
828 		scsi_synchronize_cache(&csio,
829 				       /*retries*/1,
830 				       /*cbfcnp*/dadone,
831 				       MSG_SIMPLE_Q_TAG,
832 				       /*begin_lba*/0,/* Cover the whole disk */
833 				       /*lb_count*/0,
834 				       SSD_FULL_SIZE,
835 				       5 * 60 * 1000);
836 		xpt_polled_action((union ccb *)&csio);
837 
838 		if ((csio.ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
839 			if ((csio.ccb_h.status & CAM_STATUS_MASK) ==
840 			     CAM_SCSI_STATUS_ERROR) {
841 				int asc, ascq;
842 				int sense_key, error_code;
843 
844 				scsi_extract_sense(&csio.sense_data,
845 						   &error_code,
846 						   &sense_key,
847 						   &asc, &ascq);
848 				if (sense_key != SSD_KEY_ILLEGAL_REQUEST)
849 					scsi_sense_print(&csio);
850 			} else {
851 				xpt_print(periph->path, "Synchronize cache "
852 				    "failed, status == 0x%x, scsi status == "
853 				    "0x%x\n", csio.ccb_h.status,
854 				    csio.scsi_status);
855 			}
856 		}
857 	}
858 	cam_periph_unlock(periph);
859 	return (0);
860 }
861 
862 static void
863 dainit(void)
864 {
865 	cam_status status;
866 
867 	/*
868 	 * Create our extend array for storing the devices we attach to.
869 	 */
870 	daperiphs = cam_extend_new();
871 	if (daperiphs == NULL) {
872 		kprintf("da: Failed to alloc extend array!\n");
873 		return;
874 	}
875 
876 	callout_init(&dasendorderedtag_ch);
877 
878 	/*
879 	 * Install a global async callback.  This callback will
880 	 * receive async callbacks like "new device found".
881 	 */
882 	status = xpt_register_async(AC_FOUND_DEVICE, daasync, NULL, NULL);
883 
884 	if (status != CAM_REQ_CMP) {
885 		kprintf("da: Failed to attach master async callback "
886 		       "due to status 0x%x!\n", status);
887 	} else if (da_send_ordered) {
888 
889 		/* Register our shutdown event handler */
890 		if ((EVENTHANDLER_REGISTER(shutdown_post_sync, dashutdown,
891 					   NULL, SHUTDOWN_PRI_DEFAULT)) == NULL)
892 			kprintf("%s: shutdown event registration failed!\n",
893 			    __func__);
894 	}
895 }
896 
897 static void
898 daoninvalidate(struct cam_periph *periph)
899 {
900 	struct da_softc *softc;
901 
902 	softc = (struct da_softc *)periph->softc;
903 
904 	/*
905 	 * De-register any async callbacks.
906 	 */
907 	xpt_register_async(0, daasync, periph, periph->path);
908 
909 	softc->flags |= DA_FLAG_PACK_INVALID;
910 
911 	/*
912 	 * Return all queued I/O with ENXIO.
913 	 * XXX Handle any transactions queued to the card
914 	 *     with XPT_ABORT_CCB.
915 	 */
916 	daflushbioq(&softc->bio_queue_trim, ENXIO);
917 	daflushbioq(&softc->bio_queue_wr, ENXIO);
918 	daflushbioq(&softc->bio_queue_rd, ENXIO);
919 	xpt_print(periph->path, "lost device\n");
920 }
921 
922 static void
923 daflushbioq(struct bio_queue_head *bioq, int error)
924 {
925 	struct bio *q_bio;
926 	struct buf *q_bp;
927 
928 	while ((q_bio = bioq_first(bioq)) != NULL){
929 		bioq_remove(bioq, q_bio);
930 		q_bp = q_bio->bio_buf;
931 		q_bp->b_resid = q_bp->b_bcount;
932 		q_bp->b_error = error;
933 		q_bp->b_flags |= B_ERROR;
934 		biodone(q_bio);
935 	}
936 }
937 
938 static void
939 dacleanup(struct cam_periph *periph)
940 {
941 	struct da_softc *softc;
942 
943 	softc = (struct da_softc *)periph->softc;
944 
945 	devstat_remove_entry(&softc->device_stats);
946 	cam_extend_release(daperiphs, periph->unit_number);
947 	xpt_print(periph->path, "removing device entry\n");
948 	/*
949 	 * If we can't free the sysctl tree, oh well...
950 	 */
951 	if ((softc->flags & DA_FLAG_SCTX_INIT) != 0
952 	    && sysctl_ctx_free(&softc->sysctl_ctx) != 0) {
953 		xpt_print(periph->path, "can't remove sysctl context\n");
954 	}
955 	periph->softc = NULL;
956 	if (softc->disk.d_rawdev) {
957 		cam_periph_unlock(periph);
958 		disk_destroy(&softc->disk);
959 		cam_periph_lock(periph);
960 	}
961 
962 	callout_stop(&softc->sendordered_c);
963 	kfree(softc, M_DEVBUF);
964 }
965 
966 static void
967 daasync(void *callback_arg, u_int32_t code,
968 	struct cam_path *path, void *arg)
969 {
970 	struct cam_periph *periph;
971 
972 	periph = (struct cam_periph *)callback_arg;
973 
974 	switch (code) {
975 	case AC_FOUND_DEVICE:
976 	{
977 		struct ccb_getdev *cgd;
978 		cam_status status;
979 
980 		cgd = (struct ccb_getdev *)arg;
981 		if (cgd == NULL)
982 			break;
983 
984 		if (SID_TYPE(&cgd->inq_data) != T_DIRECT
985 		    && SID_TYPE(&cgd->inq_data) != T_RBC
986 		    && SID_TYPE(&cgd->inq_data) != T_OPTICAL)
987 			break;
988 
989 		/*
990 		 * Don't complain if a valid peripheral is already attached.
991 		 */
992 		periph = cam_periph_find(cgd->ccb_h.path, "da");
993 		if (periph && (periph->flags & CAM_PERIPH_INVALID) == 0)
994 			break;
995 
996 		/*
997 		 * Allocate a peripheral instance for
998 		 * this device and start the probe
999 		 * process.
1000 		 */
1001 		status = cam_periph_alloc(daregister, daoninvalidate,
1002 					  dacleanup, dastart,
1003 					  "da", CAM_PERIPH_BIO,
1004 					  cgd->ccb_h.path, daasync,
1005 					  AC_FOUND_DEVICE, cgd);
1006 
1007 		if (status != CAM_REQ_CMP && status != CAM_REQ_INPROG) {
1008 			kprintf("%s: Unable to attach to new device "
1009 			    "due to status 0x%x\n", __func__, status);
1010 		}
1011 		break;
1012 	}
1013 	case AC_SENT_BDR:
1014 	case AC_BUS_RESET:
1015 	{
1016 		struct da_softc *softc;
1017 		struct ccb_hdr *ccbh;
1018 
1019 		softc = (struct da_softc *)periph->softc;
1020 		/*
1021 		 * Don't fail on the expected unit attention
1022 		 * that will occur.
1023 		 */
1024 		softc->flags |= DA_FLAG_RETRY_UA;
1025 		LIST_FOREACH(ccbh, &softc->pending_ccbs, periph_links.le)
1026 			ccbh->ccb_state |= DA_CCB_RETRY_UA;
1027 		/* FALLTHROUGH*/
1028 	}
1029 	default:
1030 		cam_periph_async(periph, code, path, arg);
1031 		break;
1032 	}
1033 }
1034 
1035 static void
1036 dasysctlinit(void *context, int pending)
1037 {
1038 	struct cam_periph *periph;
1039 	struct da_softc *softc;
1040 	char tmpstr[80], tmpstr2[80];
1041 
1042 	get_mplock();
1043 	periph = (struct cam_periph *)context;
1044 	if (cam_periph_acquire(periph) != CAM_REQ_CMP) {
1045 		rel_mplock();
1046 		return;
1047 	}
1048 
1049 	softc = (struct da_softc *)periph->softc;
1050 	ksnprintf(tmpstr, sizeof(tmpstr), "CAM DA unit %d", periph->unit_number);
1051 	ksnprintf(tmpstr2, sizeof(tmpstr2), "%d", periph->unit_number);
1052 
1053 	sysctl_ctx_init(&softc->sysctl_ctx);
1054 	softc->flags |= DA_FLAG_SCTX_INIT;
1055 	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1056 		SYSCTL_STATIC_CHILDREN(_kern_cam_da), OID_AUTO, tmpstr2,
1057 		CTLFLAG_RD, 0, tmpstr);
1058 	if (softc->sysctl_tree == NULL) {
1059 		kprintf("%s: unable to allocate sysctl tree\n", __func__);
1060 		cam_periph_release(periph);
1061 		rel_mplock();
1062 		return;
1063 	}
1064 
1065 	/*
1066 	 * Now register the sysctl handler, so the user can the value on
1067 	 * the fly.
1068 	 */
1069 	SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1070 		OID_AUTO, "minimum_cmd_size", CTLTYPE_INT | CTLFLAG_RW,
1071 		&softc->minimum_cmd_size, 0, dacmdsizesysctl, "I",
1072 		"Minimum CDB size");
1073 
1074 	/* Only create the option if the device supports TRIM */
1075 	if (softc->disk.d_info.d_trimflag) {
1076 		SYSCTL_ADD_INT(&softc->sysctl_ctx,
1077 		    SYSCTL_CHILDREN(softc->sysctl_tree),
1078 		    OID_AUTO,
1079 		    "trim_enabled",
1080 		    CTLFLAG_RW,
1081 		    &softc->trim_enabled,
1082 		    0,
1083 		    "Enable TRIM for this device (SSD))");
1084 	}
1085 
1086 	cam_periph_release(periph);
1087 	rel_mplock();
1088 }
1089 
1090 static int
1091 dacmdsizesysctl(SYSCTL_HANDLER_ARGS)
1092 {
1093 	int error, value;
1094 
1095 	value = *(int *)arg1;
1096 
1097 	error = sysctl_handle_int(oidp, &value, 0, req);
1098 
1099 	if ((error != 0)
1100 	 || (req->newptr == NULL))
1101 		return (error);
1102 
1103 	/*
1104 	 * Acceptable values here are 6, 10 or 12, or 16.
1105 	 */
1106 	if (value < 6)
1107 		value = 6;
1108 	else if ((value > 6)
1109 	      && (value <= 10))
1110 		value = 10;
1111 	else if ((value > 10)
1112 	      && (value <= 12))
1113 		value = 12;
1114 	else if (value > 12)
1115 		value = 16;
1116 
1117 	*(int *)arg1 = value;
1118 
1119 	return (0);
1120 }
1121 
1122 static cam_status
1123 daregister(struct cam_periph *periph, void *arg)
1124 {
1125 	struct da_softc *softc;
1126 	struct ccb_pathinq cpi;
1127 	struct ccb_getdev *cgd;
1128 	char tmpstr[80];
1129 	caddr_t match;
1130 
1131 	cgd = (struct ccb_getdev *)arg;
1132 	if (periph == NULL) {
1133 		kprintf("%s: periph was NULL!!\n", __func__);
1134 		return(CAM_REQ_CMP_ERR);
1135 	}
1136 
1137 	if (cgd == NULL) {
1138 		kprintf("%s: no getdev CCB, can't register device\n",
1139 		    __func__);
1140 		return(CAM_REQ_CMP_ERR);
1141 	}
1142 
1143 	softc = kmalloc(sizeof(*softc), M_DEVBUF, M_INTWAIT | M_ZERO);
1144 	LIST_INIT(&softc->pending_ccbs);
1145 	softc->state = DA_STATE_PROBE;
1146 	bioq_init(&softc->bio_queue_trim);
1147 	bioq_init(&softc->bio_queue_rd);
1148 	bioq_init(&softc->bio_queue_wr);
1149 	if (SID_IS_REMOVABLE(&cgd->inq_data))
1150 		softc->flags |= DA_FLAG_PACK_REMOVABLE;
1151 	if ((cgd->inq_data.flags & SID_CmdQue) != 0)
1152 		softc->flags |= DA_FLAG_TAGGED_QUEUING;
1153 
1154 	/* Used to get TRIM status from AHCI driver */
1155 	if (cgd->inq_data.vendor_specific1[0] == 1) {
1156 		/*
1157 		 * max number of lba ranges an SSD can handle in a single
1158 		 * TRIM command. vendor_specific1[1] is the num of 512-byte
1159 		 * blocks the SSD reports that can be passed in a TRIM cmd.
1160 		 */
1161 		softc->trim_max_ranges =
1162 		   min(cgd->inq_data.vendor_specific1[1] * 64, TRIM_MAX_RANGES);
1163 	}
1164 
1165 	periph->softc = softc;
1166 
1167 	cam_extend_set(daperiphs, periph->unit_number, periph);
1168 
1169 	/*
1170 	 * See if this device has any quirks.
1171 	 */
1172 	match = cam_quirkmatch((caddr_t)&cgd->inq_data,
1173 			       (caddr_t)da_quirk_table,
1174 			       NELEM(da_quirk_table),
1175 			       sizeof(*da_quirk_table), scsi_inquiry_match);
1176 
1177 	if (match != NULL)
1178 		softc->quirks = ((struct da_quirk_entry *)match)->quirks;
1179 	else
1180 		softc->quirks = DA_Q_NONE;
1181 
1182 	/*
1183 	 * Unconditionally disable the synchronize cache command for
1184 	 * usb attachments.  It's just impossible to determine if the
1185 	 * device supports it or not and if it doesn't the port can
1186 	 * brick.
1187 	 */
1188 	if (strncmp(periph->sim->sim_name, "umass", 4) == 0) {
1189 		softc->quirks |= DA_Q_NO_SYNC_CACHE;
1190 	}
1191 
1192 	TASK_INIT(&softc->sysctl_task, 0, dasysctlinit, periph);
1193 
1194 	/* Check if the SIM does not want 6 byte commands */
1195 	bzero(&cpi, sizeof(cpi));
1196 	xpt_setup_ccb(&cpi.ccb_h, periph->path, /*priority*/1);
1197 	cpi.ccb_h.func_code = XPT_PATH_INQ;
1198 	xpt_action((union ccb *)&cpi);
1199 	if (cpi.ccb_h.status == CAM_REQ_CMP && (cpi.hba_misc & PIM_NO_6_BYTE))
1200 		softc->quirks |= DA_Q_NO_6_BYTE;
1201 
1202 	/*
1203 	 * RBC devices don't have to support READ(6), only READ(10).
1204 	 */
1205 	if (softc->quirks & DA_Q_NO_6_BYTE || SID_TYPE(&cgd->inq_data) == T_RBC)
1206 		softc->minimum_cmd_size = 10;
1207 	else
1208 		softc->minimum_cmd_size = 6;
1209 
1210 	/*
1211 	 * Load the user's default, if any.
1212 	 */
1213 	ksnprintf(tmpstr, sizeof(tmpstr), "kern.cam.da.%d.minimum_cmd_size",
1214 		 periph->unit_number);
1215 	TUNABLE_INT_FETCH(tmpstr, &softc->minimum_cmd_size);
1216 
1217 	/*
1218 	 * 6, 10, 12, and 16 are the currently permissible values.
1219 	 */
1220 	if (softc->minimum_cmd_size < 6)
1221 		softc->minimum_cmd_size = 6;
1222 	else if ((softc->minimum_cmd_size > 6)
1223 	      && (softc->minimum_cmd_size <= 10))
1224 		softc->minimum_cmd_size = 10;
1225 	else if ((softc->minimum_cmd_size > 10)
1226 	      && (softc->minimum_cmd_size <= 12))
1227 		softc->minimum_cmd_size = 12;
1228 	else if (softc->minimum_cmd_size > 12)
1229 		softc->minimum_cmd_size = 16;
1230 
1231 	/*
1232 	 * The DA driver supports a blocksize, but
1233 	 * we don't know the blocksize until we do
1234 	 * a read capacity.  So, set a flag to
1235 	 * indicate that the blocksize is
1236 	 * unavailable right now.  We'll clear the
1237 	 * flag as soon as we've done a read capacity.
1238 	 */
1239 	devstat_add_entry(&softc->device_stats, "da",
1240 			  periph->unit_number, 0,
1241 	  		  DEVSTAT_BS_UNAVAILABLE,
1242 			  SID_TYPE(&cgd->inq_data) | DEVSTAT_TYPE_IF_SCSI,
1243 			  DEVSTAT_PRIORITY_DISK);
1244 
1245 	/*
1246 	 * Register this media as a disk
1247 	 */
1248 	CAM_SIM_UNLOCK(periph->sim);
1249 	disk_create(periph->unit_number, &softc->disk, &da_ops);
1250 	if (cpi.maxio == 0 || cpi.maxio > MAXPHYS)
1251 		softc->disk.d_rawdev->si_iosize_max = MAXPHYS;
1252 	else
1253 		softc->disk.d_rawdev->si_iosize_max = cpi.maxio;
1254 	if (bootverbose) {
1255 		kprintf("%s%d: si_iosize_max:%d\n",
1256 		    periph->periph_name,
1257 		    periph->unit_number,
1258 		    softc->disk.d_rawdev->si_iosize_max);
1259 	}
1260 	CAM_SIM_LOCK(periph->sim);
1261 
1262 	/*
1263 	 * Add async callbacks for bus reset and
1264 	 * bus device reset calls.  I don't bother
1265 	 * checking if this fails as, in most cases,
1266 	 * the system will function just fine without
1267 	 * them and the only alternative would be to
1268 	 * not attach the device on failure.
1269 	 */
1270 	xpt_register_async(AC_SENT_BDR | AC_BUS_RESET | AC_LOST_DEVICE,
1271 			   daasync, periph, periph->path);
1272 
1273 	/*
1274 	 * Take an exclusive refcount on the periph while dastart is called
1275 	 * to finish the probe.  The reference will be dropped in dadone at
1276 	 * the end of probe.
1277 	 */
1278 	cam_periph_hold(periph, 0);
1279 	xpt_schedule(periph, /*priority*/5);
1280 
1281 	/*
1282 	 * Schedule a periodic event to occasionally send an
1283 	 * ordered tag to a device.
1284 	 */
1285 	callout_init(&softc->sendordered_c);
1286 	callout_reset(&softc->sendordered_c,
1287 	    (DA_DEFAULT_TIMEOUT * hz) / DA_ORDEREDTAG_INTERVAL,
1288 	    dasendorderedtag, softc);
1289 
1290 
1291 
1292 	return(CAM_REQ_CMP);
1293 }
1294 
1295 static void
1296 dastart(struct cam_periph *periph, union ccb *start_ccb)
1297 {
1298 	struct da_softc *softc;
1299 
1300 	softc = (struct da_softc *)periph->softc;
1301 
1302 	switch (softc->state) {
1303 	case DA_STATE_NORMAL:
1304 	{
1305 		/* Pull a buffer from the queue and get going on it */
1306 		struct bio *bio;
1307 		struct bio *bio_rd;
1308 		struct bio *bio_wr;
1309 		struct buf *bp;
1310 		u_int8_t tag_code;
1311 		int limit;
1312 
1313 		/*
1314 		 * See if there is a buf with work for us to do..
1315 		 */
1316 		bio_rd = bioq_first(&softc->bio_queue_rd);
1317 		bio_wr = bioq_first(&softc->bio_queue_wr);
1318 
1319 		if (periph->immediate_priority <= periph->pinfo.priority) {
1320 			CAM_DEBUG_PRINT(CAM_DEBUG_SUBTRACE,
1321 					("queuing for immediate ccb\n"));
1322 			start_ccb->ccb_h.ccb_state = DA_CCB_WAITING;
1323 			SLIST_INSERT_HEAD(&periph->ccb_list, &start_ccb->ccb_h,
1324 					  periph_links.sle);
1325 			periph->immediate_priority = CAM_PRIORITY_NONE;
1326 			wakeup(&periph->ccb_list);
1327 			if (bio_rd || bio_wr) {
1328 				/*
1329 				 * Have more work to do, so ensure we stay
1330 				 * scheduled
1331 				 */
1332 				xpt_schedule(periph, /* XXX priority */1);
1333 			}
1334 			break;
1335 		}
1336 
1337 		/* Run the trim command if not already running */
1338 		if (!softc->trim_running &&
1339 		   (bio = bioq_first(&softc->bio_queue_trim)) != NULL) {
1340 			struct trim_request *req = &softc->trim_req;
1341 			struct bio *bio1;
1342 			int bps = 0, ranges = 0;
1343 
1344 			softc->trim_running = 1;
1345 			bzero(req, sizeof(*req));
1346 			bio1 = bio;
1347 			while (1) {
1348 				uint64_t lba;
1349 				int count;
1350 
1351 				bp = bio1->bio_buf;
1352 				count = bp->b_bcount / softc->params.secsize;
1353 				lba = bio1->bio_offset/softc->params.secsize;
1354 
1355 				kprintf("trim lba:%llu boff:%llu count:%d\n",
1356 				    (unsigned long long) lba,
1357 				    (unsigned long long) bio1->bio_offset,
1358 				    count);
1359 
1360 				bioq_remove(&softc->bio_queue_trim, bio1);
1361 				while (count > 0) {
1362 					int c = min(count, 0xffff);
1363 					int off = ranges * 8;
1364 
1365 					req->data[off + 0] = lba & 0xff;
1366 					req->data[off + 1] = (lba >> 8) & 0xff;
1367 					req->data[off + 2] = (lba >> 16) & 0xff;
1368 					req->data[off + 3] = (lba >> 24) & 0xff;
1369 					req->data[off + 4] = (lba >> 32) & 0xff;
1370 					req->data[off + 5] = (lba >> 40) & 0xff;
1371 					req->data[off + 6] = c & 0xff;
1372 					req->data[off + 7] = (c >> 8) & 0xff;
1373 					lba += c;
1374 					count -= c;
1375 					ranges++;
1376 				}
1377 
1378 				/* Try to merge multiple TRIM requests */
1379 				req->bios[bps++] = bio1;
1380 				bio1 = bioq_first(&softc->bio_queue_trim);
1381 				if (bio1 == NULL ||
1382 				    bio1->bio_buf->b_bcount / softc->params.secsize >
1383 				    (softc->trim_max_ranges - ranges) * 0xffff)
1384 					break;
1385 			}
1386 
1387 
1388 			cam_fill_csio(&start_ccb->csio,
1389 			    1/*retries*/,
1390 			    dadone,
1391 			    CAM_DIR_OUT,
1392 			    MSG_SIMPLE_Q_TAG,
1393 			    req->data,
1394 			    ((ranges +63)/64)*512,
1395 			    SSD_FULL_SIZE,
1396 			    sizeof(struct scsi_rw_6),
1397 			    da_default_timeout*2);
1398 
1399 			start_ccb->ccb_h.ccb_state = DA_CCB_TRIM;
1400 			LIST_INSERT_HEAD(&softc->pending_ccbs,
1401 			    &start_ccb->ccb_h, periph_links.le);
1402 			start_ccb->csio.ccb_h.func_code = XPT_TRIM;
1403 			start_ccb->ccb_h.ccb_bio = bio;
1404 			devstat_start_transaction(&softc->device_stats);
1405 			xpt_action(start_ccb);
1406 			xpt_schedule(periph, 1);
1407 			break;
1408 		}
1409 
1410 		/*
1411 		 * Select a read or write buffer to queue.  Limit the number
1412 		 * of tags dedicated to reading or writing, giving reads
1413 		 * precedence.
1414 		 *
1415 		 * Writes to modern hard drives go into the HDs cache and
1416 		 * return completion nearly instantly.  That is until the
1417 		 * cache becomes full.  When the HDs cache becomes full
1418 		 * write commands will begin to stall.  If all available
1419 		 * tags are taken up by writes which saturate the drive
1420 		 * reads will become tag-starved.
1421 		 *
1422 		 * A similar situation can occur with reads.  With many
1423 		 * parallel readers all tags can be taken up by reads
1424 		 * and prevent any writes from draining, even if the HD's
1425 		 * cache is not full.
1426 		 */
1427 		limit = periph->sim->max_tagged_dev_openings * 2 / 3 + 1;
1428 #if 0
1429 		/* DEBUGGING */
1430 		static int savets;
1431 		static long savets2;
1432 		if (1 || time_second != savets2 || (ticks != savets && (softc->outstanding_cmds_rd || softc->outstanding_cmds_wr))) {
1433 			kprintf("%d %d (%d)\n",
1434 				softc->outstanding_cmds_rd,
1435 				softc->outstanding_cmds_wr,
1436 				limit);
1437 			savets = ticks;
1438 			savets2 = time_second;
1439 		}
1440 #endif
1441 		if (bio_rd && softc->outstanding_cmds_rd < limit) {
1442 			bio = bio_rd;
1443 			bioq_remove(&softc->bio_queue_rd, bio);
1444 		} else if (bio_wr && softc->outstanding_cmds_wr < limit) {
1445 			bio = bio_wr;
1446 			bioq_remove(&softc->bio_queue_wr, bio);
1447 		} else {
1448 			if (bio_rd)
1449 				softc->flags |= DA_FLAG_RD_LIMIT;
1450 			if (bio_wr)
1451 				softc->flags |= DA_FLAG_WR_LIMIT;
1452 			xpt_release_ccb(start_ccb);
1453 			break;
1454 		}
1455 
1456 		/*
1457 		 * We can queue new work.
1458 		 */
1459 		bp = bio->bio_buf;
1460 
1461 		devstat_start_transaction(&softc->device_stats);
1462 
1463 		if ((bp->b_flags & B_ORDERED) != 0 ||
1464 		    (softc->flags & DA_FLAG_NEED_OTAG) != 0) {
1465 			softc->flags &= ~DA_FLAG_NEED_OTAG;
1466 			softc->ordered_tag_count++;
1467 			tag_code = MSG_ORDERED_Q_TAG;
1468 		} else {
1469 			tag_code = MSG_SIMPLE_Q_TAG;
1470 		}
1471 
1472 		switch(bp->b_cmd) {
1473 		case BUF_CMD_READ:
1474 		case BUF_CMD_WRITE:
1475 			/*
1476 			 * Block read/write op
1477 			 */
1478 			KKASSERT(bio->bio_offset % softc->params.secsize == 0);
1479 
1480 			scsi_read_write(
1481 				&start_ccb->csio,
1482 				da_retry_count,		/* retries */
1483 				dadone,
1484 				tag_code,
1485 				(bp->b_cmd == BUF_CMD_READ),
1486 				0,			/* byte2 */
1487 				softc->minimum_cmd_size,
1488 				bio->bio_offset / softc->params.secsize,
1489 				bp->b_bcount / softc->params.secsize,
1490 				bp->b_data,
1491 				bp->b_bcount,
1492 				SSD_FULL_SIZE,		/* sense_len */
1493 				da_default_timeout * 1000
1494 			);
1495 			break;
1496 		case BUF_CMD_FLUSH:
1497 			/*
1498 			 * Silently complete a flush request if the device
1499 			 * cannot handle it.
1500 			 */
1501 			if (softc->quirks & DA_Q_NO_SYNC_CACHE) {
1502 				xpt_release_ccb(start_ccb);
1503 				start_ccb = NULL;
1504 				devstat_end_transaction_buf(
1505 					&softc->device_stats, bp);
1506 				biodone(bio);
1507 			} else {
1508 				scsi_synchronize_cache(
1509 					&start_ccb->csio,
1510 					1,		/* retries */
1511 					dadone,		/* cbfcnp */
1512 					MSG_SIMPLE_Q_TAG,
1513 					0,		/* lba */
1514 					0,		/* count (whole disk) */
1515 					SSD_FULL_SIZE,
1516 					da_default_timeout*1000	/* timeout */
1517 				);
1518 			}
1519 			break;
1520 		case BUF_CMD_FREEBLKS:
1521 			if (softc->disk.d_info.d_trimflag & DA_FLAG_CAN_TRIM){
1522 				start_ccb->csio.ccb_h.func_code = XPT_TRIM;
1523 				break;
1524 			}
1525 		default:
1526 			xpt_release_ccb(start_ccb);
1527 			start_ccb = NULL;
1528 			panic("dastart: unrecognized bio cmd %d", bp->b_cmd);
1529 			break; /* NOT REACHED */
1530 		}
1531 
1532 		/*
1533 		 * Block out any asyncronous callbacks
1534 		 * while we touch the pending ccb list.
1535 		 */
1536 		if (start_ccb) {
1537 			start_ccb->ccb_h.ccb_state = DA_CCB_BUFFER_IO;
1538 			LIST_INSERT_HEAD(&softc->pending_ccbs,
1539 					 &start_ccb->ccb_h, periph_links.le);
1540 			if (bp->b_cmd == BUF_CMD_WRITE ||
1541 			    bp->b_cmd == BUF_CMD_FLUSH) {
1542 				++softc->outstanding_cmds_wr;
1543 			} else {
1544 				++softc->outstanding_cmds_rd;
1545 			}
1546 
1547 			/* We expect a unit attention from this device */
1548 			if ((softc->flags & DA_FLAG_RETRY_UA) != 0) {
1549 				start_ccb->ccb_h.ccb_state |= DA_CCB_RETRY_UA;
1550 				softc->flags &= ~DA_FLAG_RETRY_UA;
1551 			}
1552 
1553 			start_ccb->ccb_h.ccb_bio = bio;
1554 			xpt_action(start_ccb);
1555 		}
1556 
1557 		/*
1558 		 * Be sure we stay scheduled if we have more work to do.
1559 		 */
1560 		if (bioq_first(&softc->bio_queue_rd) ||
1561 		    bioq_first(&softc->bio_queue_wr)) {
1562 			xpt_schedule(periph, 1);
1563 		}
1564 		break;
1565 	}
1566 	case DA_STATE_PROBE:
1567 	{
1568 		struct ccb_scsiio *csio;
1569 		struct scsi_read_capacity_data *rcap;
1570 
1571 		rcap = kmalloc(sizeof(*rcap), M_SCSIDA, M_INTWAIT | M_ZERO);
1572 		csio = &start_ccb->csio;
1573 		scsi_read_capacity(csio,
1574 				   /*retries*/4,
1575 				   dadone,
1576 				   MSG_SIMPLE_Q_TAG,
1577 				   rcap,
1578 				   SSD_FULL_SIZE,
1579 				   /*timeout*/5000);
1580 		start_ccb->ccb_h.ccb_bio = NULL;
1581 		start_ccb->ccb_h.ccb_state = DA_CCB_PROBE;
1582 		xpt_action(start_ccb);
1583 		break;
1584 	}
1585 	case DA_STATE_PROBE2:
1586 	{
1587 		struct ccb_scsiio *csio;
1588 		struct scsi_read_capacity_data_16 *rcaplong;
1589 
1590 		rcaplong = kmalloc(sizeof(*rcaplong), M_SCSIDA,
1591 				   M_INTWAIT | M_ZERO);
1592 		csio = &start_ccb->csio;
1593 		scsi_read_capacity_16(csio,
1594 				    /*retries*/ 4,
1595 				    /*cbfcnp*/ dadone,
1596 				    /*tag_action*/ MSG_SIMPLE_Q_TAG,
1597 				    /*lba*/ 0,
1598 				    /*reladr*/ 0,
1599 				    /*pmi*/ 0,
1600 				    rcaplong,
1601 				    /*sense_len*/ SSD_FULL_SIZE,
1602 				    /*timeout*/ 60000);
1603 		start_ccb->ccb_h.ccb_bio = NULL;
1604 		start_ccb->ccb_h.ccb_state = DA_CCB_PROBE2;
1605 		xpt_action(start_ccb);
1606 		break;
1607 	}
1608 	}
1609 }
1610 
1611 static int
1612 cmd6workaround(union ccb *ccb)
1613 {
1614 	struct scsi_rw_6 cmd6;
1615 	struct scsi_rw_10 *cmd10;
1616 	struct da_softc *softc;
1617 	u_int8_t *cdb;
1618 	int frozen;
1619 
1620 	cdb = ccb->csio.cdb_io.cdb_bytes;
1621 
1622 	/* Translation only possible if CDB is an array and cmd is R/W6 */
1623 	if ((ccb->ccb_h.flags & CAM_CDB_POINTER) != 0 ||
1624 	    (*cdb != READ_6 && *cdb != WRITE_6))
1625 		return 0;
1626 
1627 	xpt_print(ccb->ccb_h.path, "READ(6)/WRITE(6) not supported, "
1628 	    "increasing minimum_cmd_size to 10.\n");
1629  	softc = (struct da_softc *)xpt_path_periph(ccb->ccb_h.path)->softc;
1630 	softc->minimum_cmd_size = 10;
1631 
1632 	bcopy(cdb, &cmd6, sizeof(struct scsi_rw_6));
1633 	cmd10 = (struct scsi_rw_10 *)cdb;
1634 	cmd10->opcode = (cmd6.opcode == READ_6) ? READ_10 : WRITE_10;
1635 	cmd10->byte2 = 0;
1636 	scsi_ulto4b(scsi_3btoul(cmd6.addr), cmd10->addr);
1637 	cmd10->reserved = 0;
1638 	scsi_ulto2b(cmd6.length, cmd10->length);
1639 	cmd10->control = cmd6.control;
1640 	ccb->csio.cdb_len = sizeof(*cmd10);
1641 
1642 	/* Requeue request, unfreezing queue if necessary */
1643 	frozen = (ccb->ccb_h.status & CAM_DEV_QFRZN) != 0;
1644  	ccb->ccb_h.status = CAM_REQUEUE_REQ;
1645 	xpt_action(ccb);
1646 	if (frozen) {
1647 		cam_release_devq(ccb->ccb_h.path,
1648 				 /*relsim_flags*/0,
1649 				 /*reduction*/0,
1650 				 /*timeout*/0,
1651 				 /*getcount_only*/0);
1652 	}
1653 	return (ERESTART);
1654 }
1655 
1656 static void
1657 dadone(struct cam_periph *periph, union ccb *done_ccb)
1658 {
1659 	struct da_softc *softc;
1660 	struct ccb_scsiio *csio;
1661 	struct disk_info info;
1662 
1663 	softc = (struct da_softc *)periph->softc;
1664 	csio = &done_ccb->csio;
1665 	switch (csio->ccb_h.ccb_state & DA_CCB_TYPE_MASK) {
1666 	case DA_CCB_BUFFER_IO:
1667 	case DA_CCB_TRIM:
1668 	{
1669 		struct buf *bp;
1670 		struct bio *bio;
1671 		int mustsched = 0;
1672 
1673 		bio = (struct bio *)done_ccb->ccb_h.ccb_bio;
1674 		bp = bio->bio_buf;
1675 		if ((done_ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
1676 			int error;
1677 			int sf;
1678 
1679 			if ((csio->ccb_h.ccb_state & DA_CCB_RETRY_UA) != 0)
1680 				sf = SF_RETRY_UA;
1681 			else
1682 				sf = 0;
1683 
1684 			error = daerror(done_ccb, CAM_RETRY_SELTO, sf);
1685 			if (error == ERESTART) {
1686 				/*
1687 				 * A retry was scheuled, so
1688 				 * just return.
1689 				 */
1690 				return;
1691 			}
1692 			if (error != 0) {
1693 				if (error == ENXIO) {
1694 					/*
1695 					 * Catastrophic error.  Mark our pack as
1696 					 * invalid.
1697 					 */
1698 					/*
1699 					 * XXX See if this is really a media
1700 					 * XXX change first?
1701 					 */
1702 					xpt_print(periph->path,
1703 					    "Invalidating pack\n");
1704 					softc->flags |= DA_FLAG_PACK_INVALID;
1705 				}
1706 
1707 				/*
1708 				 * Return all queued write I/O's with EIO
1709 				 * so the client can retry these I/Os in the
1710 				 * proper order should it attempt to recover.
1711 				 *
1712 				 * Leave read I/O's alone.
1713 				 */
1714 				daflushbioq(&softc->bio_queue_wr, EIO);
1715 				bp->b_error = error;
1716 				bp->b_resid = bp->b_bcount;
1717 				bp->b_flags |= B_ERROR;
1718 			} else {
1719 				bp->b_resid = csio->resid;
1720 				bp->b_error = 0;
1721 				if (bp->b_resid != 0)
1722 					bp->b_flags |= B_ERROR;
1723 			}
1724 			if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0)
1725 				cam_release_devq(done_ccb->ccb_h.path,
1726 						 /*relsim_flags*/0,
1727 						 /*reduction*/0,
1728 						 /*timeout*/0,
1729 						 /*getcount_only*/0);
1730 		} else {
1731 			if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0)
1732 				panic("REQ_CMP with QFRZN");
1733 			bp->b_resid = csio->resid;
1734 			if (csio->resid > 0)
1735 				bp->b_flags |= B_ERROR;
1736 		}
1737 
1738 		/*
1739 		 * Block out any asyncronous callbacks
1740 		 * while we touch the pending ccb list.
1741 		 */
1742 		LIST_REMOVE(&done_ccb->ccb_h, periph_links.le);
1743 		if (bp->b_cmd == BUF_CMD_WRITE || bp->b_cmd == BUF_CMD_FLUSH) {
1744 			--softc->outstanding_cmds_wr;
1745 			if (softc->flags & DA_FLAG_WR_LIMIT) {
1746 				softc->flags &= ~DA_FLAG_WR_LIMIT;
1747 				mustsched = 1;
1748 			}
1749 		} else {
1750 			--softc->outstanding_cmds_rd;
1751 			if (softc->flags & DA_FLAG_RD_LIMIT) {
1752 				softc->flags &= ~DA_FLAG_RD_LIMIT;
1753 				mustsched = 1;
1754 			}
1755 		}
1756 		if (softc->outstanding_cmds_rd +
1757 		    softc->outstanding_cmds_wr == 0) {
1758 			softc->flags |= DA_FLAG_WENT_IDLE;
1759 		}
1760 
1761 		devstat_end_transaction_buf(&softc->device_stats, bp);
1762 		if ((csio->ccb_h.ccb_state & DA_CCB_TYPE_MASK) ==
1763 		    DA_CCB_TRIM) {
1764 			struct trim_request *req =
1765 			    (struct trim_request *) csio->data_ptr;
1766 			int i;
1767 
1768 			for (i = 1; i < softc->trim_max_ranges &&
1769 			    req->bios[i]; i++) {
1770 				struct bio *bp1 = req->bios[i];
1771 
1772 				bp1->bio_buf->b_resid = bp->b_resid;
1773 				bp1->bio_buf->b_error = bp->b_error;
1774 				if (bp->b_flags & B_ERROR)
1775 					bp1->bio_buf->b_flags |= B_ERROR;
1776 				biodone(bp1);
1777 			}
1778 			softc->trim_running = 0;
1779 			biodone(bio);
1780 			xpt_schedule(periph,1);
1781 		} else
1782 			biodone(bio);
1783 
1784 
1785 		if (mustsched)
1786 			xpt_schedule(periph, /*priority*/1);
1787 
1788 		break;
1789 	}
1790 	case DA_CCB_PROBE:
1791 	case DA_CCB_PROBE2:
1792 	{
1793 		struct	   scsi_read_capacity_data *rdcap;
1794 		struct     scsi_read_capacity_data_16 *rcaplong;
1795 		char	   announce_buf[80];
1796 
1797 		rdcap = NULL;
1798 		rcaplong = NULL;
1799 		if (softc->state == DA_STATE_PROBE)
1800 			rdcap =(struct scsi_read_capacity_data *)csio->data_ptr;
1801 		else
1802 			rcaplong = (struct scsi_read_capacity_data_16 *)
1803 				    csio->data_ptr;
1804 
1805 		bzero(&info, sizeof(info));
1806 		info.d_type = DTYPE_SCSI;
1807 		info.d_serialno = xpt_path_serialno(periph->path);
1808 
1809 		if ((csio->ccb_h.status & CAM_STATUS_MASK) == CAM_REQ_CMP) {
1810 			struct disk_params *dp;
1811 			uint32_t block_size;
1812 			uint64_t maxsector;
1813 
1814 			if (softc->state == DA_STATE_PROBE) {
1815 				block_size = scsi_4btoul(rdcap->length);
1816 				maxsector = scsi_4btoul(rdcap->addr);
1817 
1818 				/*
1819 				 * According to SBC-2, if the standard 10
1820 				 * byte READ CAPACITY command returns 2^32,
1821 				 * we should issue the 16 byte version of
1822 				 * the command, since the device in question
1823 				 * has more sectors than can be represented
1824 				 * with the short version of the command.
1825 				 */
1826 				if (maxsector == 0xffffffff) {
1827 					softc->state = DA_STATE_PROBE2;
1828 					kfree(rdcap, M_SCSIDA);
1829 					xpt_release_ccb(done_ccb);
1830 					xpt_schedule(periph, /*priority*/5);
1831 					return;
1832 				}
1833 			} else {
1834 				block_size = scsi_4btoul(rcaplong->length);
1835 				maxsector = scsi_8btou64(rcaplong->addr);
1836 			}
1837 			dasetgeom(periph, block_size, maxsector);
1838 			dp = &softc->params;
1839 			ksnprintf(announce_buf, sizeof(announce_buf),
1840 				"%juMB (%ju %u byte sectors: %dH %dS/T %dC)",
1841 				(uintmax_t) (((uintmax_t)dp->secsize *
1842 				dp->sectors) / (1024*1024)),
1843 				(uintmax_t)dp->sectors,
1844 				dp->secsize, dp->heads, dp->secs_per_track,
1845 				dp->cylinders);
1846 
1847 			CAM_SIM_UNLOCK(periph->sim);
1848 			info.d_media_blksize = softc->params.secsize;
1849 			info.d_media_blocks = softc->params.sectors;
1850 			info.d_media_size = 0;
1851 			info.d_secpertrack = softc->params.secs_per_track;
1852 			info.d_nheads = softc->params.heads;
1853 			info.d_ncylinders = softc->params.cylinders;
1854 			info.d_secpercyl = softc->params.heads *
1855 						softc->params.secs_per_track;
1856 			info.d_serialno = xpt_path_serialno(periph->path);
1857 			disk_setdiskinfo(&softc->disk, &info);
1858 			CAM_SIM_LOCK(periph->sim);
1859 		} else {
1860 			int	error;
1861 
1862 			announce_buf[0] = '\0';
1863 
1864 			/*
1865 			 * Retry any UNIT ATTENTION type errors.  They
1866 			 * are expected at boot.
1867 			 */
1868 			error = daerror(done_ccb, CAM_RETRY_SELTO,
1869 					SF_RETRY_UA|SF_NO_PRINT);
1870 			if (error == ERESTART) {
1871 				/*
1872 				 * A retry was scheuled, so
1873 				 * just return.
1874 				 */
1875 				return;
1876 			} else if (error != 0) {
1877 				struct scsi_sense_data *sense;
1878 				int asc, ascq;
1879 				int sense_key, error_code;
1880 				int have_sense;
1881 				cam_status status;
1882 				struct ccb_getdev cgd;
1883 
1884 				/* Don't wedge this device's queue */
1885 				status = done_ccb->ccb_h.status;
1886 				if ((status & CAM_DEV_QFRZN) != 0)
1887 					cam_release_devq(done_ccb->ccb_h.path,
1888 							 /*relsim_flags*/0,
1889 							 /*reduction*/0,
1890 							 /*timeout*/0,
1891 							 /*getcount_only*/0);
1892 
1893 
1894 				xpt_setup_ccb(&cgd.ccb_h,
1895 					      done_ccb->ccb_h.path,
1896 					      /* priority */ 1);
1897 				cgd.ccb_h.func_code = XPT_GDEV_TYPE;
1898 				xpt_action((union ccb *)&cgd);
1899 
1900 				if (((csio->ccb_h.flags & CAM_SENSE_PHYS) != 0)
1901 				 || ((csio->ccb_h.flags & CAM_SENSE_PTR) != 0)
1902 				 || ((status & CAM_AUTOSNS_VALID) == 0))
1903 					have_sense = FALSE;
1904 				else
1905 					have_sense = TRUE;
1906 
1907 				if (have_sense) {
1908 					sense = &csio->sense_data;
1909 					scsi_extract_sense(sense, &error_code,
1910 							   &sense_key,
1911 							   &asc, &ascq);
1912 				}
1913 				/*
1914 				 * Attach to anything that claims to be a
1915 				 * direct access or optical disk device,
1916 				 * as long as it doesn't return a "Logical
1917 				 * unit not supported" (0x25) error.
1918 				 */
1919 				if ((have_sense) && (asc != 0x25)
1920 				 && (error_code == SSD_CURRENT_ERROR)) {
1921 					const char *sense_key_desc;
1922 					const char *asc_desc;
1923 
1924 					scsi_sense_desc(sense_key, asc, ascq,
1925 							&cgd.inq_data,
1926 							&sense_key_desc,
1927 							&asc_desc);
1928 					ksnprintf(announce_buf,
1929 					    sizeof(announce_buf),
1930 						"Attempt to query device "
1931 						"size failed: %s, %s",
1932 						sense_key_desc,
1933 						asc_desc);
1934 					info.d_media_blksize = 512;
1935 					disk_setdiskinfo(&softc->disk, &info);
1936 				} else {
1937 					if (have_sense)
1938 						scsi_sense_print(
1939 							&done_ccb->csio);
1940 					else {
1941 						xpt_print(periph->path,
1942 						    "got CAM status %#x\n",
1943 						    done_ccb->ccb_h.status);
1944 					}
1945 
1946 					xpt_print(periph->path, "fatal error, "
1947 					    "failed to attach to device\n");
1948 
1949 					/*
1950 					 * Free up resources.
1951 					 */
1952 					cam_periph_invalidate(periph);
1953 				}
1954 			}
1955 		}
1956 		kfree(csio->data_ptr, M_SCSIDA);
1957 		if (announce_buf[0] != '\0') {
1958 			xpt_announce_periph(periph, announce_buf);
1959 			/*
1960 			 * Create our sysctl variables, now that we know
1961 			 * we have successfully attached.
1962 			 */
1963 			taskqueue_enqueue(taskqueue_thread[mycpuid],
1964 			    &softc->sysctl_task);
1965 		}
1966 
1967 		if (softc->trim_max_ranges) {
1968 			softc->disk.d_info.d_trimflag |= DA_FLAG_CAN_TRIM;
1969 			kprintf("%s%d: supports TRIM\n",
1970 		   	    periph->periph_name,
1971 		   	    periph->unit_number);
1972 		}
1973 		softc->state = DA_STATE_NORMAL;
1974 		/*
1975 		 * Since our peripheral may be invalidated by an error
1976 		 * above or an external event, we must release our CCB
1977 		 * before releasing the probe lock on the peripheral.
1978 		 * The peripheral will only go away once the last lock
1979 		 * is removed, and we need it around for the CCB release
1980 		 * operation.
1981 		 */
1982 		xpt_release_ccb(done_ccb);
1983 		cam_periph_unhold(periph, 0);
1984 		return;
1985 	}
1986 	case DA_CCB_WAITING:
1987 	{
1988 		/* Caller will release the CCB */
1989 		wakeup(&done_ccb->ccb_h.cbfcnp);
1990 		return;
1991 	}
1992 	case DA_CCB_DUMP:
1993 		/* No-op.  We're polling */
1994 		return;
1995 	default:
1996 		break;
1997 	}
1998 	xpt_release_ccb(done_ccb);
1999 }
2000 
2001 static int
2002 daerror(union ccb *ccb, u_int32_t cam_flags, u_int32_t sense_flags)
2003 {
2004 	struct da_softc	  *softc;
2005 	struct cam_periph *periph;
2006 	int error;
2007 
2008 	periph = xpt_path_periph(ccb->ccb_h.path);
2009 	softc = (struct da_softc *)periph->softc;
2010 
2011  	/*
2012 	 * Automatically detect devices that do not support
2013  	 * READ(6)/WRITE(6) and upgrade to using 10 byte cdbs.
2014  	 */
2015 	error = 0;
2016 	if ((ccb->ccb_h.status & CAM_STATUS_MASK) == CAM_REQ_INVALID) {
2017 		error = cmd6workaround(ccb);
2018 	} else if (((ccb->ccb_h.status & CAM_STATUS_MASK) ==
2019 		   CAM_SCSI_STATUS_ERROR)
2020 	 && (ccb->ccb_h.status & CAM_AUTOSNS_VALID)
2021 	 && (ccb->csio.scsi_status == SCSI_STATUS_CHECK_COND)
2022 	 && ((ccb->ccb_h.flags & CAM_SENSE_PHYS) == 0)
2023 	 && ((ccb->ccb_h.flags & CAM_SENSE_PTR) == 0)) {
2024 		int sense_key, error_code, asc, ascq;
2025 
2026  		scsi_extract_sense(&ccb->csio.sense_data,
2027 				   &error_code, &sense_key, &asc, &ascq);
2028 		if (sense_key == SSD_KEY_ILLEGAL_REQUEST)
2029  			error = cmd6workaround(ccb);
2030 	}
2031 	if (error == ERESTART)
2032 		return (ERESTART);
2033 
2034 	/*
2035 	 * XXX
2036 	 * Until we have a better way of doing pack validation,
2037 	 * don't treat UAs as errors.
2038 	 */
2039 	sense_flags |= SF_RETRY_UA;
2040 	return(cam_periph_error(ccb, cam_flags, sense_flags,
2041 				&softc->saved_ccb));
2042 }
2043 
2044 static void
2045 daprevent(struct cam_periph *periph, int action)
2046 {
2047 	struct	da_softc *softc;
2048 	union	ccb *ccb;
2049 	int	error;
2050 
2051 	softc = (struct da_softc *)periph->softc;
2052 
2053 	if (((action == PR_ALLOW)
2054 	  && (softc->flags & DA_FLAG_PACK_LOCKED) == 0)
2055 	 || ((action == PR_PREVENT)
2056 	  && (softc->flags & DA_FLAG_PACK_LOCKED) != 0)) {
2057 		return;
2058 	}
2059 
2060 	ccb = cam_periph_getccb(periph, /*priority*/1);
2061 
2062 	scsi_prevent(&ccb->csio,
2063 		     /*retries*/1,
2064 		     /*cbcfp*/dadone,
2065 		     MSG_SIMPLE_Q_TAG,
2066 		     action,
2067 		     SSD_FULL_SIZE,
2068 		     5000);
2069 
2070 	error = cam_periph_runccb(ccb, /*error_routine*/NULL, CAM_RETRY_SELTO,
2071 				  SF_RETRY_UA, &softc->device_stats);
2072 
2073 	if (error == 0) {
2074 		if (action == PR_ALLOW)
2075 			softc->flags &= ~DA_FLAG_PACK_LOCKED;
2076 		else
2077 			softc->flags |= DA_FLAG_PACK_LOCKED;
2078 	}
2079 
2080 	xpt_release_ccb(ccb);
2081 }
2082 
2083 /*
2084  * Check media on open, e.g. card reader devices which had no initial media.
2085  */
2086 static int
2087 dacheckmedia(struct cam_periph *periph)
2088 {
2089 	struct disk_params *dp;
2090 	struct da_softc *softc;
2091 	struct disk_info info;
2092 	int error;
2093 
2094 	softc = (struct da_softc *)periph->softc;
2095 	dp = &softc->params;
2096 
2097 	error = dagetcapacity(periph);
2098 
2099 	/*
2100 	 * Only reprobe on initial open and if the media is removable.
2101 	 *
2102 	 * NOTE: If we setdiskinfo() it will take the device probe
2103 	 *	 a bit of time to probe the slices and partitions,
2104 	 *	 and mess up booting.  So avoid if nothing has changed.
2105 	 *	 XXX
2106 	 */
2107 	if (softc->flags & DA_FLAG_OPEN)
2108 		return (error);
2109 	if ((softc->flags & DA_FLAG_PACK_REMOVABLE) == 0)
2110 		return (error);
2111 
2112 	bzero(&info, sizeof(info));
2113 	info.d_type = DTYPE_SCSI;
2114 	info.d_serialno = xpt_path_serialno(periph->path);
2115 
2116 	if (error == 0) {
2117 		CAM_SIM_UNLOCK(periph->sim);
2118 		info.d_media_blksize = softc->params.secsize;
2119 		info.d_media_blocks = softc->params.sectors;
2120 		info.d_media_size = 0;
2121 		info.d_secpertrack = softc->params.secs_per_track;
2122 		info.d_nheads = softc->params.heads;
2123 		info.d_ncylinders = softc->params.cylinders;
2124 		info.d_secpercyl = softc->params.heads *
2125 					softc->params.secs_per_track;
2126 		info.d_serialno = xpt_path_serialno(periph->path);
2127 		if (info.d_media_blocks != softc->disk.d_info.d_media_blocks) {
2128 			kprintf("%s%d: open removable media: "
2129 				"%juMB (%ju %u byte sectors: %dH %dS/T %dC)\n",
2130 				periph->periph_name, periph->unit_number,
2131 				(uintmax_t)(((uintmax_t)dp->secsize *
2132 					     dp->sectors) / (1024*1024)),
2133 				(uintmax_t)dp->sectors, dp->secsize,
2134 				dp->heads, dp->secs_per_track, dp->cylinders);
2135 			disk_setdiskinfo(&softc->disk, &info);
2136 		}
2137 		CAM_SIM_LOCK(periph->sim);
2138 	} else {
2139 		kprintf("%s%d: open removable media: no media present\n",
2140 			periph->periph_name, periph->unit_number);
2141 		info.d_media_blksize = 512;
2142 		disk_setdiskinfo(&softc->disk, &info);
2143 	}
2144 	return (error);
2145 }
2146 
2147 static int
2148 dagetcapacity(struct cam_periph *periph)
2149 {
2150 	struct da_softc *softc;
2151 	union ccb *ccb;
2152 	struct scsi_read_capacity_data *rcap;
2153 	struct scsi_read_capacity_data_16 *rcaplong;
2154 	uint32_t block_len;
2155 	uint64_t maxsector;
2156 	int error;
2157 
2158 	softc = (struct da_softc *)periph->softc;
2159 	block_len = 0;
2160 	maxsector = 0;
2161 	error = 0;
2162 
2163 	/* Do a read capacity */
2164 	rcap = (struct scsi_read_capacity_data *)kmalloc(sizeof(*rcaplong),
2165 							 M_SCSIDA, M_INTWAIT);
2166 
2167 	ccb = cam_periph_getccb(periph, /*priority*/1);
2168 	scsi_read_capacity(&ccb->csio,
2169 			   /*retries*/4,
2170 			   /*cbfncp*/dadone,
2171 			   MSG_SIMPLE_Q_TAG,
2172 			   rcap,
2173 			   SSD_FULL_SIZE,
2174 			   /*timeout*/60000);
2175 	ccb->ccb_h.ccb_bio = NULL;
2176 
2177 	error = cam_periph_runccb(ccb, daerror,
2178 				  /*cam_flags*/CAM_RETRY_SELTO,
2179 				  /*sense_flags*/SF_RETRY_UA,
2180 				  &softc->device_stats);
2181 
2182 	if ((ccb->ccb_h.status & CAM_DEV_QFRZN) != 0)
2183 		cam_release_devq(ccb->ccb_h.path,
2184 				 /*relsim_flags*/0,
2185 				 /*reduction*/0,
2186 				 /*timeout*/0,
2187 				 /*getcount_only*/0);
2188 
2189 	if (error == 0) {
2190 		block_len = scsi_4btoul(rcap->length);
2191 		maxsector = scsi_4btoul(rcap->addr);
2192 
2193 		if (maxsector != 0xffffffff)
2194 			goto done;
2195 	} else
2196 		goto done;
2197 
2198 	rcaplong = (struct scsi_read_capacity_data_16 *)rcap;
2199 
2200 	scsi_read_capacity_16(&ccb->csio,
2201 			      /*retries*/ 4,
2202 			      /*cbfcnp*/ dadone,
2203 			      /*tag_action*/ MSG_SIMPLE_Q_TAG,
2204 			      /*lba*/ 0,
2205 			      /*reladr*/ 0,
2206 			      /*pmi*/ 0,
2207 			      rcaplong,
2208 			      /*sense_len*/ SSD_FULL_SIZE,
2209 			      /*timeout*/ 60000);
2210 	ccb->ccb_h.ccb_bio = NULL;
2211 
2212 	error = cam_periph_runccb(ccb, daerror,
2213 				  /*cam_flags*/CAM_RETRY_SELTO,
2214 				  /*sense_flags*/SF_RETRY_UA,
2215 				  &softc->device_stats);
2216 
2217 	if ((ccb->ccb_h.status & CAM_DEV_QFRZN) != 0)
2218 		cam_release_devq(ccb->ccb_h.path,
2219 				 /*relsim_flags*/0,
2220 				 /*reduction*/0,
2221 				 /*timeout*/0,
2222 				 /*getcount_only*/0);
2223 
2224 	if (error == 0) {
2225 		block_len = scsi_4btoul(rcaplong->length);
2226 		maxsector = scsi_8btou64(rcaplong->addr);
2227 	}
2228 
2229 done:
2230 
2231 	if (error == 0)
2232 		dasetgeom(periph, block_len, maxsector);
2233 
2234 	xpt_release_ccb(ccb);
2235 
2236 	kfree(rcap, M_SCSIDA);
2237 
2238 	return (error);
2239 }
2240 
2241 static void
2242 dasetgeom(struct cam_periph *periph, uint32_t block_len, uint64_t maxsector)
2243 {
2244 	struct ccb_calc_geometry ccg;
2245 	struct da_softc *softc;
2246 	struct disk_params *dp;
2247 
2248 	softc = (struct da_softc *)periph->softc;
2249 
2250 	dp = &softc->params;
2251 	dp->secsize = block_len;
2252 	dp->sectors = maxsector + 1;
2253 	/*
2254 	 * Have the controller provide us with a geometry
2255 	 * for this disk.  The only time the geometry
2256 	 * matters is when we boot and the controller
2257 	 * is the only one knowledgeable enough to come
2258 	 * up with something that will make this a bootable
2259 	 * device.
2260 	 */
2261 	xpt_setup_ccb(&ccg.ccb_h, periph->path, /*priority*/1);
2262 	ccg.ccb_h.func_code = XPT_CALC_GEOMETRY;
2263 	ccg.block_size = dp->secsize;
2264 	ccg.volume_size = dp->sectors;
2265 	ccg.heads = 0;
2266 	ccg.secs_per_track = 0;
2267 	ccg.cylinders = 0;
2268 	xpt_action((union ccb*)&ccg);
2269 	if ((ccg.ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
2270 		/*
2271 		 * We don't know what went wrong here- but just pick
2272 		 * a geometry so we don't have nasty things like divide
2273 		 * by zero.
2274 		 */
2275 		dp->heads = 255;
2276 		dp->secs_per_track = 255;
2277 		dp->cylinders = dp->sectors / (255 * 255);
2278 		if (dp->cylinders == 0) {
2279 			dp->cylinders = 1;
2280 		}
2281 	} else {
2282 		dp->heads = ccg.heads;
2283 		dp->secs_per_track = ccg.secs_per_track;
2284 		dp->cylinders = ccg.cylinders;
2285 	}
2286 }
2287 
2288 static void
2289 dasendorderedtag(void *arg)
2290 {
2291 	struct da_softc *softc = arg;
2292 
2293 	if (da_send_ordered) {
2294 		if ((softc->ordered_tag_count == 0)
2295 		 && ((softc->flags & DA_FLAG_WENT_IDLE) == 0)) {
2296 			softc->flags |= DA_FLAG_NEED_OTAG;
2297 		}
2298 		if (softc->outstanding_cmds_rd || softc->outstanding_cmds_wr)
2299 			softc->flags &= ~DA_FLAG_WENT_IDLE;
2300 
2301 		softc->ordered_tag_count = 0;
2302 	}
2303 	/* Queue us up again */
2304 	callout_reset(&softc->sendordered_c,
2305 	    (DA_DEFAULT_TIMEOUT * hz) / DA_ORDEREDTAG_INTERVAL,
2306 	    dasendorderedtag, softc);
2307 }
2308 
2309 /*
2310  * Step through all DA peripheral drivers, and if the device is still open,
2311  * sync the disk cache to physical media.
2312  */
2313 static void
2314 dashutdown(void * arg, int howto)
2315 {
2316 	struct cam_periph *periph;
2317 	struct da_softc *softc;
2318 
2319 	TAILQ_FOREACH(periph, &dadriver.units, unit_links) {
2320 		union ccb ccb;
2321 
2322 		cam_periph_lock(periph);
2323 		softc = (struct da_softc *)periph->softc;
2324 
2325 		/*
2326 		 * We only sync the cache if the drive is still open, and
2327 		 * if the drive is capable of it..
2328 		 */
2329 		if (((softc->flags & DA_FLAG_OPEN) == 0)
2330 		 || (softc->quirks & DA_Q_NO_SYNC_CACHE)) {
2331 			cam_periph_unlock(periph);
2332 			continue;
2333 		}
2334 
2335 		xpt_setup_ccb(&ccb.ccb_h, periph->path, /*priority*/1);
2336 
2337 		ccb.ccb_h.ccb_state = DA_CCB_DUMP;
2338 		scsi_synchronize_cache(&ccb.csio,
2339 				       /*retries*/1,
2340 				       /*cbfcnp*/dadone,
2341 				       MSG_SIMPLE_Q_TAG,
2342 				       /*begin_lba*/0, /* whole disk */
2343 				       /*lb_count*/0,
2344 				       SSD_FULL_SIZE,
2345 				       60 * 60 * 1000);
2346 
2347 		xpt_polled_action(&ccb);
2348 
2349 		if ((ccb.ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
2350 			if (((ccb.ccb_h.status & CAM_STATUS_MASK) ==
2351 			     CAM_SCSI_STATUS_ERROR)
2352 			 && (ccb.csio.scsi_status == SCSI_STATUS_CHECK_COND)){
2353 				int error_code, sense_key, asc, ascq;
2354 
2355 				scsi_extract_sense(&ccb.csio.sense_data,
2356 						   &error_code, &sense_key,
2357 						   &asc, &ascq);
2358 
2359 				if (sense_key != SSD_KEY_ILLEGAL_REQUEST)
2360 					scsi_sense_print(&ccb.csio);
2361 			} else {
2362 				xpt_print(periph->path, "Synchronize "
2363 				    "cache failed, status == 0x%x, scsi status "
2364 				    "== 0x%x\n", ccb.ccb_h.status,
2365 				    ccb.csio.scsi_status);
2366 			}
2367 		}
2368 
2369 		if ((ccb.ccb_h.status & CAM_DEV_QFRZN) != 0)
2370 			cam_release_devq(ccb.ccb_h.path,
2371 					 /*relsim_flags*/0,
2372 					 /*reduction*/0,
2373 					 /*timeout*/0,
2374 					 /*getcount_only*/0);
2375 
2376 		cam_periph_unlock(periph);
2377 	}
2378 }
2379 
2380 #else /* !_KERNEL */
2381 
2382 /*
2383  * XXX This is only left out of the kernel build to silence warnings.  If,
2384  * for some reason this function is used in the kernel, the ifdefs should
2385  * be moved so it is included both in the kernel and userland.
2386  */
2387 void
2388 scsi_format_unit(struct ccb_scsiio *csio, u_int32_t retries,
2389 		 void (*cbfcnp)(struct cam_periph *, union ccb *),
2390 		 u_int8_t tag_action, u_int8_t byte2, u_int16_t ileave,
2391 		 u_int8_t *data_ptr, u_int32_t dxfer_len, u_int8_t sense_len,
2392 		 u_int32_t timeout)
2393 {
2394 	struct scsi_format_unit *scsi_cmd;
2395 
2396 	scsi_cmd = (struct scsi_format_unit *)&csio->cdb_io.cdb_bytes;
2397 	scsi_cmd->opcode = FORMAT_UNIT;
2398 	scsi_cmd->byte2 = byte2;
2399 	scsi_ulto2b(ileave, scsi_cmd->interleave);
2400 
2401 	cam_fill_csio(csio,
2402 		      retries,
2403 		      cbfcnp,
2404 		      /*flags*/ (dxfer_len > 0) ? CAM_DIR_OUT : CAM_DIR_NONE,
2405 		      tag_action,
2406 		      data_ptr,
2407 		      dxfer_len,
2408 		      sense_len,
2409 		      sizeof(*scsi_cmd),
2410 		      timeout);
2411 }
2412 
2413 #endif /* _KERNEL */
2414