xref: /netbsd/sys/dev/raidframe/rf_states.c (revision acfd5966)
1 /*	$NetBSD: rf_states.c,v 1.53 2021/07/23 02:35:14 oster Exp $	*/
2 /*
3  * Copyright (c) 1995 Carnegie-Mellon University.
4  * All rights reserved.
5  *
6  * Author: Mark Holland, William V. Courtright II, Robby Findler
7  *
8  * Permission to use, copy, modify and distribute this software and
9  * its documentation is hereby granted, provided that both the copyright
10  * notice and this permission notice appear in all copies of the
11  * software, derivative works or modified versions, and any portions
12  * thereof, and that both notices appear in supporting documentation.
13  *
14  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
15  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
16  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
17  *
18  * Carnegie Mellon requests users of this software to return to
19  *
20  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
21  *  School of Computer Science
22  *  Carnegie Mellon University
23  *  Pittsburgh PA 15213-3890
24  *
25  * any improvements or extensions that they make and grant Carnegie the
26  * rights to redistribute these changes.
27  */
28 
29 #include <sys/cdefs.h>
30 __KERNEL_RCSID(0, "$NetBSD: rf_states.c,v 1.53 2021/07/23 02:35:14 oster Exp $");
31 
32 #include <sys/errno.h>
33 
34 #include "rf_archs.h"
35 #include "rf_threadstuff.h"
36 #include "rf_raid.h"
37 #include "rf_dag.h"
38 #include "rf_desc.h"
39 #include "rf_aselect.h"
40 #include "rf_general.h"
41 #include "rf_states.h"
42 #include "rf_dagutils.h"
43 #include "rf_driver.h"
44 #include "rf_engine.h"
45 #include "rf_map.h"
46 #include "rf_etimer.h"
47 #include "rf_kintf.h"
48 #include "rf_paritymap.h"
49 
50 #ifndef RF_DEBUG_STATES
51 #define RF_DEBUG_STATES 0
52 #endif
53 
54 /* prototypes for some of the available states.
55 
56    States must:
57 
58      - not block.
59 
60      - either schedule rf_ContinueRaidAccess as a callback and return
61        RF_TRUE, or complete all of their work and return RF_FALSE.
62 
63      - increment desc->state when they have finished their work.
64 */
65 
66 #if RF_DEBUG_STATES
67 static char *
StateName(RF_AccessState_t state)68 StateName(RF_AccessState_t state)
69 {
70 	switch (state) {
71 		case rf_QuiesceState:return "QuiesceState";
72 	case rf_MapState:
73 		return "MapState";
74 	case rf_LockState:
75 		return "LockState";
76 	case rf_CreateDAGState:
77 		return "CreateDAGState";
78 	case rf_ExecuteDAGState:
79 		return "ExecuteDAGState";
80 	case rf_ProcessDAGState:
81 		return "ProcessDAGState";
82 	case rf_CleanupState:
83 		return "CleanupState";
84 	case rf_LastState:
85 		return "LastState";
86 	case rf_IncrAccessesCountState:
87 		return "IncrAccessesCountState";
88 	case rf_DecrAccessesCountState:
89 		return "DecrAccessesCountState";
90 	default:
91 		return "!!! UnnamedState !!!";
92 	}
93 }
94 #endif
95 
96 void
rf_ContinueRaidAccess(void * v)97 rf_ContinueRaidAccess(void *v)
98 {
99 	RF_RaidAccessDesc_t *desc = v;
100 	int     suspended = RF_FALSE;
101 	int     current_state_index = desc->state;
102 	RF_AccessState_t current_state = desc->states[current_state_index];
103 #if RF_DEBUG_STATES
104 	int     unit = desc->raidPtr->raidid;
105 #endif
106 
107 	do {
108 
109 		current_state_index = desc->state;
110 		current_state = desc->states[current_state_index];
111 
112 		switch (current_state) {
113 
114 		case rf_QuiesceState:
115 			suspended = rf_State_Quiesce(desc);
116 			break;
117 		case rf_IncrAccessesCountState:
118 			suspended = rf_State_IncrAccessCount(desc);
119 			break;
120 		case rf_MapState:
121 			suspended = rf_State_Map(desc);
122 			break;
123 		case rf_LockState:
124 			suspended = rf_State_Lock(desc);
125 			break;
126 		case rf_CreateDAGState:
127 			suspended = rf_State_CreateDAG(desc);
128 			break;
129 		case rf_ExecuteDAGState:
130 			suspended = rf_State_ExecuteDAG(desc);
131 			break;
132 		case rf_ProcessDAGState:
133 			suspended = rf_State_ProcessDAG(desc);
134 			break;
135 		case rf_CleanupState:
136 			suspended = rf_State_Cleanup(desc);
137 			break;
138 		case rf_DecrAccessesCountState:
139 			suspended = rf_State_DecrAccessCount(desc);
140 			break;
141 		case rf_LastState:
142 			suspended = rf_State_LastState(desc);
143 			break;
144 		}
145 
146 		/* after this point, we cannot dereference desc since
147 		 * desc may have been freed. desc is only freed in
148 		 * LastState, so if we renter this function or loop
149 		 * back up, desc should be valid. */
150 
151 #if RF_DEBUG_STATES
152 		if (rf_printStatesDebug) {
153 			printf("raid%d: State: %-24s StateIndex: %3i desc: 0x%ld %s\n",
154 			       unit, StateName(current_state),
155 			       current_state_index, (long) desc,
156 			       suspended ? "callback scheduled" : "looping");
157 		}
158 #endif
159 	} while (!suspended && current_state != rf_LastState);
160 
161 	return;
162 }
163 
164 
165 void
rf_ContinueDagAccess(RF_DagList_t * dagList)166 rf_ContinueDagAccess(RF_DagList_t *dagList)
167 {
168 #if RF_ACC_TRACE > 0
169 	RF_AccTraceEntry_t *tracerec = &(dagList->desc->tracerec);
170 	RF_Etimer_t timer;
171 #endif
172 	RF_RaidAccessDesc_t *desc;
173 	RF_DagHeader_t *dag_h;
174 	int     i;
175 
176 	desc = dagList->desc;
177 
178 #if RF_ACC_TRACE > 0
179 	timer = tracerec->timer;
180 	RF_ETIMER_STOP(timer);
181 	RF_ETIMER_EVAL(timer);
182 	tracerec->specific.user.exec_us = RF_ETIMER_VAL_US(timer);
183 	RF_ETIMER_START(tracerec->timer);
184 #endif
185 
186 	/* skip to dag which just finished */
187 	dag_h = dagList->dags;
188 	for (i = 0; i < dagList->numDagsDone; i++) {
189 		dag_h = dag_h->next;
190 	}
191 
192 	/* check to see if retry is required */
193 	if (dag_h->status == rf_rollBackward) {
194 		/* when a dag fails, mark desc status as bad and allow
195 		 * all other dags in the desc to execute to
196 		 * completion.  then, free all dags and start over */
197 		desc->status = 1;	/* bad status */
198 #if 0
199 		printf("raid%d: DAG failure: %c addr 0x%lx "
200 		       "(%ld) nblk 0x%x (%d) buf 0x%lx state %d\n",
201 		       desc->raidPtr->raidid, desc->type,
202 		       (long) desc->raidAddress,
203 		       (long) desc->raidAddress, (int) desc->numBlocks,
204 		       (int) desc->numBlocks,
205 		       (unsigned long) (desc->bufPtr), desc->state);
206 #endif
207 	}
208 	dagList->numDagsDone++;
209 	rf_ContinueRaidAccess(desc);
210 }
211 
212 int
rf_State_LastState(RF_RaidAccessDesc_t * desc)213 rf_State_LastState(RF_RaidAccessDesc_t *desc)
214 {
215 	void    (*callbackFunc) (void *) = desc->callbackFunc;
216 	void * callbackArg = desc->callbackArg;
217 
218 	/*
219 	 * The parity_map hook has to go here, because the iodone
220 	 * callback goes straight into the kintf layer.
221 	 */
222 	if (desc->raidPtr->parity_map != NULL &&
223 	    desc->type == RF_IO_TYPE_WRITE)
224 		rf_paritymap_end(desc->raidPtr->parity_map,
225 		    desc->raidAddress, desc->numBlocks);
226 
227 	/* printf("Calling raiddone on 0x%x\n",desc->bp); */
228 	raiddone(desc->raidPtr, desc->bp); /* access came through ioctl */
229 
230 	if (callbackFunc)
231 		callbackFunc(callbackArg);
232 	rf_FreeRaidAccDesc(desc);
233 
234 	return RF_FALSE;
235 }
236 
237 int
rf_State_IncrAccessCount(RF_RaidAccessDesc_t * desc)238 rf_State_IncrAccessCount(RF_RaidAccessDesc_t *desc)
239 {
240 	RF_Raid_t *raidPtr;
241 
242 	raidPtr = desc->raidPtr;
243 	/* Bummer. We have to do this to be 100% safe w.r.t. the increment
244 	 * below */
245 	rf_lock_mutex2(raidPtr->access_suspend_mutex);
246 	raidPtr->accs_in_flight++;	/* used to detect quiescence */
247 	rf_unlock_mutex2(raidPtr->access_suspend_mutex);
248 
249 	desc->state++;
250 	return RF_FALSE;
251 }
252 
253 int
rf_State_DecrAccessCount(RF_RaidAccessDesc_t * desc)254 rf_State_DecrAccessCount(RF_RaidAccessDesc_t *desc)
255 {
256 	RF_Raid_t *raidPtr;
257 
258 	raidPtr = desc->raidPtr;
259 
260 	rf_lock_mutex2(raidPtr->access_suspend_mutex);
261 	raidPtr->accs_in_flight--;
262 	if (raidPtr->accesses_suspended && raidPtr->accs_in_flight == 0) {
263 		rf_SignalQuiescenceLock(raidPtr);
264 	}
265 	rf_unlock_mutex2(raidPtr->access_suspend_mutex);
266 
267 	desc->state++;
268 	return RF_FALSE;
269 }
270 
271 int
rf_State_Quiesce(RF_RaidAccessDesc_t * desc)272 rf_State_Quiesce(RF_RaidAccessDesc_t *desc)
273 {
274 #if RF_ACC_TRACE > 0
275 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
276 	RF_Etimer_t timer;
277 #endif
278 	RF_CallbackFuncDesc_t *cb;
279 	RF_Raid_t *raidPtr;
280 	int     suspended = RF_FALSE;
281 	int need_cb, used_cb;
282 
283 	raidPtr = desc->raidPtr;
284 
285 #if RF_ACC_TRACE > 0
286 	RF_ETIMER_START(timer);
287 	RF_ETIMER_START(desc->timer);
288 #endif
289 
290 	need_cb = 0;
291 	used_cb = 0;
292 	cb = NULL;
293 
294 	rf_lock_mutex2(raidPtr->access_suspend_mutex);
295 	/* Do an initial check to see if we might need a callback structure */
296 	if (raidPtr->accesses_suspended) {
297 		need_cb = 1;
298 	}
299 	rf_unlock_mutex2(raidPtr->access_suspend_mutex);
300 
301 	if (need_cb) {
302 		/* create a callback if we might need it...
303 		   and we likely do. */
304 		cb = rf_AllocCallbackFuncDesc(raidPtr);
305 	}
306 
307 	rf_lock_mutex2(raidPtr->access_suspend_mutex);
308 	if (raidPtr->accesses_suspended) {
309 		cb->callbackFunc = rf_ContinueRaidAccess;
310 		cb->callbackArg = desc;
311 		cb->next = raidPtr->quiesce_wait_list;
312 		raidPtr->quiesce_wait_list = cb;
313 		suspended = RF_TRUE;
314 		used_cb = 1;
315 	}
316 	rf_unlock_mutex2(raidPtr->access_suspend_mutex);
317 
318 	if ((need_cb == 1) && (used_cb == 0)) {
319 		rf_FreeCallbackFuncDesc(raidPtr, cb);
320 	}
321 
322 #if RF_ACC_TRACE > 0
323 	RF_ETIMER_STOP(timer);
324 	RF_ETIMER_EVAL(timer);
325 	tracerec->specific.user.suspend_ovhd_us += RF_ETIMER_VAL_US(timer);
326 #endif
327 
328 #if RF_DEBUG_QUIESCE
329 	if (suspended && rf_quiesceDebug)
330 		printf("Stalling access due to quiescence lock\n");
331 #endif
332 	desc->state++;
333 	return suspended;
334 }
335 
336 int
rf_State_Map(RF_RaidAccessDesc_t * desc)337 rf_State_Map(RF_RaidAccessDesc_t *desc)
338 {
339 	RF_Raid_t *raidPtr = desc->raidPtr;
340 #if RF_ACC_TRACE > 0
341 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
342 	RF_Etimer_t timer;
343 
344 	RF_ETIMER_START(timer);
345 #endif
346 
347 	if (!(desc->asmap = rf_MapAccess(raidPtr, desc->raidAddress, desc->numBlocks,
348 		    desc->bufPtr, RF_DONT_REMAP)))
349 		RF_PANIC();
350 
351 #if RF_ACC_TRACE > 0
352 	RF_ETIMER_STOP(timer);
353 	RF_ETIMER_EVAL(timer);
354 	tracerec->specific.user.map_us = RF_ETIMER_VAL_US(timer);
355 #endif
356 
357 	desc->state++;
358 	return RF_FALSE;
359 }
360 
361 int
rf_State_Lock(RF_RaidAccessDesc_t * desc)362 rf_State_Lock(RF_RaidAccessDesc_t *desc)
363 {
364 #if RF_ACC_TRACE > 0
365 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
366 	RF_Etimer_t timer;
367 #endif
368 	RF_Raid_t *raidPtr = desc->raidPtr;
369 	RF_AccessStripeMapHeader_t *asmh = desc->asmap;
370 	RF_AccessStripeMap_t *asm_p;
371 	RF_StripeNum_t lastStripeID = -1;
372 	int     suspended = RF_FALSE;
373 
374 #if RF_ACC_TRACE > 0
375 	RF_ETIMER_START(timer);
376 #endif
377 
378 	/* acquire each lock that we don't already hold */
379 	for (asm_p = asmh->stripeMap; asm_p; asm_p = asm_p->next) {
380 		RF_ASSERT(RF_IO_IS_R_OR_W(desc->type));
381 		if (!rf_suppressLocksAndLargeWrites &&
382 		    asm_p->parityInfo &&
383 		    !(desc->flags & RF_DAG_SUPPRESS_LOCKS) &&
384 		    !(asm_p->flags & RF_ASM_FLAGS_LOCK_TRIED)) {
385 			asm_p->flags |= RF_ASM_FLAGS_LOCK_TRIED;
386 				/* locks must be acquired hierarchically */
387 			RF_ASSERT(asm_p->stripeID > lastStripeID);
388 			lastStripeID = asm_p->stripeID;
389 
390 			RF_INIT_LOCK_REQ_DESC(asm_p->lockReqDesc, desc->type,
391 					      rf_ContinueRaidAccess, desc, asm_p,
392 					      raidPtr->Layout.dataSectorsPerStripe);
393 			if (rf_AcquireStripeLock(raidPtr, raidPtr->lockTable, asm_p->stripeID,
394 						 &asm_p->lockReqDesc)) {
395 				suspended = RF_TRUE;
396 				break;
397 			}
398 		}
399 		if (desc->type == RF_IO_TYPE_WRITE &&
400 		    raidPtr->status == rf_rs_reconstructing) {
401 			if (!(asm_p->flags & RF_ASM_FLAGS_FORCE_TRIED)) {
402 				int     val;
403 
404 				asm_p->flags |= RF_ASM_FLAGS_FORCE_TRIED;
405 				val = rf_ForceOrBlockRecon(raidPtr, asm_p,
406 							   rf_ContinueRaidAccess, desc);
407 				if (val == 0) {
408 					asm_p->flags |= RF_ASM_FLAGS_RECON_BLOCKED;
409 				} else {
410 					suspended = RF_TRUE;
411 					break;
412 				}
413 			} else {
414 #if RF_DEBUG_PSS > 0
415 				if (rf_pssDebug) {
416 					printf("raid%d: skipping force/block because already done, psid %ld\n",
417 					       desc->raidPtr->raidid,
418 					       (long) asm_p->stripeID);
419 				}
420 #endif
421 			}
422 		} else {
423 #if RF_DEBUG_PSS > 0
424 			if (rf_pssDebug) {
425 				printf("raid%d: skipping force/block because not write or not under recon, psid %ld\n",
426 				       desc->raidPtr->raidid,
427 				       (long) asm_p->stripeID);
428 			}
429 #endif
430 		}
431 	}
432 #if RF_ACC_TRACE > 0
433 	RF_ETIMER_STOP(timer);
434 	RF_ETIMER_EVAL(timer);
435 	tracerec->specific.user.lock_us += RF_ETIMER_VAL_US(timer);
436 #endif
437 	if (suspended)
438 		return (RF_TRUE);
439 
440 	desc->state++;
441 	return (RF_FALSE);
442 }
443 /*
444  * the following three states create, execute, and post-process dags
445  * the error recovery unit is a single dag.
446  * by default, SelectAlgorithm creates an array of dags, one per parity stripe
447  * in some tricky cases, multiple dags per stripe are created
448  *   - dags within a parity stripe are executed sequentially (arbitrary order)
449  *   - dags for distinct parity stripes are executed concurrently
450  *
451  * repeat until all dags complete successfully -or- dag selection fails
452  *
453  * while !done
454  *   create dag(s) (SelectAlgorithm)
455  *   if dag
456  *     execute dag (DispatchDAG)
457  *     if dag successful
458  *       done (SUCCESS)
459  *     else
460  *       !done (RETRY - start over with new dags)
461  *   else
462  *     done (FAIL)
463  */
464 int
rf_State_CreateDAG(RF_RaidAccessDesc_t * desc)465 rf_State_CreateDAG(RF_RaidAccessDesc_t *desc)
466 {
467 #if RF_ACC_TRACE > 0
468 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
469 	RF_Etimer_t timer;
470 #endif
471 	RF_DagHeader_t *dag_h;
472 	RF_DagList_t *dagList;
473 	struct buf *bp;
474 	int     i, selectStatus;
475 
476 	/* generate a dag for the access, and fire it off.  When the dag
477 	 * completes, we'll get re-invoked in the next state. */
478 #if RF_ACC_TRACE > 0
479 	RF_ETIMER_START(timer);
480 #endif
481 	/* SelectAlgorithm returns one or more dags */
482 	selectStatus = rf_SelectAlgorithm(desc, desc->flags | RF_DAG_SUPPRESS_LOCKS);
483 #if RF_DEBUG_VALIDATE_DAG
484 	if (rf_printDAGsDebug) {
485 		dagList = desc->dagList;
486 		for (i = 0; i < desc->numStripes; i++) {
487 			rf_PrintDAGList(dagList->dags);
488 			dagList = dagList->next;
489 		}
490 	}
491 #endif /* RF_DEBUG_VALIDATE_DAG */
492 #if RF_ACC_TRACE > 0
493 	RF_ETIMER_STOP(timer);
494 	RF_ETIMER_EVAL(timer);
495 	/* update time to create all dags */
496 	tracerec->specific.user.dag_create_us = RF_ETIMER_VAL_US(timer);
497 #endif
498 
499 	desc->status = 0;	/* good status */
500 
501 	if (selectStatus || (desc->numRetries > RF_RETRY_THRESHOLD)) {
502 		/* failed to create a dag */
503 		/* this happens when there are too many faults or incomplete
504 		 * dag libraries */
505 		if (selectStatus) {
506 			printf("raid%d: failed to create a dag. "
507 			       "Too many component failures.\n",
508 			       desc->raidPtr->raidid);
509 		} else {
510 			printf("raid%d: IO failed after %d retries.\n",
511 			       desc->raidPtr->raidid, RF_RETRY_THRESHOLD);
512 		}
513 
514 		desc->status = 1; /* bad status */
515 		/* skip straight to rf_State_Cleanup() */
516 		desc->state = rf_CleanupState;
517 		bp = (struct buf *)desc->bp;
518 		bp->b_error = EIO;
519 		bp->b_resid = bp->b_bcount;
520 	} else {
521 		/* bind dags to desc */
522 		dagList = desc->dagList;
523 		for (i = 0; i < desc->numStripes; i++) {
524 			dag_h = dagList->dags;
525 			while (dag_h) {
526 				dag_h->bp = (struct buf *) desc->bp;
527 #if RF_ACC_TRACE > 0
528 				dag_h->tracerec = tracerec;
529 #endif
530 				dag_h = dag_h->next;
531 			}
532 			dagList = dagList->next;
533 		}
534 		desc->flags |= RF_DAG_DISPATCH_RETURNED;
535 		desc->state++;	/* next state should be rf_State_ExecuteDAG */
536 	}
537 	return RF_FALSE;
538 }
539 
540 
541 
542 /* the access has an list of dagLists, one dagList per parity stripe.
543  * fire the first dag in each parity stripe (dagList).
544  * dags within a stripe (dagList) must be executed sequentially
545  *  - this preserves atomic parity update
546  * dags for independents parity groups (stripes) are fired concurrently */
547 
548 int
rf_State_ExecuteDAG(RF_RaidAccessDesc_t * desc)549 rf_State_ExecuteDAG(RF_RaidAccessDesc_t *desc)
550 {
551 	int     i;
552 	RF_DagHeader_t *dag_h;
553 	RF_DagList_t *dagList;
554 
555 	/* next state is always rf_State_ProcessDAG important to do
556 	 * this before firing the first dag (it may finish before we
557 	 * leave this routine) */
558 	desc->state++;
559 
560 	/* sweep dag array, a stripe at a time, firing the first dag
561 	 * in each stripe */
562 	dagList = desc->dagList;
563 	for (i = 0; i < desc->numStripes; i++) {
564 		RF_ASSERT(dagList->numDags > 0);
565 		RF_ASSERT(dagList->numDagsDone == 0);
566 		RF_ASSERT(dagList->numDagsFired == 0);
567 #if RF_ACC_TRACE > 0
568 		RF_ETIMER_START(dagList->tracerec.timer);
569 #endif
570 		/* fire first dag in this stripe */
571 		dag_h = dagList->dags;
572 		RF_ASSERT(dag_h);
573 		dagList->numDagsFired++;
574 		rf_DispatchDAG(dag_h, (void (*) (void *)) rf_ContinueDagAccess, dagList);
575 		dagList = dagList->next;
576 	}
577 
578 	/* the DAG will always call the callback, even if there was no
579 	 * blocking, so we are always suspended in this state */
580 	return RF_TRUE;
581 }
582 
583 
584 
585 /* rf_State_ProcessDAG is entered when a dag completes.
586  * first, check to all dags in the access have completed
587  * if not, fire as many dags as possible */
588 
589 int
rf_State_ProcessDAG(RF_RaidAccessDesc_t * desc)590 rf_State_ProcessDAG(RF_RaidAccessDesc_t *desc)
591 {
592 	RF_AccessStripeMapHeader_t *asmh = desc->asmap;
593 	RF_Raid_t *raidPtr = desc->raidPtr;
594 	RF_DagHeader_t *dag_h;
595 	int     i, j, done = RF_TRUE;
596 	RF_DagList_t *dagList, *temp;
597 
598 	/* check to see if this is the last dag */
599 	dagList = desc->dagList;
600 	for (i = 0; i < desc->numStripes; i++) {
601 		if (dagList->numDags != dagList->numDagsDone)
602 			done = RF_FALSE;
603 		dagList = dagList->next;
604 	}
605 
606 	if (done) {
607 		if (desc->status) {
608 			/* a dag failed, retry */
609 			/* free all dags */
610 			dagList = desc->dagList;
611 			for (i = 0; i < desc->numStripes; i++) {
612 				rf_FreeDAG(dagList->dags);
613 				temp = dagList;
614 				dagList = dagList->next;
615 				rf_FreeDAGList(raidPtr, temp);
616 			}
617 			desc->dagList = NULL;
618 
619 			rf_MarkFailuresInASMList(raidPtr, asmh);
620 
621 			/* note the retry so that we'll bail in
622 			   rf_State_CreateDAG() once we've retired
623 			   the IO RF_RETRY_THRESHOLD times */
624 
625 			desc->numRetries++;
626 
627 			/* back up to rf_State_CreateDAG */
628 			desc->state = desc->state - 2;
629 			return RF_FALSE;
630 		} else {
631 			/* move on to rf_State_Cleanup */
632 			desc->state++;
633 		}
634 		return RF_FALSE;
635 	} else {
636 		/* more dags to execute */
637 		/* see if any are ready to be fired.  if so, fire them */
638 		/* don't fire the initial dag in a list, it's fired in
639 		 * rf_State_ExecuteDAG */
640 		dagList = desc->dagList;
641 		for (i = 0; i < desc->numStripes; i++) {
642 			if ((dagList->numDagsDone < dagList->numDags)
643 			    && (dagList->numDagsDone == dagList->numDagsFired)
644 			    && (dagList->numDagsFired > 0)) {
645 #if RF_ACC_TRACE > 0
646 				RF_ETIMER_START(dagList->tracerec.timer);
647 #endif
648 				/* fire next dag in this stripe */
649 				/* first, skip to next dag awaiting execution */
650 				dag_h = dagList->dags;
651 				for (j = 0; j < dagList->numDagsDone; j++)
652 					dag_h = dag_h->next;
653 				dagList->numDagsFired++;
654 				rf_DispatchDAG(dag_h, (void (*) (void *)) rf_ContinueDagAccess,
655 				    dagList);
656 			}
657 			dagList = dagList->next;
658 		}
659 		return RF_TRUE;
660 	}
661 }
662 /* only make it this far if all dags complete successfully */
663 int
rf_State_Cleanup(RF_RaidAccessDesc_t * desc)664 rf_State_Cleanup(RF_RaidAccessDesc_t *desc)
665 {
666 #if RF_ACC_TRACE > 0
667 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
668 	RF_Etimer_t timer;
669 #endif
670 	RF_AccessStripeMapHeader_t *asmh = desc->asmap;
671 	RF_Raid_t *raidPtr = desc->raidPtr;
672 	RF_AccessStripeMap_t *asm_p;
673 	RF_DagList_t *dagList;
674 	int i;
675 
676 	desc->state++;
677 
678 #if RF_ACC_TRACE > 0
679 	timer = tracerec->timer;
680 	RF_ETIMER_STOP(timer);
681 	RF_ETIMER_EVAL(timer);
682 	tracerec->specific.user.dag_retry_us = RF_ETIMER_VAL_US(timer);
683 
684 	/* the RAID I/O is complete.  Clean up. */
685 	tracerec->specific.user.dag_retry_us = 0;
686 
687 	RF_ETIMER_START(timer);
688 #endif
689 	/* free all dags */
690 	dagList = desc->dagList;
691 	for (i = 0; i < desc->numStripes; i++) {
692 		rf_FreeDAG(dagList->dags);
693 		dagList = dagList->next;
694 	}
695 #if RF_ACC_TRACE > 0
696 	RF_ETIMER_STOP(timer);
697 	RF_ETIMER_EVAL(timer);
698 	tracerec->specific.user.cleanup_us = RF_ETIMER_VAL_US(timer);
699 
700 	RF_ETIMER_START(timer);
701 #endif
702 	for (asm_p = asmh->stripeMap; asm_p; asm_p = asm_p->next) {
703 		if (!rf_suppressLocksAndLargeWrites &&
704 		    asm_p->parityInfo &&
705 		    !(desc->flags & RF_DAG_SUPPRESS_LOCKS)) {
706 			RF_ASSERT_VALID_LOCKREQ(&asm_p->lockReqDesc);
707 			rf_ReleaseStripeLock(raidPtr,
708 					     raidPtr->lockTable,
709 					     asm_p->stripeID,
710 					     &asm_p->lockReqDesc);
711 		}
712 		if (asm_p->flags & RF_ASM_FLAGS_RECON_BLOCKED) {
713 			rf_UnblockRecon(raidPtr, asm_p);
714 		}
715 	}
716 #if RF_ACC_TRACE > 0
717 	RF_ETIMER_STOP(timer);
718 	RF_ETIMER_EVAL(timer);
719 	tracerec->specific.user.lock_us += RF_ETIMER_VAL_US(timer);
720 
721 	RF_ETIMER_START(timer);
722 #endif
723 	rf_FreeAccessStripeMap(raidPtr, asmh);
724 #if RF_ACC_TRACE > 0
725 	RF_ETIMER_STOP(timer);
726 	RF_ETIMER_EVAL(timer);
727 	tracerec->specific.user.cleanup_us += RF_ETIMER_VAL_US(timer);
728 
729 	RF_ETIMER_STOP(desc->timer);
730 	RF_ETIMER_EVAL(desc->timer);
731 
732 	timer = desc->tracerec.tot_timer;
733 	RF_ETIMER_STOP(timer);
734 	RF_ETIMER_EVAL(timer);
735 	desc->tracerec.total_us = RF_ETIMER_VAL_US(timer);
736 
737 	rf_LogTraceRec(raidPtr, tracerec);
738 #endif
739 	desc->flags |= RF_DAG_ACCESS_COMPLETE;
740 
741 	return RF_FALSE;
742 }
743