xref: /netbsd-src/sys/dev/raidframe/rf_engine.c (revision 5e4c038a45edbc7d63b7c2daa76e29f88b64a4e3)
1 /*	$NetBSD: rf_engine.c,v 1.12 2001/11/13 07:11:14 lukem Exp $	*/
2 /*
3  * Copyright (c) 1995 Carnegie-Mellon University.
4  * All rights reserved.
5  *
6  * Author: William V. Courtright II, Mark Holland, Rachad Youssef
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 /****************************************************************************
30  *                                                                          *
31  * engine.c -- code for DAG execution engine                                *
32  *                                                                          *
33  * Modified to work as follows (holland):                                   *
34  *   A user-thread calls into DispatchDAG, which fires off the nodes that   *
35  *   are direct successors to the header node.  DispatchDAG then returns,   *
36  *   and the rest of the I/O continues asynchronously.  As each node        *
37  *   completes, the node execution function calls FinishNode().  FinishNode *
38  *   scans the list of successors to the node and increments the antecedent *
39  *   counts.  Each node that becomes enabled is placed on a central node    *
40  *   queue.  A dedicated dag-execution thread grabs nodes off of this       *
41  *   queue and fires them.                                                  *
42  *                                                                          *
43  *   NULL nodes are never fired.                                            *
44  *                                                                          *
45  *   Terminator nodes are never fired, but rather cause the callback        *
46  *   associated with the DAG to be invoked.                                 *
47  *                                                                          *
48  *   If a node fails, the dag either rolls forward to the completion or     *
49  *   rolls back, undoing previously-completed nodes and fails atomically.   *
50  *   The direction of recovery is determined by the location of the failed  *
51  *   node in the graph.  If the failure occurred before the commit node in   *
52  *   the graph, backward recovery is used.  Otherwise, forward recovery is  *
53  *   used.                                                                  *
54  *                                                                          *
55  ****************************************************************************/
56 
57 #include <sys/cdefs.h>
58 __KERNEL_RCSID(0, "$NetBSD: rf_engine.c,v 1.12 2001/11/13 07:11:14 lukem Exp $");
59 
60 #include "rf_threadstuff.h"
61 
62 #include <sys/errno.h>
63 
64 #include "rf_dag.h"
65 #include "rf_engine.h"
66 #include "rf_etimer.h"
67 #include "rf_general.h"
68 #include "rf_dagutils.h"
69 #include "rf_shutdown.h"
70 #include "rf_raid.h"
71 
72 static void DAGExecutionThread(RF_ThreadArg_t arg);
73 
74 #define DO_INIT(_l_,_r_) { \
75   int _rc; \
76   _rc = rf_create_managed_mutex(_l_,&(_r_)->node_queue_mutex); \
77   if (_rc) { \
78     return(_rc); \
79   } \
80   _rc = rf_create_managed_cond(_l_,&(_r_)->node_queue_cond); \
81   if (_rc) { \
82     return(_rc); \
83   } \
84 }
85 
86 /* synchronization primitives for this file.  DO_WAIT should be enclosed in a while loop. */
87 
88 /*
89  * XXX Is this spl-ing really necessary?
90  */
91 #define DO_LOCK(_r_) \
92 do { \
93 	ks = splbio(); \
94 	RF_LOCK_MUTEX((_r_)->node_queue_mutex); \
95 } while (0)
96 
97 #define DO_UNLOCK(_r_) \
98 do { \
99 	RF_UNLOCK_MUTEX((_r_)->node_queue_mutex); \
100 	splx(ks); \
101 } while (0)
102 
103 #define	DO_WAIT(_r_) \
104 	RF_WAIT_COND((_r_)->node_queue, (_r_)->node_queue_mutex)
105 
106 #define	DO_SIGNAL(_r_) \
107 	RF_BROADCAST_COND((_r_)->node_queue)	/* XXX RF_SIGNAL_COND? */
108 
109 static void rf_ShutdownEngine(void *);
110 
111 static void
112 rf_ShutdownEngine(arg)
113 	void   *arg;
114 {
115 	RF_Raid_t *raidPtr;
116 
117 	raidPtr = (RF_Raid_t *) arg;
118 	raidPtr->shutdown_engine = 1;
119 	DO_SIGNAL(raidPtr);
120 }
121 
122 int
123 rf_ConfigureEngine(
124     RF_ShutdownList_t ** listp,
125     RF_Raid_t * raidPtr,
126     RF_Config_t * cfgPtr)
127 {
128 	int     rc;
129 
130 	DO_INIT(listp, raidPtr);
131 
132 	raidPtr->node_queue = NULL;
133 	raidPtr->dags_in_flight = 0;
134 
135 	rc = rf_init_managed_threadgroup(listp, &raidPtr->engine_tg);
136 	if (rc)
137 		return (rc);
138 
139 	/* we create the execution thread only once per system boot. no need
140 	 * to check return code b/c the kernel panics if it can't create the
141 	 * thread. */
142 	if (rf_engineDebug) {
143 		printf("raid%d: Creating engine thread\n", raidPtr->raidid);
144 	}
145 	if (RF_CREATE_THREAD(raidPtr->engine_thread, DAGExecutionThread, raidPtr,"raid")) {
146 		RF_ERRORMSG("RAIDFRAME: Unable to create engine thread\n");
147 		return (ENOMEM);
148 	}
149 	if (rf_engineDebug) {
150 		printf("raid%d: Created engine thread\n", raidPtr->raidid);
151 	}
152 	RF_THREADGROUP_STARTED(&raidPtr->engine_tg);
153 	/* XXX something is missing here... */
154 #ifdef debug
155 	printf("Skipping the WAIT_START!!\n");
156 #endif
157 #if 0
158 	RF_THREADGROUP_WAIT_START(&raidPtr->engine_tg);
159 #endif
160 	/* engine thread is now running and waiting for work */
161 	if (rf_engineDebug) {
162 		printf("raid%d: Engine thread running and waiting for events\n", raidPtr->raidid);
163 	}
164 	rc = rf_ShutdownCreate(listp, rf_ShutdownEngine, raidPtr);
165 	if (rc) {
166 		RF_ERRORMSG3("Unable to add to shutdown list file %s line %d rc=%d\n", __FILE__,
167 		    __LINE__, rc);
168 		rf_ShutdownEngine(NULL);
169 	}
170 	return (rc);
171 }
172 
173 static int
174 BranchDone(RF_DagNode_t * node)
175 {
176 	int     i;
177 
178 	/* return true if forward execution is completed for a node and it's
179 	 * succedents */
180 	switch (node->status) {
181 	case rf_wait:
182 		/* should never be called in this state */
183 		RF_PANIC();
184 		break;
185 	case rf_fired:
186 		/* node is currently executing, so we're not done */
187 		return (RF_FALSE);
188 	case rf_good:
189 		for (i = 0; i < node->numSuccedents; i++)	/* for each succedent */
190 			if (!BranchDone(node->succedents[i]))	/* recursively check
191 								 * branch */
192 				return RF_FALSE;
193 		return RF_TRUE;	/* node and all succedent branches aren't in
194 				 * fired state */
195 		break;
196 	case rf_bad:
197 		/* succedents can't fire */
198 		return (RF_TRUE);
199 	case rf_recover:
200 		/* should never be called in this state */
201 		RF_PANIC();
202 		break;
203 	case rf_undone:
204 	case rf_panic:
205 		/* XXX need to fix this case */
206 		/* for now, assume that we're done */
207 		return (RF_TRUE);
208 		break;
209 	default:
210 		/* illegal node status */
211 		RF_PANIC();
212 		break;
213 	}
214 }
215 
216 static int
217 NodeReady(RF_DagNode_t * node)
218 {
219 	int     ready;
220 
221 	switch (node->dagHdr->status) {
222 	case rf_enable:
223 	case rf_rollForward:
224 		if ((node->status == rf_wait) && (node->numAntecedents == node->numAntDone))
225 			ready = RF_TRUE;
226 		else
227 			ready = RF_FALSE;
228 		break;
229 	case rf_rollBackward:
230 		RF_ASSERT(node->numSuccDone <= node->numSuccedents);
231 		RF_ASSERT(node->numSuccFired <= node->numSuccedents);
232 		RF_ASSERT(node->numSuccFired <= node->numSuccDone);
233 		if ((node->status == rf_good) && (node->numSuccDone == node->numSuccedents))
234 			ready = RF_TRUE;
235 		else
236 			ready = RF_FALSE;
237 		break;
238 	default:
239 		printf("Execution engine found illegal DAG status in NodeReady\n");
240 		RF_PANIC();
241 		break;
242 	}
243 
244 	return (ready);
245 }
246 
247 
248 
249 /* user context and dag-exec-thread context:
250  * Fire a node.  The node's status field determines which function, do or undo,
251  * to be fired.
252  * This routine assumes that the node's status field has alread been set to
253  * "fired" or "recover" to indicate the direction of execution.
254  */
255 static void
256 FireNode(RF_DagNode_t * node)
257 {
258 	switch (node->status) {
259 	case rf_fired:
260 		/* fire the do function of a node */
261 		if (rf_engineDebug) {
262 			printf("raid%d: Firing node 0x%lx (%s)\n",
263 			       node->dagHdr->raidPtr->raidid,
264 			       (unsigned long) node, node->name);
265 		}
266 		if (node->flags & RF_DAGNODE_FLAG_YIELD) {
267 #if defined(__NetBSD__) && defined(_KERNEL)
268 			/* thread_block(); */
269 			/* printf("Need to block the thread here...\n");  */
270 			/* XXX thread_block is actually mentioned in
271 			 * /usr/include/vm/vm_extern.h */
272 #else
273 			thread_block();
274 #endif
275 		}
276 		(*(node->doFunc)) (node);
277 		break;
278 	case rf_recover:
279 		/* fire the undo function of a node */
280 		if (rf_engineDebug) {
281 			printf("raid%d: Firing (undo) node 0x%lx (%s)\n",
282 			       node->dagHdr->raidPtr->raidid,
283 			       (unsigned long) node, node->name);
284 		}
285 		if (node->flags & RF_DAGNODE_FLAG_YIELD)
286 #if defined(__NetBSD__) && defined(_KERNEL)
287 			/* thread_block(); */
288 			/* printf("Need to block the thread here...\n"); */
289 			/* XXX thread_block is actually mentioned in
290 			 * /usr/include/vm/vm_extern.h */
291 #else
292 			thread_block();
293 #endif
294 		(*(node->undoFunc)) (node);
295 		break;
296 	default:
297 		RF_PANIC();
298 		break;
299 	}
300 }
301 
302 
303 
304 /* user context:
305  * Attempt to fire each node in a linear array.
306  * The entire list is fired atomically.
307  */
308 static void
309 FireNodeArray(
310     int numNodes,
311     RF_DagNode_t ** nodeList)
312 {
313 	RF_DagStatus_t dstat;
314 	RF_DagNode_t *node;
315 	int     i, j;
316 
317 	/* first, mark all nodes which are ready to be fired */
318 	for (i = 0; i < numNodes; i++) {
319 		node = nodeList[i];
320 		dstat = node->dagHdr->status;
321 		RF_ASSERT((node->status == rf_wait) || (node->status == rf_good));
322 		if (NodeReady(node)) {
323 			if ((dstat == rf_enable) || (dstat == rf_rollForward)) {
324 				RF_ASSERT(node->status == rf_wait);
325 				if (node->commitNode)
326 					node->dagHdr->numCommits++;
327 				node->status = rf_fired;
328 				for (j = 0; j < node->numAntecedents; j++)
329 					node->antecedents[j]->numSuccFired++;
330 			} else {
331 				RF_ASSERT(dstat == rf_rollBackward);
332 				RF_ASSERT(node->status == rf_good);
333 				RF_ASSERT(node->commitNode == RF_FALSE);	/* only one commit node
334 										 * per graph */
335 				node->status = rf_recover;
336 			}
337 		}
338 	}
339 	/* now, fire the nodes */
340 	for (i = 0; i < numNodes; i++) {
341 		if ((nodeList[i]->status == rf_fired) || (nodeList[i]->status == rf_recover))
342 			FireNode(nodeList[i]);
343 	}
344 }
345 
346 
347 /* user context:
348  * Attempt to fire each node in a linked list.
349  * The entire list is fired atomically.
350  */
351 static void
352 FireNodeList(RF_DagNode_t * nodeList)
353 {
354 	RF_DagNode_t *node, *next;
355 	RF_DagStatus_t dstat;
356 	int     j;
357 
358 	if (nodeList) {
359 		/* first, mark all nodes which are ready to be fired */
360 		for (node = nodeList; node; node = next) {
361 			next = node->next;
362 			dstat = node->dagHdr->status;
363 			RF_ASSERT((node->status == rf_wait) || (node->status == rf_good));
364 			if (NodeReady(node)) {
365 				if ((dstat == rf_enable) || (dstat == rf_rollForward)) {
366 					RF_ASSERT(node->status == rf_wait);
367 					if (node->commitNode)
368 						node->dagHdr->numCommits++;
369 					node->status = rf_fired;
370 					for (j = 0; j < node->numAntecedents; j++)
371 						node->antecedents[j]->numSuccFired++;
372 				} else {
373 					RF_ASSERT(dstat == rf_rollBackward);
374 					RF_ASSERT(node->status == rf_good);
375 					RF_ASSERT(node->commitNode == RF_FALSE);	/* only one commit node
376 											 * per graph */
377 					node->status = rf_recover;
378 				}
379 			}
380 		}
381 		/* now, fire the nodes */
382 		for (node = nodeList; node; node = next) {
383 			next = node->next;
384 			if ((node->status == rf_fired) || (node->status == rf_recover))
385 				FireNode(node);
386 		}
387 	}
388 }
389 /* interrupt context:
390  * for each succedent
391  *    propagate required results from node to succedent
392  *    increment succedent's numAntDone
393  *    place newly-enable nodes on node queue for firing
394  *
395  * To save context switches, we don't place NIL nodes on the node queue,
396  * but rather just process them as if they had fired.  Note that NIL nodes
397  * that are the direct successors of the header will actually get fired by
398  * DispatchDAG, which is fine because no context switches are involved.
399  *
400  * Important:  when running at user level, this can be called by any
401  * disk thread, and so the increment and check of the antecedent count
402  * must be locked.  I used the node queue mutex and locked down the
403  * entire function, but this is certainly overkill.
404  */
405 static void
406 PropagateResults(
407     RF_DagNode_t * node,
408     int context)
409 {
410 	RF_DagNode_t *s, *a;
411 	RF_Raid_t *raidPtr;
412 	int     i, ks;
413 	RF_DagNode_t *finishlist = NULL;	/* a list of NIL nodes to be
414 						 * finished */
415 	RF_DagNode_t *skiplist = NULL;	/* list of nodes with failed truedata
416 					 * antecedents */
417 	RF_DagNode_t *firelist = NULL;	/* a list of nodes to be fired */
418 	RF_DagNode_t *q = NULL, *qh = NULL, *next;
419 	int     j, skipNode;
420 
421 	raidPtr = node->dagHdr->raidPtr;
422 
423 	DO_LOCK(raidPtr);
424 
425 	/* debug - validate fire counts */
426 	for (i = 0; i < node->numAntecedents; i++) {
427 		a = *(node->antecedents + i);
428 		RF_ASSERT(a->numSuccFired >= a->numSuccDone);
429 		RF_ASSERT(a->numSuccFired <= a->numSuccedents);
430 		a->numSuccDone++;
431 	}
432 
433 	switch (node->dagHdr->status) {
434 	case rf_enable:
435 	case rf_rollForward:
436 		for (i = 0; i < node->numSuccedents; i++) {
437 			s = *(node->succedents + i);
438 			RF_ASSERT(s->status == rf_wait);
439 			(s->numAntDone)++;
440 			if (s->numAntDone == s->numAntecedents) {
441 				/* look for NIL nodes */
442 				if (s->doFunc == rf_NullNodeFunc) {
443 					/* don't fire NIL nodes, just process
444 					 * them */
445 					s->next = finishlist;
446 					finishlist = s;
447 				} else {
448 					/* look to see if the node is to be
449 					 * skipped */
450 					skipNode = RF_FALSE;
451 					for (j = 0; j < s->numAntecedents; j++)
452 						if ((s->antType[j] == rf_trueData) && (s->antecedents[j]->status == rf_bad))
453 							skipNode = RF_TRUE;
454 					if (skipNode) {
455 						/* this node has one or more
456 						 * failed true data
457 						 * dependencies, so skip it */
458 						s->next = skiplist;
459 						skiplist = s;
460 					} else
461 						/* add s to list of nodes (q)
462 						 * to execute */
463 						if (context != RF_INTR_CONTEXT) {
464 							/* we only have to
465 							 * enqueue if we're at
466 							 * intr context */
467 							s->next = firelist;	/* put node on a list to
468 										 * be fired after we
469 										 * unlock */
470 							firelist = s;
471 						} else {	/* enqueue the node for
472 								 * the dag exec thread
473 								 * to fire */
474 							RF_ASSERT(NodeReady(s));
475 							if (q) {
476 								q->next = s;
477 								q = s;
478 							} else {
479 								qh = q = s;
480 								qh->next = NULL;
481 							}
482 						}
483 				}
484 			}
485 		}
486 
487 		if (q) {
488 			/* xfer our local list of nodes to the node queue */
489 			q->next = raidPtr->node_queue;
490 			raidPtr->node_queue = qh;
491 			DO_SIGNAL(raidPtr);
492 		}
493 		DO_UNLOCK(raidPtr);
494 
495 		for (; skiplist; skiplist = next) {
496 			next = skiplist->next;
497 			skiplist->status = rf_skipped;
498 			for (i = 0; i < skiplist->numAntecedents; i++) {
499 				skiplist->antecedents[i]->numSuccFired++;
500 			}
501 			if (skiplist->commitNode) {
502 				skiplist->dagHdr->numCommits++;
503 			}
504 			rf_FinishNode(skiplist, context);
505 		}
506 		for (; finishlist; finishlist = next) {
507 			/* NIL nodes: no need to fire them */
508 			next = finishlist->next;
509 			finishlist->status = rf_good;
510 			for (i = 0; i < finishlist->numAntecedents; i++) {
511 				finishlist->antecedents[i]->numSuccFired++;
512 			}
513 			if (finishlist->commitNode)
514 				finishlist->dagHdr->numCommits++;
515 			/*
516 		         * Okay, here we're calling rf_FinishNode() on nodes that
517 		         * have the null function as their work proc. Such a node
518 		         * could be the terminal node in a DAG. If so, it will
519 		         * cause the DAG to complete, which will in turn free
520 		         * memory used by the DAG, which includes the node in
521 		         * question. Thus, we must avoid referencing the node
522 		         * at all after calling rf_FinishNode() on it.
523 		         */
524 			rf_FinishNode(finishlist, context);	/* recursive call */
525 		}
526 		/* fire all nodes in firelist */
527 		FireNodeList(firelist);
528 		break;
529 
530 	case rf_rollBackward:
531 		for (i = 0; i < node->numAntecedents; i++) {
532 			a = *(node->antecedents + i);
533 			RF_ASSERT(a->status == rf_good);
534 			RF_ASSERT(a->numSuccDone <= a->numSuccedents);
535 			RF_ASSERT(a->numSuccDone <= a->numSuccFired);
536 
537 			if (a->numSuccDone == a->numSuccFired) {
538 				if (a->undoFunc == rf_NullNodeFunc) {
539 					/* don't fire NIL nodes, just process
540 					 * them */
541 					a->next = finishlist;
542 					finishlist = a;
543 				} else {
544 					if (context != RF_INTR_CONTEXT) {
545 						/* we only have to enqueue if
546 						 * we're at intr context */
547 						a->next = firelist;	/* put node on a list to
548 									 * be fired after we
549 									 * unlock */
550 						firelist = a;
551 					} else {	/* enqueue the node for
552 							 * the dag exec thread
553 							 * to fire */
554 						RF_ASSERT(NodeReady(a));
555 						if (q) {
556 							q->next = a;
557 							q = a;
558 						} else {
559 							qh = q = a;
560 							qh->next = NULL;
561 						}
562 					}
563 				}
564 			}
565 		}
566 		if (q) {
567 			/* xfer our local list of nodes to the node queue */
568 			q->next = raidPtr->node_queue;
569 			raidPtr->node_queue = qh;
570 			DO_SIGNAL(raidPtr);
571 		}
572 		DO_UNLOCK(raidPtr);
573 		for (; finishlist; finishlist = next) {	/* NIL nodes: no need to
574 							 * fire them */
575 			next = finishlist->next;
576 			finishlist->status = rf_good;
577 			/*
578 		         * Okay, here we're calling rf_FinishNode() on nodes that
579 		         * have the null function as their work proc. Such a node
580 		         * could be the first node in a DAG. If so, it will
581 		         * cause the DAG to complete, which will in turn free
582 		         * memory used by the DAG, which includes the node in
583 		         * question. Thus, we must avoid referencing the node
584 		         * at all after calling rf_FinishNode() on it.
585 		         */
586 			rf_FinishNode(finishlist, context);	/* recursive call */
587 		}
588 		/* fire all nodes in firelist */
589 		FireNodeList(firelist);
590 
591 		break;
592 	default:
593 		printf("Engine found illegal DAG status in PropagateResults()\n");
594 		RF_PANIC();
595 		break;
596 	}
597 }
598 
599 
600 
601 /*
602  * Process a fired node which has completed
603  */
604 static void
605 ProcessNode(
606     RF_DagNode_t * node,
607     int context)
608 {
609 	RF_Raid_t *raidPtr;
610 
611 	raidPtr = node->dagHdr->raidPtr;
612 
613 	switch (node->status) {
614 	case rf_good:
615 		/* normal case, don't need to do anything */
616 		break;
617 	case rf_bad:
618 		if ((node->dagHdr->numCommits > 0) || (node->dagHdr->numCommitNodes == 0)) {
619 			node->dagHdr->status = rf_rollForward;	/* crossed commit
620 								 * barrier */
621 			if (rf_engineDebug || 1) {
622 				printf("raid%d: node (%s) returned fail, rolling forward\n", raidPtr->raidid, node->name);
623 			}
624 		} else {
625 			node->dagHdr->status = rf_rollBackward;	/* never reached commit
626 								 * barrier */
627 			if (rf_engineDebug || 1) {
628 				printf("raid%d: node (%s) returned fail, rolling backward\n", raidPtr->raidid, node->name);
629 			}
630 		}
631 		break;
632 	case rf_undone:
633 		/* normal rollBackward case, don't need to do anything */
634 		break;
635 	case rf_panic:
636 		/* an undo node failed!!! */
637 		printf("UNDO of a node failed!!!/n");
638 		break;
639 	default:
640 		printf("node finished execution with an illegal status!!!\n");
641 		RF_PANIC();
642 		break;
643 	}
644 
645 	/* enqueue node's succedents (antecedents if rollBackward) for
646 	 * execution */
647 	PropagateResults(node, context);
648 }
649 
650 
651 
652 /* user context or dag-exec-thread context:
653  * This is the first step in post-processing a newly-completed node.
654  * This routine is called by each node execution function to mark the node
655  * as complete and fire off any successors that have been enabled.
656  */
657 int
658 rf_FinishNode(
659     RF_DagNode_t * node,
660     int context)
661 {
662 	/* as far as I can tell, retcode is not used -wvcii */
663 	int     retcode = RF_FALSE;
664 	node->dagHdr->numNodesCompleted++;
665 	ProcessNode(node, context);
666 
667 	return (retcode);
668 }
669 
670 
671 /* user context:
672  * submit dag for execution, return non-zero if we have to wait for completion.
673  * if and only if we return non-zero, we'll cause cbFunc to get invoked with
674  * cbArg when the DAG has completed.
675  *
676  * for now we always return 1.  If the DAG does not cause any I/O, then the callback
677  * may get invoked before DispatchDAG returns.  There's code in state 5 of ContinueRaidAccess
678  * to handle this.
679  *
680  * All we do here is fire the direct successors of the header node.  The
681  * DAG execution thread does the rest of the dag processing.
682  */
683 int
684 rf_DispatchDAG(
685     RF_DagHeader_t * dag,
686     void (*cbFunc) (void *),
687     void *cbArg)
688 {
689 	RF_Raid_t *raidPtr;
690 
691 	raidPtr = dag->raidPtr;
692 	if (dag->tracerec) {
693 		RF_ETIMER_START(dag->tracerec->timer);
694 	}
695 	if (rf_engineDebug || rf_validateDAGDebug) {
696 		if (rf_ValidateDAG(dag))
697 			RF_PANIC();
698 	}
699 	if (rf_engineDebug) {
700 		printf("raid%d: Entering DispatchDAG\n", raidPtr->raidid);
701 	}
702 	raidPtr->dags_in_flight++;	/* debug only:  blow off proper
703 					 * locking */
704 	dag->cbFunc = cbFunc;
705 	dag->cbArg = cbArg;
706 	dag->numNodesCompleted = 0;
707 	dag->status = rf_enable;
708 	FireNodeArray(dag->numSuccedents, dag->succedents);
709 	return (1);
710 }
711 /* dedicated kernel thread:
712  * the thread that handles all DAG node firing.
713  * To minimize locking and unlocking, we grab a copy of the entire node queue and then set the
714  * node queue to NULL before doing any firing of nodes.  This way we only have to release the
715  * lock once.  Of course, it's probably rare that there's more than one node in the queue at
716  * any one time, but it sometimes happens.
717  *
718  * In the kernel, this thread runs at spl0 and is not swappable.  I copied these
719  * characteristics from the aio_completion_thread.
720  */
721 
722 static void
723 DAGExecutionThread(RF_ThreadArg_t arg)
724 {
725 	RF_DagNode_t *nd, *local_nq, *term_nq, *fire_nq;
726 	RF_Raid_t *raidPtr;
727 	int     ks;
728 	int     s;
729 
730 	raidPtr = (RF_Raid_t *) arg;
731 
732 	if (rf_engineDebug) {
733 		printf("raid%d: Engine thread is running\n", raidPtr->raidid);
734 	}
735 
736 	s = splbio();
737 
738 	RF_THREADGROUP_RUNNING(&raidPtr->engine_tg);
739 
740 	DO_LOCK(raidPtr);
741 	while (!raidPtr->shutdown_engine) {
742 
743 		while (raidPtr->node_queue != NULL) {
744 			local_nq = raidPtr->node_queue;
745 			fire_nq = NULL;
746 			term_nq = NULL;
747 			raidPtr->node_queue = NULL;
748 			DO_UNLOCK(raidPtr);
749 
750 			/* first, strip out the terminal nodes */
751 			while (local_nq) {
752 				nd = local_nq;
753 				local_nq = local_nq->next;
754 				switch (nd->dagHdr->status) {
755 				case rf_enable:
756 				case rf_rollForward:
757 					if (nd->numSuccedents == 0) {
758 						/* end of the dag, add to
759 						 * callback list */
760 						nd->next = term_nq;
761 						term_nq = nd;
762 					} else {
763 						/* not the end, add to the
764 						 * fire queue */
765 						nd->next = fire_nq;
766 						fire_nq = nd;
767 					}
768 					break;
769 				case rf_rollBackward:
770 					if (nd->numAntecedents == 0) {
771 						/* end of the dag, add to the
772 						 * callback list */
773 						nd->next = term_nq;
774 						term_nq = nd;
775 					} else {
776 						/* not the end, add to the
777 						 * fire queue */
778 						nd->next = fire_nq;
779 						fire_nq = nd;
780 					}
781 					break;
782 				default:
783 					RF_PANIC();
784 					break;
785 				}
786 			}
787 
788 			/* execute callback of dags which have reached the
789 			 * terminal node */
790 			while (term_nq) {
791 				nd = term_nq;
792 				term_nq = term_nq->next;
793 				nd->next = NULL;
794 				(nd->dagHdr->cbFunc) (nd->dagHdr->cbArg);
795 				raidPtr->dags_in_flight--;	/* debug only */
796 			}
797 
798 			/* fire remaining nodes */
799 			FireNodeList(fire_nq);
800 
801 			DO_LOCK(raidPtr);
802 		}
803 		while (!raidPtr->shutdown_engine && raidPtr->node_queue == NULL)
804 			DO_WAIT(raidPtr);
805 	}
806 	DO_UNLOCK(raidPtr);
807 
808 	RF_THREADGROUP_DONE(&raidPtr->engine_tg);
809 
810 	splx(s);
811 	kthread_exit(0);
812 }
813