xref: /freebsd/sys/dev/mps/mps.c (revision 430f7286a566b1407c7b32ce13585caf5aa59b92)
1 /*-
2  * Copyright (c) 2009 Yahoo! Inc.
3  * Copyright (c) 2011-2015 LSI Corp.
4  * Copyright (c) 2013-2015 Avago Technologies
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  * Avago Technologies (LSI) MPT-Fusion Host Adapter FreeBSD
29  *
30  * $FreeBSD$
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 /* Communications core for Avago Technologies (LSI) MPT2 */
37 
38 /* TODO Move headers to mpsvar */
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/lock.h>
45 #include <sys/mutex.h>
46 #include <sys/module.h>
47 #include <sys/bus.h>
48 #include <sys/conf.h>
49 #include <sys/bio.h>
50 #include <sys/malloc.h>
51 #include <sys/uio.h>
52 #include <sys/sysctl.h>
53 #include <sys/queue.h>
54 #include <sys/kthread.h>
55 #include <sys/taskqueue.h>
56 #include <sys/endian.h>
57 #include <sys/eventhandler.h>
58 
59 #include <machine/bus.h>
60 #include <machine/resource.h>
61 #include <sys/rman.h>
62 #include <sys/proc.h>
63 
64 #include <dev/pci/pcivar.h>
65 
66 #include <cam/cam.h>
67 #include <cam/scsi/scsi_all.h>
68 
69 #include <dev/mps/mpi/mpi2_type.h>
70 #include <dev/mps/mpi/mpi2.h>
71 #include <dev/mps/mpi/mpi2_ioc.h>
72 #include <dev/mps/mpi/mpi2_sas.h>
73 #include <dev/mps/mpi/mpi2_cnfg.h>
74 #include <dev/mps/mpi/mpi2_init.h>
75 #include <dev/mps/mpi/mpi2_tool.h>
76 #include <dev/mps/mps_ioctl.h>
77 #include <dev/mps/mpsvar.h>
78 #include <dev/mps/mps_table.h>
79 
80 static int mps_diag_reset(struct mps_softc *sc, int sleep_flag);
81 static int mps_init_queues(struct mps_softc *sc);
82 static int mps_message_unit_reset(struct mps_softc *sc, int sleep_flag);
83 static int mps_transition_operational(struct mps_softc *sc);
84 static int mps_iocfacts_allocate(struct mps_softc *sc, uint8_t attaching);
85 static void mps_iocfacts_free(struct mps_softc *sc);
86 static void mps_startup(void *arg);
87 static int mps_send_iocinit(struct mps_softc *sc);
88 static int mps_alloc_queues(struct mps_softc *sc);
89 static int mps_alloc_replies(struct mps_softc *sc);
90 static int mps_alloc_requests(struct mps_softc *sc);
91 static int mps_attach_log(struct mps_softc *sc);
92 static __inline void mps_complete_command(struct mps_softc *sc,
93     struct mps_command *cm);
94 static void mps_dispatch_event(struct mps_softc *sc, uintptr_t data,
95     MPI2_EVENT_NOTIFICATION_REPLY *reply);
96 static void mps_config_complete(struct mps_softc *sc, struct mps_command *cm);
97 static void mps_periodic(void *);
98 static int mps_reregister_events(struct mps_softc *sc);
99 static void mps_enqueue_request(struct mps_softc *sc, struct mps_command *cm);
100 static int mps_get_iocfacts(struct mps_softc *sc, MPI2_IOC_FACTS_REPLY *facts);
101 static int mps_wait_db_ack(struct mps_softc *sc, int timeout, int sleep_flag);
102 SYSCTL_NODE(_hw, OID_AUTO, mps, CTLFLAG_RD, 0, "MPS Driver Parameters");
103 
104 MALLOC_DEFINE(M_MPT2, "mps", "mpt2 driver memory");
105 
106 /*
107  * Do a "Diagnostic Reset" aka a hard reset.  This should get the chip out of
108  * any state and back to its initialization state machine.
109  */
110 static char mpt2_reset_magic[] = { 0x00, 0x0f, 0x04, 0x0b, 0x02, 0x07, 0x0d };
111 
112 /* Added this union to smoothly convert le64toh cm->cm_desc.Words.
113  * Compiler only support unint64_t to be passed as argument.
114  * Otherwise it will through below error
115  * "aggregate value used where an integer was expected"
116  */
117 
118 typedef union _reply_descriptor {
119         u64 word;
120         struct {
121                 u32 low;
122                 u32 high;
123         } u;
124 }reply_descriptor,address_descriptor;
125 
126 /* Rate limit chain-fail messages to 1 per minute */
127 static struct timeval mps_chainfail_interval = { 60, 0 };
128 
129 /*
130  * sleep_flag can be either CAN_SLEEP or NO_SLEEP.
131  * If this function is called from process context, it can sleep
132  * and there is no harm to sleep, in case if this fuction is called
133  * from Interrupt handler, we can not sleep and need NO_SLEEP flag set.
134  * based on sleep flags driver will call either msleep, pause or DELAY.
135  * msleep and pause are of same variant, but pause is used when mps_mtx
136  * is not hold by driver.
137  *
138  */
139 static int
140 mps_diag_reset(struct mps_softc *sc,int sleep_flag)
141 {
142 	uint32_t reg;
143 	int i, error, tries = 0;
144 	uint8_t first_wait_done = FALSE;
145 
146 	mps_dprint(sc, MPS_TRACE, "%s\n", __func__);
147 
148 	/* Clear any pending interrupts */
149 	mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0);
150 
151 	/*Force NO_SLEEP for threads prohibited to sleep
152  	* e.a Thread from interrupt handler are prohibited to sleep.
153  	*/
154 	if (curthread->td_no_sleeping != 0)
155 		sleep_flag = NO_SLEEP;
156 
157 	/* Push the magic sequence */
158 	error = ETIMEDOUT;
159 	while (tries++ < 20) {
160 		for (i = 0; i < sizeof(mpt2_reset_magic); i++)
161 			mps_regwrite(sc, MPI2_WRITE_SEQUENCE_OFFSET,
162 			    mpt2_reset_magic[i]);
163 		/* wait 100 msec */
164 		if (mtx_owned(&sc->mps_mtx) && sleep_flag == CAN_SLEEP)
165 			msleep(&sc->msleep_fake_chan, &sc->mps_mtx, 0,
166 			    "mpsdiag", hz/10);
167 		else if (sleep_flag == CAN_SLEEP)
168 			pause("mpsdiag", hz/10);
169 		else
170 			DELAY(100 * 1000);
171 
172 		reg = mps_regread(sc, MPI2_HOST_DIAGNOSTIC_OFFSET);
173 		if (reg & MPI2_DIAG_DIAG_WRITE_ENABLE) {
174 			error = 0;
175 			break;
176 		}
177 	}
178 	if (error)
179 		return (error);
180 
181 	/* Send the actual reset.  XXX need to refresh the reg? */
182 	mps_regwrite(sc, MPI2_HOST_DIAGNOSTIC_OFFSET,
183 	    reg | MPI2_DIAG_RESET_ADAPTER);
184 
185 	/* Wait up to 300 seconds in 50ms intervals */
186 	error = ETIMEDOUT;
187 	for (i = 0; i < 6000; i++) {
188 		/*
189 		 * Wait 50 msec. If this is the first time through, wait 256
190 		 * msec to satisfy Diag Reset timing requirements.
191 		 */
192 		if (first_wait_done) {
193 			if (mtx_owned(&sc->mps_mtx) && sleep_flag == CAN_SLEEP)
194 				msleep(&sc->msleep_fake_chan, &sc->mps_mtx, 0,
195 				    "mpsdiag", hz/20);
196 			else if (sleep_flag == CAN_SLEEP)
197 				pause("mpsdiag", hz/20);
198 			else
199 				DELAY(50 * 1000);
200 		} else {
201 			DELAY(256 * 1000);
202 			first_wait_done = TRUE;
203 		}
204 		/*
205 		 * Check for the RESET_ADAPTER bit to be cleared first, then
206 		 * wait for the RESET state to be cleared, which takes a little
207 		 * longer.
208 		 */
209 		reg = mps_regread(sc, MPI2_HOST_DIAGNOSTIC_OFFSET);
210 		if (reg & MPI2_DIAG_RESET_ADAPTER) {
211 			continue;
212 		}
213 		reg = mps_regread(sc, MPI2_DOORBELL_OFFSET);
214 		if ((reg & MPI2_IOC_STATE_MASK) != MPI2_IOC_STATE_RESET) {
215 			error = 0;
216 			break;
217 		}
218 	}
219 	if (error)
220 		return (error);
221 
222 	mps_regwrite(sc, MPI2_WRITE_SEQUENCE_OFFSET, 0x0);
223 
224 	return (0);
225 }
226 
227 static int
228 mps_message_unit_reset(struct mps_softc *sc, int sleep_flag)
229 {
230 
231 	MPS_FUNCTRACE(sc);
232 
233 	mps_regwrite(sc, MPI2_DOORBELL_OFFSET,
234 	    MPI2_FUNCTION_IOC_MESSAGE_UNIT_RESET <<
235 	    MPI2_DOORBELL_FUNCTION_SHIFT);
236 
237 	if (mps_wait_db_ack(sc, 5, sleep_flag) != 0) {
238 		mps_dprint(sc, MPS_FAULT, "Doorbell handshake failed : <%s>\n",
239 				__func__);
240 		return (ETIMEDOUT);
241 	}
242 
243 	return (0);
244 }
245 
246 static int
247 mps_transition_ready(struct mps_softc *sc)
248 {
249 	uint32_t reg, state;
250 	int error, tries = 0;
251 	int sleep_flags;
252 
253 	MPS_FUNCTRACE(sc);
254 	/* If we are in attach call, do not sleep */
255 	sleep_flags = (sc->mps_flags & MPS_FLAGS_ATTACH_DONE)
256 					? CAN_SLEEP:NO_SLEEP;
257 	error = 0;
258 	while (tries++ < 1200) {
259 		reg = mps_regread(sc, MPI2_DOORBELL_OFFSET);
260 		mps_dprint(sc, MPS_INIT, "Doorbell= 0x%x\n", reg);
261 
262 		/*
263 		 * Ensure the IOC is ready to talk.  If it's not, try
264 		 * resetting it.
265 		 */
266 		if (reg & MPI2_DOORBELL_USED) {
267 			mps_diag_reset(sc, sleep_flags);
268 			DELAY(50000);
269 			continue;
270 		}
271 
272 		/* Is the adapter owned by another peer? */
273 		if ((reg & MPI2_DOORBELL_WHO_INIT_MASK) ==
274 		    (MPI2_WHOINIT_PCI_PEER << MPI2_DOORBELL_WHO_INIT_SHIFT)) {
275 			device_printf(sc->mps_dev, "IOC is under the control "
276 			    "of another peer host, aborting initialization.\n");
277 			return (ENXIO);
278 		}
279 
280 		state = reg & MPI2_IOC_STATE_MASK;
281 		if (state == MPI2_IOC_STATE_READY) {
282 			/* Ready to go! */
283 			error = 0;
284 			break;
285 		} else if (state == MPI2_IOC_STATE_FAULT) {
286 			mps_dprint(sc, MPS_FAULT, "IOC in fault state 0x%x, resetting\n",
287 			    state & MPI2_DOORBELL_FAULT_CODE_MASK);
288 			mps_diag_reset(sc, sleep_flags);
289 		} else if (state == MPI2_IOC_STATE_OPERATIONAL) {
290 			/* Need to take ownership */
291 			mps_message_unit_reset(sc, sleep_flags);
292 		} else if (state == MPI2_IOC_STATE_RESET) {
293 			/* Wait a bit, IOC might be in transition */
294 			mps_dprint(sc, MPS_FAULT,
295 			    "IOC in unexpected reset state\n");
296 		} else {
297 			mps_dprint(sc, MPS_FAULT,
298 			    "IOC in unknown state 0x%x\n", state);
299 			error = EINVAL;
300 			break;
301 		}
302 
303 		/* Wait 50ms for things to settle down. */
304 		DELAY(50000);
305 	}
306 
307 	if (error)
308 		device_printf(sc->mps_dev, "Cannot transition IOC to ready\n");
309 
310 	return (error);
311 }
312 
313 static int
314 mps_transition_operational(struct mps_softc *sc)
315 {
316 	uint32_t reg, state;
317 	int error;
318 
319 	MPS_FUNCTRACE(sc);
320 
321 	error = 0;
322 	reg = mps_regread(sc, MPI2_DOORBELL_OFFSET);
323 	mps_dprint(sc, MPS_INIT, "Doorbell= 0x%x\n", reg);
324 
325 	state = reg & MPI2_IOC_STATE_MASK;
326 	if (state != MPI2_IOC_STATE_READY) {
327 		if ((error = mps_transition_ready(sc)) != 0) {
328 			mps_dprint(sc, MPS_FAULT,
329 			    "%s failed to transition ready\n", __func__);
330 			return (error);
331 		}
332 	}
333 
334 	error = mps_send_iocinit(sc);
335 	return (error);
336 }
337 
338 /*
339  * This is called during attach and when re-initializing due to a Diag Reset.
340  * IOC Facts is used to allocate many of the structures needed by the driver.
341  * If called from attach, de-allocation is not required because the driver has
342  * not allocated any structures yet, but if called from a Diag Reset, previously
343  * allocated structures based on IOC Facts will need to be freed and re-
344  * allocated bases on the latest IOC Facts.
345  */
346 static int
347 mps_iocfacts_allocate(struct mps_softc *sc, uint8_t attaching)
348 {
349 	int error;
350 	Mpi2IOCFactsReply_t saved_facts;
351 	uint8_t saved_mode, reallocating;
352 
353 	mps_dprint(sc, MPS_TRACE, "%s\n", __func__);
354 
355 	/* Save old IOC Facts and then only reallocate if Facts have changed */
356 	if (!attaching) {
357 		bcopy(sc->facts, &saved_facts, sizeof(MPI2_IOC_FACTS_REPLY));
358 	}
359 
360 	/*
361 	 * Get IOC Facts.  In all cases throughout this function, panic if doing
362 	 * a re-initialization and only return the error if attaching so the OS
363 	 * can handle it.
364 	 */
365 	if ((error = mps_get_iocfacts(sc, sc->facts)) != 0) {
366 		if (attaching) {
367 			mps_dprint(sc, MPS_FAULT, "%s failed to get IOC Facts "
368 			    "with error %d\n", __func__, error);
369 			return (error);
370 		} else {
371 			panic("%s failed to get IOC Facts with error %d\n",
372 			    __func__, error);
373 		}
374 	}
375 
376 	mps_print_iocfacts(sc, sc->facts);
377 
378 	snprintf(sc->fw_version, sizeof(sc->fw_version),
379 	    "%02d.%02d.%02d.%02d",
380 	    sc->facts->FWVersion.Struct.Major,
381 	    sc->facts->FWVersion.Struct.Minor,
382 	    sc->facts->FWVersion.Struct.Unit,
383 	    sc->facts->FWVersion.Struct.Dev);
384 
385 	mps_printf(sc, "Firmware: %s, Driver: %s\n", sc->fw_version,
386 	    MPS_DRIVER_VERSION);
387 	mps_printf(sc, "IOCCapabilities: %b\n", sc->facts->IOCCapabilities,
388 	    "\20" "\3ScsiTaskFull" "\4DiagTrace" "\5SnapBuf" "\6ExtBuf"
389 	    "\7EEDP" "\10BiDirTarg" "\11Multicast" "\14TransRetry" "\15IR"
390 	    "\16EventReplay" "\17RaidAccel" "\20MSIXIndex" "\21HostDisc");
391 
392 	/*
393 	 * If the chip doesn't support event replay then a hard reset will be
394 	 * required to trigger a full discovery.  Do the reset here then
395 	 * retransition to Ready.  A hard reset might have already been done,
396 	 * but it doesn't hurt to do it again.  Only do this if attaching, not
397 	 * for a Diag Reset.
398 	 */
399 	if (attaching) {
400 		if ((sc->facts->IOCCapabilities &
401 		    MPI2_IOCFACTS_CAPABILITY_EVENT_REPLAY) == 0) {
402 			mps_diag_reset(sc, NO_SLEEP);
403 			if ((error = mps_transition_ready(sc)) != 0) {
404 				mps_dprint(sc, MPS_FAULT, "%s failed to "
405 				    "transition to ready with error %d\n",
406 				    __func__, error);
407 				return (error);
408 			}
409 		}
410 	}
411 
412 	/*
413 	 * Set flag if IR Firmware is loaded.  If the RAID Capability has
414 	 * changed from the previous IOC Facts, log a warning, but only if
415 	 * checking this after a Diag Reset and not during attach.
416 	 */
417 	saved_mode = sc->ir_firmware;
418 	if (sc->facts->IOCCapabilities &
419 	    MPI2_IOCFACTS_CAPABILITY_INTEGRATED_RAID)
420 		sc->ir_firmware = 1;
421 	if (!attaching) {
422 		if (sc->ir_firmware != saved_mode) {
423 			mps_dprint(sc, MPS_FAULT, "%s new IR/IT mode in IOC "
424 			    "Facts does not match previous mode\n", __func__);
425 		}
426 	}
427 
428 	/* Only deallocate and reallocate if relevant IOC Facts have changed */
429 	reallocating = FALSE;
430 	if ((!attaching) &&
431 	    ((saved_facts.MsgVersion != sc->facts->MsgVersion) ||
432 	    (saved_facts.HeaderVersion != sc->facts->HeaderVersion) ||
433 	    (saved_facts.MaxChainDepth != sc->facts->MaxChainDepth) ||
434 	    (saved_facts.RequestCredit != sc->facts->RequestCredit) ||
435 	    (saved_facts.ProductID != sc->facts->ProductID) ||
436 	    (saved_facts.IOCCapabilities != sc->facts->IOCCapabilities) ||
437 	    (saved_facts.IOCRequestFrameSize !=
438 	    sc->facts->IOCRequestFrameSize) ||
439 	    (saved_facts.MaxTargets != sc->facts->MaxTargets) ||
440 	    (saved_facts.MaxSasExpanders != sc->facts->MaxSasExpanders) ||
441 	    (saved_facts.MaxEnclosures != sc->facts->MaxEnclosures) ||
442 	    (saved_facts.HighPriorityCredit != sc->facts->HighPriorityCredit) ||
443 	    (saved_facts.MaxReplyDescriptorPostQueueDepth !=
444 	    sc->facts->MaxReplyDescriptorPostQueueDepth) ||
445 	    (saved_facts.ReplyFrameSize != sc->facts->ReplyFrameSize) ||
446 	    (saved_facts.MaxVolumes != sc->facts->MaxVolumes) ||
447 	    (saved_facts.MaxPersistentEntries !=
448 	    sc->facts->MaxPersistentEntries))) {
449 		reallocating = TRUE;
450 	}
451 
452 	/*
453 	 * Some things should be done if attaching or re-allocating after a Diag
454 	 * Reset, but are not needed after a Diag Reset if the FW has not
455 	 * changed.
456 	 */
457 	if (attaching || reallocating) {
458 		/*
459 		 * Check if controller supports FW diag buffers and set flag to
460 		 * enable each type.
461 		 */
462 		if (sc->facts->IOCCapabilities &
463 		    MPI2_IOCFACTS_CAPABILITY_DIAG_TRACE_BUFFER)
464 			sc->fw_diag_buffer_list[MPI2_DIAG_BUF_TYPE_TRACE].
465 			    enabled = TRUE;
466 		if (sc->facts->IOCCapabilities &
467 		    MPI2_IOCFACTS_CAPABILITY_SNAPSHOT_BUFFER)
468 			sc->fw_diag_buffer_list[MPI2_DIAG_BUF_TYPE_SNAPSHOT].
469 			    enabled = TRUE;
470 		if (sc->facts->IOCCapabilities &
471 		    MPI2_IOCFACTS_CAPABILITY_EXTENDED_BUFFER)
472 			sc->fw_diag_buffer_list[MPI2_DIAG_BUF_TYPE_EXTENDED].
473 			    enabled = TRUE;
474 
475 		/*
476 		 * Set flag if EEDP is supported and if TLR is supported.
477 		 */
478 		if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_EEDP)
479 			sc->eedp_enabled = TRUE;
480 		if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_TLR)
481 			sc->control_TLR = TRUE;
482 
483 		/*
484 		 * Size the queues. Since the reply queues always need one free
485 		 * entry, we'll just deduct one reply message here.
486 		 */
487 		sc->num_reqs = MIN(MPS_REQ_FRAMES, sc->facts->RequestCredit);
488 		sc->num_replies = MIN(MPS_REPLY_FRAMES + MPS_EVT_REPLY_FRAMES,
489 		    sc->facts->MaxReplyDescriptorPostQueueDepth) - 1;
490 
491 		/*
492 		 * Initialize all Tail Queues
493 		 */
494 		TAILQ_INIT(&sc->req_list);
495 		TAILQ_INIT(&sc->high_priority_req_list);
496 		TAILQ_INIT(&sc->chain_list);
497 		TAILQ_INIT(&sc->tm_list);
498 	}
499 
500 	/*
501 	 * If doing a Diag Reset and the FW is significantly different
502 	 * (reallocating will be set above in IOC Facts comparison), then all
503 	 * buffers based on the IOC Facts will need to be freed before they are
504 	 * reallocated.
505 	 */
506 	if (reallocating) {
507 		mps_iocfacts_free(sc);
508 		mpssas_realloc_targets(sc, saved_facts.MaxTargets);
509 	}
510 
511 	/*
512 	 * Any deallocation has been completed.  Now start reallocating
513 	 * if needed.  Will only need to reallocate if attaching or if the new
514 	 * IOC Facts are different from the previous IOC Facts after a Diag
515 	 * Reset. Targets have already been allocated above if needed.
516 	 */
517 	if (attaching || reallocating) {
518 		if (((error = mps_alloc_queues(sc)) != 0) ||
519 		    ((error = mps_alloc_replies(sc)) != 0) ||
520 		    ((error = mps_alloc_requests(sc)) != 0)) {
521 			if (attaching ) {
522 				mps_dprint(sc, MPS_FAULT, "%s failed to alloc "
523 				    "queues with error %d\n", __func__, error);
524 				mps_free(sc);
525 				return (error);
526 			} else {
527 				panic("%s failed to alloc queues with error "
528 				    "%d\n", __func__, error);
529 			}
530 		}
531 	}
532 
533 	/* Always initialize the queues */
534 	bzero(sc->free_queue, sc->fqdepth * 4);
535 	mps_init_queues(sc);
536 
537 	/*
538 	 * Always get the chip out of the reset state, but only panic if not
539 	 * attaching.  If attaching and there is an error, that is handled by
540 	 * the OS.
541 	 */
542 	error = mps_transition_operational(sc);
543 	if (error != 0) {
544 		if (attaching) {
545 			mps_printf(sc, "%s failed to transition to operational "
546 			    "with error %d\n", __func__, error);
547 			mps_free(sc);
548 			return (error);
549 		} else {
550 			panic("%s failed to transition to operational with "
551 			    "error %d\n", __func__, error);
552 		}
553 	}
554 
555 	/*
556 	 * Finish the queue initialization.
557 	 * These are set here instead of in mps_init_queues() because the
558 	 * IOC resets these values during the state transition in
559 	 * mps_transition_operational().  The free index is set to 1
560 	 * because the corresponding index in the IOC is set to 0, and the
561 	 * IOC treats the queues as full if both are set to the same value.
562 	 * Hence the reason that the queue can't hold all of the possible
563 	 * replies.
564 	 */
565 	sc->replypostindex = 0;
566 	mps_regwrite(sc, MPI2_REPLY_FREE_HOST_INDEX_OFFSET, sc->replyfreeindex);
567 	mps_regwrite(sc, MPI2_REPLY_POST_HOST_INDEX_OFFSET, 0);
568 
569 	/*
570 	 * Attach the subsystems so they can prepare their event masks.
571 	 */
572 	/* XXX Should be dynamic so that IM/IR and user modules can attach */
573 	if (attaching) {
574 		if (((error = mps_attach_log(sc)) != 0) ||
575 		    ((error = mps_attach_sas(sc)) != 0) ||
576 		    ((error = mps_attach_user(sc)) != 0)) {
577 			mps_printf(sc, "%s failed to attach all subsystems: "
578 			    "error %d\n", __func__, error);
579 			mps_free(sc);
580 			return (error);
581 		}
582 
583 		if ((error = mps_pci_setup_interrupts(sc)) != 0) {
584 			mps_printf(sc, "%s failed to setup interrupts\n",
585 			    __func__);
586 			mps_free(sc);
587 			return (error);
588 		}
589 	}
590 
591 	/*
592 	 * Set flag if this is a WD controller.  This shouldn't ever change, but
593 	 * reset it after a Diag Reset, just in case.
594 	 */
595 	sc->WD_available = FALSE;
596 	if (pci_get_device(sc->mps_dev) == MPI2_MFGPAGE_DEVID_SSS6200)
597 		sc->WD_available = TRUE;
598 
599 	return (error);
600 }
601 
602 /*
603  * This is called if memory is being free (during detach for example) and when
604  * buffers need to be reallocated due to a Diag Reset.
605  */
606 static void
607 mps_iocfacts_free(struct mps_softc *sc)
608 {
609 	struct mps_command *cm;
610 	int i;
611 
612 	mps_dprint(sc, MPS_TRACE, "%s\n", __func__);
613 
614 	if (sc->free_busaddr != 0)
615 		bus_dmamap_unload(sc->queues_dmat, sc->queues_map);
616 	if (sc->free_queue != NULL)
617 		bus_dmamem_free(sc->queues_dmat, sc->free_queue,
618 		    sc->queues_map);
619 	if (sc->queues_dmat != NULL)
620 		bus_dma_tag_destroy(sc->queues_dmat);
621 
622 	if (sc->chain_busaddr != 0)
623 		bus_dmamap_unload(sc->chain_dmat, sc->chain_map);
624 	if (sc->chain_frames != NULL)
625 		bus_dmamem_free(sc->chain_dmat, sc->chain_frames,
626 		    sc->chain_map);
627 	if (sc->chain_dmat != NULL)
628 		bus_dma_tag_destroy(sc->chain_dmat);
629 
630 	if (sc->sense_busaddr != 0)
631 		bus_dmamap_unload(sc->sense_dmat, sc->sense_map);
632 	if (sc->sense_frames != NULL)
633 		bus_dmamem_free(sc->sense_dmat, sc->sense_frames,
634 		    sc->sense_map);
635 	if (sc->sense_dmat != NULL)
636 		bus_dma_tag_destroy(sc->sense_dmat);
637 
638 	if (sc->reply_busaddr != 0)
639 		bus_dmamap_unload(sc->reply_dmat, sc->reply_map);
640 	if (sc->reply_frames != NULL)
641 		bus_dmamem_free(sc->reply_dmat, sc->reply_frames,
642 		    sc->reply_map);
643 	if (sc->reply_dmat != NULL)
644 		bus_dma_tag_destroy(sc->reply_dmat);
645 
646 	if (sc->req_busaddr != 0)
647 		bus_dmamap_unload(sc->req_dmat, sc->req_map);
648 	if (sc->req_frames != NULL)
649 		bus_dmamem_free(sc->req_dmat, sc->req_frames, sc->req_map);
650 	if (sc->req_dmat != NULL)
651 		bus_dma_tag_destroy(sc->req_dmat);
652 
653 	if (sc->chains != NULL)
654 		free(sc->chains, M_MPT2);
655 	if (sc->commands != NULL) {
656 		for (i = 1; i < sc->num_reqs; i++) {
657 			cm = &sc->commands[i];
658 			bus_dmamap_destroy(sc->buffer_dmat, cm->cm_dmamap);
659 		}
660 		free(sc->commands, M_MPT2);
661 	}
662 	if (sc->buffer_dmat != NULL)
663 		bus_dma_tag_destroy(sc->buffer_dmat);
664 }
665 
666 /*
667  * The terms diag reset and hard reset are used interchangeably in the MPI
668  * docs to mean resetting the controller chip.  In this code diag reset
669  * cleans everything up, and the hard reset function just sends the reset
670  * sequence to the chip.  This should probably be refactored so that every
671  * subsystem gets a reset notification of some sort, and can clean up
672  * appropriately.
673  */
674 int
675 mps_reinit(struct mps_softc *sc)
676 {
677 	int error;
678 	struct mpssas_softc *sassc;
679 
680 	sassc = sc->sassc;
681 
682 	MPS_FUNCTRACE(sc);
683 
684 	mtx_assert(&sc->mps_mtx, MA_OWNED);
685 
686 	if (sc->mps_flags & MPS_FLAGS_DIAGRESET) {
687 		mps_dprint(sc, MPS_INIT, "%s reset already in progress\n",
688 			   __func__);
689 		return 0;
690 	}
691 
692 	mps_dprint(sc, MPS_INFO, "Reinitializing controller,\n");
693 	/* make sure the completion callbacks can recognize they're getting
694 	 * a NULL cm_reply due to a reset.
695 	 */
696 	sc->mps_flags |= MPS_FLAGS_DIAGRESET;
697 
698 	/*
699 	 * Mask interrupts here.
700 	 */
701 	mps_dprint(sc, MPS_INIT, "%s mask interrupts\n", __func__);
702 	mps_mask_intr(sc);
703 
704 	error = mps_diag_reset(sc, CAN_SLEEP);
705 	if (error != 0) {
706 		/* XXXSL No need to panic here */
707 		panic("%s hard reset failed with error %d\n",
708 		    __func__, error);
709 	}
710 
711 	/* Restore the PCI state, including the MSI-X registers */
712 	mps_pci_restore(sc);
713 
714 	/* Give the I/O subsystem special priority to get itself prepared */
715 	mpssas_handle_reinit(sc);
716 
717 	/*
718 	 * Get IOC Facts and allocate all structures based on this information.
719 	 * The attach function will also call mps_iocfacts_allocate at startup.
720 	 * If relevant values have changed in IOC Facts, this function will free
721 	 * all of the memory based on IOC Facts and reallocate that memory.
722 	 */
723 	if ((error = mps_iocfacts_allocate(sc, FALSE)) != 0) {
724 		panic("%s IOC Facts based allocation failed with error %d\n",
725 		    __func__, error);
726 	}
727 
728 	/*
729 	 * Mapping structures will be re-allocated after getting IOC Page8, so
730 	 * free these structures here.
731 	 */
732 	mps_mapping_exit(sc);
733 
734 	/*
735 	 * The static page function currently read is IOC Page8.  Others can be
736 	 * added in future.  It's possible that the values in IOC Page8 have
737 	 * changed after a Diag Reset due to user modification, so always read
738 	 * these.  Interrupts are masked, so unmask them before getting config
739 	 * pages.
740 	 */
741 	mps_unmask_intr(sc);
742 	sc->mps_flags &= ~MPS_FLAGS_DIAGRESET;
743 	mps_base_static_config_pages(sc);
744 
745 	/*
746 	 * Some mapping info is based in IOC Page8 data, so re-initialize the
747 	 * mapping tables.
748 	 */
749 	mps_mapping_initialize(sc);
750 
751 	/*
752 	 * Restart will reload the event masks clobbered by the reset, and
753 	 * then enable the port.
754 	 */
755 	mps_reregister_events(sc);
756 
757 	/* the end of discovery will release the simq, so we're done. */
758 	mps_dprint(sc, MPS_INFO, "%s finished sc %p post %u free %u\n",
759 	    __func__, sc, sc->replypostindex, sc->replyfreeindex);
760 
761 	mpssas_release_simq_reinit(sassc);
762 
763 	return 0;
764 }
765 
766 /* Wait for the chip to ACK a word that we've put into its FIFO
767  * Wait for <timeout> seconds. In single loop wait for busy loop
768  * for 500 microseconds.
769  * Total is [ 0.5 * (2000 * <timeout>) ] in miliseconds.
770  * */
771 static int
772 mps_wait_db_ack(struct mps_softc *sc, int timeout, int sleep_flag)
773 {
774 
775 	u32 cntdn, count;
776 	u32 int_status;
777 	u32 doorbell;
778 
779 	count = 0;
780 	cntdn = (sleep_flag == CAN_SLEEP) ? 1000*timeout : 2000*timeout;
781 	do {
782 		int_status = mps_regread(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET);
783 		if (!(int_status & MPI2_HIS_SYS2IOC_DB_STATUS)) {
784 			mps_dprint(sc, MPS_INIT,
785 			"%s: successful count(%d), timeout(%d)\n",
786 			__func__, count, timeout);
787 		return 0;
788 		} else if (int_status & MPI2_HIS_IOC2SYS_DB_STATUS) {
789 			doorbell = mps_regread(sc, MPI2_DOORBELL_OFFSET);
790 			if ((doorbell & MPI2_IOC_STATE_MASK) ==
791 				MPI2_IOC_STATE_FAULT) {
792 				mps_dprint(sc, MPS_FAULT,
793 					"fault_state(0x%04x)!\n", doorbell);
794 				return (EFAULT);
795 			}
796 		} else if (int_status == 0xFFFFFFFF)
797 			goto out;
798 
799 		/* If it can sleep, sleep for 1 milisecond, else busy loop for
800 		* 0.5 milisecond */
801 		if (mtx_owned(&sc->mps_mtx) && sleep_flag == CAN_SLEEP)
802 			msleep(&sc->msleep_fake_chan, &sc->mps_mtx, 0,
803 			"mpsdba", hz/1000);
804 		else if (sleep_flag == CAN_SLEEP)
805 			pause("mpsdba", hz/1000);
806 		else
807 			DELAY(500);
808 		count++;
809 	} while (--cntdn);
810 
811 	out:
812 	mps_dprint(sc, MPS_FAULT, "%s: failed due to timeout count(%d), "
813 		"int_status(%x)!\n", __func__, count, int_status);
814 	return (ETIMEDOUT);
815 
816 }
817 
818 /* Wait for the chip to signal that the next word in its FIFO can be fetched */
819 static int
820 mps_wait_db_int(struct mps_softc *sc)
821 {
822 	int retry;
823 
824 	for (retry = 0; retry < MPS_DB_MAX_WAIT; retry++) {
825 		if ((mps_regread(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET) &
826 		    MPI2_HIS_IOC2SYS_DB_STATUS) != 0)
827 			return (0);
828 		DELAY(2000);
829 	}
830 	return (ETIMEDOUT);
831 }
832 
833 /* Step through the synchronous command state machine, i.e. "Doorbell mode" */
834 static int
835 mps_request_sync(struct mps_softc *sc, void *req, MPI2_DEFAULT_REPLY *reply,
836     int req_sz, int reply_sz, int timeout)
837 {
838 	uint32_t *data32;
839 	uint16_t *data16;
840 	int i, count, ioc_sz, residual;
841 	int sleep_flags = CAN_SLEEP;
842 
843 	if (curthread->td_no_sleeping != 0)
844 		sleep_flags = NO_SLEEP;
845 
846 	/* Step 1 */
847 	mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0);
848 
849 	/* Step 2 */
850 	if (mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_USED)
851 		return (EBUSY);
852 
853 	/* Step 3
854 	 * Announce that a message is coming through the doorbell.  Messages
855 	 * are pushed at 32bit words, so round up if needed.
856 	 */
857 	count = (req_sz + 3) / 4;
858 	mps_regwrite(sc, MPI2_DOORBELL_OFFSET,
859 	    (MPI2_FUNCTION_HANDSHAKE << MPI2_DOORBELL_FUNCTION_SHIFT) |
860 	    (count << MPI2_DOORBELL_ADD_DWORDS_SHIFT));
861 
862 	/* Step 4 */
863 	if (mps_wait_db_int(sc) ||
864 	    (mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_USED) == 0) {
865 		mps_dprint(sc, MPS_FAULT, "Doorbell failed to activate\n");
866 		return (ENXIO);
867 	}
868 	mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0);
869 	if (mps_wait_db_ack(sc, 5, sleep_flags) != 0) {
870 		mps_dprint(sc, MPS_FAULT, "Doorbell handshake failed\n");
871 		return (ENXIO);
872 	}
873 
874 	/* Step 5 */
875 	/* Clock out the message data synchronously in 32-bit dwords*/
876 	data32 = (uint32_t *)req;
877 	for (i = 0; i < count; i++) {
878 		mps_regwrite(sc, MPI2_DOORBELL_OFFSET, htole32(data32[i]));
879 		if (mps_wait_db_ack(sc, 5, sleep_flags) != 0) {
880 			mps_dprint(sc, MPS_FAULT,
881 			    "Timeout while writing doorbell\n");
882 			return (ENXIO);
883 		}
884 	}
885 
886 	/* Step 6 */
887 	/* Clock in the reply in 16-bit words.  The total length of the
888 	 * message is always in the 4th byte, so clock out the first 2 words
889 	 * manually, then loop the rest.
890 	 */
891 	data16 = (uint16_t *)reply;
892 	if (mps_wait_db_int(sc) != 0) {
893 		mps_dprint(sc, MPS_FAULT, "Timeout reading doorbell 0\n");
894 		return (ENXIO);
895 	}
896 	data16[0] =
897 	    mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_DATA_MASK;
898 	mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0);
899 	if (mps_wait_db_int(sc) != 0) {
900 		mps_dprint(sc, MPS_FAULT, "Timeout reading doorbell 1\n");
901 		return (ENXIO);
902 	}
903 	data16[1] =
904 	    mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_DATA_MASK;
905 	mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0);
906 
907 	/* Number of 32bit words in the message */
908 	ioc_sz = reply->MsgLength;
909 
910 	/*
911 	 * Figure out how many 16bit words to clock in without overrunning.
912 	 * The precision loss with dividing reply_sz can safely be
913 	 * ignored because the messages can only be multiples of 32bits.
914 	 */
915 	residual = 0;
916 	count = MIN((reply_sz / 4), ioc_sz) * 2;
917 	if (count < ioc_sz * 2) {
918 		residual = ioc_sz * 2 - count;
919 		mps_dprint(sc, MPS_ERROR, "Driver error, throwing away %d "
920 		    "residual message words\n", residual);
921 	}
922 
923 	for (i = 2; i < count; i++) {
924 		if (mps_wait_db_int(sc) != 0) {
925 			mps_dprint(sc, MPS_FAULT,
926 			    "Timeout reading doorbell %d\n", i);
927 			return (ENXIO);
928 		}
929 		data16[i] = mps_regread(sc, MPI2_DOORBELL_OFFSET) &
930 		    MPI2_DOORBELL_DATA_MASK;
931 		mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0);
932 	}
933 
934 	/*
935 	 * Pull out residual words that won't fit into the provided buffer.
936 	 * This keeps the chip from hanging due to a driver programming
937 	 * error.
938 	 */
939 	while (residual--) {
940 		if (mps_wait_db_int(sc) != 0) {
941 			mps_dprint(sc, MPS_FAULT,
942 			    "Timeout reading doorbell\n");
943 			return (ENXIO);
944 		}
945 		(void)mps_regread(sc, MPI2_DOORBELL_OFFSET);
946 		mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0);
947 	}
948 
949 	/* Step 7 */
950 	if (mps_wait_db_int(sc) != 0) {
951 		mps_dprint(sc, MPS_FAULT, "Timeout waiting to exit doorbell\n");
952 		return (ENXIO);
953 	}
954 	if (mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_USED)
955 		mps_dprint(sc, MPS_FAULT, "Warning, doorbell still active\n");
956 	mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0);
957 
958 	return (0);
959 }
960 
961 static void
962 mps_enqueue_request(struct mps_softc *sc, struct mps_command *cm)
963 {
964 	reply_descriptor rd;
965 	MPS_FUNCTRACE(sc);
966 	mps_dprint(sc, MPS_TRACE, "SMID %u cm %p ccb %p\n",
967 	    cm->cm_desc.Default.SMID, cm, cm->cm_ccb);
968 
969 	if (sc->mps_flags & MPS_FLAGS_ATTACH_DONE && !(sc->mps_flags & MPS_FLAGS_SHUTDOWN))
970 		mtx_assert(&sc->mps_mtx, MA_OWNED);
971 
972 	if (++sc->io_cmds_active > sc->io_cmds_highwater)
973 		sc->io_cmds_highwater++;
974 	rd.u.low = cm->cm_desc.Words.Low;
975 	rd.u.high = cm->cm_desc.Words.High;
976 	rd.word = htole64(rd.word);
977 	/* TODO-We may need to make below regwrite atomic */
978 	mps_regwrite(sc, MPI2_REQUEST_DESCRIPTOR_POST_LOW_OFFSET,
979 	    rd.u.low);
980 	mps_regwrite(sc, MPI2_REQUEST_DESCRIPTOR_POST_HIGH_OFFSET,
981 	    rd.u.high);
982 }
983 
984 /*
985  * Just the FACTS, ma'am.
986  */
987 static int
988 mps_get_iocfacts(struct mps_softc *sc, MPI2_IOC_FACTS_REPLY *facts)
989 {
990 	MPI2_DEFAULT_REPLY *reply;
991 	MPI2_IOC_FACTS_REQUEST request;
992 	int error, req_sz, reply_sz;
993 
994 	MPS_FUNCTRACE(sc);
995 
996 	req_sz = sizeof(MPI2_IOC_FACTS_REQUEST);
997 	reply_sz = sizeof(MPI2_IOC_FACTS_REPLY);
998 	reply = (MPI2_DEFAULT_REPLY *)facts;
999 
1000 	bzero(&request, req_sz);
1001 	request.Function = MPI2_FUNCTION_IOC_FACTS;
1002 	error = mps_request_sync(sc, &request, reply, req_sz, reply_sz, 5);
1003 
1004 	return (error);
1005 }
1006 
1007 static int
1008 mps_send_iocinit(struct mps_softc *sc)
1009 {
1010 	MPI2_IOC_INIT_REQUEST	init;
1011 	MPI2_DEFAULT_REPLY	reply;
1012 	int req_sz, reply_sz, error;
1013 	struct timeval now;
1014 	uint64_t time_in_msec;
1015 
1016 	MPS_FUNCTRACE(sc);
1017 
1018 	req_sz = sizeof(MPI2_IOC_INIT_REQUEST);
1019 	reply_sz = sizeof(MPI2_IOC_INIT_REPLY);
1020 	bzero(&init, req_sz);
1021 	bzero(&reply, reply_sz);
1022 
1023 	/*
1024 	 * Fill in the init block.  Note that most addresses are
1025 	 * deliberately in the lower 32bits of memory.  This is a micro-
1026 	 * optimzation for PCI/PCIX, though it's not clear if it helps PCIe.
1027 	 */
1028 	init.Function = MPI2_FUNCTION_IOC_INIT;
1029 	init.WhoInit = MPI2_WHOINIT_HOST_DRIVER;
1030 	init.MsgVersion = htole16(MPI2_VERSION);
1031 	init.HeaderVersion = htole16(MPI2_HEADER_VERSION);
1032 	init.SystemRequestFrameSize = htole16(sc->facts->IOCRequestFrameSize);
1033 	init.ReplyDescriptorPostQueueDepth = htole16(sc->pqdepth);
1034 	init.ReplyFreeQueueDepth = htole16(sc->fqdepth);
1035 	init.SenseBufferAddressHigh = 0;
1036 	init.SystemReplyAddressHigh = 0;
1037 	init.SystemRequestFrameBaseAddress.High = 0;
1038 	init.SystemRequestFrameBaseAddress.Low = htole32((uint32_t)sc->req_busaddr);
1039 	init.ReplyDescriptorPostQueueAddress.High = 0;
1040 	init.ReplyDescriptorPostQueueAddress.Low = htole32((uint32_t)sc->post_busaddr);
1041 	init.ReplyFreeQueueAddress.High = 0;
1042 	init.ReplyFreeQueueAddress.Low = htole32((uint32_t)sc->free_busaddr);
1043 	getmicrotime(&now);
1044 	time_in_msec = (now.tv_sec * 1000 + now.tv_usec/1000);
1045 	init.TimeStamp.High = htole32((time_in_msec >> 32) & 0xFFFFFFFF);
1046 	init.TimeStamp.Low = htole32(time_in_msec & 0xFFFFFFFF);
1047 
1048 	error = mps_request_sync(sc, &init, &reply, req_sz, reply_sz, 5);
1049 	if ((reply.IOCStatus & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS)
1050 		error = ENXIO;
1051 
1052 	mps_dprint(sc, MPS_INIT, "IOCInit status= 0x%x\n", reply.IOCStatus);
1053 	return (error);
1054 }
1055 
1056 void
1057 mps_memaddr_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
1058 {
1059 	bus_addr_t *addr;
1060 
1061 	addr = arg;
1062 	*addr = segs[0].ds_addr;
1063 }
1064 
1065 static int
1066 mps_alloc_queues(struct mps_softc *sc)
1067 {
1068 	bus_addr_t queues_busaddr;
1069 	uint8_t *queues;
1070 	int qsize, fqsize, pqsize;
1071 
1072 	/*
1073 	 * The reply free queue contains 4 byte entries in multiples of 16 and
1074 	 * aligned on a 16 byte boundary. There must always be an unused entry.
1075 	 * This queue supplies fresh reply frames for the firmware to use.
1076 	 *
1077 	 * The reply descriptor post queue contains 8 byte entries in
1078 	 * multiples of 16 and aligned on a 16 byte boundary.  This queue
1079 	 * contains filled-in reply frames sent from the firmware to the host.
1080 	 *
1081 	 * These two queues are allocated together for simplicity.
1082 	 */
1083 	sc->fqdepth = roundup2(sc->num_replies + 1, 16);
1084 	sc->pqdepth = roundup2(sc->num_replies + 1, 16);
1085 	fqsize= sc->fqdepth * 4;
1086 	pqsize = sc->pqdepth * 8;
1087 	qsize = fqsize + pqsize;
1088 
1089         if (bus_dma_tag_create( sc->mps_parent_dmat,    /* parent */
1090 				16, 0,			/* algnmnt, boundary */
1091 				BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1092 				BUS_SPACE_MAXADDR,	/* highaddr */
1093 				NULL, NULL,		/* filter, filterarg */
1094                                 qsize,			/* maxsize */
1095                                 1,			/* nsegments */
1096                                 qsize,			/* maxsegsize */
1097                                 0,			/* flags */
1098                                 NULL, NULL,		/* lockfunc, lockarg */
1099                                 &sc->queues_dmat)) {
1100 		device_printf(sc->mps_dev, "Cannot allocate queues DMA tag\n");
1101 		return (ENOMEM);
1102         }
1103         if (bus_dmamem_alloc(sc->queues_dmat, (void **)&queues, BUS_DMA_NOWAIT,
1104 	    &sc->queues_map)) {
1105 		device_printf(sc->mps_dev, "Cannot allocate queues memory\n");
1106 		return (ENOMEM);
1107         }
1108         bzero(queues, qsize);
1109         bus_dmamap_load(sc->queues_dmat, sc->queues_map, queues, qsize,
1110 	    mps_memaddr_cb, &queues_busaddr, 0);
1111 
1112 	sc->free_queue = (uint32_t *)queues;
1113 	sc->free_busaddr = queues_busaddr;
1114 	sc->post_queue = (MPI2_REPLY_DESCRIPTORS_UNION *)(queues + fqsize);
1115 	sc->post_busaddr = queues_busaddr + fqsize;
1116 
1117 	return (0);
1118 }
1119 
1120 static int
1121 mps_alloc_replies(struct mps_softc *sc)
1122 {
1123 	int rsize, num_replies;
1124 
1125 	/*
1126 	 * sc->num_replies should be one less than sc->fqdepth.  We need to
1127 	 * allocate space for sc->fqdepth replies, but only sc->num_replies
1128 	 * replies can be used at once.
1129 	 */
1130 	num_replies = max(sc->fqdepth, sc->num_replies);
1131 
1132 	rsize = sc->facts->ReplyFrameSize * num_replies * 4;
1133         if (bus_dma_tag_create( sc->mps_parent_dmat,    /* parent */
1134 				4, 0,			/* algnmnt, boundary */
1135 				BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1136 				BUS_SPACE_MAXADDR,	/* highaddr */
1137 				NULL, NULL,		/* filter, filterarg */
1138                                 rsize,			/* maxsize */
1139                                 1,			/* nsegments */
1140                                 rsize,			/* maxsegsize */
1141                                 0,			/* flags */
1142                                 NULL, NULL,		/* lockfunc, lockarg */
1143                                 &sc->reply_dmat)) {
1144 		device_printf(sc->mps_dev, "Cannot allocate replies DMA tag\n");
1145 		return (ENOMEM);
1146         }
1147         if (bus_dmamem_alloc(sc->reply_dmat, (void **)&sc->reply_frames,
1148 	    BUS_DMA_NOWAIT, &sc->reply_map)) {
1149 		device_printf(sc->mps_dev, "Cannot allocate replies memory\n");
1150 		return (ENOMEM);
1151         }
1152         bzero(sc->reply_frames, rsize);
1153         bus_dmamap_load(sc->reply_dmat, sc->reply_map, sc->reply_frames, rsize,
1154 	    mps_memaddr_cb, &sc->reply_busaddr, 0);
1155 
1156 	return (0);
1157 }
1158 
1159 static int
1160 mps_alloc_requests(struct mps_softc *sc)
1161 {
1162 	struct mps_command *cm;
1163 	struct mps_chain *chain;
1164 	int i, rsize, nsegs;
1165 
1166 	rsize = sc->facts->IOCRequestFrameSize * sc->num_reqs * 4;
1167         if (bus_dma_tag_create( sc->mps_parent_dmat,    /* parent */
1168 				16, 0,			/* algnmnt, boundary */
1169 				BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1170 				BUS_SPACE_MAXADDR,	/* highaddr */
1171 				NULL, NULL,		/* filter, filterarg */
1172                                 rsize,			/* maxsize */
1173                                 1,			/* nsegments */
1174                                 rsize,			/* maxsegsize */
1175                                 0,			/* flags */
1176                                 NULL, NULL,		/* lockfunc, lockarg */
1177                                 &sc->req_dmat)) {
1178 		device_printf(sc->mps_dev, "Cannot allocate request DMA tag\n");
1179 		return (ENOMEM);
1180         }
1181         if (bus_dmamem_alloc(sc->req_dmat, (void **)&sc->req_frames,
1182 	    BUS_DMA_NOWAIT, &sc->req_map)) {
1183 		device_printf(sc->mps_dev, "Cannot allocate request memory\n");
1184 		return (ENOMEM);
1185         }
1186         bzero(sc->req_frames, rsize);
1187         bus_dmamap_load(sc->req_dmat, sc->req_map, sc->req_frames, rsize,
1188 	    mps_memaddr_cb, &sc->req_busaddr, 0);
1189 
1190 	rsize = sc->facts->IOCRequestFrameSize * sc->max_chains * 4;
1191         if (bus_dma_tag_create( sc->mps_parent_dmat,    /* parent */
1192 				16, 0,			/* algnmnt, boundary */
1193 				BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1194 				BUS_SPACE_MAXADDR,	/* highaddr */
1195 				NULL, NULL,		/* filter, filterarg */
1196                                 rsize,			/* maxsize */
1197                                 1,			/* nsegments */
1198                                 rsize,			/* maxsegsize */
1199                                 0,			/* flags */
1200                                 NULL, NULL,		/* lockfunc, lockarg */
1201                                 &sc->chain_dmat)) {
1202 		device_printf(sc->mps_dev, "Cannot allocate chain DMA tag\n");
1203 		return (ENOMEM);
1204         }
1205         if (bus_dmamem_alloc(sc->chain_dmat, (void **)&sc->chain_frames,
1206 	    BUS_DMA_NOWAIT, &sc->chain_map)) {
1207 		device_printf(sc->mps_dev, "Cannot allocate chain memory\n");
1208 		return (ENOMEM);
1209         }
1210         bzero(sc->chain_frames, rsize);
1211         bus_dmamap_load(sc->chain_dmat, sc->chain_map, sc->chain_frames, rsize,
1212 	    mps_memaddr_cb, &sc->chain_busaddr, 0);
1213 
1214 	rsize = MPS_SENSE_LEN * sc->num_reqs;
1215         if (bus_dma_tag_create( sc->mps_parent_dmat,    /* parent */
1216 				1, 0,			/* algnmnt, boundary */
1217 				BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1218 				BUS_SPACE_MAXADDR,	/* highaddr */
1219 				NULL, NULL,		/* filter, filterarg */
1220                                 rsize,			/* maxsize */
1221                                 1,			/* nsegments */
1222                                 rsize,			/* maxsegsize */
1223                                 0,			/* flags */
1224                                 NULL, NULL,		/* lockfunc, lockarg */
1225                                 &sc->sense_dmat)) {
1226 		device_printf(sc->mps_dev, "Cannot allocate sense DMA tag\n");
1227 		return (ENOMEM);
1228         }
1229         if (bus_dmamem_alloc(sc->sense_dmat, (void **)&sc->sense_frames,
1230 	    BUS_DMA_NOWAIT, &sc->sense_map)) {
1231 		device_printf(sc->mps_dev, "Cannot allocate sense memory\n");
1232 		return (ENOMEM);
1233         }
1234         bzero(sc->sense_frames, rsize);
1235         bus_dmamap_load(sc->sense_dmat, sc->sense_map, sc->sense_frames, rsize,
1236 	    mps_memaddr_cb, &sc->sense_busaddr, 0);
1237 
1238 	sc->chains = malloc(sizeof(struct mps_chain) * sc->max_chains, M_MPT2,
1239 	    M_WAITOK | M_ZERO);
1240 	if(!sc->chains) {
1241 		device_printf(sc->mps_dev,
1242 		"Cannot allocate chains memory %s %d\n",
1243 		 __func__, __LINE__);
1244 		return (ENOMEM);
1245 	}
1246 	for (i = 0; i < sc->max_chains; i++) {
1247 		chain = &sc->chains[i];
1248 		chain->chain = (MPI2_SGE_IO_UNION *)(sc->chain_frames +
1249 		    i * sc->facts->IOCRequestFrameSize * 4);
1250 		chain->chain_busaddr = sc->chain_busaddr +
1251 		    i * sc->facts->IOCRequestFrameSize * 4;
1252 		mps_free_chain(sc, chain);
1253 		sc->chain_free_lowwater++;
1254 	}
1255 
1256 	/* XXX Need to pick a more precise value */
1257 	nsegs = (MAXPHYS / PAGE_SIZE) + 1;
1258         if (bus_dma_tag_create( sc->mps_parent_dmat,    /* parent */
1259 				1, 0,			/* algnmnt, boundary */
1260 				BUS_SPACE_MAXADDR,	/* lowaddr */
1261 				BUS_SPACE_MAXADDR,	/* highaddr */
1262 				NULL, NULL,		/* filter, filterarg */
1263                                 BUS_SPACE_MAXSIZE_32BIT,/* maxsize */
1264                                 nsegs,			/* nsegments */
1265                                 BUS_SPACE_MAXSIZE_24BIT,/* maxsegsize */
1266                                 BUS_DMA_ALLOCNOW,	/* flags */
1267                                 busdma_lock_mutex,	/* lockfunc */
1268 				&sc->mps_mtx,		/* lockarg */
1269                                 &sc->buffer_dmat)) {
1270 		device_printf(sc->mps_dev, "Cannot allocate buffer DMA tag\n");
1271 		return (ENOMEM);
1272         }
1273 
1274 	/*
1275 	 * SMID 0 cannot be used as a free command per the firmware spec.
1276 	 * Just drop that command instead of risking accounting bugs.
1277 	 */
1278 	sc->commands = malloc(sizeof(struct mps_command) * sc->num_reqs,
1279 	    M_MPT2, M_WAITOK | M_ZERO);
1280 	if(!sc->commands) {
1281 		device_printf(sc->mps_dev, "Cannot allocate memory %s %d\n",
1282 		 __func__, __LINE__);
1283 		return (ENOMEM);
1284 	}
1285 	for (i = 1; i < sc->num_reqs; i++) {
1286 		cm = &sc->commands[i];
1287 		cm->cm_req = sc->req_frames +
1288 		    i * sc->facts->IOCRequestFrameSize * 4;
1289 		cm->cm_req_busaddr = sc->req_busaddr +
1290 		    i * sc->facts->IOCRequestFrameSize * 4;
1291 		cm->cm_sense = &sc->sense_frames[i];
1292 		cm->cm_sense_busaddr = sc->sense_busaddr + i * MPS_SENSE_LEN;
1293 		cm->cm_desc.Default.SMID = i;
1294 		cm->cm_sc = sc;
1295 		TAILQ_INIT(&cm->cm_chain_list);
1296 		callout_init_mtx(&cm->cm_callout, &sc->mps_mtx, 0);
1297 
1298 		/* XXX Is a failure here a critical problem? */
1299 		if (bus_dmamap_create(sc->buffer_dmat, 0, &cm->cm_dmamap) == 0)
1300 			if (i <= sc->facts->HighPriorityCredit)
1301 				mps_free_high_priority_command(sc, cm);
1302 			else
1303 				mps_free_command(sc, cm);
1304 		else {
1305 			panic("failed to allocate command %d\n", i);
1306 			sc->num_reqs = i;
1307 			break;
1308 		}
1309 	}
1310 
1311 	return (0);
1312 }
1313 
1314 static int
1315 mps_init_queues(struct mps_softc *sc)
1316 {
1317 	int i;
1318 
1319 	memset((uint8_t *)sc->post_queue, 0xff, sc->pqdepth * 8);
1320 
1321 	/*
1322 	 * According to the spec, we need to use one less reply than we
1323 	 * have space for on the queue.  So sc->num_replies (the number we
1324 	 * use) should be less than sc->fqdepth (allocated size).
1325 	 */
1326 	if (sc->num_replies >= sc->fqdepth)
1327 		return (EINVAL);
1328 
1329 	/*
1330 	 * Initialize all of the free queue entries.
1331 	 */
1332 	for (i = 0; i < sc->fqdepth; i++)
1333 		sc->free_queue[i] = sc->reply_busaddr + (i * sc->facts->ReplyFrameSize * 4);
1334 	sc->replyfreeindex = sc->num_replies;
1335 
1336 	return (0);
1337 }
1338 
1339 /* Get the driver parameter tunables.  Lowest priority are the driver defaults.
1340  * Next are the global settings, if they exist.  Highest are the per-unit
1341  * settings, if they exist.
1342  */
1343 static void
1344 mps_get_tunables(struct mps_softc *sc)
1345 {
1346 	char tmpstr[80];
1347 
1348 	/* XXX default to some debugging for now */
1349 	sc->mps_debug = MPS_INFO|MPS_FAULT;
1350 	sc->disable_msix = 0;
1351 	sc->disable_msi = 0;
1352 	sc->max_chains = MPS_CHAIN_FRAMES;
1353 	sc->enable_ssu = MPS_SSU_ENABLE_SSD_DISABLE_HDD;
1354 	sc->spinup_wait_time = DEFAULT_SPINUP_WAIT;
1355 
1356 	/*
1357 	 * Grab the global variables.
1358 	 */
1359 	TUNABLE_INT_FETCH("hw.mps.debug_level", &sc->mps_debug);
1360 	TUNABLE_INT_FETCH("hw.mps.disable_msix", &sc->disable_msix);
1361 	TUNABLE_INT_FETCH("hw.mps.disable_msi", &sc->disable_msi);
1362 	TUNABLE_INT_FETCH("hw.mps.max_chains", &sc->max_chains);
1363 	TUNABLE_INT_FETCH("hw.mps.enable_ssu", &sc->enable_ssu);
1364 	TUNABLE_INT_FETCH("hw.mps.spinup_wait_time", &sc->spinup_wait_time);
1365 
1366 	/* Grab the unit-instance variables */
1367 	snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.debug_level",
1368 	    device_get_unit(sc->mps_dev));
1369 	TUNABLE_INT_FETCH(tmpstr, &sc->mps_debug);
1370 
1371 	snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.disable_msix",
1372 	    device_get_unit(sc->mps_dev));
1373 	TUNABLE_INT_FETCH(tmpstr, &sc->disable_msix);
1374 
1375 	snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.disable_msi",
1376 	    device_get_unit(sc->mps_dev));
1377 	TUNABLE_INT_FETCH(tmpstr, &sc->disable_msi);
1378 
1379 	snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_chains",
1380 	    device_get_unit(sc->mps_dev));
1381 	TUNABLE_INT_FETCH(tmpstr, &sc->max_chains);
1382 
1383 	bzero(sc->exclude_ids, sizeof(sc->exclude_ids));
1384 	snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.exclude_ids",
1385 	    device_get_unit(sc->mps_dev));
1386 	TUNABLE_STR_FETCH(tmpstr, sc->exclude_ids, sizeof(sc->exclude_ids));
1387 
1388 	snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.enable_ssu",
1389 	    device_get_unit(sc->mps_dev));
1390 	TUNABLE_INT_FETCH(tmpstr, &sc->enable_ssu);
1391 
1392 	snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.spinup_wait_time",
1393 	    device_get_unit(sc->mps_dev));
1394 	TUNABLE_INT_FETCH(tmpstr, &sc->spinup_wait_time);
1395 }
1396 
1397 static void
1398 mps_setup_sysctl(struct mps_softc *sc)
1399 {
1400 	struct sysctl_ctx_list	*sysctl_ctx = NULL;
1401 	struct sysctl_oid	*sysctl_tree = NULL;
1402 	char tmpstr[80], tmpstr2[80];
1403 
1404 	/*
1405 	 * Setup the sysctl variable so the user can change the debug level
1406 	 * on the fly.
1407 	 */
1408 	snprintf(tmpstr, sizeof(tmpstr), "MPS controller %d",
1409 	    device_get_unit(sc->mps_dev));
1410 	snprintf(tmpstr2, sizeof(tmpstr2), "%d", device_get_unit(sc->mps_dev));
1411 
1412 	sysctl_ctx = device_get_sysctl_ctx(sc->mps_dev);
1413 	if (sysctl_ctx != NULL)
1414 		sysctl_tree = device_get_sysctl_tree(sc->mps_dev);
1415 
1416 	if (sysctl_tree == NULL) {
1417 		sysctl_ctx_init(&sc->sysctl_ctx);
1418 		sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
1419 		    SYSCTL_STATIC_CHILDREN(_hw_mps), OID_AUTO, tmpstr2,
1420 		    CTLFLAG_RD, 0, tmpstr);
1421 		if (sc->sysctl_tree == NULL)
1422 			return;
1423 		sysctl_ctx = &sc->sysctl_ctx;
1424 		sysctl_tree = sc->sysctl_tree;
1425 	}
1426 
1427 	SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1428 	    OID_AUTO, "debug_level", CTLFLAG_RW, &sc->mps_debug, 0,
1429 	    "mps debug level");
1430 
1431 	SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1432 	    OID_AUTO, "disable_msix", CTLFLAG_RD, &sc->disable_msix, 0,
1433 	    "Disable the use of MSI-X interrupts");
1434 
1435 	SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1436 	    OID_AUTO, "disable_msi", CTLFLAG_RD, &sc->disable_msi, 0,
1437 	    "Disable the use of MSI interrupts");
1438 
1439 	SYSCTL_ADD_STRING(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1440 	    OID_AUTO, "firmware_version", CTLFLAG_RW, sc->fw_version,
1441 	    strlen(sc->fw_version), "firmware version");
1442 
1443 	SYSCTL_ADD_STRING(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1444 	    OID_AUTO, "driver_version", CTLFLAG_RW, MPS_DRIVER_VERSION,
1445 	    strlen(MPS_DRIVER_VERSION), "driver version");
1446 
1447 	SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1448 	    OID_AUTO, "io_cmds_active", CTLFLAG_RD,
1449 	    &sc->io_cmds_active, 0, "number of currently active commands");
1450 
1451 	SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1452 	    OID_AUTO, "io_cmds_highwater", CTLFLAG_RD,
1453 	    &sc->io_cmds_highwater, 0, "maximum active commands seen");
1454 
1455 	SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1456 	    OID_AUTO, "chain_free", CTLFLAG_RD,
1457 	    &sc->chain_free, 0, "number of free chain elements");
1458 
1459 	SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1460 	    OID_AUTO, "chain_free_lowwater", CTLFLAG_RD,
1461 	    &sc->chain_free_lowwater, 0,"lowest number of free chain elements");
1462 
1463 	SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1464 	    OID_AUTO, "max_chains", CTLFLAG_RD,
1465 	    &sc->max_chains, 0,"maximum chain frames that will be allocated");
1466 
1467 	SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1468 	    OID_AUTO, "enable_ssu", CTLFLAG_RW, &sc->enable_ssu, 0,
1469 	    "enable SSU to SATA SSD/HDD at shutdown");
1470 
1471 	SYSCTL_ADD_UQUAD(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1472 	    OID_AUTO, "chain_alloc_fail", CTLFLAG_RD,
1473 	    &sc->chain_alloc_fail, "chain allocation failures");
1474 
1475 	SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1476 	    OID_AUTO, "spinup_wait_time", CTLFLAG_RD,
1477 	    &sc->spinup_wait_time, DEFAULT_SPINUP_WAIT, "seconds to wait for "
1478 	    "spinup after SATA ID error");
1479 
1480 	SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1481 	    OID_AUTO, "mapping_table_dump", CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
1482 	    mps_mapping_dump, "A", "Mapping Table Dump");
1483 
1484 	SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree),
1485 	    OID_AUTO, "encl_table_dump", CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
1486 	    mps_mapping_encl_dump, "A", "Enclosure Table Dump");
1487 }
1488 
1489 int
1490 mps_attach(struct mps_softc *sc)
1491 {
1492 	int error;
1493 
1494 	mps_get_tunables(sc);
1495 
1496 	MPS_FUNCTRACE(sc);
1497 
1498 	mtx_init(&sc->mps_mtx, "MPT2SAS lock", NULL, MTX_DEF);
1499 	callout_init_mtx(&sc->periodic, &sc->mps_mtx, 0);
1500 	TAILQ_INIT(&sc->event_list);
1501 	timevalclear(&sc->lastfail);
1502 
1503 	if ((error = mps_transition_ready(sc)) != 0) {
1504 		mps_printf(sc, "%s failed to transition ready\n", __func__);
1505 		return (error);
1506 	}
1507 
1508 	sc->facts = malloc(sizeof(MPI2_IOC_FACTS_REPLY), M_MPT2,
1509 	    M_ZERO|M_NOWAIT);
1510 	if(!sc->facts) {
1511 		device_printf(sc->mps_dev, "Cannot allocate memory %s %d\n",
1512 		 __func__, __LINE__);
1513 		return (ENOMEM);
1514 	}
1515 
1516 	/*
1517 	 * Get IOC Facts and allocate all structures based on this information.
1518 	 * A Diag Reset will also call mps_iocfacts_allocate and re-read the IOC
1519 	 * Facts. If relevant values have changed in IOC Facts, this function
1520 	 * will free all of the memory based on IOC Facts and reallocate that
1521 	 * memory.  If this fails, any allocated memory should already be freed.
1522 	 */
1523 	if ((error = mps_iocfacts_allocate(sc, TRUE)) != 0) {
1524 		mps_dprint(sc, MPS_FAULT, "%s IOC Facts based allocation "
1525 		    "failed with error %d\n", __func__, error);
1526 		return (error);
1527 	}
1528 
1529 	/* Start the periodic watchdog check on the IOC Doorbell */
1530 	mps_periodic(sc);
1531 
1532 	/*
1533 	 * The portenable will kick off discovery events that will drive the
1534 	 * rest of the initialization process.  The CAM/SAS module will
1535 	 * hold up the boot sequence until discovery is complete.
1536 	 */
1537 	sc->mps_ich.ich_func = mps_startup;
1538 	sc->mps_ich.ich_arg = sc;
1539 	if (config_intrhook_establish(&sc->mps_ich) != 0) {
1540 		mps_dprint(sc, MPS_ERROR, "Cannot establish MPS config hook\n");
1541 		error = EINVAL;
1542 	}
1543 
1544 	/*
1545 	 * Allow IR to shutdown gracefully when shutdown occurs.
1546 	 */
1547 	sc->shutdown_eh = EVENTHANDLER_REGISTER(shutdown_final,
1548 	    mpssas_ir_shutdown, sc, SHUTDOWN_PRI_DEFAULT);
1549 
1550 	if (sc->shutdown_eh == NULL)
1551 		mps_dprint(sc, MPS_ERROR, "shutdown event registration "
1552 		    "failed\n");
1553 
1554 	mps_setup_sysctl(sc);
1555 
1556 	sc->mps_flags |= MPS_FLAGS_ATTACH_DONE;
1557 
1558 	return (error);
1559 }
1560 
1561 /* Run through any late-start handlers. */
1562 static void
1563 mps_startup(void *arg)
1564 {
1565 	struct mps_softc *sc;
1566 
1567 	sc = (struct mps_softc *)arg;
1568 
1569 	mps_lock(sc);
1570 	mps_unmask_intr(sc);
1571 
1572 	/* initialize device mapping tables */
1573 	mps_base_static_config_pages(sc);
1574 	mps_mapping_initialize(sc);
1575 	mpssas_startup(sc);
1576 	mps_unlock(sc);
1577 }
1578 
1579 /* Periodic watchdog.  Is called with the driver lock already held. */
1580 static void
1581 mps_periodic(void *arg)
1582 {
1583 	struct mps_softc *sc;
1584 	uint32_t db;
1585 
1586 	sc = (struct mps_softc *)arg;
1587 	if (sc->mps_flags & MPS_FLAGS_SHUTDOWN)
1588 		return;
1589 
1590 	db = mps_regread(sc, MPI2_DOORBELL_OFFSET);
1591 	if ((db & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_FAULT) {
1592 		mps_dprint(sc, MPS_FAULT, "IOC Fault 0x%08x, Resetting\n", db);
1593 		mps_reinit(sc);
1594 	}
1595 
1596 	callout_reset(&sc->periodic, MPS_PERIODIC_DELAY * hz, mps_periodic, sc);
1597 }
1598 
1599 static void
1600 mps_log_evt_handler(struct mps_softc *sc, uintptr_t data,
1601     MPI2_EVENT_NOTIFICATION_REPLY *event)
1602 {
1603 	MPI2_EVENT_DATA_LOG_ENTRY_ADDED *entry;
1604 
1605 	mps_print_event(sc, event);
1606 
1607 	switch (event->Event) {
1608 	case MPI2_EVENT_LOG_DATA:
1609 		mps_dprint(sc, MPS_EVENT, "MPI2_EVENT_LOG_DATA:\n");
1610 		if (sc->mps_debug & MPS_EVENT)
1611 			hexdump(event->EventData, event->EventDataLength, NULL, 0);
1612 		break;
1613 	case MPI2_EVENT_LOG_ENTRY_ADDED:
1614 		entry = (MPI2_EVENT_DATA_LOG_ENTRY_ADDED *)event->EventData;
1615 		mps_dprint(sc, MPS_EVENT, "MPI2_EVENT_LOG_ENTRY_ADDED event "
1616 		    "0x%x Sequence %d:\n", entry->LogEntryQualifier,
1617 		     entry->LogSequence);
1618 		break;
1619 	default:
1620 		break;
1621 	}
1622 	return;
1623 }
1624 
1625 static int
1626 mps_attach_log(struct mps_softc *sc)
1627 {
1628 	u32 events[MPI2_EVENT_NOTIFY_EVENTMASK_WORDS];
1629 
1630 	bzero(events, 16);
1631 	setbit(events, MPI2_EVENT_LOG_DATA);
1632 	setbit(events, MPI2_EVENT_LOG_ENTRY_ADDED);
1633 
1634 	mps_register_events(sc, events, mps_log_evt_handler, NULL,
1635 	    &sc->mps_log_eh);
1636 
1637 	return (0);
1638 }
1639 
1640 static int
1641 mps_detach_log(struct mps_softc *sc)
1642 {
1643 
1644 	if (sc->mps_log_eh != NULL)
1645 		mps_deregister_events(sc, sc->mps_log_eh);
1646 	return (0);
1647 }
1648 
1649 /*
1650  * Free all of the driver resources and detach submodules.  Should be called
1651  * without the lock held.
1652  */
1653 int
1654 mps_free(struct mps_softc *sc)
1655 {
1656 	int error;
1657 
1658 	/* Turn off the watchdog */
1659 	mps_lock(sc);
1660 	sc->mps_flags |= MPS_FLAGS_SHUTDOWN;
1661 	mps_unlock(sc);
1662 	/* Lock must not be held for this */
1663 	callout_drain(&sc->periodic);
1664 
1665 	if (((error = mps_detach_log(sc)) != 0) ||
1666 	    ((error = mps_detach_sas(sc)) != 0))
1667 		return (error);
1668 
1669 	mps_detach_user(sc);
1670 
1671 	/* Put the IOC back in the READY state. */
1672 	mps_lock(sc);
1673 	if ((error = mps_transition_ready(sc)) != 0) {
1674 		mps_unlock(sc);
1675 		return (error);
1676 	}
1677 	mps_unlock(sc);
1678 
1679 	if (sc->facts != NULL)
1680 		free(sc->facts, M_MPT2);
1681 
1682 	/*
1683 	 * Free all buffers that are based on IOC Facts.  A Diag Reset may need
1684 	 * to free these buffers too.
1685 	 */
1686 	mps_iocfacts_free(sc);
1687 
1688 	if (sc->sysctl_tree != NULL)
1689 		sysctl_ctx_free(&sc->sysctl_ctx);
1690 
1691 	/* Deregister the shutdown function */
1692 	if (sc->shutdown_eh != NULL)
1693 		EVENTHANDLER_DEREGISTER(shutdown_final, sc->shutdown_eh);
1694 
1695 	mtx_destroy(&sc->mps_mtx);
1696 
1697 	return (0);
1698 }
1699 
1700 static __inline void
1701 mps_complete_command(struct mps_softc *sc, struct mps_command *cm)
1702 {
1703 	MPS_FUNCTRACE(sc);
1704 
1705 	if (cm == NULL) {
1706 		mps_dprint(sc, MPS_ERROR, "Completing NULL command\n");
1707 		return;
1708 	}
1709 
1710 	if (cm->cm_flags & MPS_CM_FLAGS_POLLED)
1711 		cm->cm_flags |= MPS_CM_FLAGS_COMPLETE;
1712 
1713 	if (cm->cm_complete != NULL) {
1714 		mps_dprint(sc, MPS_TRACE,
1715 			   "%s cm %p calling cm_complete %p data %p reply %p\n",
1716 			   __func__, cm, cm->cm_complete, cm->cm_complete_data,
1717 			   cm->cm_reply);
1718 		cm->cm_complete(sc, cm);
1719 	}
1720 
1721 	if (cm->cm_flags & MPS_CM_FLAGS_WAKEUP) {
1722 		mps_dprint(sc, MPS_TRACE, "waking up %p\n", cm);
1723 		wakeup(cm);
1724 	}
1725 
1726 	if (cm->cm_sc->io_cmds_active != 0) {
1727 		cm->cm_sc->io_cmds_active--;
1728 	} else {
1729 		mps_dprint(sc, MPS_ERROR, "Warning: io_cmds_active is "
1730 		    "out of sync - resynching to 0\n");
1731 	}
1732 }
1733 
1734 
1735 static void
1736 mps_sas_log_info(struct mps_softc *sc , u32 log_info)
1737 {
1738 	union loginfo_type {
1739 		u32     loginfo;
1740 		struct {
1741 			u32     subcode:16;
1742 			u32     code:8;
1743 			u32     originator:4;
1744 			u32     bus_type:4;
1745 		} dw;
1746 	};
1747 	union loginfo_type sas_loginfo;
1748 	char *originator_str = NULL;
1749 
1750 	sas_loginfo.loginfo = log_info;
1751 	if (sas_loginfo.dw.bus_type != 3 /*SAS*/)
1752 		return;
1753 
1754 	/* each nexus loss loginfo */
1755 	if (log_info == 0x31170000)
1756 		return;
1757 
1758 	/* eat the loginfos associated with task aborts */
1759 	if ((log_info == 30050000 || log_info ==
1760 	    0x31140000 || log_info == 0x31130000))
1761 		return;
1762 
1763 	switch (sas_loginfo.dw.originator) {
1764 	case 0:
1765 		originator_str = "IOP";
1766 		break;
1767 	case 1:
1768 		originator_str = "PL";
1769 		break;
1770 	case 2:
1771 		originator_str = "IR";
1772 		break;
1773 }
1774 
1775 	mps_dprint(sc, MPS_LOG, "log_info(0x%08x): originator(%s), "
1776 	"code(0x%02x), sub_code(0x%04x)\n", log_info,
1777 	originator_str, sas_loginfo.dw.code,
1778 	sas_loginfo.dw.subcode);
1779 }
1780 
1781 static void
1782 mps_display_reply_info(struct mps_softc *sc, uint8_t *reply)
1783 {
1784 	MPI2DefaultReply_t *mpi_reply;
1785 	u16 sc_status;
1786 
1787 	mpi_reply = (MPI2DefaultReply_t*)reply;
1788 	sc_status = le16toh(mpi_reply->IOCStatus);
1789 	if (sc_status & MPI2_IOCSTATUS_FLAG_LOG_INFO_AVAILABLE)
1790 		mps_sas_log_info(sc, le32toh(mpi_reply->IOCLogInfo));
1791 }
1792 void
1793 mps_intr(void *data)
1794 {
1795 	struct mps_softc *sc;
1796 	uint32_t status;
1797 
1798 	sc = (struct mps_softc *)data;
1799 	mps_dprint(sc, MPS_TRACE, "%s\n", __func__);
1800 
1801 	/*
1802 	 * Check interrupt status register to flush the bus.  This is
1803 	 * needed for both INTx interrupts and driver-driven polling
1804 	 */
1805 	status = mps_regread(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET);
1806 	if ((status & MPI2_HIS_REPLY_DESCRIPTOR_INTERRUPT) == 0)
1807 		return;
1808 
1809 	mps_lock(sc);
1810 	mps_intr_locked(data);
1811 	mps_unlock(sc);
1812 	return;
1813 }
1814 
1815 /*
1816  * In theory, MSI/MSIX interrupts shouldn't need to read any registers on the
1817  * chip.  Hopefully this theory is correct.
1818  */
1819 void
1820 mps_intr_msi(void *data)
1821 {
1822 	struct mps_softc *sc;
1823 
1824 	sc = (struct mps_softc *)data;
1825 	mps_dprint(sc, MPS_TRACE, "%s\n", __func__);
1826 	mps_lock(sc);
1827 	mps_intr_locked(data);
1828 	mps_unlock(sc);
1829 	return;
1830 }
1831 
1832 /*
1833  * The locking is overly broad and simplistic, but easy to deal with for now.
1834  */
1835 void
1836 mps_intr_locked(void *data)
1837 {
1838 	MPI2_REPLY_DESCRIPTORS_UNION *desc;
1839 	struct mps_softc *sc;
1840 	struct mps_command *cm = NULL;
1841 	uint8_t flags;
1842 	u_int pq;
1843 	MPI2_DIAG_RELEASE_REPLY *rel_rep;
1844 	mps_fw_diagnostic_buffer_t *pBuffer;
1845 
1846 	sc = (struct mps_softc *)data;
1847 
1848 	pq = sc->replypostindex;
1849 	mps_dprint(sc, MPS_TRACE,
1850 	    "%s sc %p starting with replypostindex %u\n",
1851 	    __func__, sc, sc->replypostindex);
1852 
1853 	for ( ;; ) {
1854 		cm = NULL;
1855 		desc = &sc->post_queue[sc->replypostindex];
1856 		flags = desc->Default.ReplyFlags &
1857 		    MPI2_RPY_DESCRIPT_FLAGS_TYPE_MASK;
1858 		if ((flags == MPI2_RPY_DESCRIPT_FLAGS_UNUSED)
1859 		 || (le32toh(desc->Words.High) == 0xffffffff))
1860 			break;
1861 
1862 		/* increment the replypostindex now, so that event handlers
1863 		 * and cm completion handlers which decide to do a diag
1864 		 * reset can zero it without it getting incremented again
1865 		 * afterwards, and we break out of this loop on the next
1866 		 * iteration since the reply post queue has been cleared to
1867 		 * 0xFF and all descriptors look unused (which they are).
1868 		 */
1869 		if (++sc->replypostindex >= sc->pqdepth)
1870 			sc->replypostindex = 0;
1871 
1872 		switch (flags) {
1873 		case MPI2_RPY_DESCRIPT_FLAGS_SCSI_IO_SUCCESS:
1874 			cm = &sc->commands[le16toh(desc->SCSIIOSuccess.SMID)];
1875 			cm->cm_reply = NULL;
1876 			break;
1877 		case MPI2_RPY_DESCRIPT_FLAGS_ADDRESS_REPLY:
1878 		{
1879 			uint32_t baddr;
1880 			uint8_t *reply;
1881 
1882 			/*
1883 			 * Re-compose the reply address from the address
1884 			 * sent back from the chip.  The ReplyFrameAddress
1885 			 * is the lower 32 bits of the physical address of
1886 			 * particular reply frame.  Convert that address to
1887 			 * host format, and then use that to provide the
1888 			 * offset against the virtual address base
1889 			 * (sc->reply_frames).
1890 			 */
1891 			baddr = le32toh(desc->AddressReply.ReplyFrameAddress);
1892 			reply = sc->reply_frames +
1893 				(baddr - ((uint32_t)sc->reply_busaddr));
1894 			/*
1895 			 * Make sure the reply we got back is in a valid
1896 			 * range.  If not, go ahead and panic here, since
1897 			 * we'll probably panic as soon as we deference the
1898 			 * reply pointer anyway.
1899 			 */
1900 			if ((reply < sc->reply_frames)
1901 			 || (reply > (sc->reply_frames +
1902 			     (sc->fqdepth * sc->facts->ReplyFrameSize * 4)))) {
1903 				printf("%s: WARNING: reply %p out of range!\n",
1904 				       __func__, reply);
1905 				printf("%s: reply_frames %p, fqdepth %d, "
1906 				       "frame size %d\n", __func__,
1907 				       sc->reply_frames, sc->fqdepth,
1908 				       sc->facts->ReplyFrameSize * 4);
1909 				printf("%s: baddr %#x,\n", __func__, baddr);
1910 				/* LSI-TODO. See Linux Code. Need Graceful exit*/
1911 				panic("Reply address out of range");
1912 			}
1913 			if (le16toh(desc->AddressReply.SMID) == 0) {
1914 				if (((MPI2_DEFAULT_REPLY *)reply)->Function ==
1915 				    MPI2_FUNCTION_DIAG_BUFFER_POST) {
1916 					/*
1917 					 * If SMID is 0 for Diag Buffer Post,
1918 					 * this implies that the reply is due to
1919 					 * a release function with a status that
1920 					 * the buffer has been released.  Set
1921 					 * the buffer flags accordingly.
1922 					 */
1923 					rel_rep =
1924 					    (MPI2_DIAG_RELEASE_REPLY *)reply;
1925 					if (le16toh(rel_rep->IOCStatus) ==
1926 					    MPI2_IOCSTATUS_DIAGNOSTIC_RELEASED)
1927 					    {
1928 						pBuffer =
1929 						    &sc->fw_diag_buffer_list[
1930 						    rel_rep->BufferType];
1931 						pBuffer->valid_data = TRUE;
1932 						pBuffer->owned_by_firmware =
1933 						    FALSE;
1934 						pBuffer->immediate = FALSE;
1935 					}
1936 				} else
1937 					mps_dispatch_event(sc, baddr,
1938 					    (MPI2_EVENT_NOTIFICATION_REPLY *)
1939 					    reply);
1940 			} else {
1941 				cm = &sc->commands[le16toh(desc->AddressReply.SMID)];
1942 				cm->cm_reply = reply;
1943 				cm->cm_reply_data =
1944 				    le32toh(desc->AddressReply.ReplyFrameAddress);
1945 			}
1946 			break;
1947 		}
1948 		case MPI2_RPY_DESCRIPT_FLAGS_TARGETASSIST_SUCCESS:
1949 		case MPI2_RPY_DESCRIPT_FLAGS_TARGET_COMMAND_BUFFER:
1950 		case MPI2_RPY_DESCRIPT_FLAGS_RAID_ACCELERATOR_SUCCESS:
1951 		default:
1952 			/* Unhandled */
1953 			mps_dprint(sc, MPS_ERROR, "Unhandled reply 0x%x\n",
1954 			    desc->Default.ReplyFlags);
1955 			cm = NULL;
1956 			break;
1957 		}
1958 
1959 
1960 		if (cm != NULL) {
1961 			// Print Error reply frame
1962 			if (cm->cm_reply)
1963 				mps_display_reply_info(sc,cm->cm_reply);
1964 			mps_complete_command(sc, cm);
1965 		}
1966 
1967 		desc->Words.Low = 0xffffffff;
1968 		desc->Words.High = 0xffffffff;
1969 	}
1970 
1971 	if (pq != sc->replypostindex) {
1972 		mps_dprint(sc, MPS_TRACE,
1973 		    "%s sc %p writing postindex %d\n",
1974 		    __func__, sc, sc->replypostindex);
1975 		mps_regwrite(sc, MPI2_REPLY_POST_HOST_INDEX_OFFSET, sc->replypostindex);
1976 	}
1977 
1978 	return;
1979 }
1980 
1981 static void
1982 mps_dispatch_event(struct mps_softc *sc, uintptr_t data,
1983     MPI2_EVENT_NOTIFICATION_REPLY *reply)
1984 {
1985 	struct mps_event_handle *eh;
1986 	int event, handled = 0;
1987 
1988 	event = le16toh(reply->Event);
1989 	TAILQ_FOREACH(eh, &sc->event_list, eh_list) {
1990 		if (isset(eh->mask, event)) {
1991 			eh->callback(sc, data, reply);
1992 			handled++;
1993 		}
1994 	}
1995 
1996 	if (handled == 0)
1997 		mps_dprint(sc, MPS_EVENT, "Unhandled event 0x%x\n", le16toh(event));
1998 
1999 	/*
2000 	 * This is the only place that the event/reply should be freed.
2001 	 * Anything wanting to hold onto the event data should have
2002 	 * already copied it into their own storage.
2003 	 */
2004 	mps_free_reply(sc, data);
2005 }
2006 
2007 static void
2008 mps_reregister_events_complete(struct mps_softc *sc, struct mps_command *cm)
2009 {
2010 	mps_dprint(sc, MPS_TRACE, "%s\n", __func__);
2011 
2012 	if (cm->cm_reply)
2013 		mps_print_event(sc,
2014 			(MPI2_EVENT_NOTIFICATION_REPLY *)cm->cm_reply);
2015 
2016 	mps_free_command(sc, cm);
2017 
2018 	/* next, send a port enable */
2019 	mpssas_startup(sc);
2020 }
2021 
2022 /*
2023  * For both register_events and update_events, the caller supplies a bitmap
2024  * of events that it _wants_.  These functions then turn that into a bitmask
2025  * suitable for the controller.
2026  */
2027 int
2028 mps_register_events(struct mps_softc *sc, u32 *mask,
2029     mps_evt_callback_t *cb, void *data, struct mps_event_handle **handle)
2030 {
2031 	struct mps_event_handle *eh;
2032 	int error = 0;
2033 
2034 	eh = malloc(sizeof(struct mps_event_handle), M_MPT2, M_WAITOK|M_ZERO);
2035 	if(!eh) {
2036 		device_printf(sc->mps_dev, "Cannot allocate memory %s %d\n",
2037 		 __func__, __LINE__);
2038 		return (ENOMEM);
2039 	}
2040 	eh->callback = cb;
2041 	eh->data = data;
2042 	TAILQ_INSERT_TAIL(&sc->event_list, eh, eh_list);
2043 	if (mask != NULL)
2044 		error = mps_update_events(sc, eh, mask);
2045 	*handle = eh;
2046 
2047 	return (error);
2048 }
2049 
2050 int
2051 mps_update_events(struct mps_softc *sc, struct mps_event_handle *handle,
2052     u32 *mask)
2053 {
2054 	MPI2_EVENT_NOTIFICATION_REQUEST *evtreq;
2055 	MPI2_EVENT_NOTIFICATION_REPLY *reply;
2056 	struct mps_command *cm;
2057 	int error, i;
2058 
2059 	mps_dprint(sc, MPS_TRACE, "%s\n", __func__);
2060 
2061 	if ((mask != NULL) && (handle != NULL))
2062 		bcopy(mask, &handle->mask[0], sizeof(u32) *
2063 				MPI2_EVENT_NOTIFY_EVENTMASK_WORDS);
2064 
2065 	for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++)
2066 		sc->event_mask[i] = -1;
2067 
2068 	for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++)
2069 		sc->event_mask[i] &= ~handle->mask[i];
2070 
2071 
2072 	if ((cm = mps_alloc_command(sc)) == NULL)
2073 		return (EBUSY);
2074 	evtreq = (MPI2_EVENT_NOTIFICATION_REQUEST *)cm->cm_req;
2075 	evtreq->Function = MPI2_FUNCTION_EVENT_NOTIFICATION;
2076 	evtreq->MsgFlags = 0;
2077 	evtreq->SASBroadcastPrimitiveMasks = 0;
2078 #ifdef MPS_DEBUG_ALL_EVENTS
2079 	{
2080 		u_char fullmask[16];
2081 		memset(fullmask, 0x00, 16);
2082 		bcopy(fullmask, &evtreq->EventMasks[0], sizeof(u32) *
2083 				MPI2_EVENT_NOTIFY_EVENTMASK_WORDS);
2084 	}
2085 #else
2086         for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++)
2087                 evtreq->EventMasks[i] =
2088                     htole32(sc->event_mask[i]);
2089 #endif
2090 	cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE;
2091 	cm->cm_data = NULL;
2092 
2093 	error = mps_wait_command(sc, cm, 60, 0);
2094 	reply = (MPI2_EVENT_NOTIFICATION_REPLY *)cm->cm_reply;
2095 	if ((reply == NULL) ||
2096 	    (reply->IOCStatus & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS)
2097 		error = ENXIO;
2098 	mps_print_event(sc, reply);
2099 	mps_dprint(sc, MPS_TRACE, "%s finished error %d\n", __func__, error);
2100 
2101 	mps_free_command(sc, cm);
2102 	return (error);
2103 }
2104 
2105 static int
2106 mps_reregister_events(struct mps_softc *sc)
2107 {
2108 	MPI2_EVENT_NOTIFICATION_REQUEST *evtreq;
2109 	struct mps_command *cm;
2110 	struct mps_event_handle *eh;
2111 	int error, i;
2112 
2113 	mps_dprint(sc, MPS_TRACE, "%s\n", __func__);
2114 
2115 	/* first, reregister events */
2116 
2117 	for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++)
2118 		sc->event_mask[i] = -1;
2119 
2120 	TAILQ_FOREACH(eh, &sc->event_list, eh_list) {
2121 		for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++)
2122 			sc->event_mask[i] &= ~eh->mask[i];
2123 	}
2124 
2125 	if ((cm = mps_alloc_command(sc)) == NULL)
2126 		return (EBUSY);
2127 	evtreq = (MPI2_EVENT_NOTIFICATION_REQUEST *)cm->cm_req;
2128 	evtreq->Function = MPI2_FUNCTION_EVENT_NOTIFICATION;
2129 	evtreq->MsgFlags = 0;
2130 	evtreq->SASBroadcastPrimitiveMasks = 0;
2131 #ifdef MPS_DEBUG_ALL_EVENTS
2132 	{
2133 		u_char fullmask[16];
2134 		memset(fullmask, 0x00, 16);
2135 		bcopy(fullmask, &evtreq->EventMasks[0], sizeof(u32) *
2136 			MPI2_EVENT_NOTIFY_EVENTMASK_WORDS);
2137 	}
2138 #else
2139         for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++)
2140                 evtreq->EventMasks[i] =
2141                     htole32(sc->event_mask[i]);
2142 #endif
2143 	cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE;
2144 	cm->cm_data = NULL;
2145 	cm->cm_complete = mps_reregister_events_complete;
2146 
2147 	error = mps_map_command(sc, cm);
2148 
2149 	mps_dprint(sc, MPS_TRACE, "%s finished with error %d\n", __func__,
2150 	    error);
2151 	return (error);
2152 }
2153 
2154 void
2155 mps_deregister_events(struct mps_softc *sc, struct mps_event_handle *handle)
2156 {
2157 
2158 	TAILQ_REMOVE(&sc->event_list, handle, eh_list);
2159 	free(handle, M_MPT2);
2160 }
2161 
2162 /*
2163  * Add a chain element as the next SGE for the specified command.
2164  * Reset cm_sge and cm_sgesize to indicate all the available space.
2165  */
2166 static int
2167 mps_add_chain(struct mps_command *cm)
2168 {
2169 	MPI2_SGE_CHAIN32 *sgc;
2170 	struct mps_chain *chain;
2171 	int space;
2172 
2173 	if (cm->cm_sglsize < MPS_SGC_SIZE)
2174 		panic("MPS: Need SGE Error Code\n");
2175 
2176 	chain = mps_alloc_chain(cm->cm_sc);
2177 	if (chain == NULL)
2178 		return (ENOBUFS);
2179 
2180 	space = (int)cm->cm_sc->facts->IOCRequestFrameSize * 4;
2181 
2182 	/*
2183 	 * Note: a double-linked list is used to make it easier to
2184 	 * walk for debugging.
2185 	 */
2186 	TAILQ_INSERT_TAIL(&cm->cm_chain_list, chain, chain_link);
2187 
2188 	sgc = (MPI2_SGE_CHAIN32 *)&cm->cm_sge->MpiChain;
2189 	sgc->Length = htole16(space);
2190 	sgc->NextChainOffset = 0;
2191 	/* TODO Looks like bug in Setting sgc->Flags.
2192 	 *	sgc->Flags = ( MPI2_SGE_FLAGS_CHAIN_ELEMENT | MPI2_SGE_FLAGS_64_BIT_ADDRESSING |
2193 	 *	            MPI2_SGE_FLAGS_SYSTEM_ADDRESS) << MPI2_SGE_FLAGS_SHIFT
2194 	 *	This is fine.. because we are not using simple element. In case of
2195 	 *	MPI2_SGE_CHAIN32, we have separate Length and Flags feild.
2196  	 */
2197 	sgc->Flags = MPI2_SGE_FLAGS_CHAIN_ELEMENT;
2198 	sgc->Address = htole32(chain->chain_busaddr);
2199 
2200 	cm->cm_sge = (MPI2_SGE_IO_UNION *)&chain->chain->MpiSimple;
2201 	cm->cm_sglsize = space;
2202 	return (0);
2203 }
2204 
2205 /*
2206  * Add one scatter-gather element (chain, simple, transaction context)
2207  * to the scatter-gather list for a command.  Maintain cm_sglsize and
2208  * cm_sge as the remaining size and pointer to the next SGE to fill
2209  * in, respectively.
2210  */
2211 int
2212 mps_push_sge(struct mps_command *cm, void *sgep, size_t len, int segsleft)
2213 {
2214 	MPI2_SGE_TRANSACTION_UNION *tc = sgep;
2215 	MPI2_SGE_SIMPLE64 *sge = sgep;
2216 	int error, type;
2217 	uint32_t saved_buf_len, saved_address_low, saved_address_high;
2218 
2219 	type = (tc->Flags & MPI2_SGE_FLAGS_ELEMENT_MASK);
2220 
2221 #ifdef INVARIANTS
2222 	switch (type) {
2223 	case MPI2_SGE_FLAGS_TRANSACTION_ELEMENT: {
2224 		if (len != tc->DetailsLength + 4)
2225 			panic("TC %p length %u or %zu?", tc,
2226 			    tc->DetailsLength + 4, len);
2227 		}
2228 		break;
2229 	case MPI2_SGE_FLAGS_CHAIN_ELEMENT:
2230 		/* Driver only uses 32-bit chain elements */
2231 		if (len != MPS_SGC_SIZE)
2232 			panic("CHAIN %p length %u or %zu?", sgep,
2233 			    MPS_SGC_SIZE, len);
2234 		break;
2235 	case MPI2_SGE_FLAGS_SIMPLE_ELEMENT:
2236 		/* Driver only uses 64-bit SGE simple elements */
2237 		if (len != MPS_SGE64_SIZE)
2238 			panic("SGE simple %p length %u or %zu?", sge,
2239 			    MPS_SGE64_SIZE, len);
2240 		if (((le32toh(sge->FlagsLength) >> MPI2_SGE_FLAGS_SHIFT) &
2241 		    MPI2_SGE_FLAGS_ADDRESS_SIZE) == 0)
2242 			panic("SGE simple %p not marked 64-bit?", sge);
2243 
2244 		break;
2245 	default:
2246 		panic("Unexpected SGE %p, flags %02x", tc, tc->Flags);
2247 	}
2248 #endif
2249 
2250 	/*
2251 	 * case 1: 1 more segment, enough room for it
2252 	 * case 2: 2 more segments, enough room for both
2253 	 * case 3: >=2 more segments, only enough room for 1 and a chain
2254 	 * case 4: >=1 more segment, enough room for only a chain
2255 	 * case 5: >=1 more segment, no room for anything (error)
2256          */
2257 
2258 	/*
2259 	 * There should be room for at least a chain element, or this
2260 	 * code is buggy.  Case (5).
2261 	 */
2262 	if (cm->cm_sglsize < MPS_SGC_SIZE)
2263 		panic("MPS: Need SGE Error Code\n");
2264 
2265 	if (segsleft >= 2 &&
2266 	    cm->cm_sglsize < len + MPS_SGC_SIZE + MPS_SGE64_SIZE) {
2267 		/*
2268 		 * There are 2 or more segments left to add, and only
2269 		 * enough room for 1 and a chain.  Case (3).
2270 		 *
2271 		 * Mark as last element in this chain if necessary.
2272 		 */
2273 		if (type == MPI2_SGE_FLAGS_SIMPLE_ELEMENT) {
2274 			sge->FlagsLength |= htole32(
2275 			    MPI2_SGE_FLAGS_LAST_ELEMENT << MPI2_SGE_FLAGS_SHIFT);
2276 		}
2277 
2278 		/*
2279 		 * Add the item then a chain.  Do the chain now,
2280 		 * rather than on the next iteration, to simplify
2281 		 * understanding the code.
2282 		 */
2283 		cm->cm_sglsize -= len;
2284 		bcopy(sgep, cm->cm_sge, len);
2285 		cm->cm_sge = (MPI2_SGE_IO_UNION *)((uintptr_t)cm->cm_sge + len);
2286 		return (mps_add_chain(cm));
2287 	}
2288 
2289 	if (segsleft >= 1 && cm->cm_sglsize < len + MPS_SGC_SIZE) {
2290 		/*
2291 		 * 1 or more segment, enough room for only a chain.
2292 		 * Hope the previous element wasn't a Simple entry
2293 		 * that needed to be marked with
2294 		 * MPI2_SGE_FLAGS_LAST_ELEMENT.  Case (4).
2295 		 */
2296 		if ((error = mps_add_chain(cm)) != 0)
2297 			return (error);
2298 	}
2299 
2300 #ifdef INVARIANTS
2301 	/* Case 1: 1 more segment, enough room for it. */
2302 	if (segsleft == 1 && cm->cm_sglsize < len)
2303 		panic("1 seg left and no room? %u versus %zu",
2304 		    cm->cm_sglsize, len);
2305 
2306 	/* Case 2: 2 more segments, enough room for both */
2307 	if (segsleft == 2 && cm->cm_sglsize < len + MPS_SGE64_SIZE)
2308 		panic("2 segs left and no room? %u versus %zu",
2309 		    cm->cm_sglsize, len);
2310 #endif
2311 
2312 	if (segsleft == 1 && type == MPI2_SGE_FLAGS_SIMPLE_ELEMENT) {
2313 		/*
2314 		 * If this is a bi-directional request, need to account for that
2315 		 * here.  Save the pre-filled sge values.  These will be used
2316 		 * either for the 2nd SGL or for a single direction SGL.  If
2317 		 * cm_out_len is non-zero, this is a bi-directional request, so
2318 		 * fill in the OUT SGL first, then the IN SGL, otherwise just
2319 		 * fill in the IN SGL.  Note that at this time, when filling in
2320 		 * 2 SGL's for a bi-directional request, they both use the same
2321 		 * DMA buffer (same cm command).
2322 		 */
2323 		saved_buf_len = le32toh(sge->FlagsLength) & 0x00FFFFFF;
2324 		saved_address_low = sge->Address.Low;
2325 		saved_address_high = sge->Address.High;
2326 		if (cm->cm_out_len) {
2327 			sge->FlagsLength = htole32(cm->cm_out_len |
2328 			    ((uint32_t)(MPI2_SGE_FLAGS_SIMPLE_ELEMENT |
2329 			    MPI2_SGE_FLAGS_END_OF_BUFFER |
2330 			    MPI2_SGE_FLAGS_HOST_TO_IOC |
2331 			    MPI2_SGE_FLAGS_64_BIT_ADDRESSING) <<
2332 			    MPI2_SGE_FLAGS_SHIFT));
2333 			cm->cm_sglsize -= len;
2334 			bcopy(sgep, cm->cm_sge, len);
2335 			cm->cm_sge = (MPI2_SGE_IO_UNION *)((uintptr_t)cm->cm_sge
2336 			    + len);
2337 		}
2338 		saved_buf_len |=
2339 		    ((uint32_t)(MPI2_SGE_FLAGS_SIMPLE_ELEMENT |
2340 		    MPI2_SGE_FLAGS_END_OF_BUFFER |
2341 		    MPI2_SGE_FLAGS_LAST_ELEMENT |
2342 		    MPI2_SGE_FLAGS_END_OF_LIST |
2343 		    MPI2_SGE_FLAGS_64_BIT_ADDRESSING) <<
2344 		    MPI2_SGE_FLAGS_SHIFT);
2345 		if (cm->cm_flags & MPS_CM_FLAGS_DATAIN) {
2346 			saved_buf_len |=
2347 			    ((uint32_t)(MPI2_SGE_FLAGS_IOC_TO_HOST) <<
2348 			    MPI2_SGE_FLAGS_SHIFT);
2349 		} else {
2350 			saved_buf_len |=
2351 			    ((uint32_t)(MPI2_SGE_FLAGS_HOST_TO_IOC) <<
2352 			    MPI2_SGE_FLAGS_SHIFT);
2353 		}
2354 		sge->FlagsLength = htole32(saved_buf_len);
2355 		sge->Address.Low = saved_address_low;
2356 		sge->Address.High = saved_address_high;
2357 	}
2358 
2359 	cm->cm_sglsize -= len;
2360 	bcopy(sgep, cm->cm_sge, len);
2361 	cm->cm_sge = (MPI2_SGE_IO_UNION *)((uintptr_t)cm->cm_sge + len);
2362 	return (0);
2363 }
2364 
2365 /*
2366  * Add one dma segment to the scatter-gather list for a command.
2367  */
2368 int
2369 mps_add_dmaseg(struct mps_command *cm, vm_paddr_t pa, size_t len, u_int flags,
2370     int segsleft)
2371 {
2372 	MPI2_SGE_SIMPLE64 sge;
2373 
2374 	/*
2375 	 * This driver always uses 64-bit address elements for simplicity.
2376 	 */
2377 	bzero(&sge, sizeof(sge));
2378 	flags |= MPI2_SGE_FLAGS_SIMPLE_ELEMENT |
2379 	    MPI2_SGE_FLAGS_64_BIT_ADDRESSING;
2380 	sge.FlagsLength = htole32(len | (flags << MPI2_SGE_FLAGS_SHIFT));
2381 	mps_from_u64(pa, &sge.Address);
2382 
2383 	return (mps_push_sge(cm, &sge, sizeof sge, segsleft));
2384 }
2385 
2386 static void
2387 mps_data_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
2388 {
2389 	struct mps_softc *sc;
2390 	struct mps_command *cm;
2391 	u_int i, dir, sflags;
2392 
2393 	cm = (struct mps_command *)arg;
2394 	sc = cm->cm_sc;
2395 
2396 	/*
2397 	 * In this case, just print out a warning and let the chip tell the
2398 	 * user they did the wrong thing.
2399 	 */
2400 	if ((cm->cm_max_segs != 0) && (nsegs > cm->cm_max_segs)) {
2401 		mps_dprint(sc, MPS_ERROR,
2402 			   "%s: warning: busdma returned %d segments, "
2403 			   "more than the %d allowed\n", __func__, nsegs,
2404 			   cm->cm_max_segs);
2405 	}
2406 
2407 	/*
2408 	 * Set up DMA direction flags.  Bi-directional requests are also handled
2409 	 * here.  In that case, both direction flags will be set.
2410 	 */
2411 	sflags = 0;
2412 	if (cm->cm_flags & MPS_CM_FLAGS_SMP_PASS) {
2413 		/*
2414 		 * We have to add a special case for SMP passthrough, there
2415 		 * is no easy way to generically handle it.  The first
2416 		 * S/G element is used for the command (therefore the
2417 		 * direction bit needs to be set).  The second one is used
2418 		 * for the reply.  We'll leave it to the caller to make
2419 		 * sure we only have two buffers.
2420 		 */
2421 		/*
2422 		 * Even though the busdma man page says it doesn't make
2423 		 * sense to have both direction flags, it does in this case.
2424 		 * We have one s/g element being accessed in each direction.
2425 		 */
2426 		dir = BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD;
2427 
2428 		/*
2429 		 * Set the direction flag on the first buffer in the SMP
2430 		 * passthrough request.  We'll clear it for the second one.
2431 		 */
2432 		sflags |= MPI2_SGE_FLAGS_DIRECTION |
2433 			  MPI2_SGE_FLAGS_END_OF_BUFFER;
2434 	} else if (cm->cm_flags & MPS_CM_FLAGS_DATAOUT) {
2435 		sflags |= MPI2_SGE_FLAGS_HOST_TO_IOC;
2436 		dir = BUS_DMASYNC_PREWRITE;
2437 	} else
2438 		dir = BUS_DMASYNC_PREREAD;
2439 
2440 	for (i = 0; i < nsegs; i++) {
2441 		if ((cm->cm_flags & MPS_CM_FLAGS_SMP_PASS) && (i != 0)) {
2442 			sflags &= ~MPI2_SGE_FLAGS_DIRECTION;
2443 		}
2444 		error = mps_add_dmaseg(cm, segs[i].ds_addr, segs[i].ds_len,
2445 		    sflags, nsegs - i);
2446 		if (error != 0) {
2447 			/* Resource shortage, roll back! */
2448 			if (ratecheck(&sc->lastfail, &mps_chainfail_interval))
2449 				mps_dprint(sc, MPS_INFO, "Out of chain frames, "
2450 				    "consider increasing hw.mps.max_chains.\n");
2451 			cm->cm_flags |= MPS_CM_FLAGS_CHAIN_FAILED;
2452 			mps_complete_command(sc, cm);
2453 			return;
2454 		}
2455 	}
2456 
2457 	bus_dmamap_sync(sc->buffer_dmat, cm->cm_dmamap, dir);
2458 	mps_enqueue_request(sc, cm);
2459 
2460 	return;
2461 }
2462 
2463 static void
2464 mps_data_cb2(void *arg, bus_dma_segment_t *segs, int nsegs, bus_size_t mapsize,
2465 	     int error)
2466 {
2467 	mps_data_cb(arg, segs, nsegs, error);
2468 }
2469 
2470 /*
2471  * This is the routine to enqueue commands ansynchronously.
2472  * Note that the only error path here is from bus_dmamap_load(), which can
2473  * return EINPROGRESS if it is waiting for resources.  Other than this, it's
2474  * assumed that if you have a command in-hand, then you have enough credits
2475  * to use it.
2476  */
2477 int
2478 mps_map_command(struct mps_softc *sc, struct mps_command *cm)
2479 {
2480 	int error = 0;
2481 
2482 	if (cm->cm_flags & MPS_CM_FLAGS_USE_UIO) {
2483 		error = bus_dmamap_load_uio(sc->buffer_dmat, cm->cm_dmamap,
2484 		    &cm->cm_uio, mps_data_cb2, cm, 0);
2485 	} else if (cm->cm_flags & MPS_CM_FLAGS_USE_CCB) {
2486 		error = bus_dmamap_load_ccb(sc->buffer_dmat, cm->cm_dmamap,
2487 		    cm->cm_data, mps_data_cb, cm, 0);
2488 	} else if ((cm->cm_data != NULL) && (cm->cm_length != 0)) {
2489 		error = bus_dmamap_load(sc->buffer_dmat, cm->cm_dmamap,
2490 		    cm->cm_data, cm->cm_length, mps_data_cb, cm, 0);
2491 	} else {
2492 		/* Add a zero-length element as needed */
2493 		if (cm->cm_sge != NULL)
2494 			mps_add_dmaseg(cm, 0, 0, 0, 1);
2495 		mps_enqueue_request(sc, cm);
2496 	}
2497 
2498 	return (error);
2499 }
2500 
2501 /*
2502  * This is the routine to enqueue commands synchronously.  An error of
2503  * EINPROGRESS from mps_map_command() is ignored since the command will
2504  * be executed and enqueued automatically.  Other errors come from msleep().
2505  */
2506 int
2507 mps_wait_command(struct mps_softc *sc, struct mps_command *cm, int timeout,
2508     int sleep_flag)
2509 {
2510 	int error, rc;
2511 	struct timeval cur_time, start_time;
2512 
2513 	if (sc->mps_flags & MPS_FLAGS_DIAGRESET)
2514 		return  EBUSY;
2515 
2516 	cm->cm_complete = NULL;
2517 	cm->cm_flags |= MPS_CM_FLAGS_POLLED;
2518 	error = mps_map_command(sc, cm);
2519 	if ((error != 0) && (error != EINPROGRESS))
2520 		return (error);
2521 
2522 	/*
2523 	 * Check for context and wait for 50 mSec at a time until time has
2524 	 * expired or the command has finished.  If msleep can't be used, need
2525 	 * to poll.
2526 	 */
2527 	if (curthread->td_no_sleeping != 0)
2528 		sleep_flag = NO_SLEEP;
2529 	getmicrotime(&start_time);
2530 	if (mtx_owned(&sc->mps_mtx) && sleep_flag == CAN_SLEEP) {
2531 		cm->cm_flags |= MPS_CM_FLAGS_WAKEUP;
2532 		error = msleep(cm, &sc->mps_mtx, 0, "mpswait", timeout*hz);
2533 	} else {
2534 		while ((cm->cm_flags & MPS_CM_FLAGS_COMPLETE) == 0) {
2535 			mps_intr_locked(sc);
2536 			if (sleep_flag == CAN_SLEEP)
2537 				pause("mpswait", hz/20);
2538 			else
2539 				DELAY(50000);
2540 
2541 			getmicrotime(&cur_time);
2542 			if ((cur_time.tv_sec - start_time.tv_sec) > timeout) {
2543 				error = EWOULDBLOCK;
2544 				break;
2545 			}
2546 		}
2547 	}
2548 
2549 	if (error == EWOULDBLOCK) {
2550 		mps_dprint(sc, MPS_FAULT, "Calling Reinit from %s\n", __func__);
2551 		rc = mps_reinit(sc);
2552 		mps_dprint(sc, MPS_FAULT, "Reinit %s\n", (rc == 0) ? "success" :
2553 		    "failed");
2554 		error = ETIMEDOUT;
2555 	}
2556 	return (error);
2557 }
2558 
2559 /*
2560  * The MPT driver had a verbose interface for config pages.  In this driver,
2561  * reduce it to much simpler terms, similar to the Linux driver.
2562  */
2563 int
2564 mps_read_config_page(struct mps_softc *sc, struct mps_config_params *params)
2565 {
2566 	MPI2_CONFIG_REQUEST *req;
2567 	struct mps_command *cm;
2568 	int error;
2569 
2570 	if (sc->mps_flags & MPS_FLAGS_BUSY) {
2571 		return (EBUSY);
2572 	}
2573 
2574 	cm = mps_alloc_command(sc);
2575 	if (cm == NULL) {
2576 		return (EBUSY);
2577 	}
2578 
2579 	req = (MPI2_CONFIG_REQUEST *)cm->cm_req;
2580 	req->Function = MPI2_FUNCTION_CONFIG;
2581 	req->Action = params->action;
2582 	req->SGLFlags = 0;
2583 	req->ChainOffset = 0;
2584 	req->PageAddress = params->page_address;
2585 	if (params->hdr.Struct.PageType == MPI2_CONFIG_PAGETYPE_EXTENDED) {
2586 		MPI2_CONFIG_EXTENDED_PAGE_HEADER *hdr;
2587 
2588 		hdr = &params->hdr.Ext;
2589 		req->ExtPageType = hdr->ExtPageType;
2590 		req->ExtPageLength = hdr->ExtPageLength;
2591 		req->Header.PageType = MPI2_CONFIG_PAGETYPE_EXTENDED;
2592 		req->Header.PageLength = 0; /* Must be set to zero */
2593 		req->Header.PageNumber = hdr->PageNumber;
2594 		req->Header.PageVersion = hdr->PageVersion;
2595 	} else {
2596 		MPI2_CONFIG_PAGE_HEADER *hdr;
2597 
2598 		hdr = &params->hdr.Struct;
2599 		req->Header.PageType = hdr->PageType;
2600 		req->Header.PageNumber = hdr->PageNumber;
2601 		req->Header.PageLength = hdr->PageLength;
2602 		req->Header.PageVersion = hdr->PageVersion;
2603 	}
2604 
2605 	cm->cm_data = params->buffer;
2606 	cm->cm_length = params->length;
2607 	if (cm->cm_data != NULL) {
2608 		cm->cm_sge = &req->PageBufferSGE;
2609 		cm->cm_sglsize = sizeof(MPI2_SGE_IO_UNION);
2610 		cm->cm_flags = MPS_CM_FLAGS_SGE_SIMPLE | MPS_CM_FLAGS_DATAIN;
2611 	} else
2612 		cm->cm_sge = NULL;
2613 	cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE;
2614 
2615 	cm->cm_complete_data = params;
2616 	if (params->callback != NULL) {
2617 		cm->cm_complete = mps_config_complete;
2618 		return (mps_map_command(sc, cm));
2619 	} else {
2620 		error = mps_wait_command(sc, cm, 0, CAN_SLEEP);
2621 		if (error) {
2622 			mps_dprint(sc, MPS_FAULT,
2623 			    "Error %d reading config page\n", error);
2624 			mps_free_command(sc, cm);
2625 			return (error);
2626 		}
2627 		mps_config_complete(sc, cm);
2628 	}
2629 
2630 	return (0);
2631 }
2632 
2633 int
2634 mps_write_config_page(struct mps_softc *sc, struct mps_config_params *params)
2635 {
2636 	return (EINVAL);
2637 }
2638 
2639 static void
2640 mps_config_complete(struct mps_softc *sc, struct mps_command *cm)
2641 {
2642 	MPI2_CONFIG_REPLY *reply;
2643 	struct mps_config_params *params;
2644 
2645 	MPS_FUNCTRACE(sc);
2646 	params = cm->cm_complete_data;
2647 
2648 	if (cm->cm_data != NULL) {
2649 		bus_dmamap_sync(sc->buffer_dmat, cm->cm_dmamap,
2650 		    BUS_DMASYNC_POSTREAD);
2651 		bus_dmamap_unload(sc->buffer_dmat, cm->cm_dmamap);
2652 	}
2653 
2654 	/*
2655 	 * XXX KDM need to do more error recovery?  This results in the
2656 	 * device in question not getting probed.
2657 	 */
2658 	if ((cm->cm_flags & MPS_CM_FLAGS_ERROR_MASK) != 0) {
2659 		params->status = MPI2_IOCSTATUS_BUSY;
2660 		goto done;
2661 	}
2662 
2663 	reply = (MPI2_CONFIG_REPLY *)cm->cm_reply;
2664 	if (reply == NULL) {
2665 		params->status = MPI2_IOCSTATUS_BUSY;
2666 		goto done;
2667 	}
2668 	params->status = reply->IOCStatus;
2669 	if (params->hdr.Struct.PageType == MPI2_CONFIG_PAGETYPE_EXTENDED) {
2670 		params->hdr.Ext.ExtPageType = reply->ExtPageType;
2671 		params->hdr.Ext.ExtPageLength = reply->ExtPageLength;
2672 		params->hdr.Ext.PageType = reply->Header.PageType;
2673 		params->hdr.Ext.PageNumber = reply->Header.PageNumber;
2674 		params->hdr.Ext.PageVersion = reply->Header.PageVersion;
2675 	} else {
2676 		params->hdr.Struct.PageType = reply->Header.PageType;
2677 		params->hdr.Struct.PageNumber = reply->Header.PageNumber;
2678 		params->hdr.Struct.PageLength = reply->Header.PageLength;
2679 		params->hdr.Struct.PageVersion = reply->Header.PageVersion;
2680 	}
2681 
2682 done:
2683 	mps_free_command(sc, cm);
2684 	if (params->callback != NULL)
2685 		params->callback(sc, params);
2686 
2687 	return;
2688 }
2689