xref: /freebsd/sys/dev/mpr/mpr_sas.c (revision e17f5b1d)
1 /*-
2  * Copyright (c) 2009 Yahoo! Inc.
3  * Copyright (c) 2011-2015 LSI Corp.
4  * Copyright (c) 2013-2016 Avago Technologies
5  * Copyright 2000-2020 Broadcom Inc.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  * Broadcom Inc. (LSI) MPT-Fusion Host Adapter FreeBSD
30  *
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 /* Communications core for Avago Technologies (LSI) MPT3 */
37 
38 /* TODO Move headers to mprvar */
39 #include <sys/types.h>
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/selinfo.h>
44 #include <sys/module.h>
45 #include <sys/bus.h>
46 #include <sys/conf.h>
47 #include <sys/bio.h>
48 #include <sys/malloc.h>
49 #include <sys/uio.h>
50 #include <sys/sysctl.h>
51 #include <sys/endian.h>
52 #include <sys/queue.h>
53 #include <sys/kthread.h>
54 #include <sys/taskqueue.h>
55 #include <sys/sbuf.h>
56 
57 #include <machine/bus.h>
58 #include <machine/resource.h>
59 #include <sys/rman.h>
60 
61 #include <machine/stdarg.h>
62 
63 #include <cam/cam.h>
64 #include <cam/cam_ccb.h>
65 #include <cam/cam_debug.h>
66 #include <cam/cam_sim.h>
67 #include <cam/cam_xpt_sim.h>
68 #include <cam/cam_xpt_periph.h>
69 #include <cam/cam_periph.h>
70 #include <cam/scsi/scsi_all.h>
71 #include <cam/scsi/scsi_message.h>
72 #include <cam/scsi/smp_all.h>
73 
74 #include <dev/nvme/nvme.h>
75 
76 #include <dev/mpr/mpi/mpi2_type.h>
77 #include <dev/mpr/mpi/mpi2.h>
78 #include <dev/mpr/mpi/mpi2_ioc.h>
79 #include <dev/mpr/mpi/mpi2_sas.h>
80 #include <dev/mpr/mpi/mpi2_pci.h>
81 #include <dev/mpr/mpi/mpi2_cnfg.h>
82 #include <dev/mpr/mpi/mpi2_init.h>
83 #include <dev/mpr/mpi/mpi2_tool.h>
84 #include <dev/mpr/mpr_ioctl.h>
85 #include <dev/mpr/mprvar.h>
86 #include <dev/mpr/mpr_table.h>
87 #include <dev/mpr/mpr_sas.h>
88 
89 #define MPRSAS_DISCOVERY_TIMEOUT	20
90 #define MPRSAS_MAX_DISCOVERY_TIMEOUTS	10 /* 200 seconds */
91 
92 /*
93  * static array to check SCSI OpCode for EEDP protection bits
94  */
95 #define	PRO_R MPI2_SCSIIO_EEDPFLAGS_CHECK_REMOVE_OP
96 #define	PRO_W MPI2_SCSIIO_EEDPFLAGS_INSERT_OP
97 #define	PRO_V MPI2_SCSIIO_EEDPFLAGS_INSERT_OP
98 static uint8_t op_code_prot[256] = {
99 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
100 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
101 	0, 0, 0, 0, 0, 0, 0, 0, PRO_R, 0, PRO_W, 0, 0, 0, PRO_W, PRO_V,
102 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
103 	0, PRO_W, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
104 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
105 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
106 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
107 	0, 0, 0, 0, 0, 0, 0, 0, PRO_R, 0, PRO_W, 0, 0, 0, PRO_W, PRO_V,
108 	0, 0, 0, PRO_W, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
109 	0, 0, 0, 0, 0, 0, 0, 0, PRO_R, 0, PRO_W, 0, 0, 0, PRO_W, PRO_V,
110 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
111 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
112 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
113 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
114 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
115 };
116 
117 MALLOC_DEFINE(M_MPRSAS, "MPRSAS", "MPR SAS memory");
118 
119 static void mprsas_remove_device(struct mpr_softc *, struct mpr_command *);
120 static void mprsas_remove_complete(struct mpr_softc *, struct mpr_command *);
121 static void mprsas_action(struct cam_sim *sim, union ccb *ccb);
122 static void mprsas_poll(struct cam_sim *sim);
123 static void mprsas_scsiio_timeout(void *data);
124 static void mprsas_abort_complete(struct mpr_softc *sc, struct mpr_command *cm);
125 static void mprsas_action_scsiio(struct mprsas_softc *, union ccb *);
126 static void mprsas_scsiio_complete(struct mpr_softc *, struct mpr_command *);
127 static void mprsas_action_resetdev(struct mprsas_softc *, union ccb *);
128 static void mprsas_resetdev_complete(struct mpr_softc *, struct mpr_command *);
129 static int mprsas_send_abort(struct mpr_softc *sc, struct mpr_command *tm,
130     struct mpr_command *cm);
131 static void mprsas_async(void *callback_arg, uint32_t code,
132     struct cam_path *path, void *arg);
133 static int mprsas_send_portenable(struct mpr_softc *sc);
134 static void mprsas_portenable_complete(struct mpr_softc *sc,
135     struct mpr_command *cm);
136 
137 static void mprsas_smpio_complete(struct mpr_softc *sc, struct mpr_command *cm);
138 static void mprsas_send_smpcmd(struct mprsas_softc *sassc, union ccb *ccb,
139     uint64_t sasaddr);
140 static void mprsas_action_smpio(struct mprsas_softc *sassc, union ccb *ccb);
141 
142 struct mprsas_target *
143 mprsas_find_target_by_handle(struct mprsas_softc *sassc, int start,
144     uint16_t handle)
145 {
146 	struct mprsas_target *target;
147 	int i;
148 
149 	for (i = start; i < sassc->maxtargets; i++) {
150 		target = &sassc->targets[i];
151 		if (target->handle == handle)
152 			return (target);
153 	}
154 
155 	return (NULL);
156 }
157 
158 /* we need to freeze the simq during attach and diag reset, to avoid failing
159  * commands before device handles have been found by discovery.  Since
160  * discovery involves reading config pages and possibly sending commands,
161  * discovery actions may continue even after we receive the end of discovery
162  * event, so refcount discovery actions instead of assuming we can unfreeze
163  * the simq when we get the event.
164  */
165 void
166 mprsas_startup_increment(struct mprsas_softc *sassc)
167 {
168 	MPR_FUNCTRACE(sassc->sc);
169 
170 	if ((sassc->flags & MPRSAS_IN_STARTUP) != 0) {
171 		if (sassc->startup_refcount++ == 0) {
172 			/* just starting, freeze the simq */
173 			mpr_dprint(sassc->sc, MPR_INIT,
174 			    "%s freezing simq\n", __func__);
175 			xpt_hold_boot();
176 			xpt_freeze_simq(sassc->sim, 1);
177 		}
178 		mpr_dprint(sassc->sc, MPR_INIT, "%s refcount %u\n", __func__,
179 		    sassc->startup_refcount);
180 	}
181 }
182 
183 void
184 mprsas_release_simq_reinit(struct mprsas_softc *sassc)
185 {
186 	if (sassc->flags & MPRSAS_QUEUE_FROZEN) {
187 		sassc->flags &= ~MPRSAS_QUEUE_FROZEN;
188 		xpt_release_simq(sassc->sim, 1);
189 		mpr_dprint(sassc->sc, MPR_INFO, "Unfreezing SIM queue\n");
190 	}
191 }
192 
193 void
194 mprsas_startup_decrement(struct mprsas_softc *sassc)
195 {
196 	MPR_FUNCTRACE(sassc->sc);
197 
198 	if ((sassc->flags & MPRSAS_IN_STARTUP) != 0) {
199 		if (--sassc->startup_refcount == 0) {
200 			/* finished all discovery-related actions, release
201 			 * the simq and rescan for the latest topology.
202 			 */
203 			mpr_dprint(sassc->sc, MPR_INIT,
204 			    "%s releasing simq\n", __func__);
205 			sassc->flags &= ~MPRSAS_IN_STARTUP;
206 			xpt_release_simq(sassc->sim, 1);
207 			xpt_release_boot();
208 		}
209 		mpr_dprint(sassc->sc, MPR_INIT, "%s refcount %u\n", __func__,
210 		    sassc->startup_refcount);
211 	}
212 }
213 
214 /*
215  * The firmware requires us to stop sending commands when we're doing task
216  * management.
217  * use.
218  * XXX The logic for serializing the device has been made lazy and moved to
219  * mprsas_prepare_for_tm().
220  */
221 struct mpr_command *
222 mprsas_alloc_tm(struct mpr_softc *sc)
223 {
224 	MPI2_SCSI_TASK_MANAGE_REQUEST *req;
225 	struct mpr_command *tm;
226 
227 	MPR_FUNCTRACE(sc);
228 	tm = mpr_alloc_high_priority_command(sc);
229 	if (tm == NULL)
230 		return (NULL);
231 
232 	req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req;
233 	req->Function = MPI2_FUNCTION_SCSI_TASK_MGMT;
234 	return tm;
235 }
236 
237 void
238 mprsas_free_tm(struct mpr_softc *sc, struct mpr_command *tm)
239 {
240 	int target_id = 0xFFFFFFFF;
241 
242 	MPR_FUNCTRACE(sc);
243 	if (tm == NULL)
244 		return;
245 
246 	/*
247 	 * For TM's the devq is frozen for the device.  Unfreeze it here and
248 	 * free the resources used for freezing the devq.  Must clear the
249 	 * INRESET flag as well or scsi I/O will not work.
250 	 */
251 	if (tm->cm_targ != NULL) {
252 		tm->cm_targ->flags &= ~MPRSAS_TARGET_INRESET;
253 		target_id = tm->cm_targ->tid;
254 	}
255 	if (tm->cm_ccb) {
256 		mpr_dprint(sc, MPR_INFO, "Unfreezing devq for target ID %d\n",
257 		    target_id);
258 		xpt_release_devq(tm->cm_ccb->ccb_h.path, 1, TRUE);
259 		xpt_free_path(tm->cm_ccb->ccb_h.path);
260 		xpt_free_ccb(tm->cm_ccb);
261 	}
262 
263 	mpr_free_high_priority_command(sc, tm);
264 }
265 
266 void
267 mprsas_rescan_target(struct mpr_softc *sc, struct mprsas_target *targ)
268 {
269 	struct mprsas_softc *sassc = sc->sassc;
270 	path_id_t pathid;
271 	target_id_t targetid;
272 	union ccb *ccb;
273 
274 	MPR_FUNCTRACE(sc);
275 	pathid = cam_sim_path(sassc->sim);
276 	if (targ == NULL)
277 		targetid = CAM_TARGET_WILDCARD;
278 	else
279 		targetid = targ - sassc->targets;
280 
281 	/*
282 	 * Allocate a CCB and schedule a rescan.
283 	 */
284 	ccb = xpt_alloc_ccb_nowait();
285 	if (ccb == NULL) {
286 		mpr_dprint(sc, MPR_ERROR, "unable to alloc CCB for rescan\n");
287 		return;
288 	}
289 
290 	if (xpt_create_path(&ccb->ccb_h.path, NULL, pathid, targetid,
291 	    CAM_LUN_WILDCARD) != CAM_REQ_CMP) {
292 		mpr_dprint(sc, MPR_ERROR, "unable to create path for rescan\n");
293 		xpt_free_ccb(ccb);
294 		return;
295 	}
296 
297 	if (targetid == CAM_TARGET_WILDCARD)
298 		ccb->ccb_h.func_code = XPT_SCAN_BUS;
299 	else
300 		ccb->ccb_h.func_code = XPT_SCAN_TGT;
301 
302 	mpr_dprint(sc, MPR_TRACE, "%s targetid %u\n", __func__, targetid);
303 	xpt_rescan(ccb);
304 }
305 
306 static void
307 mprsas_log_command(struct mpr_command *cm, u_int level, const char *fmt, ...)
308 {
309 	struct sbuf sb;
310 	va_list ap;
311 	char str[224];
312 	char path_str[64];
313 
314 	if (cm == NULL)
315 		return;
316 
317 	/* No need to be in here if debugging isn't enabled */
318 	if ((cm->cm_sc->mpr_debug & level) == 0)
319 		return;
320 
321 	sbuf_new(&sb, str, sizeof(str), 0);
322 
323 	va_start(ap, fmt);
324 
325 	if (cm->cm_ccb != NULL) {
326 		xpt_path_string(cm->cm_ccb->csio.ccb_h.path, path_str,
327 		    sizeof(path_str));
328 		sbuf_cat(&sb, path_str);
329 		if (cm->cm_ccb->ccb_h.func_code == XPT_SCSI_IO) {
330 			scsi_command_string(&cm->cm_ccb->csio, &sb);
331 			sbuf_printf(&sb, "length %d ",
332 			    cm->cm_ccb->csio.dxfer_len);
333 		}
334 	} else {
335 		sbuf_printf(&sb, "(noperiph:%s%d:%u:%u:%u): ",
336 		    cam_sim_name(cm->cm_sc->sassc->sim),
337 		    cam_sim_unit(cm->cm_sc->sassc->sim),
338 		    cam_sim_bus(cm->cm_sc->sassc->sim),
339 		    cm->cm_targ ? cm->cm_targ->tid : 0xFFFFFFFF,
340 		    cm->cm_lun);
341 	}
342 
343 	sbuf_printf(&sb, "SMID %u ", cm->cm_desc.Default.SMID);
344 	sbuf_vprintf(&sb, fmt, ap);
345 	sbuf_finish(&sb);
346 	mpr_print_field(cm->cm_sc, "%s", sbuf_data(&sb));
347 
348 	va_end(ap);
349 }
350 
351 static void
352 mprsas_remove_volume(struct mpr_softc *sc, struct mpr_command *tm)
353 {
354 	MPI2_SCSI_TASK_MANAGE_REPLY *reply;
355 	struct mprsas_target *targ;
356 	uint16_t handle;
357 
358 	MPR_FUNCTRACE(sc);
359 
360 	reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply;
361 	handle = (uint16_t)(uintptr_t)tm->cm_complete_data;
362 	targ = tm->cm_targ;
363 
364 	if (reply == NULL) {
365 		/* XXX retry the remove after the diag reset completes? */
366 		mpr_dprint(sc, MPR_FAULT, "%s NULL reply resetting device "
367 		    "0x%04x\n", __func__, handle);
368 		mprsas_free_tm(sc, tm);
369 		return;
370 	}
371 
372 	if ((le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) !=
373 	    MPI2_IOCSTATUS_SUCCESS) {
374 		mpr_dprint(sc, MPR_ERROR, "IOCStatus = 0x%x while resetting "
375 		    "device 0x%x\n", le16toh(reply->IOCStatus), handle);
376 	}
377 
378 	mpr_dprint(sc, MPR_XINFO, "Reset aborted %u commands\n",
379 	    le32toh(reply->TerminationCount));
380 	mpr_free_reply(sc, tm->cm_reply_data);
381 	tm->cm_reply = NULL;	/* Ensures the reply won't get re-freed */
382 
383 	mpr_dprint(sc, MPR_XINFO, "clearing target %u handle 0x%04x\n",
384 	    targ->tid, handle);
385 
386 	/*
387 	 * Don't clear target if remove fails because things will get confusing.
388 	 * Leave the devname and sasaddr intact so that we know to avoid reusing
389 	 * this target id if possible, and so we can assign the same target id
390 	 * to this device if it comes back in the future.
391 	 */
392 	if ((le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) ==
393 	    MPI2_IOCSTATUS_SUCCESS) {
394 		targ = tm->cm_targ;
395 		targ->handle = 0x0;
396 		targ->encl_handle = 0x0;
397 		targ->encl_level_valid = 0x0;
398 		targ->encl_level = 0x0;
399 		targ->connector_name[0] = ' ';
400 		targ->connector_name[1] = ' ';
401 		targ->connector_name[2] = ' ';
402 		targ->connector_name[3] = ' ';
403 		targ->encl_slot = 0x0;
404 		targ->exp_dev_handle = 0x0;
405 		targ->phy_num = 0x0;
406 		targ->linkrate = 0x0;
407 		targ->devinfo = 0x0;
408 		targ->flags = 0x0;
409 		targ->scsi_req_desc_type = 0;
410 	}
411 
412 	mprsas_free_tm(sc, tm);
413 }
414 
415 
416 /*
417  * No Need to call "MPI2_SAS_OP_REMOVE_DEVICE" For Volume removal.
418  * Otherwise Volume Delete is same as Bare Drive Removal.
419  */
420 void
421 mprsas_prepare_volume_remove(struct mprsas_softc *sassc, uint16_t handle)
422 {
423 	MPI2_SCSI_TASK_MANAGE_REQUEST *req;
424 	struct mpr_softc *sc;
425 	struct mpr_command *cm;
426 	struct mprsas_target *targ = NULL;
427 
428 	MPR_FUNCTRACE(sassc->sc);
429 	sc = sassc->sc;
430 
431 	targ = mprsas_find_target_by_handle(sassc, 0, handle);
432 	if (targ == NULL) {
433 		/* FIXME: what is the action? */
434 		/* We don't know about this device? */
435 		mpr_dprint(sc, MPR_ERROR,
436 		   "%s %d : invalid handle 0x%x \n", __func__,__LINE__, handle);
437 		return;
438 	}
439 
440 	targ->flags |= MPRSAS_TARGET_INREMOVAL;
441 
442 	cm = mprsas_alloc_tm(sc);
443 	if (cm == NULL) {
444 		mpr_dprint(sc, MPR_ERROR,
445 		    "%s: command alloc failure\n", __func__);
446 		return;
447 	}
448 
449 	mprsas_rescan_target(sc, targ);
450 
451 	req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)cm->cm_req;
452 	req->DevHandle = targ->handle;
453 	req->TaskType = MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET;
454 
455 	if (!targ->is_nvme || sc->custom_nvme_tm_handling) {
456 		/* SAS Hard Link Reset / SATA Link Reset */
457 		req->MsgFlags = MPI2_SCSITASKMGMT_MSGFLAGS_LINK_RESET;
458 	} else {
459 		/* PCIe Protocol Level Reset*/
460 		req->MsgFlags =
461 		    MPI26_SCSITASKMGMT_MSGFLAGS_PROTOCOL_LVL_RST_PCIE;
462 	}
463 
464 	cm->cm_targ = targ;
465 	cm->cm_data = NULL;
466 	cm->cm_complete = mprsas_remove_volume;
467 	cm->cm_complete_data = (void *)(uintptr_t)handle;
468 
469 	mpr_dprint(sc, MPR_INFO, "%s: Sending reset for target ID %d\n",
470 	    __func__, targ->tid);
471 	mprsas_prepare_for_tm(sc, cm, targ, CAM_LUN_WILDCARD);
472 
473 	mpr_map_command(sc, cm);
474 }
475 
476 /*
477  * The firmware performs debounce on the link to avoid transient link errors
478  * and false removals.  When it does decide that link has been lost and a
479  * device needs to go away, it expects that the host will perform a target reset
480  * and then an op remove.  The reset has the side-effect of aborting any
481  * outstanding requests for the device, which is required for the op-remove to
482  * succeed.  It's not clear if the host should check for the device coming back
483  * alive after the reset.
484  */
485 void
486 mprsas_prepare_remove(struct mprsas_softc *sassc, uint16_t handle)
487 {
488 	MPI2_SCSI_TASK_MANAGE_REQUEST *req;
489 	struct mpr_softc *sc;
490 	struct mpr_command *tm;
491 	struct mprsas_target *targ = NULL;
492 
493 	MPR_FUNCTRACE(sassc->sc);
494 
495 	sc = sassc->sc;
496 
497 	targ = mprsas_find_target_by_handle(sassc, 0, handle);
498 	if (targ == NULL) {
499 		/* FIXME: what is the action? */
500 		/* We don't know about this device? */
501 		mpr_dprint(sc, MPR_ERROR, "%s : invalid handle 0x%x \n",
502 		    __func__, handle);
503 		return;
504 	}
505 
506 	targ->flags |= MPRSAS_TARGET_INREMOVAL;
507 
508 	tm = mprsas_alloc_tm(sc);
509 	if (tm == NULL) {
510 		mpr_dprint(sc, MPR_ERROR, "%s: command alloc failure\n",
511 		    __func__);
512 		return;
513 	}
514 
515 	mprsas_rescan_target(sc, targ);
516 
517 	req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req;
518 	memset(req, 0, sizeof(*req));
519 	req->DevHandle = htole16(targ->handle);
520 	req->TaskType = MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET;
521 
522 	/* SAS Hard Link Reset / SATA Link Reset */
523 	req->MsgFlags = MPI2_SCSITASKMGMT_MSGFLAGS_LINK_RESET;
524 
525 	tm->cm_targ = targ;
526 	tm->cm_data = NULL;
527 	tm->cm_complete = mprsas_remove_device;
528 	tm->cm_complete_data = (void *)(uintptr_t)handle;
529 
530 	mpr_dprint(sc, MPR_INFO, "%s: Sending reset for target ID %d\n",
531 	    __func__, targ->tid);
532 	mprsas_prepare_for_tm(sc, tm, targ, CAM_LUN_WILDCARD);
533 
534 	mpr_map_command(sc, tm);
535 }
536 
537 static void
538 mprsas_remove_device(struct mpr_softc *sc, struct mpr_command *tm)
539 {
540 	MPI2_SCSI_TASK_MANAGE_REPLY *reply;
541 	MPI2_SAS_IOUNIT_CONTROL_REQUEST *req;
542 	struct mprsas_target *targ;
543 	uint16_t handle;
544 
545 	MPR_FUNCTRACE(sc);
546 
547 	reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply;
548 	handle = (uint16_t)(uintptr_t)tm->cm_complete_data;
549 	targ = tm->cm_targ;
550 
551 	/*
552 	 * Currently there should be no way we can hit this case.  It only
553 	 * happens when we have a failure to allocate chain frames, and
554 	 * task management commands don't have S/G lists.
555 	 */
556 	if ((tm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) {
557 		mpr_dprint(sc, MPR_ERROR, "%s: cm_flags = %#x for remove of "
558 		    "handle %#04x! This should not happen!\n", __func__,
559 		    tm->cm_flags, handle);
560 	}
561 
562 	if (reply == NULL) {
563 		/* XXX retry the remove after the diag reset completes? */
564 		mpr_dprint(sc, MPR_FAULT, "%s NULL reply resetting device "
565 		    "0x%04x\n", __func__, handle);
566 		mprsas_free_tm(sc, tm);
567 		return;
568 	}
569 
570 	if ((le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) !=
571 	    MPI2_IOCSTATUS_SUCCESS) {
572 		mpr_dprint(sc, MPR_ERROR, "IOCStatus = 0x%x while resetting "
573 		    "device 0x%x\n", le16toh(reply->IOCStatus), handle);
574 	}
575 
576 	mpr_dprint(sc, MPR_XINFO, "Reset aborted %u commands\n",
577 	    le32toh(reply->TerminationCount));
578 	mpr_free_reply(sc, tm->cm_reply_data);
579 	tm->cm_reply = NULL;	/* Ensures the reply won't get re-freed */
580 
581 	/* Reuse the existing command */
582 	req = (MPI2_SAS_IOUNIT_CONTROL_REQUEST *)tm->cm_req;
583 	memset(req, 0, sizeof(*req));
584 	req->Function = MPI2_FUNCTION_SAS_IO_UNIT_CONTROL;
585 	req->Operation = MPI2_SAS_OP_REMOVE_DEVICE;
586 	req->DevHandle = htole16(handle);
587 	tm->cm_data = NULL;
588 	tm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE;
589 	tm->cm_complete = mprsas_remove_complete;
590 	tm->cm_complete_data = (void *)(uintptr_t)handle;
591 
592 	/*
593 	 * Wait to send the REMOVE_DEVICE until all the commands have cleared.
594 	 * They should be aborted or time out and we'll kick thus off there
595 	 * if so.
596 	 */
597 	if (TAILQ_FIRST(&targ->commands) == NULL) {
598 		mpr_dprint(sc, MPR_INFO, "No pending commands: starting remove_device\n");
599 		mpr_map_command(sc, tm);
600 		targ->pending_remove_tm = NULL;
601 	} else {
602 		targ->pending_remove_tm = tm;
603 	}
604 
605 	mpr_dprint(sc, MPR_INFO, "clearing target %u handle 0x%04x\n",
606 	    targ->tid, handle);
607 	if (targ->encl_level_valid) {
608 		mpr_dprint(sc, MPR_INFO, "At enclosure level %d, slot %d, "
609 		    "connector name (%4s)\n", targ->encl_level, targ->encl_slot,
610 		    targ->connector_name);
611 	}
612 }
613 
614 static void
615 mprsas_remove_complete(struct mpr_softc *sc, struct mpr_command *tm)
616 {
617 	MPI2_SAS_IOUNIT_CONTROL_REPLY *reply;
618 	uint16_t handle;
619 	struct mprsas_target *targ;
620 	struct mprsas_lun *lun;
621 
622 	MPR_FUNCTRACE(sc);
623 
624 	reply = (MPI2_SAS_IOUNIT_CONTROL_REPLY *)tm->cm_reply;
625 	handle = (uint16_t)(uintptr_t)tm->cm_complete_data;
626 
627 	targ = tm->cm_targ;
628 
629 	/*
630 	 * At this point, we should have no pending commands for the target.
631 	 * The remove target has just completed.
632 	 */
633 	KASSERT(TAILQ_FIRST(&targ->commands) == NULL,
634 	    ("%s: no commands should be pending\n", __func__));
635 
636 	/*
637 	 * Currently there should be no way we can hit this case.  It only
638 	 * happens when we have a failure to allocate chain frames, and
639 	 * task management commands don't have S/G lists.
640 	 */
641 	if ((tm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) {
642 		mpr_dprint(sc, MPR_XINFO, "%s: cm_flags = %#x for remove of "
643 		    "handle %#04x! This should not happen!\n", __func__,
644 		    tm->cm_flags, handle);
645 		mprsas_free_tm(sc, tm);
646 		return;
647 	}
648 
649 	if (reply == NULL) {
650 		/* most likely a chip reset */
651 		mpr_dprint(sc, MPR_FAULT, "%s NULL reply removing device "
652 		    "0x%04x\n", __func__, handle);
653 		mprsas_free_tm(sc, tm);
654 		return;
655 	}
656 
657 	mpr_dprint(sc, MPR_XINFO, "%s on handle 0x%04x, IOCStatus= 0x%x\n",
658 	    __func__, handle, le16toh(reply->IOCStatus));
659 
660 	/*
661 	 * Don't clear target if remove fails because things will get confusing.
662 	 * Leave the devname and sasaddr intact so that we know to avoid reusing
663 	 * this target id if possible, and so we can assign the same target id
664 	 * to this device if it comes back in the future.
665 	 */
666 	if ((le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) ==
667 	    MPI2_IOCSTATUS_SUCCESS) {
668 		targ->handle = 0x0;
669 		targ->encl_handle = 0x0;
670 		targ->encl_level_valid = 0x0;
671 		targ->encl_level = 0x0;
672 		targ->connector_name[0] = ' ';
673 		targ->connector_name[1] = ' ';
674 		targ->connector_name[2] = ' ';
675 		targ->connector_name[3] = ' ';
676 		targ->encl_slot = 0x0;
677 		targ->exp_dev_handle = 0x0;
678 		targ->phy_num = 0x0;
679 		targ->linkrate = 0x0;
680 		targ->devinfo = 0x0;
681 		targ->flags = 0x0;
682 		targ->scsi_req_desc_type = 0;
683 
684 		while (!SLIST_EMPTY(&targ->luns)) {
685 			lun = SLIST_FIRST(&targ->luns);
686 			SLIST_REMOVE_HEAD(&targ->luns, lun_link);
687 			free(lun, M_MPR);
688 		}
689 	}
690 
691 	mprsas_free_tm(sc, tm);
692 }
693 
694 static int
695 mprsas_register_events(struct mpr_softc *sc)
696 {
697 	uint8_t events[16];
698 
699 	bzero(events, 16);
700 	setbit(events, MPI2_EVENT_SAS_DEVICE_STATUS_CHANGE);
701 	setbit(events, MPI2_EVENT_SAS_DISCOVERY);
702 	setbit(events, MPI2_EVENT_SAS_BROADCAST_PRIMITIVE);
703 	setbit(events, MPI2_EVENT_SAS_INIT_DEVICE_STATUS_CHANGE);
704 	setbit(events, MPI2_EVENT_SAS_INIT_TABLE_OVERFLOW);
705 	setbit(events, MPI2_EVENT_SAS_TOPOLOGY_CHANGE_LIST);
706 	setbit(events, MPI2_EVENT_SAS_ENCL_DEVICE_STATUS_CHANGE);
707 	setbit(events, MPI2_EVENT_IR_CONFIGURATION_CHANGE_LIST);
708 	setbit(events, MPI2_EVENT_IR_VOLUME);
709 	setbit(events, MPI2_EVENT_IR_PHYSICAL_DISK);
710 	setbit(events, MPI2_EVENT_IR_OPERATION_STATUS);
711 	setbit(events, MPI2_EVENT_TEMP_THRESHOLD);
712 	setbit(events, MPI2_EVENT_SAS_DEVICE_DISCOVERY_ERROR);
713 	if (sc->facts->MsgVersion >= MPI2_VERSION_02_06) {
714 		setbit(events, MPI2_EVENT_ACTIVE_CABLE_EXCEPTION);
715 		if (sc->mpr_flags & MPR_FLAGS_GEN35_IOC) {
716 			setbit(events, MPI2_EVENT_PCIE_DEVICE_STATUS_CHANGE);
717 			setbit(events, MPI2_EVENT_PCIE_ENUMERATION);
718 			setbit(events, MPI2_EVENT_PCIE_TOPOLOGY_CHANGE_LIST);
719 		}
720 	}
721 
722 	mpr_register_events(sc, events, mprsas_evt_handler, NULL,
723 	    &sc->sassc->mprsas_eh);
724 
725 	return (0);
726 }
727 
728 int
729 mpr_attach_sas(struct mpr_softc *sc)
730 {
731 	struct mprsas_softc *sassc;
732 	cam_status status;
733 	int unit, error = 0, reqs;
734 
735 	MPR_FUNCTRACE(sc);
736 	mpr_dprint(sc, MPR_INIT, "%s entered\n", __func__);
737 
738 	sassc = malloc(sizeof(struct mprsas_softc), M_MPR, M_WAITOK|M_ZERO);
739 	if (!sassc) {
740 		mpr_dprint(sc, MPR_INIT|MPR_ERROR,
741 		    "Cannot allocate SAS subsystem memory\n");
742 		return (ENOMEM);
743 	}
744 
745 	/*
746 	 * XXX MaxTargets could change during a reinit.  Since we don't
747 	 * resize the targets[] array during such an event, cache the value
748 	 * of MaxTargets here so that we don't get into trouble later.  This
749 	 * should move into the reinit logic.
750 	 */
751 	sassc->maxtargets = sc->facts->MaxTargets + sc->facts->MaxVolumes;
752 	sassc->targets = malloc(sizeof(struct mprsas_target) *
753 	    sassc->maxtargets, M_MPR, M_WAITOK|M_ZERO);
754 	if (!sassc->targets) {
755 		mpr_dprint(sc, MPR_INIT|MPR_ERROR,
756 		    "Cannot allocate SAS target memory\n");
757 		free(sassc, M_MPR);
758 		return (ENOMEM);
759 	}
760 	sc->sassc = sassc;
761 	sassc->sc = sc;
762 
763 	reqs = sc->num_reqs - sc->num_prireqs - 1;
764 	if ((sassc->devq = cam_simq_alloc(reqs)) == NULL) {
765 		mpr_dprint(sc, MPR_INIT|MPR_ERROR, "Cannot allocate SIMQ\n");
766 		error = ENOMEM;
767 		goto out;
768 	}
769 
770 	unit = device_get_unit(sc->mpr_dev);
771 	sassc->sim = cam_sim_alloc(mprsas_action, mprsas_poll, "mpr", sassc,
772 	    unit, &sc->mpr_mtx, reqs, reqs, sassc->devq);
773 	if (sassc->sim == NULL) {
774 		mpr_dprint(sc, MPR_INIT|MPR_ERROR, "Cannot allocate SIM\n");
775 		error = EINVAL;
776 		goto out;
777 	}
778 
779 	TAILQ_INIT(&sassc->ev_queue);
780 
781 	/* Initialize taskqueue for Event Handling */
782 	TASK_INIT(&sassc->ev_task, 0, mprsas_firmware_event_work, sc);
783 	sassc->ev_tq = taskqueue_create("mpr_taskq", M_NOWAIT | M_ZERO,
784 	    taskqueue_thread_enqueue, &sassc->ev_tq);
785 	taskqueue_start_threads(&sassc->ev_tq, 1, PRIBIO, "%s taskq",
786 	    device_get_nameunit(sc->mpr_dev));
787 
788 	mpr_lock(sc);
789 
790 	/*
791 	 * XXX There should be a bus for every port on the adapter, but since
792 	 * we're just going to fake the topology for now, we'll pretend that
793 	 * everything is just a target on a single bus.
794 	 */
795 	if ((error = xpt_bus_register(sassc->sim, sc->mpr_dev, 0)) != 0) {
796 		mpr_dprint(sc, MPR_INIT|MPR_ERROR,
797 		    "Error %d registering SCSI bus\n", error);
798 		mpr_unlock(sc);
799 		goto out;
800 	}
801 
802 	/*
803 	 * Assume that discovery events will start right away.
804 	 *
805 	 * Hold off boot until discovery is complete.
806 	 */
807 	sassc->flags |= MPRSAS_IN_STARTUP | MPRSAS_IN_DISCOVERY;
808 	sc->sassc->startup_refcount = 0;
809 	mprsas_startup_increment(sassc);
810 
811 	callout_init(&sassc->discovery_callout, 1 /*mpsafe*/);
812 
813 	/*
814 	 * Register for async events so we can determine the EEDP
815 	 * capabilities of devices.
816 	 */
817 	status = xpt_create_path(&sassc->path, /*periph*/NULL,
818 	    cam_sim_path(sc->sassc->sim), CAM_TARGET_WILDCARD,
819 	    CAM_LUN_WILDCARD);
820 	if (status != CAM_REQ_CMP) {
821 		mpr_dprint(sc, MPR_INIT|MPR_ERROR,
822 		    "Error %#x creating sim path\n", status);
823 		sassc->path = NULL;
824 	} else {
825 		int event;
826 
827 		event = AC_ADVINFO_CHANGED | AC_FOUND_DEVICE;
828 		status = xpt_register_async(event, mprsas_async, sc,
829 					    sassc->path);
830 
831 		if (status != CAM_REQ_CMP) {
832 			mpr_dprint(sc, MPR_ERROR,
833 			    "Error %#x registering async handler for "
834 			    "AC_ADVINFO_CHANGED events\n", status);
835 			xpt_free_path(sassc->path);
836 			sassc->path = NULL;
837 		}
838 	}
839 	if (status != CAM_REQ_CMP) {
840 		/*
841 		 * EEDP use is the exception, not the rule.
842 		 * Warn the user, but do not fail to attach.
843 		 */
844 		mpr_printf(sc, "EEDP capabilities disabled.\n");
845 	}
846 
847 	mpr_unlock(sc);
848 
849 	mprsas_register_events(sc);
850 out:
851 	if (error)
852 		mpr_detach_sas(sc);
853 
854 	mpr_dprint(sc, MPR_INIT, "%s exit, error= %d\n", __func__, error);
855 	return (error);
856 }
857 
858 int
859 mpr_detach_sas(struct mpr_softc *sc)
860 {
861 	struct mprsas_softc *sassc;
862 	struct mprsas_lun *lun, *lun_tmp;
863 	struct mprsas_target *targ;
864 	int i;
865 
866 	MPR_FUNCTRACE(sc);
867 
868 	if (sc->sassc == NULL)
869 		return (0);
870 
871 	sassc = sc->sassc;
872 	mpr_deregister_events(sc, sassc->mprsas_eh);
873 
874 	/*
875 	 * Drain and free the event handling taskqueue with the lock
876 	 * unheld so that any parallel processing tasks drain properly
877 	 * without deadlocking.
878 	 */
879 	if (sassc->ev_tq != NULL)
880 		taskqueue_free(sassc->ev_tq);
881 
882 	/* Make sure CAM doesn't wedge if we had to bail out early. */
883 	mpr_lock(sc);
884 
885 	while (sassc->startup_refcount != 0)
886 		mprsas_startup_decrement(sassc);
887 
888 	/* Deregister our async handler */
889 	if (sassc->path != NULL) {
890 		xpt_register_async(0, mprsas_async, sc, sassc->path);
891 		xpt_free_path(sassc->path);
892 		sassc->path = NULL;
893 	}
894 
895 	if (sassc->flags & MPRSAS_IN_STARTUP)
896 		xpt_release_simq(sassc->sim, 1);
897 
898 	if (sassc->sim != NULL) {
899 		xpt_bus_deregister(cam_sim_path(sassc->sim));
900 		cam_sim_free(sassc->sim, FALSE);
901 	}
902 
903 	mpr_unlock(sc);
904 
905 	if (sassc->devq != NULL)
906 		cam_simq_free(sassc->devq);
907 
908 	for (i = 0; i < sassc->maxtargets; i++) {
909 		targ = &sassc->targets[i];
910 		SLIST_FOREACH_SAFE(lun, &targ->luns, lun_link, lun_tmp) {
911 			free(lun, M_MPR);
912 		}
913 	}
914 	free(sassc->targets, M_MPR);
915 	free(sassc, M_MPR);
916 	sc->sassc = NULL;
917 
918 	return (0);
919 }
920 
921 void
922 mprsas_discovery_end(struct mprsas_softc *sassc)
923 {
924 	struct mpr_softc *sc = sassc->sc;
925 
926 	MPR_FUNCTRACE(sc);
927 
928 	if (sassc->flags & MPRSAS_DISCOVERY_TIMEOUT_PENDING)
929 		callout_stop(&sassc->discovery_callout);
930 
931 	/*
932 	 * After discovery has completed, check the mapping table for any
933 	 * missing devices and update their missing counts. Only do this once
934 	 * whenever the driver is initialized so that missing counts aren't
935 	 * updated unnecessarily. Note that just because discovery has
936 	 * completed doesn't mean that events have been processed yet. The
937 	 * check_devices function is a callout timer that checks if ALL devices
938 	 * are missing. If so, it will wait a little longer for events to
939 	 * complete and keep resetting itself until some device in the mapping
940 	 * table is not missing, meaning that event processing has started.
941 	 */
942 	if (sc->track_mapping_events) {
943 		mpr_dprint(sc, MPR_XINFO | MPR_MAPPING, "Discovery has "
944 		    "completed. Check for missing devices in the mapping "
945 		    "table.\n");
946 		callout_reset(&sc->device_check_callout,
947 		    MPR_MISSING_CHECK_DELAY * hz, mpr_mapping_check_devices,
948 		    sc);
949 	}
950 }
951 
952 static void
953 mprsas_action(struct cam_sim *sim, union ccb *ccb)
954 {
955 	struct mprsas_softc *sassc;
956 
957 	sassc = cam_sim_softc(sim);
958 
959 	MPR_FUNCTRACE(sassc->sc);
960 	mpr_dprint(sassc->sc, MPR_TRACE, "ccb func_code 0x%x\n",
961 	    ccb->ccb_h.func_code);
962 	mtx_assert(&sassc->sc->mpr_mtx, MA_OWNED);
963 
964 	switch (ccb->ccb_h.func_code) {
965 	case XPT_PATH_INQ:
966 	{
967 		struct ccb_pathinq *cpi = &ccb->cpi;
968 		struct mpr_softc *sc = sassc->sc;
969 
970 		cpi->version_num = 1;
971 		cpi->hba_inquiry = PI_SDTR_ABLE|PI_TAG_ABLE|PI_WIDE_16;
972 		cpi->target_sprt = 0;
973 		cpi->hba_misc = PIM_NOBUSRESET | PIM_UNMAPPED | PIM_NOSCAN;
974 		cpi->hba_eng_cnt = 0;
975 		cpi->max_target = sassc->maxtargets - 1;
976 		cpi->max_lun = 255;
977 
978 		/*
979 		 * initiator_id is set here to an ID outside the set of valid
980 		 * target IDs (including volumes).
981 		 */
982 		cpi->initiator_id = sassc->maxtargets;
983 		strlcpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN);
984 		strlcpy(cpi->hba_vid, "Avago Tech", HBA_IDLEN);
985 		strlcpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN);
986 		cpi->unit_number = cam_sim_unit(sim);
987 		cpi->bus_id = cam_sim_bus(sim);
988 		/*
989 		 * XXXSLM-I think this needs to change based on config page or
990 		 * something instead of hardcoded to 150000.
991 		 */
992 		cpi->base_transfer_speed = 150000;
993 		cpi->transport = XPORT_SAS;
994 		cpi->transport_version = 0;
995 		cpi->protocol = PROTO_SCSI;
996 		cpi->protocol_version = SCSI_REV_SPC;
997 		cpi->maxio = sc->maxio;
998 		mprsas_set_ccbstatus(ccb, CAM_REQ_CMP);
999 		break;
1000 	}
1001 	case XPT_GET_TRAN_SETTINGS:
1002 	{
1003 		struct ccb_trans_settings	*cts;
1004 		struct ccb_trans_settings_sas	*sas;
1005 		struct ccb_trans_settings_scsi	*scsi;
1006 		struct mprsas_target *targ;
1007 
1008 		cts = &ccb->cts;
1009 		sas = &cts->xport_specific.sas;
1010 		scsi = &cts->proto_specific.scsi;
1011 
1012 		KASSERT(cts->ccb_h.target_id < sassc->maxtargets,
1013 		    ("Target %d out of bounds in XPT_GET_TRAN_SETTINGS\n",
1014 		    cts->ccb_h.target_id));
1015 		targ = &sassc->targets[cts->ccb_h.target_id];
1016 		if (targ->handle == 0x0) {
1017 			mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE);
1018 			break;
1019 		}
1020 
1021 		cts->protocol_version = SCSI_REV_SPC2;
1022 		cts->transport = XPORT_SAS;
1023 		cts->transport_version = 0;
1024 
1025 		sas->valid = CTS_SAS_VALID_SPEED;
1026 		switch (targ->linkrate) {
1027 		case 0x08:
1028 			sas->bitrate = 150000;
1029 			break;
1030 		case 0x09:
1031 			sas->bitrate = 300000;
1032 			break;
1033 		case 0x0a:
1034 			sas->bitrate = 600000;
1035 			break;
1036 		case 0x0b:
1037 			sas->bitrate = 1200000;
1038 			break;
1039 		default:
1040 			sas->valid = 0;
1041 		}
1042 
1043 		cts->protocol = PROTO_SCSI;
1044 		scsi->valid = CTS_SCSI_VALID_TQ;
1045 		scsi->flags = CTS_SCSI_FLAGS_TAG_ENB;
1046 
1047 		mprsas_set_ccbstatus(ccb, CAM_REQ_CMP);
1048 		break;
1049 	}
1050 	case XPT_CALC_GEOMETRY:
1051 		cam_calc_geometry(&ccb->ccg, /*extended*/1);
1052 		mprsas_set_ccbstatus(ccb, CAM_REQ_CMP);
1053 		break;
1054 	case XPT_RESET_DEV:
1055 		mpr_dprint(sassc->sc, MPR_XINFO, "mprsas_action "
1056 		    "XPT_RESET_DEV\n");
1057 		mprsas_action_resetdev(sassc, ccb);
1058 		return;
1059 	case XPT_RESET_BUS:
1060 	case XPT_ABORT:
1061 	case XPT_TERM_IO:
1062 		mpr_dprint(sassc->sc, MPR_XINFO, "mprsas_action faking success "
1063 		    "for abort or reset\n");
1064 		mprsas_set_ccbstatus(ccb, CAM_REQ_CMP);
1065 		break;
1066 	case XPT_SCSI_IO:
1067 		mprsas_action_scsiio(sassc, ccb);
1068 		return;
1069 	case XPT_SMP_IO:
1070 		mprsas_action_smpio(sassc, ccb);
1071 		return;
1072 	default:
1073 		mprsas_set_ccbstatus(ccb, CAM_FUNC_NOTAVAIL);
1074 		break;
1075 	}
1076 	xpt_done(ccb);
1077 
1078 }
1079 
1080 static void
1081 mprsas_announce_reset(struct mpr_softc *sc, uint32_t ac_code,
1082     target_id_t target_id, lun_id_t lun_id)
1083 {
1084 	path_id_t path_id = cam_sim_path(sc->sassc->sim);
1085 	struct cam_path *path;
1086 
1087 	mpr_dprint(sc, MPR_XINFO, "%s code %x target %d lun %jx\n", __func__,
1088 	    ac_code, target_id, (uintmax_t)lun_id);
1089 
1090 	if (xpt_create_path(&path, NULL,
1091 		path_id, target_id, lun_id) != CAM_REQ_CMP) {
1092 		mpr_dprint(sc, MPR_ERROR, "unable to create path for reset "
1093 		    "notification\n");
1094 		return;
1095 	}
1096 
1097 	xpt_async(ac_code, path, NULL);
1098 	xpt_free_path(path);
1099 }
1100 
1101 static void
1102 mprsas_complete_all_commands(struct mpr_softc *sc)
1103 {
1104 	struct mpr_command *cm;
1105 	int i;
1106 	int completed;
1107 
1108 	MPR_FUNCTRACE(sc);
1109 	mtx_assert(&sc->mpr_mtx, MA_OWNED);
1110 
1111 	/* complete all commands with a NULL reply */
1112 	for (i = 1; i < sc->num_reqs; i++) {
1113 		cm = &sc->commands[i];
1114 		if (cm->cm_state == MPR_CM_STATE_FREE)
1115 			continue;
1116 
1117 		cm->cm_state = MPR_CM_STATE_BUSY;
1118 		cm->cm_reply = NULL;
1119 		completed = 0;
1120 
1121 		if (cm->cm_flags & MPR_CM_FLAGS_SATA_ID_TIMEOUT) {
1122 			MPASS(cm->cm_data);
1123 			free(cm->cm_data, M_MPR);
1124 			cm->cm_data = NULL;
1125 		}
1126 
1127 		if (cm->cm_flags & MPR_CM_FLAGS_POLLED)
1128 			cm->cm_flags |= MPR_CM_FLAGS_COMPLETE;
1129 
1130 		if (cm->cm_complete != NULL) {
1131 			mprsas_log_command(cm, MPR_RECOVERY,
1132 			    "completing cm %p state %x ccb %p for diag reset\n",
1133 			    cm, cm->cm_state, cm->cm_ccb);
1134 			cm->cm_complete(sc, cm);
1135 			completed = 1;
1136 		} else if (cm->cm_flags & MPR_CM_FLAGS_WAKEUP) {
1137 			mprsas_log_command(cm, MPR_RECOVERY,
1138 			    "waking up cm %p state %x ccb %p for diag reset\n",
1139 			    cm, cm->cm_state, cm->cm_ccb);
1140 			wakeup(cm);
1141 			completed = 1;
1142 		}
1143 
1144 		if ((completed == 0) && (cm->cm_state != MPR_CM_STATE_FREE)) {
1145 			/* this should never happen, but if it does, log */
1146 			mprsas_log_command(cm, MPR_RECOVERY,
1147 			    "cm %p state %x flags 0x%x ccb %p during diag "
1148 			    "reset\n", cm, cm->cm_state, cm->cm_flags,
1149 			    cm->cm_ccb);
1150 		}
1151 	}
1152 
1153 	sc->io_cmds_active = 0;
1154 }
1155 
1156 void
1157 mprsas_handle_reinit(struct mpr_softc *sc)
1158 {
1159 	int i;
1160 
1161 	/* Go back into startup mode and freeze the simq, so that CAM
1162 	 * doesn't send any commands until after we've rediscovered all
1163 	 * targets and found the proper device handles for them.
1164 	 *
1165 	 * After the reset, portenable will trigger discovery, and after all
1166 	 * discovery-related activities have finished, the simq will be
1167 	 * released.
1168 	 */
1169 	mpr_dprint(sc, MPR_INIT, "%s startup\n", __func__);
1170 	sc->sassc->flags |= MPRSAS_IN_STARTUP;
1171 	sc->sassc->flags |= MPRSAS_IN_DISCOVERY;
1172 	mprsas_startup_increment(sc->sassc);
1173 
1174 	/* notify CAM of a bus reset */
1175 	mprsas_announce_reset(sc, AC_BUS_RESET, CAM_TARGET_WILDCARD,
1176 	    CAM_LUN_WILDCARD);
1177 
1178 	/* complete and cleanup after all outstanding commands */
1179 	mprsas_complete_all_commands(sc);
1180 
1181 	mpr_dprint(sc, MPR_INIT, "%s startup %u after command completion\n",
1182 	    __func__, sc->sassc->startup_refcount);
1183 
1184 	/* zero all the target handles, since they may change after the
1185 	 * reset, and we have to rediscover all the targets and use the new
1186 	 * handles.
1187 	 */
1188 	for (i = 0; i < sc->sassc->maxtargets; i++) {
1189 		if (sc->sassc->targets[i].outstanding != 0)
1190 			mpr_dprint(sc, MPR_INIT, "target %u outstanding %u\n",
1191 			    i, sc->sassc->targets[i].outstanding);
1192 		sc->sassc->targets[i].handle = 0x0;
1193 		sc->sassc->targets[i].exp_dev_handle = 0x0;
1194 		sc->sassc->targets[i].outstanding = 0;
1195 		sc->sassc->targets[i].flags = MPRSAS_TARGET_INDIAGRESET;
1196 	}
1197 }
1198 static void
1199 mprsas_tm_timeout(void *data)
1200 {
1201 	struct mpr_command *tm = data;
1202 	struct mpr_softc *sc = tm->cm_sc;
1203 
1204 	mtx_assert(&sc->mpr_mtx, MA_OWNED);
1205 
1206 	mprsas_log_command(tm, MPR_INFO|MPR_RECOVERY, "task mgmt %p timed "
1207 	    "out\n", tm);
1208 
1209 	KASSERT(tm->cm_state == MPR_CM_STATE_INQUEUE,
1210 	    ("command not inqueue\n"));
1211 
1212 	tm->cm_state = MPR_CM_STATE_BUSY;
1213 	mpr_reinit(sc);
1214 }
1215 
1216 static void
1217 mprsas_logical_unit_reset_complete(struct mpr_softc *sc, struct mpr_command *tm)
1218 {
1219 	MPI2_SCSI_TASK_MANAGE_REPLY *reply;
1220 	MPI2_SCSI_TASK_MANAGE_REQUEST *req;
1221 	unsigned int cm_count = 0;
1222 	struct mpr_command *cm;
1223 	struct mprsas_target *targ;
1224 
1225 	callout_stop(&tm->cm_callout);
1226 
1227 	req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req;
1228 	reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply;
1229 	targ = tm->cm_targ;
1230 
1231 	/*
1232 	 * Currently there should be no way we can hit this case.  It only
1233 	 * happens when we have a failure to allocate chain frames, and
1234 	 * task management commands don't have S/G lists.
1235 	 */
1236 	if ((tm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) {
1237 		mpr_dprint(sc, MPR_RECOVERY|MPR_ERROR,
1238 		    "%s: cm_flags = %#x for LUN reset! "
1239 		    "This should not happen!\n", __func__, tm->cm_flags);
1240 		mprsas_free_tm(sc, tm);
1241 		return;
1242 	}
1243 
1244 	if (reply == NULL) {
1245 		mpr_dprint(sc, MPR_RECOVERY, "NULL reset reply for tm %p\n",
1246 		    tm);
1247 		if ((sc->mpr_flags & MPR_FLAGS_DIAGRESET) != 0) {
1248 			/* this completion was due to a reset, just cleanup */
1249 			mpr_dprint(sc, MPR_RECOVERY, "Hardware undergoing "
1250 			    "reset, ignoring NULL LUN reset reply\n");
1251 			targ->tm = NULL;
1252 			mprsas_free_tm(sc, tm);
1253 		}
1254 		else {
1255 			/* we should have gotten a reply. */
1256 			mpr_dprint(sc, MPR_INFO|MPR_RECOVERY, "NULL reply on "
1257 			    "LUN reset attempt, resetting controller\n");
1258 			mpr_reinit(sc);
1259 		}
1260 		return;
1261 	}
1262 
1263 	mpr_dprint(sc, MPR_RECOVERY,
1264 	    "logical unit reset status 0x%x code 0x%x count %u\n",
1265 	    le16toh(reply->IOCStatus), le32toh(reply->ResponseCode),
1266 	    le32toh(reply->TerminationCount));
1267 
1268 	/*
1269 	 * See if there are any outstanding commands for this LUN.
1270 	 * This could be made more efficient by using a per-LU data
1271 	 * structure of some sort.
1272 	 */
1273 	TAILQ_FOREACH(cm, &targ->commands, cm_link) {
1274 		if (cm->cm_lun == tm->cm_lun)
1275 			cm_count++;
1276 	}
1277 
1278 	if (cm_count == 0) {
1279 		mpr_dprint(sc, MPR_RECOVERY|MPR_INFO,
1280 		    "Finished recovery after LUN reset for target %u\n",
1281 		    targ->tid);
1282 
1283 		mprsas_announce_reset(sc, AC_SENT_BDR, targ->tid,
1284 		    tm->cm_lun);
1285 
1286 		/*
1287 		 * We've finished recovery for this logical unit.  check and
1288 		 * see if some other logical unit has a timedout command
1289 		 * that needs to be processed.
1290 		 */
1291 		cm = TAILQ_FIRST(&targ->timedout_commands);
1292 		if (cm) {
1293 			mpr_dprint(sc, MPR_INFO|MPR_RECOVERY,
1294 			   "More commands to abort for target %u\n", targ->tid);
1295 			mprsas_send_abort(sc, tm, cm);
1296 		} else {
1297 			targ->tm = NULL;
1298 			mprsas_free_tm(sc, tm);
1299 		}
1300 	} else {
1301 		/* if we still have commands for this LUN, the reset
1302 		 * effectively failed, regardless of the status reported.
1303 		 * Escalate to a target reset.
1304 		 */
1305 		mpr_dprint(sc, MPR_INFO|MPR_RECOVERY,
1306 		    "logical unit reset complete for target %u, but still "
1307 		    "have %u command(s), sending target reset\n", targ->tid,
1308 		    cm_count);
1309 		if (!targ->is_nvme || sc->custom_nvme_tm_handling)
1310 			mprsas_send_reset(sc, tm,
1311 			    MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET);
1312 		else
1313 			mpr_reinit(sc);
1314 	}
1315 }
1316 
1317 static void
1318 mprsas_target_reset_complete(struct mpr_softc *sc, struct mpr_command *tm)
1319 {
1320 	MPI2_SCSI_TASK_MANAGE_REPLY *reply;
1321 	MPI2_SCSI_TASK_MANAGE_REQUEST *req;
1322 	struct mprsas_target *targ;
1323 
1324 	callout_stop(&tm->cm_callout);
1325 
1326 	req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req;
1327 	reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply;
1328 	targ = tm->cm_targ;
1329 
1330 	/*
1331 	 * Currently there should be no way we can hit this case.  It only
1332 	 * happens when we have a failure to allocate chain frames, and
1333 	 * task management commands don't have S/G lists.
1334 	 */
1335 	if ((tm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) {
1336 		mpr_dprint(sc, MPR_ERROR, "%s: cm_flags = %#x for target "
1337 		    "reset! This should not happen!\n", __func__, tm->cm_flags);
1338 		mprsas_free_tm(sc, tm);
1339 		return;
1340 	}
1341 
1342 	if (reply == NULL) {
1343 		mpr_dprint(sc, MPR_RECOVERY,
1344 		    "NULL target reset reply for tm %p TaskMID %u\n",
1345 		    tm, le16toh(req->TaskMID));
1346 		if ((sc->mpr_flags & MPR_FLAGS_DIAGRESET) != 0) {
1347 			/* this completion was due to a reset, just cleanup */
1348 			mpr_dprint(sc, MPR_RECOVERY, "Hardware undergoing "
1349 			    "reset, ignoring NULL target reset reply\n");
1350 			targ->tm = NULL;
1351 			mprsas_free_tm(sc, tm);
1352 		}
1353 		else {
1354 			/* we should have gotten a reply. */
1355 			mpr_dprint(sc, MPR_INFO|MPR_RECOVERY, "NULL reply on "
1356 			    "target reset attempt, resetting controller\n");
1357 			mpr_reinit(sc);
1358 		}
1359 		return;
1360 	}
1361 
1362 	mpr_dprint(sc, MPR_RECOVERY,
1363 	    "target reset status 0x%x code 0x%x count %u\n",
1364 	    le16toh(reply->IOCStatus), le32toh(reply->ResponseCode),
1365 	    le32toh(reply->TerminationCount));
1366 
1367 	if (targ->outstanding == 0) {
1368 		/*
1369 		 * We've finished recovery for this target and all
1370 		 * of its logical units.
1371 		 */
1372 		mpr_dprint(sc, MPR_RECOVERY|MPR_INFO,
1373 		    "Finished reset recovery for target %u\n", targ->tid);
1374 
1375 		mprsas_announce_reset(sc, AC_SENT_BDR, tm->cm_targ->tid,
1376 		    CAM_LUN_WILDCARD);
1377 
1378 		targ->tm = NULL;
1379 		mprsas_free_tm(sc, tm);
1380 	} else {
1381 		/*
1382 		 * After a target reset, if this target still has
1383 		 * outstanding commands, the reset effectively failed,
1384 		 * regardless of the status reported.  escalate.
1385 		 */
1386 		mpr_dprint(sc, MPR_INFO|MPR_RECOVERY,
1387 		    "Target reset complete for target %u, but still have %u "
1388 		    "command(s), resetting controller\n", targ->tid,
1389 		    targ->outstanding);
1390 		mpr_reinit(sc);
1391 	}
1392 }
1393 
1394 #define MPR_RESET_TIMEOUT 30
1395 
1396 int
1397 mprsas_send_reset(struct mpr_softc *sc, struct mpr_command *tm, uint8_t type)
1398 {
1399 	MPI2_SCSI_TASK_MANAGE_REQUEST *req;
1400 	struct mprsas_target *target;
1401 	int err, timeout;
1402 
1403 	target = tm->cm_targ;
1404 	if (target->handle == 0) {
1405 		mpr_dprint(sc, MPR_ERROR, "%s null devhandle for target_id "
1406 		    "%d\n", __func__, target->tid);
1407 		return -1;
1408 	}
1409 
1410 	req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req;
1411 	req->DevHandle = htole16(target->handle);
1412 	req->TaskType = type;
1413 
1414 	if (!target->is_nvme || sc->custom_nvme_tm_handling) {
1415 		timeout = MPR_RESET_TIMEOUT;
1416 		/*
1417 		 * Target reset method =
1418 		 *     SAS Hard Link Reset / SATA Link Reset
1419 		 */
1420 		req->MsgFlags = MPI2_SCSITASKMGMT_MSGFLAGS_LINK_RESET;
1421 	} else {
1422 		timeout = (target->controller_reset_timeout) ? (
1423 		    target->controller_reset_timeout) : (MPR_RESET_TIMEOUT);
1424 		/* PCIe Protocol Level Reset*/
1425 		req->MsgFlags =
1426 		    MPI26_SCSITASKMGMT_MSGFLAGS_PROTOCOL_LVL_RST_PCIE;
1427 	}
1428 
1429 	if (type == MPI2_SCSITASKMGMT_TASKTYPE_LOGICAL_UNIT_RESET) {
1430 		/* XXX Need to handle invalid LUNs */
1431 		MPR_SET_LUN(req->LUN, tm->cm_lun);
1432 		tm->cm_targ->logical_unit_resets++;
1433 		mpr_dprint(sc, MPR_RECOVERY|MPR_INFO,
1434 		    "Sending logical unit reset to target %u lun %d\n",
1435 		    target->tid, tm->cm_lun);
1436 		tm->cm_complete = mprsas_logical_unit_reset_complete;
1437 		mprsas_prepare_for_tm(sc, tm, target, tm->cm_lun);
1438 	} else if (type == MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET) {
1439 		tm->cm_targ->target_resets++;
1440 		mpr_dprint(sc, MPR_RECOVERY|MPR_INFO,
1441 		    "Sending target reset to target %u\n", target->tid);
1442 		tm->cm_complete = mprsas_target_reset_complete;
1443 		mprsas_prepare_for_tm(sc, tm, target, CAM_LUN_WILDCARD);
1444 	}
1445 	else {
1446 		mpr_dprint(sc, MPR_ERROR, "unexpected reset type 0x%x\n", type);
1447 		return -1;
1448 	}
1449 
1450 	if (target->encl_level_valid) {
1451 		mpr_dprint(sc, MPR_RECOVERY|MPR_INFO,
1452 		    "At enclosure level %d, slot %d, connector name (%4s)\n",
1453 		    target->encl_level, target->encl_slot,
1454 		    target->connector_name);
1455 	}
1456 
1457 	tm->cm_data = NULL;
1458 	tm->cm_complete_data = (void *)tm;
1459 
1460 	callout_reset(&tm->cm_callout, timeout * hz,
1461 	    mprsas_tm_timeout, tm);
1462 
1463 	err = mpr_map_command(sc, tm);
1464 	if (err)
1465 		mpr_dprint(sc, MPR_ERROR|MPR_RECOVERY,
1466 		    "error %d sending reset type %u\n", err, type);
1467 
1468 	return err;
1469 }
1470 
1471 
1472 static void
1473 mprsas_abort_complete(struct mpr_softc *sc, struct mpr_command *tm)
1474 {
1475 	struct mpr_command *cm;
1476 	MPI2_SCSI_TASK_MANAGE_REPLY *reply;
1477 	MPI2_SCSI_TASK_MANAGE_REQUEST *req;
1478 	struct mprsas_target *targ;
1479 
1480 	callout_stop(&tm->cm_callout);
1481 
1482 	req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req;
1483 	reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply;
1484 	targ = tm->cm_targ;
1485 
1486 	/*
1487 	 * Currently there should be no way we can hit this case.  It only
1488 	 * happens when we have a failure to allocate chain frames, and
1489 	 * task management commands don't have S/G lists.
1490 	 */
1491 	if ((tm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) {
1492 		mpr_dprint(sc, MPR_RECOVERY|MPR_ERROR,
1493 		    "cm_flags = %#x for abort %p TaskMID %u!\n",
1494 		    tm->cm_flags, tm, le16toh(req->TaskMID));
1495 		mprsas_free_tm(sc, tm);
1496 		return;
1497 	}
1498 
1499 	if (reply == NULL) {
1500 		mpr_dprint(sc, MPR_RECOVERY,
1501 		    "NULL abort reply for tm %p TaskMID %u\n",
1502 		    tm, le16toh(req->TaskMID));
1503 		if ((sc->mpr_flags & MPR_FLAGS_DIAGRESET) != 0) {
1504 			/* this completion was due to a reset, just cleanup */
1505 			mpr_dprint(sc, MPR_RECOVERY, "Hardware undergoing "
1506 			    "reset, ignoring NULL abort reply\n");
1507 			targ->tm = NULL;
1508 			mprsas_free_tm(sc, tm);
1509 		} else {
1510 			/* we should have gotten a reply. */
1511 			mpr_dprint(sc, MPR_INFO|MPR_RECOVERY, "NULL reply on "
1512 			    "abort attempt, resetting controller\n");
1513 			mpr_reinit(sc);
1514 		}
1515 		return;
1516 	}
1517 
1518 	mpr_dprint(sc, MPR_RECOVERY,
1519 	    "abort TaskMID %u status 0x%x code 0x%x count %u\n",
1520 	    le16toh(req->TaskMID),
1521 	    le16toh(reply->IOCStatus), le32toh(reply->ResponseCode),
1522 	    le32toh(reply->TerminationCount));
1523 
1524 	cm = TAILQ_FIRST(&tm->cm_targ->timedout_commands);
1525 	if (cm == NULL) {
1526 		/*
1527 		 * if there are no more timedout commands, we're done with
1528 		 * error recovery for this target.
1529 		 */
1530 		mpr_dprint(sc, MPR_INFO|MPR_RECOVERY,
1531 		    "Finished abort recovery for target %u\n", targ->tid);
1532 		targ->tm = NULL;
1533 		mprsas_free_tm(sc, tm);
1534 	} else if (le16toh(req->TaskMID) != cm->cm_desc.Default.SMID) {
1535 		/* abort success, but we have more timedout commands to abort */
1536 		mpr_dprint(sc, MPR_INFO|MPR_RECOVERY,
1537 		    "Continuing abort recovery for target %u\n", targ->tid);
1538 		mprsas_send_abort(sc, tm, cm);
1539 	} else {
1540 		/*
1541 		 * we didn't get a command completion, so the abort
1542 		 * failed as far as we're concerned.  escalate.
1543 		 */
1544 		mpr_dprint(sc, MPR_INFO|MPR_RECOVERY,
1545 		    "Abort failed for target %u, sending logical unit reset\n",
1546 		    targ->tid);
1547 
1548 		mprsas_send_reset(sc, tm,
1549 		    MPI2_SCSITASKMGMT_TASKTYPE_LOGICAL_UNIT_RESET);
1550 	}
1551 }
1552 
1553 #define MPR_ABORT_TIMEOUT 5
1554 
1555 static int
1556 mprsas_send_abort(struct mpr_softc *sc, struct mpr_command *tm,
1557     struct mpr_command *cm)
1558 {
1559 	MPI2_SCSI_TASK_MANAGE_REQUEST *req;
1560 	struct mprsas_target *targ;
1561 	int err, timeout;
1562 
1563 	targ = cm->cm_targ;
1564 	if (targ->handle == 0) {
1565 		mpr_dprint(sc, MPR_ERROR|MPR_RECOVERY,
1566 		   "%s null devhandle for target_id %d\n",
1567 		    __func__, cm->cm_ccb->ccb_h.target_id);
1568 		return -1;
1569 	}
1570 
1571 	mprsas_log_command(cm, MPR_RECOVERY|MPR_INFO,
1572 	    "Aborting command %p\n", cm);
1573 
1574 	req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req;
1575 	req->DevHandle = htole16(targ->handle);
1576 	req->TaskType = MPI2_SCSITASKMGMT_TASKTYPE_ABORT_TASK;
1577 
1578 	/* XXX Need to handle invalid LUNs */
1579 	MPR_SET_LUN(req->LUN, cm->cm_ccb->ccb_h.target_lun);
1580 
1581 	req->TaskMID = htole16(cm->cm_desc.Default.SMID);
1582 
1583 	tm->cm_data = NULL;
1584 	tm->cm_complete = mprsas_abort_complete;
1585 	tm->cm_complete_data = (void *)tm;
1586 	tm->cm_targ = cm->cm_targ;
1587 	tm->cm_lun = cm->cm_lun;
1588 
1589 	if (!targ->is_nvme || sc->custom_nvme_tm_handling)
1590 		timeout	= MPR_ABORT_TIMEOUT;
1591 	else
1592 		timeout = sc->nvme_abort_timeout;
1593 
1594 	callout_reset(&tm->cm_callout, timeout * hz,
1595 	    mprsas_tm_timeout, tm);
1596 
1597 	targ->aborts++;
1598 
1599 	mprsas_prepare_for_tm(sc, tm, targ, tm->cm_lun);
1600 
1601 	err = mpr_map_command(sc, tm);
1602 	if (err)
1603 		mpr_dprint(sc, MPR_ERROR|MPR_RECOVERY,
1604 		    "error %d sending abort for cm %p SMID %u\n",
1605 		    err, cm, req->TaskMID);
1606 	return err;
1607 }
1608 
1609 static void
1610 mprsas_scsiio_timeout(void *data)
1611 {
1612 	sbintime_t elapsed, now;
1613 	union ccb *ccb;
1614 	struct mpr_softc *sc;
1615 	struct mpr_command *cm;
1616 	struct mprsas_target *targ;
1617 
1618 	cm = (struct mpr_command *)data;
1619 	sc = cm->cm_sc;
1620 	ccb = cm->cm_ccb;
1621 	now = sbinuptime();
1622 
1623 	MPR_FUNCTRACE(sc);
1624 	mtx_assert(&sc->mpr_mtx, MA_OWNED);
1625 
1626 	mpr_dprint(sc, MPR_XINFO|MPR_RECOVERY, "Timeout checking cm %p\n", cm);
1627 
1628 	/*
1629 	 * Run the interrupt handler to make sure it's not pending.  This
1630 	 * isn't perfect because the command could have already completed
1631 	 * and been re-used, though this is unlikely.
1632 	 */
1633 	mpr_intr_locked(sc);
1634 	if (cm->cm_flags & MPR_CM_FLAGS_ON_RECOVERY) {
1635 		mprsas_log_command(cm, MPR_XINFO,
1636 		    "SCSI command %p almost timed out\n", cm);
1637 		return;
1638 	}
1639 
1640 	if (cm->cm_ccb == NULL) {
1641 		mpr_dprint(sc, MPR_ERROR, "command timeout with NULL ccb\n");
1642 		return;
1643 	}
1644 
1645 	targ = cm->cm_targ;
1646 	targ->timeouts++;
1647 
1648 	elapsed = now - ccb->ccb_h.qos.sim_data;
1649 	mprsas_log_command(cm, MPR_INFO|MPR_RECOVERY,
1650 	    "Command timeout on target %u(0x%04x), %d set, %d.%d elapsed\n",
1651 	    targ->tid, targ->handle, ccb->ccb_h.timeout,
1652 	    sbintime_getsec(elapsed), elapsed & 0xffffffff);
1653 	if (targ->encl_level_valid) {
1654 		mpr_dprint(sc, MPR_INFO|MPR_RECOVERY,
1655 		    "At enclosure level %d, slot %d, connector name (%4s)\n",
1656 		    targ->encl_level, targ->encl_slot, targ->connector_name);
1657 	}
1658 
1659 	/* XXX first, check the firmware state, to see if it's still
1660 	 * operational.  if not, do a diag reset.
1661 	 */
1662 	mprsas_set_ccbstatus(cm->cm_ccb, CAM_CMD_TIMEOUT);
1663 	cm->cm_flags |= MPR_CM_FLAGS_ON_RECOVERY | MPR_CM_FLAGS_TIMEDOUT;
1664 	TAILQ_INSERT_TAIL(&targ->timedout_commands, cm, cm_recovery);
1665 
1666 	if (targ->tm != NULL) {
1667 		/* target already in recovery, just queue up another
1668 		 * timedout command to be processed later.
1669 		 */
1670 		mpr_dprint(sc, MPR_RECOVERY, "queued timedout cm %p for "
1671 		    "processing by tm %p\n", cm, targ->tm);
1672 	}
1673 	else if ((targ->tm = mprsas_alloc_tm(sc)) != NULL) {
1674 
1675 		/* start recovery by aborting the first timedout command */
1676 		mpr_dprint(sc, MPR_RECOVERY|MPR_INFO,
1677 		    "Sending abort to target %u for SMID %d\n", targ->tid,
1678 		    cm->cm_desc.Default.SMID);
1679 		mpr_dprint(sc, MPR_RECOVERY, "timedout cm %p allocated tm %p\n",
1680 		    cm, targ->tm);
1681 		mprsas_send_abort(sc, targ->tm, cm);
1682 	}
1683 	else {
1684 		/* XXX queue this target up for recovery once a TM becomes
1685 		 * available.  The firmware only has a limited number of
1686 		 * HighPriority credits for the high priority requests used
1687 		 * for task management, and we ran out.
1688 		 *
1689 		 * Isilon: don't worry about this for now, since we have
1690 		 * more credits than disks in an enclosure, and limit
1691 		 * ourselves to one TM per target for recovery.
1692 		 */
1693 		mpr_dprint(sc, MPR_ERROR|MPR_RECOVERY,
1694 		    "timedout cm %p failed to allocate a tm\n", cm);
1695 	}
1696 }
1697 
1698 /**
1699  * mprsas_build_nvme_unmap - Build Native NVMe DSM command equivalent
1700  *			     to SCSI Unmap.
1701  * Return 0 - for success,
1702  *	  1 - to immediately return back the command with success status to CAM
1703  *	  negative value - to fallback to firmware path i.e. issue scsi unmap
1704  *			   to FW without any translation.
1705  */
1706 static int
1707 mprsas_build_nvme_unmap(struct mpr_softc *sc, struct mpr_command *cm,
1708     union ccb *ccb, struct mprsas_target *targ)
1709 {
1710 	Mpi26NVMeEncapsulatedRequest_t *req = NULL;
1711 	struct ccb_scsiio *csio;
1712 	struct unmap_parm_list *plist;
1713 	struct nvme_dsm_range *nvme_dsm_ranges = NULL;
1714 	struct nvme_command *c;
1715 	int i, res;
1716 	uint16_t ndesc, list_len, data_length;
1717 	struct mpr_prp_page *prp_page_info;
1718 	uint64_t nvme_dsm_ranges_dma_handle;
1719 
1720 	csio = &ccb->csio;
1721 	list_len = (scsiio_cdb_ptr(csio)[7] << 8 | scsiio_cdb_ptr(csio)[8]);
1722 	if (!list_len) {
1723 		mpr_dprint(sc, MPR_ERROR, "Parameter list length is Zero\n");
1724 		return -EINVAL;
1725 	}
1726 
1727 	plist = malloc(csio->dxfer_len, M_MPR, M_ZERO|M_NOWAIT);
1728 	if (!plist) {
1729 		mpr_dprint(sc, MPR_ERROR, "Unable to allocate memory to "
1730 		    "save UNMAP data\n");
1731 		return -ENOMEM;
1732 	}
1733 
1734 	/* Copy SCSI unmap data to a local buffer */
1735 	bcopy(csio->data_ptr, plist, csio->dxfer_len);
1736 
1737 	/* return back the unmap command to CAM with success status,
1738 	 * if number of descripts is zero.
1739 	 */
1740 	ndesc = be16toh(plist->unmap_blk_desc_data_len) >> 4;
1741 	if (!ndesc) {
1742 		mpr_dprint(sc, MPR_XINFO, "Number of descriptors in "
1743 		    "UNMAP cmd is Zero\n");
1744 		res = 1;
1745 		goto out;
1746 	}
1747 
1748 	data_length = ndesc * sizeof(struct nvme_dsm_range);
1749 	if (data_length > targ->MDTS) {
1750 		mpr_dprint(sc, MPR_ERROR, "data length: %d is greater than "
1751 		    "Device's MDTS: %d\n", data_length, targ->MDTS);
1752 		res = -EINVAL;
1753 		goto out;
1754 	}
1755 
1756 	prp_page_info = mpr_alloc_prp_page(sc);
1757 	KASSERT(prp_page_info != NULL, ("%s: There is no PRP Page for "
1758 	    "UNMAP command.\n", __func__));
1759 
1760 	/*
1761 	 * Insert the allocated PRP page into the command's PRP page list. This
1762 	 * will be freed when the command is freed.
1763 	 */
1764 	TAILQ_INSERT_TAIL(&cm->cm_prp_page_list, prp_page_info, prp_page_link);
1765 
1766 	nvme_dsm_ranges = (struct nvme_dsm_range *)prp_page_info->prp_page;
1767 	nvme_dsm_ranges_dma_handle = prp_page_info->prp_page_busaddr;
1768 
1769 	bzero(nvme_dsm_ranges, data_length);
1770 
1771 	/* Convert SCSI unmap's descriptor data to NVMe DSM specific Range data
1772 	 * for each descriptors contained in SCSI UNMAP data.
1773 	 */
1774 	for (i = 0; i < ndesc; i++) {
1775 		nvme_dsm_ranges[i].length =
1776 		    htole32(be32toh(plist->desc[i].nlb));
1777 		nvme_dsm_ranges[i].starting_lba =
1778 		    htole64(be64toh(plist->desc[i].slba));
1779 		nvme_dsm_ranges[i].attributes = 0;
1780 	}
1781 
1782 	/* Build MPI2.6's NVMe Encapsulated Request Message */
1783 	req = (Mpi26NVMeEncapsulatedRequest_t *)cm->cm_req;
1784 	bzero(req, sizeof(*req));
1785 	req->DevHandle = htole16(targ->handle);
1786 	req->Function = MPI2_FUNCTION_NVME_ENCAPSULATED;
1787 	req->Flags = MPI26_NVME_FLAGS_WRITE;
1788 	req->ErrorResponseBaseAddress.High =
1789 	    htole32((uint32_t)((uint64_t)cm->cm_sense_busaddr >> 32));
1790 	req->ErrorResponseBaseAddress.Low =
1791 	    htole32(cm->cm_sense_busaddr);
1792 	req->ErrorResponseAllocationLength =
1793 	    htole16(sizeof(struct nvme_completion));
1794 	req->EncapsulatedCommandLength =
1795 	    htole16(sizeof(struct nvme_command));
1796 	req->DataLength = htole32(data_length);
1797 
1798 	/* Build NVMe DSM command */
1799 	c = (struct nvme_command *) req->NVMe_Command;
1800 	c->opc = NVME_OPC_DATASET_MANAGEMENT;
1801 	c->nsid = htole32(csio->ccb_h.target_lun + 1);
1802 	c->cdw10 = htole32(ndesc - 1);
1803 	c->cdw11 = htole32(NVME_DSM_ATTR_DEALLOCATE);
1804 
1805 	cm->cm_length = data_length;
1806 	cm->cm_data = NULL;
1807 
1808 	cm->cm_complete = mprsas_scsiio_complete;
1809 	cm->cm_complete_data = ccb;
1810 	cm->cm_targ = targ;
1811 	cm->cm_lun = csio->ccb_h.target_lun;
1812 	cm->cm_ccb = ccb;
1813 
1814 	cm->cm_desc.Default.RequestFlags =
1815 	    MPI26_REQ_DESCRIPT_FLAGS_PCIE_ENCAPSULATED;
1816 
1817 	csio->ccb_h.qos.sim_data = sbinuptime();
1818 	callout_reset_sbt(&cm->cm_callout, SBT_1MS * ccb->ccb_h.timeout, 0,
1819 	    mprsas_scsiio_timeout, cm, 0);
1820 
1821 	targ->issued++;
1822 	targ->outstanding++;
1823 	TAILQ_INSERT_TAIL(&targ->commands, cm, cm_link);
1824 	ccb->ccb_h.status |= CAM_SIM_QUEUED;
1825 
1826 	mprsas_log_command(cm, MPR_XINFO, "%s cm %p ccb %p outstanding %u\n",
1827 	    __func__, cm, ccb, targ->outstanding);
1828 
1829 	mpr_build_nvme_prp(sc, cm, req,
1830 	    (void *)(uintptr_t)nvme_dsm_ranges_dma_handle, 0, data_length);
1831 	mpr_map_command(sc, cm);
1832 
1833 out:
1834 	free(plist, M_MPR);
1835 	return 0;
1836 }
1837 
1838 static void
1839 mprsas_action_scsiio(struct mprsas_softc *sassc, union ccb *ccb)
1840 {
1841 	MPI2_SCSI_IO_REQUEST *req;
1842 	struct ccb_scsiio *csio;
1843 	struct mpr_softc *sc;
1844 	struct mprsas_target *targ;
1845 	struct mprsas_lun *lun;
1846 	struct mpr_command *cm;
1847 	uint8_t i, lba_byte, *ref_tag_addr, scsi_opcode;
1848 	uint16_t eedp_flags;
1849 	uint32_t mpi_control;
1850 	int rc;
1851 
1852 	sc = sassc->sc;
1853 	MPR_FUNCTRACE(sc);
1854 	mtx_assert(&sc->mpr_mtx, MA_OWNED);
1855 
1856 	csio = &ccb->csio;
1857 	KASSERT(csio->ccb_h.target_id < sassc->maxtargets,
1858 	    ("Target %d out of bounds in XPT_SCSI_IO\n",
1859 	     csio->ccb_h.target_id));
1860 	targ = &sassc->targets[csio->ccb_h.target_id];
1861 	mpr_dprint(sc, MPR_TRACE, "ccb %p target flag %x\n", ccb, targ->flags);
1862 	if (targ->handle == 0x0) {
1863 		mpr_dprint(sc, MPR_ERROR, "%s NULL handle for target %u\n",
1864 		    __func__, csio->ccb_h.target_id);
1865 		mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE);
1866 		xpt_done(ccb);
1867 		return;
1868 	}
1869 	if (targ->flags & MPR_TARGET_FLAGS_RAID_COMPONENT) {
1870 		mpr_dprint(sc, MPR_ERROR, "%s Raid component no SCSI IO "
1871 		    "supported %u\n", __func__, csio->ccb_h.target_id);
1872 		mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE);
1873 		xpt_done(ccb);
1874 		return;
1875 	}
1876 	/*
1877 	 * Sometimes, it is possible to get a command that is not "In
1878 	 * Progress" and was actually aborted by the upper layer.  Check for
1879 	 * this here and complete the command without error.
1880 	 */
1881 	if (mprsas_get_ccbstatus(ccb) != CAM_REQ_INPROG) {
1882 		mpr_dprint(sc, MPR_TRACE, "%s Command is not in progress for "
1883 		    "target %u\n", __func__, csio->ccb_h.target_id);
1884 		xpt_done(ccb);
1885 		return;
1886 	}
1887 	/*
1888 	 * If devinfo is 0 this will be a volume.  In that case don't tell CAM
1889 	 * that the volume has timed out.  We want volumes to be enumerated
1890 	 * until they are deleted/removed, not just failed. In either event,
1891 	 * we're removing the target due to a firmware event telling us
1892 	 * the device is now gone (as opposed to some transient event). Since
1893 	 * we're opting to remove failed devices from the OS's view, we need
1894 	 * to propagate that status up the stack.
1895 	 */
1896 	if (targ->flags & MPRSAS_TARGET_INREMOVAL) {
1897 		if (targ->devinfo == 0)
1898 			mprsas_set_ccbstatus(ccb, CAM_REQ_CMP);
1899 		else
1900 			mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE);
1901 		xpt_done(ccb);
1902 		return;
1903 	}
1904 
1905 	if ((sc->mpr_flags & MPR_FLAGS_SHUTDOWN) != 0) {
1906 		mpr_dprint(sc, MPR_INFO, "%s shutting down\n", __func__);
1907 		mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE);
1908 		xpt_done(ccb);
1909 		return;
1910 	}
1911 
1912 	/*
1913 	 * If target has a reset in progress, freeze the devq and return.  The
1914 	 * devq will be released when the TM reset is finished.
1915 	 */
1916 	if (targ->flags & MPRSAS_TARGET_INRESET) {
1917 		ccb->ccb_h.status = CAM_BUSY | CAM_DEV_QFRZN;
1918 		mpr_dprint(sc, MPR_INFO, "%s: Freezing devq for target ID %d\n",
1919 		    __func__, targ->tid);
1920 		xpt_freeze_devq(ccb->ccb_h.path, 1);
1921 		xpt_done(ccb);
1922 		return;
1923 	}
1924 
1925 	cm = mpr_alloc_command(sc);
1926 	if (cm == NULL || (sc->mpr_flags & MPR_FLAGS_DIAGRESET)) {
1927 		if (cm != NULL) {
1928 			mpr_free_command(sc, cm);
1929 		}
1930 		if ((sassc->flags & MPRSAS_QUEUE_FROZEN) == 0) {
1931 			xpt_freeze_simq(sassc->sim, 1);
1932 			sassc->flags |= MPRSAS_QUEUE_FROZEN;
1933 		}
1934 		ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
1935 		ccb->ccb_h.status |= CAM_REQUEUE_REQ;
1936 		xpt_done(ccb);
1937 		return;
1938 	}
1939 
1940 	/* For NVME device's issue UNMAP command directly to NVME drives by
1941 	 * constructing equivalent native NVMe DataSetManagement command.
1942 	 */
1943 	scsi_opcode = scsiio_cdb_ptr(csio)[0];
1944 	if (scsi_opcode == UNMAP &&
1945 	    targ->is_nvme &&
1946 	    (csio->ccb_h.flags & CAM_DATA_MASK) == CAM_DATA_VADDR) {
1947 		rc = mprsas_build_nvme_unmap(sc, cm, ccb, targ);
1948 		if (rc == 1) { /* return command to CAM with success status */
1949 			mpr_free_command(sc, cm);
1950 			mprsas_set_ccbstatus(ccb, CAM_REQ_CMP);
1951 			xpt_done(ccb);
1952 			return;
1953 		} else if (!rc) /* Issued NVMe Encapsulated Request Message */
1954 			return;
1955 	}
1956 
1957 	req = (MPI2_SCSI_IO_REQUEST *)cm->cm_req;
1958 	bzero(req, sizeof(*req));
1959 	req->DevHandle = htole16(targ->handle);
1960 	req->Function = MPI2_FUNCTION_SCSI_IO_REQUEST;
1961 	req->MsgFlags = 0;
1962 	req->SenseBufferLowAddress = htole32(cm->cm_sense_busaddr);
1963 	req->SenseBufferLength = MPR_SENSE_LEN;
1964 	req->SGLFlags = 0;
1965 	req->ChainOffset = 0;
1966 	req->SGLOffset0 = 24;	/* 32bit word offset to the SGL */
1967 	req->SGLOffset1= 0;
1968 	req->SGLOffset2= 0;
1969 	req->SGLOffset3= 0;
1970 	req->SkipCount = 0;
1971 	req->DataLength = htole32(csio->dxfer_len);
1972 	req->BidirectionalDataLength = 0;
1973 	req->IoFlags = htole16(csio->cdb_len);
1974 	req->EEDPFlags = 0;
1975 
1976 	/* Note: BiDirectional transfers are not supported */
1977 	switch (csio->ccb_h.flags & CAM_DIR_MASK) {
1978 	case CAM_DIR_IN:
1979 		mpi_control = MPI2_SCSIIO_CONTROL_READ;
1980 		cm->cm_flags |= MPR_CM_FLAGS_DATAIN;
1981 		break;
1982 	case CAM_DIR_OUT:
1983 		mpi_control = MPI2_SCSIIO_CONTROL_WRITE;
1984 		cm->cm_flags |= MPR_CM_FLAGS_DATAOUT;
1985 		break;
1986 	case CAM_DIR_NONE:
1987 	default:
1988 		mpi_control = MPI2_SCSIIO_CONTROL_NODATATRANSFER;
1989 		break;
1990 	}
1991 
1992 	if (csio->cdb_len == 32)
1993 		mpi_control |= 4 << MPI2_SCSIIO_CONTROL_ADDCDBLEN_SHIFT;
1994 	/*
1995 	 * It looks like the hardware doesn't require an explicit tag
1996 	 * number for each transaction.  SAM Task Management not supported
1997 	 * at the moment.
1998 	 */
1999 	switch (csio->tag_action) {
2000 	case MSG_HEAD_OF_Q_TAG:
2001 		mpi_control |= MPI2_SCSIIO_CONTROL_HEADOFQ;
2002 		break;
2003 	case MSG_ORDERED_Q_TAG:
2004 		mpi_control |= MPI2_SCSIIO_CONTROL_ORDEREDQ;
2005 		break;
2006 	case MSG_ACA_TASK:
2007 		mpi_control |= MPI2_SCSIIO_CONTROL_ACAQ;
2008 		break;
2009 	case CAM_TAG_ACTION_NONE:
2010 	case MSG_SIMPLE_Q_TAG:
2011 	default:
2012 		mpi_control |= MPI2_SCSIIO_CONTROL_SIMPLEQ;
2013 		break;
2014 	}
2015 	mpi_control |= sc->mapping_table[csio->ccb_h.target_id].TLR_bits;
2016 	req->Control = htole32(mpi_control);
2017 
2018 	if (MPR_SET_LUN(req->LUN, csio->ccb_h.target_lun) != 0) {
2019 		mpr_free_command(sc, cm);
2020 		mprsas_set_ccbstatus(ccb, CAM_LUN_INVALID);
2021 		xpt_done(ccb);
2022 		return;
2023 	}
2024 
2025 	if (csio->ccb_h.flags & CAM_CDB_POINTER)
2026 		bcopy(csio->cdb_io.cdb_ptr, &req->CDB.CDB32[0], csio->cdb_len);
2027 	else {
2028 		KASSERT(csio->cdb_len <= IOCDBLEN,
2029 		    ("cdb_len %d is greater than IOCDBLEN but CAM_CDB_POINTER "
2030 		    "is not set", csio->cdb_len));
2031 		bcopy(csio->cdb_io.cdb_bytes, &req->CDB.CDB32[0],csio->cdb_len);
2032 	}
2033 	req->IoFlags = htole16(csio->cdb_len);
2034 
2035 	/*
2036 	 * Check if EEDP is supported and enabled.  If it is then check if the
2037 	 * SCSI opcode could be using EEDP.  If so, make sure the LUN exists and
2038 	 * is formatted for EEDP support.  If all of this is true, set CDB up
2039 	 * for EEDP transfer.
2040 	 */
2041 	eedp_flags = op_code_prot[req->CDB.CDB32[0]];
2042 	if (sc->eedp_enabled && eedp_flags) {
2043 		SLIST_FOREACH(lun, &targ->luns, lun_link) {
2044 			if (lun->lun_id == csio->ccb_h.target_lun) {
2045 				break;
2046 			}
2047 		}
2048 
2049 		if ((lun != NULL) && (lun->eedp_formatted)) {
2050 			req->EEDPBlockSize = htole16(lun->eedp_block_size);
2051 			eedp_flags |= (MPI2_SCSIIO_EEDPFLAGS_INC_PRI_REFTAG |
2052 			    MPI2_SCSIIO_EEDPFLAGS_CHECK_REFTAG |
2053 			    MPI2_SCSIIO_EEDPFLAGS_CHECK_GUARD);
2054 			if (sc->mpr_flags & MPR_FLAGS_GEN35_IOC) {
2055 				eedp_flags |=
2056 				    MPI25_SCSIIO_EEDPFLAGS_APPTAG_DISABLE_MODE;
2057 			}
2058 			req->EEDPFlags = htole16(eedp_flags);
2059 
2060 			/*
2061 			 * If CDB less than 32, fill in Primary Ref Tag with
2062 			 * low 4 bytes of LBA.  If CDB is 32, tag stuff is
2063 			 * already there.  Also, set protection bit.  FreeBSD
2064 			 * currently does not support CDBs bigger than 16, but
2065 			 * the code doesn't hurt, and will be here for the
2066 			 * future.
2067 			 */
2068 			if (csio->cdb_len != 32) {
2069 				lba_byte = (csio->cdb_len == 16) ? 6 : 2;
2070 				ref_tag_addr = (uint8_t *)&req->CDB.EEDP32.
2071 				    PrimaryReferenceTag;
2072 				for (i = 0; i < 4; i++) {
2073 					*ref_tag_addr =
2074 					    req->CDB.CDB32[lba_byte + i];
2075 					ref_tag_addr++;
2076 				}
2077 				req->CDB.EEDP32.PrimaryReferenceTag =
2078 				    htole32(req->
2079 				    CDB.EEDP32.PrimaryReferenceTag);
2080 				req->CDB.EEDP32.PrimaryApplicationTagMask =
2081 				    0xFFFF;
2082 				req->CDB.CDB32[1] =
2083 				    (req->CDB.CDB32[1] & 0x1F) | 0x20;
2084 			} else {
2085 				eedp_flags |=
2086 				    MPI2_SCSIIO_EEDPFLAGS_INC_PRI_APPTAG;
2087 				req->EEDPFlags = htole16(eedp_flags);
2088 				req->CDB.CDB32[10] = (req->CDB.CDB32[10] &
2089 				    0x1F) | 0x20;
2090 			}
2091 		}
2092 	}
2093 
2094 	cm->cm_length = csio->dxfer_len;
2095 	if (cm->cm_length != 0) {
2096 		cm->cm_data = ccb;
2097 		cm->cm_flags |= MPR_CM_FLAGS_USE_CCB;
2098 	} else {
2099 		cm->cm_data = NULL;
2100 	}
2101 	cm->cm_sge = &req->SGL;
2102 	cm->cm_sglsize = (32 - 24) * 4;
2103 	cm->cm_complete = mprsas_scsiio_complete;
2104 	cm->cm_complete_data = ccb;
2105 	cm->cm_targ = targ;
2106 	cm->cm_lun = csio->ccb_h.target_lun;
2107 	cm->cm_ccb = ccb;
2108 	/*
2109 	 * If using FP desc type, need to set a bit in IoFlags (SCSI IO is 0)
2110 	 * and set descriptor type.
2111 	 */
2112 	if (targ->scsi_req_desc_type ==
2113 	    MPI25_REQ_DESCRIPT_FLAGS_FAST_PATH_SCSI_IO) {
2114 		req->IoFlags |= MPI25_SCSIIO_IOFLAGS_FAST_PATH;
2115 		cm->cm_desc.FastPathSCSIIO.RequestFlags =
2116 		    MPI25_REQ_DESCRIPT_FLAGS_FAST_PATH_SCSI_IO;
2117 		if (!sc->atomic_desc_capable) {
2118 			cm->cm_desc.FastPathSCSIIO.DevHandle =
2119 			    htole16(targ->handle);
2120 		}
2121 	} else {
2122 		cm->cm_desc.SCSIIO.RequestFlags =
2123 		    MPI2_REQ_DESCRIPT_FLAGS_SCSI_IO;
2124 		if (!sc->atomic_desc_capable)
2125 			cm->cm_desc.SCSIIO.DevHandle = htole16(targ->handle);
2126 	}
2127 
2128 	csio->ccb_h.qos.sim_data = sbinuptime();
2129 	callout_reset_sbt(&cm->cm_callout, SBT_1MS * ccb->ccb_h.timeout, 0,
2130 	    mprsas_scsiio_timeout, cm, 0);
2131 
2132 	targ->issued++;
2133 	targ->outstanding++;
2134 	TAILQ_INSERT_TAIL(&targ->commands, cm, cm_link);
2135 	ccb->ccb_h.status |= CAM_SIM_QUEUED;
2136 
2137 	mprsas_log_command(cm, MPR_XINFO, "%s cm %p ccb %p outstanding %u\n",
2138 	    __func__, cm, ccb, targ->outstanding);
2139 
2140 	mpr_map_command(sc, cm);
2141 	return;
2142 }
2143 
2144 /**
2145  * mpr_sc_failed_io_info - translated non-succesfull SCSI_IO request
2146  */
2147 static void
2148 mpr_sc_failed_io_info(struct mpr_softc *sc, struct ccb_scsiio *csio,
2149     Mpi2SCSIIOReply_t *mpi_reply, struct mprsas_target *targ)
2150 {
2151 	u32 response_info;
2152 	u8 *response_bytes;
2153 	u16 ioc_status = le16toh(mpi_reply->IOCStatus) &
2154 	    MPI2_IOCSTATUS_MASK;
2155 	u8 scsi_state = mpi_reply->SCSIState;
2156 	u8 scsi_status = mpi_reply->SCSIStatus;
2157 	char *desc_ioc_state = NULL;
2158 	char *desc_scsi_status = NULL;
2159 	u32 log_info = le32toh(mpi_reply->IOCLogInfo);
2160 
2161 	if (log_info == 0x31170000)
2162 		return;
2163 
2164 	desc_ioc_state = mpr_describe_table(mpr_iocstatus_string,
2165 	     ioc_status);
2166 	desc_scsi_status = mpr_describe_table(mpr_scsi_status_string,
2167 	    scsi_status);
2168 
2169 	mpr_dprint(sc, MPR_XINFO, "\thandle(0x%04x), ioc_status(%s)(0x%04x)\n",
2170 	    le16toh(mpi_reply->DevHandle), desc_ioc_state, ioc_status);
2171 	if (targ->encl_level_valid) {
2172 		mpr_dprint(sc, MPR_XINFO, "At enclosure level %d, slot %d, "
2173 		    "connector name (%4s)\n", targ->encl_level, targ->encl_slot,
2174 		    targ->connector_name);
2175 	}
2176 
2177 	/*
2178 	 * We can add more detail about underflow data here
2179 	 * TO-DO
2180 	 */
2181 	mpr_dprint(sc, MPR_XINFO, "\tscsi_status(%s)(0x%02x), "
2182 	    "scsi_state %b\n", desc_scsi_status, scsi_status,
2183 	    scsi_state, "\20" "\1AutosenseValid" "\2AutosenseFailed"
2184 	    "\3NoScsiStatus" "\4Terminated" "\5Response InfoValid");
2185 
2186 	if (sc->mpr_debug & MPR_XINFO &&
2187 	    scsi_state & MPI2_SCSI_STATE_AUTOSENSE_VALID) {
2188 		mpr_dprint(sc, MPR_XINFO, "-> Sense Buffer Data : Start :\n");
2189 		scsi_sense_print(csio);
2190 		mpr_dprint(sc, MPR_XINFO, "-> Sense Buffer Data : End :\n");
2191 	}
2192 
2193 	if (scsi_state & MPI2_SCSI_STATE_RESPONSE_INFO_VALID) {
2194 		response_info = le32toh(mpi_reply->ResponseInfo);
2195 		response_bytes = (u8 *)&response_info;
2196 		mpr_dprint(sc, MPR_XINFO, "response code(0x%01x): %s\n",
2197 		    response_bytes[0],
2198 		    mpr_describe_table(mpr_scsi_taskmgmt_string,
2199 		    response_bytes[0]));
2200 	}
2201 }
2202 
2203 /** mprsas_nvme_trans_status_code
2204  *
2205  * Convert Native NVMe command error status to
2206  * equivalent SCSI error status.
2207  *
2208  * Returns appropriate scsi_status
2209  */
2210 static u8
2211 mprsas_nvme_trans_status_code(uint16_t nvme_status,
2212     struct mpr_command *cm)
2213 {
2214 	u8 status = MPI2_SCSI_STATUS_GOOD;
2215 	int skey, asc, ascq;
2216 	union ccb *ccb = cm->cm_complete_data;
2217 	int returned_sense_len;
2218 	uint8_t sct, sc;
2219 
2220 	sct = NVME_STATUS_GET_SCT(nvme_status);
2221 	sc = NVME_STATUS_GET_SC(nvme_status);
2222 
2223 	status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2224 	skey = SSD_KEY_ILLEGAL_REQUEST;
2225 	asc = SCSI_ASC_NO_SENSE;
2226 	ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
2227 
2228 	switch (sct) {
2229 	case NVME_SCT_GENERIC:
2230 		switch (sc) {
2231 		case NVME_SC_SUCCESS:
2232 			status = MPI2_SCSI_STATUS_GOOD;
2233 			skey = SSD_KEY_NO_SENSE;
2234 			asc = SCSI_ASC_NO_SENSE;
2235 			ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
2236 			break;
2237 		case NVME_SC_INVALID_OPCODE:
2238 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2239 			skey = SSD_KEY_ILLEGAL_REQUEST;
2240 			asc = SCSI_ASC_ILLEGAL_COMMAND;
2241 			ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
2242 			break;
2243 		case NVME_SC_INVALID_FIELD:
2244 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2245 			skey = SSD_KEY_ILLEGAL_REQUEST;
2246 			asc = SCSI_ASC_INVALID_CDB;
2247 			ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
2248 			break;
2249 		case NVME_SC_DATA_TRANSFER_ERROR:
2250 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2251 			skey = SSD_KEY_MEDIUM_ERROR;
2252 			asc = SCSI_ASC_NO_SENSE;
2253 			ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
2254 			break;
2255 		case NVME_SC_ABORTED_POWER_LOSS:
2256 			status = MPI2_SCSI_STATUS_TASK_ABORTED;
2257 			skey = SSD_KEY_ABORTED_COMMAND;
2258 			asc = SCSI_ASC_WARNING;
2259 			ascq = SCSI_ASCQ_POWER_LOSS_EXPECTED;
2260 			break;
2261 		case NVME_SC_INTERNAL_DEVICE_ERROR:
2262 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2263 			skey = SSD_KEY_HARDWARE_ERROR;
2264 			asc = SCSI_ASC_INTERNAL_TARGET_FAILURE;
2265 			ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
2266 			break;
2267 		case NVME_SC_ABORTED_BY_REQUEST:
2268 		case NVME_SC_ABORTED_SQ_DELETION:
2269 		case NVME_SC_ABORTED_FAILED_FUSED:
2270 		case NVME_SC_ABORTED_MISSING_FUSED:
2271 			status = MPI2_SCSI_STATUS_TASK_ABORTED;
2272 			skey = SSD_KEY_ABORTED_COMMAND;
2273 			asc = SCSI_ASC_NO_SENSE;
2274 			ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
2275 			break;
2276 		case NVME_SC_INVALID_NAMESPACE_OR_FORMAT:
2277 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2278 			skey = SSD_KEY_ILLEGAL_REQUEST;
2279 			asc = SCSI_ASC_ACCESS_DENIED_INVALID_LUN_ID;
2280 			ascq = SCSI_ASCQ_INVALID_LUN_ID;
2281 			break;
2282 		case NVME_SC_LBA_OUT_OF_RANGE:
2283 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2284 			skey = SSD_KEY_ILLEGAL_REQUEST;
2285 			asc = SCSI_ASC_ILLEGAL_BLOCK;
2286 			ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
2287 			break;
2288 		case NVME_SC_CAPACITY_EXCEEDED:
2289 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2290 			skey = SSD_KEY_MEDIUM_ERROR;
2291 			asc = SCSI_ASC_NO_SENSE;
2292 			ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
2293 			break;
2294 		case NVME_SC_NAMESPACE_NOT_READY:
2295 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2296 			skey = SSD_KEY_NOT_READY;
2297 			asc = SCSI_ASC_LUN_NOT_READY;
2298 			ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
2299 			break;
2300 		}
2301 		break;
2302 	case NVME_SCT_COMMAND_SPECIFIC:
2303 		switch (sc) {
2304 		case NVME_SC_INVALID_FORMAT:
2305 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2306 			skey = SSD_KEY_ILLEGAL_REQUEST;
2307 			asc = SCSI_ASC_FORMAT_COMMAND_FAILED;
2308 			ascq = SCSI_ASCQ_FORMAT_COMMAND_FAILED;
2309 			break;
2310 		case NVME_SC_CONFLICTING_ATTRIBUTES:
2311 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2312 			skey = SSD_KEY_ILLEGAL_REQUEST;
2313 			asc = SCSI_ASC_INVALID_CDB;
2314 			ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
2315 			break;
2316 		}
2317 		break;
2318 	case NVME_SCT_MEDIA_ERROR:
2319 		switch (sc) {
2320 		case NVME_SC_WRITE_FAULTS:
2321 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2322 			skey = SSD_KEY_MEDIUM_ERROR;
2323 			asc = SCSI_ASC_PERIPHERAL_DEV_WRITE_FAULT;
2324 			ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
2325 			break;
2326 		case NVME_SC_UNRECOVERED_READ_ERROR:
2327 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2328 			skey = SSD_KEY_MEDIUM_ERROR;
2329 			asc = SCSI_ASC_UNRECOVERED_READ_ERROR;
2330 			ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
2331 			break;
2332 		case NVME_SC_GUARD_CHECK_ERROR:
2333 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2334 			skey = SSD_KEY_MEDIUM_ERROR;
2335 			asc = SCSI_ASC_LOG_BLOCK_GUARD_CHECK_FAILED;
2336 			ascq = SCSI_ASCQ_LOG_BLOCK_GUARD_CHECK_FAILED;
2337 			break;
2338 		case NVME_SC_APPLICATION_TAG_CHECK_ERROR:
2339 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2340 			skey = SSD_KEY_MEDIUM_ERROR;
2341 			asc = SCSI_ASC_LOG_BLOCK_APPTAG_CHECK_FAILED;
2342 			ascq = SCSI_ASCQ_LOG_BLOCK_APPTAG_CHECK_FAILED;
2343 			break;
2344 		case NVME_SC_REFERENCE_TAG_CHECK_ERROR:
2345 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2346 			skey = SSD_KEY_MEDIUM_ERROR;
2347 			asc = SCSI_ASC_LOG_BLOCK_REFTAG_CHECK_FAILED;
2348 			ascq = SCSI_ASCQ_LOG_BLOCK_REFTAG_CHECK_FAILED;
2349 			break;
2350 		case NVME_SC_COMPARE_FAILURE:
2351 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2352 			skey = SSD_KEY_MISCOMPARE;
2353 			asc = SCSI_ASC_MISCOMPARE_DURING_VERIFY;
2354 			ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
2355 			break;
2356 		case NVME_SC_ACCESS_DENIED:
2357 			status = MPI2_SCSI_STATUS_CHECK_CONDITION;
2358 			skey = SSD_KEY_ILLEGAL_REQUEST;
2359 			asc = SCSI_ASC_ACCESS_DENIED_INVALID_LUN_ID;
2360 			ascq = SCSI_ASCQ_INVALID_LUN_ID;
2361 			break;
2362 		}
2363 		break;
2364 	}
2365 
2366 	returned_sense_len = sizeof(struct scsi_sense_data);
2367 	if (returned_sense_len < ccb->csio.sense_len)
2368 		ccb->csio.sense_resid = ccb->csio.sense_len -
2369 		    returned_sense_len;
2370 	else
2371 		ccb->csio.sense_resid = 0;
2372 
2373 	scsi_set_sense_data(&ccb->csio.sense_data, SSD_TYPE_FIXED,
2374 	    1, skey, asc, ascq, SSD_ELEM_NONE);
2375 	ccb->ccb_h.status |= CAM_AUTOSNS_VALID;
2376 
2377 	return status;
2378 }
2379 
2380 /** mprsas_complete_nvme_unmap
2381  *
2382  * Complete native NVMe command issued using NVMe Encapsulated
2383  * Request Message.
2384  */
2385 static u8
2386 mprsas_complete_nvme_unmap(struct mpr_softc *sc, struct mpr_command *cm)
2387 {
2388 	Mpi26NVMeEncapsulatedErrorReply_t *mpi_reply;
2389 	struct nvme_completion *nvme_completion = NULL;
2390 	u8 scsi_status = MPI2_SCSI_STATUS_GOOD;
2391 
2392 	mpi_reply =(Mpi26NVMeEncapsulatedErrorReply_t *)cm->cm_reply;
2393 	if (le16toh(mpi_reply->ErrorResponseCount)){
2394 		nvme_completion = (struct nvme_completion *)cm->cm_sense;
2395 		scsi_status = mprsas_nvme_trans_status_code(
2396 		    nvme_completion->status, cm);
2397 	}
2398 	return scsi_status;
2399 }
2400 
2401 static void
2402 mprsas_scsiio_complete(struct mpr_softc *sc, struct mpr_command *cm)
2403 {
2404 	MPI2_SCSI_IO_REPLY *rep;
2405 	union ccb *ccb;
2406 	struct ccb_scsiio *csio;
2407 	struct mprsas_softc *sassc;
2408 	struct scsi_vpd_supported_page_list *vpd_list = NULL;
2409 	u8 *TLR_bits, TLR_on, *scsi_cdb;
2410 	int dir = 0, i;
2411 	u16 alloc_len;
2412 	struct mprsas_target *target;
2413 	target_id_t target_id;
2414 
2415 	MPR_FUNCTRACE(sc);
2416 	mpr_dprint(sc, MPR_TRACE,
2417 	    "cm %p SMID %u ccb %p reply %p outstanding %u\n", cm,
2418 	    cm->cm_desc.Default.SMID, cm->cm_ccb, cm->cm_reply,
2419 	    cm->cm_targ->outstanding);
2420 
2421 	callout_stop(&cm->cm_callout);
2422 	mtx_assert(&sc->mpr_mtx, MA_OWNED);
2423 
2424 	sassc = sc->sassc;
2425 	ccb = cm->cm_complete_data;
2426 	csio = &ccb->csio;
2427 	target_id = csio->ccb_h.target_id;
2428 	rep = (MPI2_SCSI_IO_REPLY *)cm->cm_reply;
2429 	/*
2430 	 * XXX KDM if the chain allocation fails, does it matter if we do
2431 	 * the sync and unload here?  It is simpler to do it in every case,
2432 	 * assuming it doesn't cause problems.
2433 	 */
2434 	if (cm->cm_data != NULL) {
2435 		if (cm->cm_flags & MPR_CM_FLAGS_DATAIN)
2436 			dir = BUS_DMASYNC_POSTREAD;
2437 		else if (cm->cm_flags & MPR_CM_FLAGS_DATAOUT)
2438 			dir = BUS_DMASYNC_POSTWRITE;
2439 		bus_dmamap_sync(sc->buffer_dmat, cm->cm_dmamap, dir);
2440 		bus_dmamap_unload(sc->buffer_dmat, cm->cm_dmamap);
2441 	}
2442 
2443 	cm->cm_targ->completed++;
2444 	cm->cm_targ->outstanding--;
2445 	TAILQ_REMOVE(&cm->cm_targ->commands, cm, cm_link);
2446 	ccb->ccb_h.status &= ~(CAM_STATUS_MASK | CAM_SIM_QUEUED);
2447 
2448 	if (cm->cm_flags & MPR_CM_FLAGS_ON_RECOVERY) {
2449 		TAILQ_REMOVE(&cm->cm_targ->timedout_commands, cm, cm_recovery);
2450 		KASSERT(cm->cm_state == MPR_CM_STATE_BUSY,
2451 		    ("Not busy for CM_FLAGS_TIMEDOUT: %d\n", cm->cm_state));
2452 		cm->cm_flags &= ~MPR_CM_FLAGS_ON_RECOVERY;
2453 		if (cm->cm_reply != NULL)
2454 			mprsas_log_command(cm, MPR_RECOVERY,
2455 			    "completed timedout cm %p ccb %p during recovery "
2456 			    "ioc %x scsi %x state %x xfer %u\n", cm, cm->cm_ccb,
2457 			    le16toh(rep->IOCStatus), rep->SCSIStatus,
2458 			    rep->SCSIState, le32toh(rep->TransferCount));
2459 		else
2460 			mprsas_log_command(cm, MPR_RECOVERY,
2461 			    "completed timedout cm %p ccb %p during recovery\n",
2462 			    cm, cm->cm_ccb);
2463 	} else if (cm->cm_targ->tm != NULL) {
2464 		if (cm->cm_reply != NULL)
2465 			mprsas_log_command(cm, MPR_RECOVERY,
2466 			    "completed cm %p ccb %p during recovery "
2467 			    "ioc %x scsi %x state %x xfer %u\n",
2468 			    cm, cm->cm_ccb, le16toh(rep->IOCStatus),
2469 			    rep->SCSIStatus, rep->SCSIState,
2470 			    le32toh(rep->TransferCount));
2471 		else
2472 			mprsas_log_command(cm, MPR_RECOVERY,
2473 			    "completed cm %p ccb %p during recovery\n",
2474 			    cm, cm->cm_ccb);
2475 	} else if ((sc->mpr_flags & MPR_FLAGS_DIAGRESET) != 0) {
2476 		mprsas_log_command(cm, MPR_RECOVERY,
2477 		    "reset completed cm %p ccb %p\n", cm, cm->cm_ccb);
2478 	}
2479 
2480 	if ((cm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) {
2481 		/*
2482 		 * We ran into an error after we tried to map the command,
2483 		 * so we're getting a callback without queueing the command
2484 		 * to the hardware.  So we set the status here, and it will
2485 		 * be retained below.  We'll go through the "fast path",
2486 		 * because there can be no reply when we haven't actually
2487 		 * gone out to the hardware.
2488 		 */
2489 		mprsas_set_ccbstatus(ccb, CAM_REQUEUE_REQ);
2490 
2491 		/*
2492 		 * Currently the only error included in the mask is
2493 		 * MPR_CM_FLAGS_CHAIN_FAILED, which means we're out of
2494 		 * chain frames.  We need to freeze the queue until we get
2495 		 * a command that completed without this error, which will
2496 		 * hopefully have some chain frames attached that we can
2497 		 * use.  If we wanted to get smarter about it, we would
2498 		 * only unfreeze the queue in this condition when we're
2499 		 * sure that we're getting some chain frames back.  That's
2500 		 * probably unnecessary.
2501 		 */
2502 		if ((sassc->flags & MPRSAS_QUEUE_FROZEN) == 0) {
2503 			xpt_freeze_simq(sassc->sim, 1);
2504 			sassc->flags |= MPRSAS_QUEUE_FROZEN;
2505 			mpr_dprint(sc, MPR_XINFO, "Error sending command, "
2506 			    "freezing SIM queue\n");
2507 		}
2508 	}
2509 
2510 	/*
2511 	 * Point to the SCSI CDB, which is dependent on the CAM_CDB_POINTER
2512 	 * flag, and use it in a few places in the rest of this function for
2513 	 * convenience. Use the macro if available.
2514 	 */
2515 	scsi_cdb = scsiio_cdb_ptr(csio);
2516 
2517 	/*
2518 	 * If this is a Start Stop Unit command and it was issued by the driver
2519 	 * during shutdown, decrement the refcount to account for all of the
2520 	 * commands that were sent.  All SSU commands should be completed before
2521 	 * shutdown completes, meaning SSU_refcount will be 0 after SSU_started
2522 	 * is TRUE.
2523 	 */
2524 	if (sc->SSU_started && (scsi_cdb[0] == START_STOP_UNIT)) {
2525 		mpr_dprint(sc, MPR_INFO, "Decrementing SSU count.\n");
2526 		sc->SSU_refcount--;
2527 	}
2528 
2529 	/* Take the fast path to completion */
2530 	if (cm->cm_reply == NULL) {
2531 		if (mprsas_get_ccbstatus(ccb) == CAM_REQ_INPROG) {
2532 			if ((sc->mpr_flags & MPR_FLAGS_DIAGRESET) != 0)
2533 				mprsas_set_ccbstatus(ccb, CAM_SCSI_BUS_RESET);
2534 			else {
2535 				mprsas_set_ccbstatus(ccb, CAM_REQ_CMP);
2536 				csio->scsi_status = SCSI_STATUS_OK;
2537 			}
2538 			if (sassc->flags & MPRSAS_QUEUE_FROZEN) {
2539 				ccb->ccb_h.status |= CAM_RELEASE_SIMQ;
2540 				sassc->flags &= ~MPRSAS_QUEUE_FROZEN;
2541 				mpr_dprint(sc, MPR_XINFO,
2542 				    "Unfreezing SIM queue\n");
2543 			}
2544 		}
2545 
2546 		/*
2547 		 * There are two scenarios where the status won't be
2548 		 * CAM_REQ_CMP.  The first is if MPR_CM_FLAGS_ERROR_MASK is
2549 		 * set, the second is in the MPR_FLAGS_DIAGRESET above.
2550 		 */
2551 		if (mprsas_get_ccbstatus(ccb) != CAM_REQ_CMP) {
2552 			/*
2553 			 * Freeze the dev queue so that commands are
2554 			 * executed in the correct order after error
2555 			 * recovery.
2556 			 */
2557 			ccb->ccb_h.status |= CAM_DEV_QFRZN;
2558 			xpt_freeze_devq(ccb->ccb_h.path, /*count*/ 1);
2559 		}
2560 		mpr_free_command(sc, cm);
2561 		xpt_done(ccb);
2562 		return;
2563 	}
2564 
2565 	target = &sassc->targets[target_id];
2566 	if (scsi_cdb[0] == UNMAP &&
2567 	    target->is_nvme &&
2568 	    (csio->ccb_h.flags & CAM_DATA_MASK) == CAM_DATA_VADDR) {
2569 		rep->SCSIStatus = mprsas_complete_nvme_unmap(sc, cm);
2570 		csio->scsi_status = rep->SCSIStatus;
2571 	}
2572 
2573 	mprsas_log_command(cm, MPR_XINFO,
2574 	    "ioc %x scsi %x state %x xfer %u\n",
2575 	    le16toh(rep->IOCStatus), rep->SCSIStatus, rep->SCSIState,
2576 	    le32toh(rep->TransferCount));
2577 
2578 	switch (le16toh(rep->IOCStatus) & MPI2_IOCSTATUS_MASK) {
2579 	case MPI2_IOCSTATUS_SCSI_DATA_UNDERRUN:
2580 		csio->resid = cm->cm_length - le32toh(rep->TransferCount);
2581 		/* FALLTHROUGH */
2582 	case MPI2_IOCSTATUS_SUCCESS:
2583 	case MPI2_IOCSTATUS_SCSI_RECOVERED_ERROR:
2584 		if ((le16toh(rep->IOCStatus) & MPI2_IOCSTATUS_MASK) ==
2585 		    MPI2_IOCSTATUS_SCSI_RECOVERED_ERROR)
2586 			mprsas_log_command(cm, MPR_XINFO, "recovered error\n");
2587 
2588 		/* Completion failed at the transport level. */
2589 		if (rep->SCSIState & (MPI2_SCSI_STATE_NO_SCSI_STATUS |
2590 		    MPI2_SCSI_STATE_TERMINATED)) {
2591 			mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR);
2592 			break;
2593 		}
2594 
2595 		/* In a modern packetized environment, an autosense failure
2596 		 * implies that there's not much else that can be done to
2597 		 * recover the command.
2598 		 */
2599 		if (rep->SCSIState & MPI2_SCSI_STATE_AUTOSENSE_FAILED) {
2600 			mprsas_set_ccbstatus(ccb, CAM_AUTOSENSE_FAIL);
2601 			break;
2602 		}
2603 
2604 		/*
2605 		 * CAM doesn't care about SAS Response Info data, but if this is
2606 		 * the state check if TLR should be done.  If not, clear the
2607 		 * TLR_bits for the target.
2608 		 */
2609 		if ((rep->SCSIState & MPI2_SCSI_STATE_RESPONSE_INFO_VALID) &&
2610 		    ((le32toh(rep->ResponseInfo) & MPI2_SCSI_RI_MASK_REASONCODE)
2611 		    == MPR_SCSI_RI_INVALID_FRAME)) {
2612 			sc->mapping_table[target_id].TLR_bits =
2613 			    (u8)MPI2_SCSIIO_CONTROL_NO_TLR;
2614 		}
2615 
2616 		/*
2617 		 * Intentionally override the normal SCSI status reporting
2618 		 * for these two cases.  These are likely to happen in a
2619 		 * multi-initiator environment, and we want to make sure that
2620 		 * CAM retries these commands rather than fail them.
2621 		 */
2622 		if ((rep->SCSIStatus == MPI2_SCSI_STATUS_COMMAND_TERMINATED) ||
2623 		    (rep->SCSIStatus == MPI2_SCSI_STATUS_TASK_ABORTED)) {
2624 			mprsas_set_ccbstatus(ccb, CAM_REQ_ABORTED);
2625 			break;
2626 		}
2627 
2628 		/* Handle normal status and sense */
2629 		csio->scsi_status = rep->SCSIStatus;
2630 		if (rep->SCSIStatus == MPI2_SCSI_STATUS_GOOD)
2631 			mprsas_set_ccbstatus(ccb, CAM_REQ_CMP);
2632 		else
2633 			mprsas_set_ccbstatus(ccb, CAM_SCSI_STATUS_ERROR);
2634 
2635 		if (rep->SCSIState & MPI2_SCSI_STATE_AUTOSENSE_VALID) {
2636 			int sense_len, returned_sense_len;
2637 
2638 			returned_sense_len = min(le32toh(rep->SenseCount),
2639 			    sizeof(struct scsi_sense_data));
2640 			if (returned_sense_len < csio->sense_len)
2641 				csio->sense_resid = csio->sense_len -
2642 				    returned_sense_len;
2643 			else
2644 				csio->sense_resid = 0;
2645 
2646 			sense_len = min(returned_sense_len,
2647 			    csio->sense_len - csio->sense_resid);
2648 			bzero(&csio->sense_data, sizeof(csio->sense_data));
2649 			bcopy(cm->cm_sense, &csio->sense_data, sense_len);
2650 			ccb->ccb_h.status |= CAM_AUTOSNS_VALID;
2651 		}
2652 
2653 		/*
2654 		 * Check if this is an INQUIRY command.  If it's a VPD inquiry,
2655 		 * and it's page code 0 (Supported Page List), and there is
2656 		 * inquiry data, and this is for a sequential access device, and
2657 		 * the device is an SSP target, and TLR is supported by the
2658 		 * controller, turn the TLR_bits value ON if page 0x90 is
2659 		 * supported.
2660 		 */
2661 		if ((scsi_cdb[0] == INQUIRY) &&
2662 		    (scsi_cdb[1] & SI_EVPD) &&
2663 		    (scsi_cdb[2] == SVPD_SUPPORTED_PAGE_LIST) &&
2664 		    ((csio->ccb_h.flags & CAM_DATA_MASK) == CAM_DATA_VADDR) &&
2665 		    (csio->data_ptr != NULL) &&
2666 		    ((csio->data_ptr[0] & 0x1f) == T_SEQUENTIAL) &&
2667 		    (sc->control_TLR) &&
2668 		    (sc->mapping_table[target_id].device_info &
2669 		    MPI2_SAS_DEVICE_INFO_SSP_TARGET)) {
2670 			vpd_list = (struct scsi_vpd_supported_page_list *)
2671 			    csio->data_ptr;
2672 			TLR_bits = &sc->mapping_table[target_id].TLR_bits;
2673 			*TLR_bits = (u8)MPI2_SCSIIO_CONTROL_NO_TLR;
2674 			TLR_on = (u8)MPI2_SCSIIO_CONTROL_TLR_ON;
2675 			alloc_len = ((u16)scsi_cdb[3] << 8) + scsi_cdb[4];
2676 			alloc_len -= csio->resid;
2677 			for (i = 0; i < MIN(vpd_list->length, alloc_len); i++) {
2678 				if (vpd_list->list[i] == 0x90) {
2679 					*TLR_bits = TLR_on;
2680 					break;
2681 				}
2682 			}
2683 		}
2684 
2685 		/*
2686 		 * If this is a SATA direct-access end device, mark it so that
2687 		 * a SCSI StartStopUnit command will be sent to it when the
2688 		 * driver is being shutdown.
2689 		 */
2690 		if ((scsi_cdb[0] == INQUIRY) &&
2691 		    (csio->data_ptr != NULL) &&
2692 		    ((csio->data_ptr[0] & 0x1f) == T_DIRECT) &&
2693 		    (sc->mapping_table[target_id].device_info &
2694 		    MPI2_SAS_DEVICE_INFO_SATA_DEVICE) &&
2695 		    ((sc->mapping_table[target_id].device_info &
2696 		    MPI2_SAS_DEVICE_INFO_MASK_DEVICE_TYPE) ==
2697 		    MPI2_SAS_DEVICE_INFO_END_DEVICE)) {
2698 			target = &sassc->targets[target_id];
2699 			target->supports_SSU = TRUE;
2700 			mpr_dprint(sc, MPR_XINFO, "Target %d supports SSU\n",
2701 			    target_id);
2702 		}
2703 		break;
2704 	case MPI2_IOCSTATUS_SCSI_INVALID_DEVHANDLE:
2705 	case MPI2_IOCSTATUS_SCSI_DEVICE_NOT_THERE:
2706 		/*
2707 		 * If devinfo is 0 this will be a volume.  In that case don't
2708 		 * tell CAM that the volume is not there.  We want volumes to
2709 		 * be enumerated until they are deleted/removed, not just
2710 		 * failed.
2711 		 */
2712 		if (cm->cm_targ->devinfo == 0)
2713 			mprsas_set_ccbstatus(ccb, CAM_REQ_CMP);
2714 		else
2715 			mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE);
2716 		break;
2717 	case MPI2_IOCSTATUS_INVALID_SGL:
2718 		mpr_print_scsiio_cmd(sc, cm);
2719 		mprsas_set_ccbstatus(ccb, CAM_UNREC_HBA_ERROR);
2720 		break;
2721 	case MPI2_IOCSTATUS_SCSI_TASK_TERMINATED:
2722 		/*
2723 		 * This is one of the responses that comes back when an I/O
2724 		 * has been aborted.  If it is because of a timeout that we
2725 		 * initiated, just set the status to CAM_CMD_TIMEOUT.
2726 		 * Otherwise set it to CAM_REQ_ABORTED.  The effect on the
2727 		 * command is the same (it gets retried, subject to the
2728 		 * retry counter), the only difference is what gets printed
2729 		 * on the console.
2730 		 */
2731 		if (cm->cm_flags & MPR_CM_FLAGS_TIMEDOUT)
2732 			mprsas_set_ccbstatus(ccb, CAM_CMD_TIMEOUT);
2733 		else
2734 			mprsas_set_ccbstatus(ccb, CAM_REQ_ABORTED);
2735 		break;
2736 	case MPI2_IOCSTATUS_SCSI_DATA_OVERRUN:
2737 		/* resid is ignored for this condition */
2738 		csio->resid = 0;
2739 		mprsas_set_ccbstatus(ccb, CAM_DATA_RUN_ERR);
2740 		break;
2741 	case MPI2_IOCSTATUS_SCSI_IOC_TERMINATED:
2742 	case MPI2_IOCSTATUS_SCSI_EXT_TERMINATED:
2743 		/*
2744 		 * These can sometimes be transient transport-related
2745 		 * errors, and sometimes persistent drive-related errors.
2746 		 * We used to retry these without decrementing the retry
2747 		 * count by returning CAM_REQUEUE_REQ.  Unfortunately, if
2748 		 * we hit a persistent drive problem that returns one of
2749 		 * these error codes, we would retry indefinitely.  So,
2750 		 * return CAM_REQ_CMP_ERROR so that we decrement the retry
2751 		 * count and avoid infinite retries.  We're taking the
2752 		 * potential risk of flagging false failures in the event
2753 		 * of a topology-related error (e.g. a SAS expander problem
2754 		 * causes a command addressed to a drive to fail), but
2755 		 * avoiding getting into an infinite retry loop. However,
2756 		 * if we get them while were moving a device, we should
2757 		 * fail the request as 'not there' because the device
2758 		 * is effectively gone.
2759 		 */
2760 		if (cm->cm_targ->flags & MPRSAS_TARGET_INREMOVAL)
2761 			mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE);
2762 		else
2763 			mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR);
2764 		mpr_dprint(sc, MPR_INFO,
2765 		    "Controller reported %s tgt %u SMID %u loginfo %x%s\n",
2766 		    mpr_describe_table(mpr_iocstatus_string,
2767 		    le16toh(rep->IOCStatus) & MPI2_IOCSTATUS_MASK),
2768 		    target_id, cm->cm_desc.Default.SMID,
2769 		    le32toh(rep->IOCLogInfo),
2770 		    (cm->cm_targ->flags & MPRSAS_TARGET_INREMOVAL) ? " departing" : "");
2771 		mpr_dprint(sc, MPR_XINFO,
2772 		    "SCSIStatus %x SCSIState %x xfercount %u\n",
2773 		    rep->SCSIStatus, rep->SCSIState,
2774 		    le32toh(rep->TransferCount));
2775 		break;
2776 	case MPI2_IOCSTATUS_INVALID_FUNCTION:
2777 	case MPI2_IOCSTATUS_INTERNAL_ERROR:
2778 	case MPI2_IOCSTATUS_INVALID_VPID:
2779 	case MPI2_IOCSTATUS_INVALID_FIELD:
2780 	case MPI2_IOCSTATUS_INVALID_STATE:
2781 	case MPI2_IOCSTATUS_OP_STATE_NOT_SUPPORTED:
2782 	case MPI2_IOCSTATUS_SCSI_IO_DATA_ERROR:
2783 	case MPI2_IOCSTATUS_SCSI_PROTOCOL_ERROR:
2784 	case MPI2_IOCSTATUS_SCSI_RESIDUAL_MISMATCH:
2785 	case MPI2_IOCSTATUS_SCSI_TASK_MGMT_FAILED:
2786 	default:
2787 		mprsas_log_command(cm, MPR_XINFO,
2788 		    "completed ioc %x loginfo %x scsi %x state %x xfer %u\n",
2789 		    le16toh(rep->IOCStatus), le32toh(rep->IOCLogInfo),
2790 		    rep->SCSIStatus, rep->SCSIState,
2791 		    le32toh(rep->TransferCount));
2792 		csio->resid = cm->cm_length;
2793 
2794 		if (scsi_cdb[0] == UNMAP &&
2795 		    target->is_nvme &&
2796 		    (csio->ccb_h.flags & CAM_DATA_MASK) == CAM_DATA_VADDR)
2797 			mprsas_set_ccbstatus(ccb, CAM_REQ_CMP);
2798 		else
2799 			mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR);
2800 
2801 		break;
2802 	}
2803 
2804 	mpr_sc_failed_io_info(sc, csio, rep, cm->cm_targ);
2805 
2806 	if (sassc->flags & MPRSAS_QUEUE_FROZEN) {
2807 		ccb->ccb_h.status |= CAM_RELEASE_SIMQ;
2808 		sassc->flags &= ~MPRSAS_QUEUE_FROZEN;
2809 		mpr_dprint(sc, MPR_XINFO, "Command completed, unfreezing SIM "
2810 		    "queue\n");
2811 	}
2812 
2813 	if (mprsas_get_ccbstatus(ccb) != CAM_REQ_CMP) {
2814 		ccb->ccb_h.status |= CAM_DEV_QFRZN;
2815 		xpt_freeze_devq(ccb->ccb_h.path, /*count*/ 1);
2816 	}
2817 
2818 	/*
2819 	 * Check to see if we're removing the device. If so, and this is the
2820 	 * last command on the queue, proceed with the deferred removal of the
2821 	 * device.  Note, for removing a volume, this won't trigger because
2822 	 * pending_remove_tm will be NULL.
2823 	 */
2824 	if (cm->cm_targ->flags & MPRSAS_TARGET_INREMOVAL) {
2825 		if (TAILQ_FIRST(&cm->cm_targ->commands) == NULL &&
2826 		    cm->cm_targ->pending_remove_tm != NULL) {
2827 			mpr_dprint(sc, MPR_INFO, "Last pending command complete: starting remove_device\n");
2828 			mpr_map_command(sc, cm->cm_targ->pending_remove_tm);
2829 			cm->cm_targ->pending_remove_tm = NULL;
2830 		}
2831 	}
2832 
2833 	mpr_free_command(sc, cm);
2834 	xpt_done(ccb);
2835 }
2836 
2837 static void
2838 mprsas_smpio_complete(struct mpr_softc *sc, struct mpr_command *cm)
2839 {
2840 	MPI2_SMP_PASSTHROUGH_REPLY *rpl;
2841 	MPI2_SMP_PASSTHROUGH_REQUEST *req;
2842 	uint64_t sasaddr;
2843 	union ccb *ccb;
2844 
2845 	ccb = cm->cm_complete_data;
2846 
2847 	/*
2848 	 * Currently there should be no way we can hit this case.  It only
2849 	 * happens when we have a failure to allocate chain frames, and SMP
2850 	 * commands require two S/G elements only.  That should be handled
2851 	 * in the standard request size.
2852 	 */
2853 	if ((cm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) {
2854 		mpr_dprint(sc, MPR_ERROR, "%s: cm_flags = %#x on SMP "
2855 		    "request!\n", __func__, cm->cm_flags);
2856 		mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR);
2857 		goto bailout;
2858         }
2859 
2860 	rpl = (MPI2_SMP_PASSTHROUGH_REPLY *)cm->cm_reply;
2861 	if (rpl == NULL) {
2862 		mpr_dprint(sc, MPR_ERROR, "%s: NULL cm_reply!\n", __func__);
2863 		mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR);
2864 		goto bailout;
2865 	}
2866 
2867 	req = (MPI2_SMP_PASSTHROUGH_REQUEST *)cm->cm_req;
2868 	sasaddr = le32toh(req->SASAddress.Low);
2869 	sasaddr |= ((uint64_t)(le32toh(req->SASAddress.High))) << 32;
2870 
2871 	if ((le16toh(rpl->IOCStatus) & MPI2_IOCSTATUS_MASK) !=
2872 	    MPI2_IOCSTATUS_SUCCESS ||
2873 	    rpl->SASStatus != MPI2_SASSTATUS_SUCCESS) {
2874 		mpr_dprint(sc, MPR_XINFO, "%s: IOCStatus %04x SASStatus %02x\n",
2875 		    __func__, le16toh(rpl->IOCStatus), rpl->SASStatus);
2876 		mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR);
2877 		goto bailout;
2878 	}
2879 
2880 	mpr_dprint(sc, MPR_XINFO, "%s: SMP request to SAS address %#jx "
2881 	    "completed successfully\n", __func__, (uintmax_t)sasaddr);
2882 
2883 	if (ccb->smpio.smp_response[2] == SMP_FR_ACCEPTED)
2884 		mprsas_set_ccbstatus(ccb, CAM_REQ_CMP);
2885 	else
2886 		mprsas_set_ccbstatus(ccb, CAM_SMP_STATUS_ERROR);
2887 
2888 bailout:
2889 	/*
2890 	 * We sync in both directions because we had DMAs in the S/G list
2891 	 * in both directions.
2892 	 */
2893 	bus_dmamap_sync(sc->buffer_dmat, cm->cm_dmamap,
2894 			BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2895 	bus_dmamap_unload(sc->buffer_dmat, cm->cm_dmamap);
2896 	mpr_free_command(sc, cm);
2897 	xpt_done(ccb);
2898 }
2899 
2900 static void
2901 mprsas_send_smpcmd(struct mprsas_softc *sassc, union ccb *ccb, uint64_t sasaddr)
2902 {
2903 	struct mpr_command *cm;
2904 	uint8_t *request, *response;
2905 	MPI2_SMP_PASSTHROUGH_REQUEST *req;
2906 	struct mpr_softc *sc;
2907 	struct sglist *sg;
2908 	int error;
2909 
2910 	sc = sassc->sc;
2911 	sg = NULL;
2912 	error = 0;
2913 
2914 	switch (ccb->ccb_h.flags & CAM_DATA_MASK) {
2915 	case CAM_DATA_PADDR:
2916 	case CAM_DATA_SG_PADDR:
2917 		/*
2918 		 * XXX We don't yet support physical addresses here.
2919 		 */
2920 		mpr_dprint(sc, MPR_ERROR, "%s: physical addresses not "
2921 		    "supported\n", __func__);
2922 		mprsas_set_ccbstatus(ccb, CAM_REQ_INVALID);
2923 		xpt_done(ccb);
2924 		return;
2925 	case CAM_DATA_SG:
2926 		/*
2927 		 * The chip does not support more than one buffer for the
2928 		 * request or response.
2929 		 */
2930 		if ((ccb->smpio.smp_request_sglist_cnt > 1)
2931 		    || (ccb->smpio.smp_response_sglist_cnt > 1)) {
2932 			mpr_dprint(sc, MPR_ERROR, "%s: multiple request or "
2933 			    "response buffer segments not supported for SMP\n",
2934 			    __func__);
2935 			mprsas_set_ccbstatus(ccb, CAM_REQ_INVALID);
2936 			xpt_done(ccb);
2937 			return;
2938 		}
2939 
2940 		/*
2941 		 * The CAM_SCATTER_VALID flag was originally implemented
2942 		 * for the XPT_SCSI_IO CCB, which only has one data pointer.
2943 		 * We have two.  So, just take that flag to mean that we
2944 		 * might have S/G lists, and look at the S/G segment count
2945 		 * to figure out whether that is the case for each individual
2946 		 * buffer.
2947 		 */
2948 		if (ccb->smpio.smp_request_sglist_cnt != 0) {
2949 			bus_dma_segment_t *req_sg;
2950 
2951 			req_sg = (bus_dma_segment_t *)ccb->smpio.smp_request;
2952 			request = (uint8_t *)(uintptr_t)req_sg[0].ds_addr;
2953 		} else
2954 			request = ccb->smpio.smp_request;
2955 
2956 		if (ccb->smpio.smp_response_sglist_cnt != 0) {
2957 			bus_dma_segment_t *rsp_sg;
2958 
2959 			rsp_sg = (bus_dma_segment_t *)ccb->smpio.smp_response;
2960 			response = (uint8_t *)(uintptr_t)rsp_sg[0].ds_addr;
2961 		} else
2962 			response = ccb->smpio.smp_response;
2963 		break;
2964 	case CAM_DATA_VADDR:
2965 		request = ccb->smpio.smp_request;
2966 		response = ccb->smpio.smp_response;
2967 		break;
2968 	default:
2969 		mprsas_set_ccbstatus(ccb, CAM_REQ_INVALID);
2970 		xpt_done(ccb);
2971 		return;
2972 	}
2973 
2974 	cm = mpr_alloc_command(sc);
2975 	if (cm == NULL) {
2976 		mpr_dprint(sc, MPR_ERROR, "%s: cannot allocate command\n",
2977 		    __func__);
2978 		mprsas_set_ccbstatus(ccb, CAM_RESRC_UNAVAIL);
2979 		xpt_done(ccb);
2980 		return;
2981 	}
2982 
2983 	req = (MPI2_SMP_PASSTHROUGH_REQUEST *)cm->cm_req;
2984 	bzero(req, sizeof(*req));
2985 	req->Function = MPI2_FUNCTION_SMP_PASSTHROUGH;
2986 
2987 	/* Allow the chip to use any route to this SAS address. */
2988 	req->PhysicalPort = 0xff;
2989 
2990 	req->RequestDataLength = htole16(ccb->smpio.smp_request_len);
2991 	req->SGLFlags =
2992 	    MPI2_SGLFLAGS_SYSTEM_ADDRESS_SPACE | MPI2_SGLFLAGS_SGL_TYPE_MPI;
2993 
2994 	mpr_dprint(sc, MPR_XINFO, "%s: sending SMP request to SAS address "
2995 	    "%#jx\n", __func__, (uintmax_t)sasaddr);
2996 
2997 	mpr_init_sge(cm, req, &req->SGL);
2998 
2999 	/*
3000 	 * Set up a uio to pass into mpr_map_command().  This allows us to
3001 	 * do one map command, and one busdma call in there.
3002 	 */
3003 	cm->cm_uio.uio_iov = cm->cm_iovec;
3004 	cm->cm_uio.uio_iovcnt = 2;
3005 	cm->cm_uio.uio_segflg = UIO_SYSSPACE;
3006 
3007 	/*
3008 	 * The read/write flag isn't used by busdma, but set it just in
3009 	 * case.  This isn't exactly accurate, either, since we're going in
3010 	 * both directions.
3011 	 */
3012 	cm->cm_uio.uio_rw = UIO_WRITE;
3013 
3014 	cm->cm_iovec[0].iov_base = request;
3015 	cm->cm_iovec[0].iov_len = le16toh(req->RequestDataLength);
3016 	cm->cm_iovec[1].iov_base = response;
3017 	cm->cm_iovec[1].iov_len = ccb->smpio.smp_response_len;
3018 
3019 	cm->cm_uio.uio_resid = cm->cm_iovec[0].iov_len +
3020 			       cm->cm_iovec[1].iov_len;
3021 
3022 	/*
3023 	 * Trigger a warning message in mpr_data_cb() for the user if we
3024 	 * wind up exceeding two S/G segments.  The chip expects one
3025 	 * segment for the request and another for the response.
3026 	 */
3027 	cm->cm_max_segs = 2;
3028 
3029 	cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE;
3030 	cm->cm_complete = mprsas_smpio_complete;
3031 	cm->cm_complete_data = ccb;
3032 
3033 	/*
3034 	 * Tell the mapping code that we're using a uio, and that this is
3035 	 * an SMP passthrough request.  There is a little special-case
3036 	 * logic there (in mpr_data_cb()) to handle the bidirectional
3037 	 * transfer.
3038 	 */
3039 	cm->cm_flags |= MPR_CM_FLAGS_USE_UIO | MPR_CM_FLAGS_SMP_PASS |
3040 			MPR_CM_FLAGS_DATAIN | MPR_CM_FLAGS_DATAOUT;
3041 
3042 	/* The chip data format is little endian. */
3043 	req->SASAddress.High = htole32(sasaddr >> 32);
3044 	req->SASAddress.Low = htole32(sasaddr);
3045 
3046 	/*
3047 	 * XXX Note that we don't have a timeout/abort mechanism here.
3048 	 * From the manual, it looks like task management requests only
3049 	 * work for SCSI IO and SATA passthrough requests.  We may need to
3050 	 * have a mechanism to retry requests in the event of a chip reset
3051 	 * at least.  Hopefully the chip will insure that any errors short
3052 	 * of that are relayed back to the driver.
3053 	 */
3054 	error = mpr_map_command(sc, cm);
3055 	if ((error != 0) && (error != EINPROGRESS)) {
3056 		mpr_dprint(sc, MPR_ERROR, "%s: error %d returned from "
3057 		    "mpr_map_command()\n", __func__, error);
3058 		goto bailout_error;
3059 	}
3060 
3061 	return;
3062 
3063 bailout_error:
3064 	mpr_free_command(sc, cm);
3065 	mprsas_set_ccbstatus(ccb, CAM_RESRC_UNAVAIL);
3066 	xpt_done(ccb);
3067 	return;
3068 }
3069 
3070 static void
3071 mprsas_action_smpio(struct mprsas_softc *sassc, union ccb *ccb)
3072 {
3073 	struct mpr_softc *sc;
3074 	struct mprsas_target *targ;
3075 	uint64_t sasaddr = 0;
3076 
3077 	sc = sassc->sc;
3078 
3079 	/*
3080 	 * Make sure the target exists.
3081 	 */
3082 	KASSERT(ccb->ccb_h.target_id < sassc->maxtargets,
3083 	    ("Target %d out of bounds in XPT_SMP_IO\n", ccb->ccb_h.target_id));
3084 	targ = &sassc->targets[ccb->ccb_h.target_id];
3085 	if (targ->handle == 0x0) {
3086 		mpr_dprint(sc, MPR_ERROR, "%s: target %d does not exist!\n",
3087 		    __func__, ccb->ccb_h.target_id);
3088 		mprsas_set_ccbstatus(ccb, CAM_SEL_TIMEOUT);
3089 		xpt_done(ccb);
3090 		return;
3091 	}
3092 
3093 	/*
3094 	 * If this device has an embedded SMP target, we'll talk to it
3095 	 * directly.
3096 	 * figure out what the expander's address is.
3097 	 */
3098 	if ((targ->devinfo & MPI2_SAS_DEVICE_INFO_SMP_TARGET) != 0)
3099 		sasaddr = targ->sasaddr;
3100 
3101 	/*
3102 	 * If we don't have a SAS address for the expander yet, try
3103 	 * grabbing it from the page 0x83 information cached in the
3104 	 * transport layer for this target.  LSI expanders report the
3105 	 * expander SAS address as the port-associated SAS address in
3106 	 * Inquiry VPD page 0x83.  Maxim expanders don't report it in page
3107 	 * 0x83.
3108 	 *
3109 	 * XXX KDM disable this for now, but leave it commented out so that
3110 	 * it is obvious that this is another possible way to get the SAS
3111 	 * address.
3112 	 *
3113 	 * The parent handle method below is a little more reliable, and
3114 	 * the other benefit is that it works for devices other than SES
3115 	 * devices.  So you can send a SMP request to a da(4) device and it
3116 	 * will get routed to the expander that device is attached to.
3117 	 * (Assuming the da(4) device doesn't contain an SMP target...)
3118 	 */
3119 #if 0
3120 	if (sasaddr == 0)
3121 		sasaddr = xpt_path_sas_addr(ccb->ccb_h.path);
3122 #endif
3123 
3124 	/*
3125 	 * If we still don't have a SAS address for the expander, look for
3126 	 * the parent device of this device, which is probably the expander.
3127 	 */
3128 	if (sasaddr == 0) {
3129 #ifdef OLD_MPR_PROBE
3130 		struct mprsas_target *parent_target;
3131 #endif
3132 
3133 		if (targ->parent_handle == 0x0) {
3134 			mpr_dprint(sc, MPR_ERROR, "%s: handle %d does not have "
3135 			    "a valid parent handle!\n", __func__, targ->handle);
3136 			mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE);
3137 			goto bailout;
3138 		}
3139 #ifdef OLD_MPR_PROBE
3140 		parent_target = mprsas_find_target_by_handle(sassc, 0,
3141 		    targ->parent_handle);
3142 
3143 		if (parent_target == NULL) {
3144 			mpr_dprint(sc, MPR_ERROR, "%s: handle %d does not have "
3145 			    "a valid parent target!\n", __func__, targ->handle);
3146 			mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE);
3147 			goto bailout;
3148 		}
3149 
3150 		if ((parent_target->devinfo &
3151 		     MPI2_SAS_DEVICE_INFO_SMP_TARGET) == 0) {
3152 			mpr_dprint(sc, MPR_ERROR, "%s: handle %d parent %d "
3153 			    "does not have an SMP target!\n", __func__,
3154 			    targ->handle, parent_target->handle);
3155 			mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE);
3156 			goto bailout;
3157 		}
3158 
3159 		sasaddr = parent_target->sasaddr;
3160 #else /* OLD_MPR_PROBE */
3161 		if ((targ->parent_devinfo &
3162 		     MPI2_SAS_DEVICE_INFO_SMP_TARGET) == 0) {
3163 			mpr_dprint(sc, MPR_ERROR, "%s: handle %d parent %d "
3164 			    "does not have an SMP target!\n", __func__,
3165 			    targ->handle, targ->parent_handle);
3166 			mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE);
3167 			goto bailout;
3168 
3169 		}
3170 		if (targ->parent_sasaddr == 0x0) {
3171 			mpr_dprint(sc, MPR_ERROR, "%s: handle %d parent handle "
3172 			    "%d does not have a valid SAS address!\n", __func__,
3173 			    targ->handle, targ->parent_handle);
3174 			mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE);
3175 			goto bailout;
3176 		}
3177 
3178 		sasaddr = targ->parent_sasaddr;
3179 #endif /* OLD_MPR_PROBE */
3180 
3181 	}
3182 
3183 	if (sasaddr == 0) {
3184 		mpr_dprint(sc, MPR_INFO, "%s: unable to find SAS address for "
3185 		    "handle %d\n", __func__, targ->handle);
3186 		mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE);
3187 		goto bailout;
3188 	}
3189 	mprsas_send_smpcmd(sassc, ccb, sasaddr);
3190 
3191 	return;
3192 
3193 bailout:
3194 	xpt_done(ccb);
3195 
3196 }
3197 
3198 static void
3199 mprsas_action_resetdev(struct mprsas_softc *sassc, union ccb *ccb)
3200 {
3201 	MPI2_SCSI_TASK_MANAGE_REQUEST *req;
3202 	struct mpr_softc *sc;
3203 	struct mpr_command *tm;
3204 	struct mprsas_target *targ;
3205 
3206 	MPR_FUNCTRACE(sassc->sc);
3207 	mtx_assert(&sassc->sc->mpr_mtx, MA_OWNED);
3208 
3209 	KASSERT(ccb->ccb_h.target_id < sassc->maxtargets, ("Target %d out of "
3210 	    "bounds in XPT_RESET_DEV\n", ccb->ccb_h.target_id));
3211 	sc = sassc->sc;
3212 	tm = mprsas_alloc_tm(sc);
3213 	if (tm == NULL) {
3214 		mpr_dprint(sc, MPR_ERROR, "command alloc failure in "
3215 		    "mprsas_action_resetdev\n");
3216 		mprsas_set_ccbstatus(ccb, CAM_RESRC_UNAVAIL);
3217 		xpt_done(ccb);
3218 		return;
3219 	}
3220 
3221 	targ = &sassc->targets[ccb->ccb_h.target_id];
3222 	req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req;
3223 	req->DevHandle = htole16(targ->handle);
3224 	req->TaskType = MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET;
3225 
3226 	if (!targ->is_nvme || sc->custom_nvme_tm_handling) {
3227 		/* SAS Hard Link Reset / SATA Link Reset */
3228 		req->MsgFlags = MPI2_SCSITASKMGMT_MSGFLAGS_LINK_RESET;
3229 	} else {
3230 		/* PCIe Protocol Level Reset*/
3231 		req->MsgFlags =
3232 		    MPI26_SCSITASKMGMT_MSGFLAGS_PROTOCOL_LVL_RST_PCIE;
3233 	}
3234 
3235 	tm->cm_data = NULL;
3236 	tm->cm_complete = mprsas_resetdev_complete;
3237 	tm->cm_complete_data = ccb;
3238 
3239 	mpr_dprint(sc, MPR_INFO, "%s: Sending reset for target ID %d\n",
3240 	    __func__, targ->tid);
3241 	tm->cm_targ = targ;
3242 
3243 	mprsas_prepare_for_tm(sc, tm, targ, CAM_LUN_WILDCARD);
3244 	mpr_map_command(sc, tm);
3245 }
3246 
3247 static void
3248 mprsas_resetdev_complete(struct mpr_softc *sc, struct mpr_command *tm)
3249 {
3250 	MPI2_SCSI_TASK_MANAGE_REPLY *resp;
3251 	union ccb *ccb;
3252 
3253 	MPR_FUNCTRACE(sc);
3254 	mtx_assert(&sc->mpr_mtx, MA_OWNED);
3255 
3256 	resp = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply;
3257 	ccb = tm->cm_complete_data;
3258 
3259 	/*
3260 	 * Currently there should be no way we can hit this case.  It only
3261 	 * happens when we have a failure to allocate chain frames, and
3262 	 * task management commands don't have S/G lists.
3263 	 */
3264 	if ((tm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) {
3265 		MPI2_SCSI_TASK_MANAGE_REQUEST *req;
3266 
3267 		req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req;
3268 
3269 		mpr_dprint(sc, MPR_ERROR, "%s: cm_flags = %#x for reset of "
3270 		    "handle %#04x! This should not happen!\n", __func__,
3271 		    tm->cm_flags, req->DevHandle);
3272 		mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR);
3273 		goto bailout;
3274 	}
3275 
3276 	mpr_dprint(sc, MPR_XINFO, "%s: IOCStatus = 0x%x ResponseCode = 0x%x\n",
3277 	    __func__, le16toh(resp->IOCStatus), le32toh(resp->ResponseCode));
3278 
3279 	if (le32toh(resp->ResponseCode) == MPI2_SCSITASKMGMT_RSP_TM_COMPLETE) {
3280 		mprsas_set_ccbstatus(ccb, CAM_REQ_CMP);
3281 		mprsas_announce_reset(sc, AC_SENT_BDR, tm->cm_targ->tid,
3282 		    CAM_LUN_WILDCARD);
3283 	}
3284 	else
3285 		mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR);
3286 
3287 bailout:
3288 
3289 	mprsas_free_tm(sc, tm);
3290 	xpt_done(ccb);
3291 }
3292 
3293 static void
3294 mprsas_poll(struct cam_sim *sim)
3295 {
3296 	struct mprsas_softc *sassc;
3297 
3298 	sassc = cam_sim_softc(sim);
3299 
3300 	if (sassc->sc->mpr_debug & MPR_TRACE) {
3301 		/* frequent debug messages during a panic just slow
3302 		 * everything down too much.
3303 		 */
3304 		mpr_dprint(sassc->sc, MPR_XINFO, "%s clearing MPR_TRACE\n",
3305 		    __func__);
3306 		sassc->sc->mpr_debug &= ~MPR_TRACE;
3307 	}
3308 
3309 	mpr_intr_locked(sassc->sc);
3310 }
3311 
3312 static void
3313 mprsas_async(void *callback_arg, uint32_t code, struct cam_path *path,
3314     void *arg)
3315 {
3316 	struct mpr_softc *sc;
3317 
3318 	sc = (struct mpr_softc *)callback_arg;
3319 
3320 	switch (code) {
3321 	case AC_ADVINFO_CHANGED: {
3322 		struct mprsas_target *target;
3323 		struct mprsas_softc *sassc;
3324 		struct scsi_read_capacity_data_long rcap_buf;
3325 		struct ccb_dev_advinfo cdai;
3326 		struct mprsas_lun *lun;
3327 		lun_id_t lunid;
3328 		int found_lun;
3329 		uintptr_t buftype;
3330 
3331 		buftype = (uintptr_t)arg;
3332 
3333 		found_lun = 0;
3334 		sassc = sc->sassc;
3335 
3336 		/*
3337 		 * We're only interested in read capacity data changes.
3338 		 */
3339 		if (buftype != CDAI_TYPE_RCAPLONG)
3340 			break;
3341 
3342 		/*
3343 		 * We should have a handle for this, but check to make sure.
3344 		 */
3345 		KASSERT(xpt_path_target_id(path) < sassc->maxtargets,
3346 		    ("Target %d out of bounds in mprsas_async\n",
3347 		    xpt_path_target_id(path)));
3348 		target = &sassc->targets[xpt_path_target_id(path)];
3349 		if (target->handle == 0)
3350 			break;
3351 
3352 		lunid = xpt_path_lun_id(path);
3353 
3354 		SLIST_FOREACH(lun, &target->luns, lun_link) {
3355 			if (lun->lun_id == lunid) {
3356 				found_lun = 1;
3357 				break;
3358 			}
3359 		}
3360 
3361 		if (found_lun == 0) {
3362 			lun = malloc(sizeof(struct mprsas_lun), M_MPR,
3363 			    M_NOWAIT | M_ZERO);
3364 			if (lun == NULL) {
3365 				mpr_dprint(sc, MPR_ERROR, "Unable to alloc "
3366 				    "LUN for EEDP support.\n");
3367 				break;
3368 			}
3369 			lun->lun_id = lunid;
3370 			SLIST_INSERT_HEAD(&target->luns, lun, lun_link);
3371 		}
3372 
3373 		bzero(&rcap_buf, sizeof(rcap_buf));
3374 		xpt_setup_ccb(&cdai.ccb_h, path, CAM_PRIORITY_NORMAL);
3375 		cdai.ccb_h.func_code = XPT_DEV_ADVINFO;
3376 		cdai.ccb_h.flags = CAM_DIR_IN;
3377 		cdai.buftype = CDAI_TYPE_RCAPLONG;
3378 		cdai.flags = CDAI_FLAG_NONE;
3379 		cdai.bufsiz = sizeof(rcap_buf);
3380 		cdai.buf = (uint8_t *)&rcap_buf;
3381 		xpt_action((union ccb *)&cdai);
3382 		if ((cdai.ccb_h.status & CAM_DEV_QFRZN) != 0)
3383 			cam_release_devq(cdai.ccb_h.path, 0, 0, 0, FALSE);
3384 
3385 		if ((mprsas_get_ccbstatus((union ccb *)&cdai) == CAM_REQ_CMP)
3386 		    && (rcap_buf.prot & SRC16_PROT_EN)) {
3387 			switch (rcap_buf.prot & SRC16_P_TYPE) {
3388 			case SRC16_PTYPE_1:
3389 			case SRC16_PTYPE_3:
3390 				lun->eedp_formatted = TRUE;
3391 				lun->eedp_block_size =
3392 				    scsi_4btoul(rcap_buf.length);
3393 				break;
3394 			case SRC16_PTYPE_2:
3395 			default:
3396 				lun->eedp_formatted = FALSE;
3397 				lun->eedp_block_size = 0;
3398 				break;
3399 			}
3400 		} else {
3401 			lun->eedp_formatted = FALSE;
3402 			lun->eedp_block_size = 0;
3403 		}
3404 		break;
3405 	}
3406 	case AC_FOUND_DEVICE:
3407 	default:
3408 		break;
3409 	}
3410 }
3411 
3412 /*
3413  * Set the INRESET flag for this target so that no I/O will be sent to
3414  * the target until the reset has completed.  If an I/O request does
3415  * happen, the devq will be frozen.  The CCB holds the path which is
3416  * used to release the devq.  The devq is released and the CCB is freed
3417  * when the TM completes.
3418  */
3419 void
3420 mprsas_prepare_for_tm(struct mpr_softc *sc, struct mpr_command *tm,
3421     struct mprsas_target *target, lun_id_t lun_id)
3422 {
3423 	union ccb *ccb;
3424 	path_id_t path_id;
3425 
3426 	ccb = xpt_alloc_ccb_nowait();
3427 	if (ccb) {
3428 		path_id = cam_sim_path(sc->sassc->sim);
3429 		if (xpt_create_path(&ccb->ccb_h.path, xpt_periph, path_id,
3430 		    target->tid, lun_id) != CAM_REQ_CMP) {
3431 			xpt_free_ccb(ccb);
3432 		} else {
3433 			tm->cm_ccb = ccb;
3434 			tm->cm_targ = target;
3435 			target->flags |= MPRSAS_TARGET_INRESET;
3436 		}
3437 	}
3438 }
3439 
3440 int
3441 mprsas_startup(struct mpr_softc *sc)
3442 {
3443 	/*
3444 	 * Send the port enable message and set the wait_for_port_enable flag.
3445 	 * This flag helps to keep the simq frozen until all discovery events
3446 	 * are processed.
3447 	 */
3448 	sc->wait_for_port_enable = 1;
3449 	mprsas_send_portenable(sc);
3450 	return (0);
3451 }
3452 
3453 static int
3454 mprsas_send_portenable(struct mpr_softc *sc)
3455 {
3456 	MPI2_PORT_ENABLE_REQUEST *request;
3457 	struct mpr_command *cm;
3458 
3459 	MPR_FUNCTRACE(sc);
3460 
3461 	if ((cm = mpr_alloc_command(sc)) == NULL)
3462 		return (EBUSY);
3463 	request = (MPI2_PORT_ENABLE_REQUEST *)cm->cm_req;
3464 	request->Function = MPI2_FUNCTION_PORT_ENABLE;
3465 	request->MsgFlags = 0;
3466 	request->VP_ID = 0;
3467 	cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE;
3468 	cm->cm_complete = mprsas_portenable_complete;
3469 	cm->cm_data = NULL;
3470 	cm->cm_sge = NULL;
3471 
3472 	mpr_map_command(sc, cm);
3473 	mpr_dprint(sc, MPR_XINFO,
3474 	    "mpr_send_portenable finished cm %p req %p complete %p\n",
3475 	    cm, cm->cm_req, cm->cm_complete);
3476 	return (0);
3477 }
3478 
3479 static void
3480 mprsas_portenable_complete(struct mpr_softc *sc, struct mpr_command *cm)
3481 {
3482 	MPI2_PORT_ENABLE_REPLY *reply;
3483 	struct mprsas_softc *sassc;
3484 
3485 	MPR_FUNCTRACE(sc);
3486 	sassc = sc->sassc;
3487 
3488 	/*
3489 	 * Currently there should be no way we can hit this case.  It only
3490 	 * happens when we have a failure to allocate chain frames, and
3491 	 * port enable commands don't have S/G lists.
3492 	 */
3493 	if ((cm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) {
3494 		mpr_dprint(sc, MPR_ERROR, "%s: cm_flags = %#x for port enable! "
3495 		    "This should not happen!\n", __func__, cm->cm_flags);
3496 	}
3497 
3498 	reply = (MPI2_PORT_ENABLE_REPLY *)cm->cm_reply;
3499 	if (reply == NULL)
3500 		mpr_dprint(sc, MPR_FAULT, "Portenable NULL reply\n");
3501 	else if (le16toh(reply->IOCStatus & MPI2_IOCSTATUS_MASK) !=
3502 	    MPI2_IOCSTATUS_SUCCESS)
3503 		mpr_dprint(sc, MPR_FAULT, "Portenable failed\n");
3504 
3505 	mpr_free_command(sc, cm);
3506 	/*
3507 	 * Done waiting for port enable to complete.  Decrement the refcount.
3508 	 * If refcount is 0, discovery is complete and a rescan of the bus can
3509 	 * take place.
3510 	 */
3511 	sc->wait_for_port_enable = 0;
3512 	sc->port_enable_complete = 1;
3513 	wakeup(&sc->port_enable_complete);
3514 	mprsas_startup_decrement(sassc);
3515 }
3516 
3517 int
3518 mprsas_check_id(struct mprsas_softc *sassc, int id)
3519 {
3520 	struct mpr_softc *sc = sassc->sc;
3521 	char *ids;
3522 	char *name;
3523 
3524 	ids = &sc->exclude_ids[0];
3525 	while((name = strsep(&ids, ",")) != NULL) {
3526 		if (name[0] == '\0')
3527 			continue;
3528 		if (strtol(name, NULL, 0) == (long)id)
3529 			return (1);
3530 	}
3531 
3532 	return (0);
3533 }
3534 
3535 void
3536 mprsas_realloc_targets(struct mpr_softc *sc, int maxtargets)
3537 {
3538 	struct mprsas_softc *sassc;
3539 	struct mprsas_lun *lun, *lun_tmp;
3540 	struct mprsas_target *targ;
3541 	int i;
3542 
3543 	sassc = sc->sassc;
3544 	/*
3545 	 * The number of targets is based on IOC Facts, so free all of
3546 	 * the allocated LUNs for each target and then the target buffer
3547 	 * itself.
3548 	 */
3549 	for (i=0; i< maxtargets; i++) {
3550 		targ = &sassc->targets[i];
3551 		SLIST_FOREACH_SAFE(lun, &targ->luns, lun_link, lun_tmp) {
3552 			free(lun, M_MPR);
3553 		}
3554 	}
3555 	free(sassc->targets, M_MPR);
3556 
3557 	sassc->targets = malloc(sizeof(struct mprsas_target) * maxtargets,
3558 	    M_MPR, M_WAITOK|M_ZERO);
3559 	if (!sassc->targets) {
3560 		panic("%s failed to alloc targets with error %d\n",
3561 		    __func__, ENOMEM);
3562 	}
3563 }
3564