xref: /netbsd-src/sys/dev/raidframe/rf_engine.c (revision 3b435a73967be44dfb4a27315acd72bfacde430c)
1 /*	$NetBSD: rf_engine.c,v 1.5 1999/03/14 21:53:31 oster 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 occured before the commit node in   *
52  *   the graph, backward recovery is used.  Otherwise, forward recovery is  *
53  *   used.                                                                  *
54  *                                                                          *
55  ****************************************************************************/
56 
57 #include "rf_threadstuff.h"
58 
59 #include <sys/errno.h>
60 
61 #include "rf_dag.h"
62 #include "rf_engine.h"
63 #include "rf_threadid.h"
64 #include "rf_etimer.h"
65 #include "rf_general.h"
66 #include "rf_dagutils.h"
67 #include "rf_shutdown.h"
68 #include "rf_raid.h"
69 
70 static void DAGExecutionThread(RF_ThreadArg_t arg);
71 
72 #define DO_INIT(_l_,_r_) { \
73   int _rc; \
74   _rc = rf_create_managed_mutex(_l_,&(_r_)->node_queue_mutex); \
75   if (_rc) { \
76     return(_rc); \
77   } \
78   _rc = rf_create_managed_cond(_l_,&(_r_)->node_queue_cond); \
79   if (_rc) { \
80     return(_rc); \
81   } \
82 }
83 
84 /* synchronization primitives for this file.  DO_WAIT should be enclosed in a while loop. */
85 
86 /*
87  * XXX Is this spl-ing really necessary?
88  */
89 #define DO_LOCK(_r_)      { ks = splbio(); RF_LOCK_MUTEX((_r_)->node_queue_mutex); }
90 #define DO_UNLOCK(_r_)    { RF_UNLOCK_MUTEX((_r_)->node_queue_mutex); splx(ks); }
91 #define DO_WAIT(_r_)   tsleep(&(_r_)->node_queue, PRIBIO, "raidframe nq",0)
92 #define DO_SIGNAL(_r_)    wakeup(&(_r_)->node_queue)
93 
94 static void rf_ShutdownEngine(void *);
95 
96 static void
97 rf_ShutdownEngine(arg)
98 	void   *arg;
99 {
100 	RF_Raid_t *raidPtr;
101 
102 	raidPtr = (RF_Raid_t *) arg;
103 	raidPtr->shutdown_engine = 1;
104 	DO_SIGNAL(raidPtr);
105 }
106 
107 int
108 rf_ConfigureEngine(
109     RF_ShutdownList_t ** listp,
110     RF_Raid_t * raidPtr,
111     RF_Config_t * cfgPtr)
112 {
113 	int     rc, tid = 0;
114 
115 	if (rf_engineDebug) {
116 		rf_get_threadid(tid);
117 	}
118 	DO_INIT(listp, raidPtr);
119 
120 	raidPtr->node_queue = NULL;
121 	raidPtr->dags_in_flight = 0;
122 
123 	rc = rf_init_managed_threadgroup(listp, &raidPtr->engine_tg);
124 	if (rc)
125 		return (rc);
126 
127 	/* we create the execution thread only once per system boot. no need
128 	 * to check return code b/c the kernel panics if it can't create the
129 	 * thread. */
130 	if (rf_engineDebug) {
131 		printf("[%d] Creating engine thread\n", tid);
132 	}
133 	if (RF_CREATE_THREAD(raidPtr->engine_thread, DAGExecutionThread, raidPtr)) {
134 		RF_ERRORMSG("RAIDFRAME: Unable to create engine thread\n");
135 		return (ENOMEM);
136 	}
137 	if (rf_engineDebug) {
138 		printf("[%d] Created engine thread\n", tid);
139 	}
140 	RF_THREADGROUP_STARTED(&raidPtr->engine_tg);
141 	/* XXX something is missing here... */
142 #ifdef debug
143 	printf("Skipping the WAIT_START!!\n");
144 #endif
145 #if 0
146 	RF_THREADGROUP_WAIT_START(&raidPtr->engine_tg);
147 #endif
148 	/* engine thread is now running and waiting for work */
149 	if (rf_engineDebug) {
150 		printf("[%d] Engine thread running and waiting for events\n", tid);
151 	}
152 	rc = rf_ShutdownCreate(listp, rf_ShutdownEngine, raidPtr);
153 	if (rc) {
154 		RF_ERRORMSG3("Unable to add to shutdown list file %s line %d rc=%d\n", __FILE__,
155 		    __LINE__, rc);
156 		rf_ShutdownEngine(NULL);
157 	}
158 	return (rc);
159 }
160 
161 static int
162 BranchDone(RF_DagNode_t * node)
163 {
164 	int     i;
165 
166 	/* return true if forward execution is completed for a node and it's
167 	 * succedents */
168 	switch (node->status) {
169 	case rf_wait:
170 		/* should never be called in this state */
171 		RF_PANIC();
172 		break;
173 	case rf_fired:
174 		/* node is currently executing, so we're not done */
175 		return (RF_FALSE);
176 	case rf_good:
177 		for (i = 0; i < node->numSuccedents; i++)	/* for each succedent */
178 			if (!BranchDone(node->succedents[i]))	/* recursively check
179 								 * branch */
180 				return RF_FALSE;
181 		return RF_TRUE;	/* node and all succedent branches aren't in
182 				 * fired state */
183 		break;
184 	case rf_bad:
185 		/* succedents can't fire */
186 		return (RF_TRUE);
187 	case rf_recover:
188 		/* should never be called in this state */
189 		RF_PANIC();
190 		break;
191 	case rf_undone:
192 	case rf_panic:
193 		/* XXX need to fix this case */
194 		/* for now, assume that we're done */
195 		return (RF_TRUE);
196 		break;
197 	default:
198 		/* illegal node status */
199 		RF_PANIC();
200 		break;
201 	}
202 }
203 
204 static int
205 NodeReady(RF_DagNode_t * node)
206 {
207 	int     ready;
208 
209 	switch (node->dagHdr->status) {
210 	case rf_enable:
211 	case rf_rollForward:
212 		if ((node->status == rf_wait) && (node->numAntecedents == node->numAntDone))
213 			ready = RF_TRUE;
214 		else
215 			ready = RF_FALSE;
216 		break;
217 	case rf_rollBackward:
218 		RF_ASSERT(node->numSuccDone <= node->numSuccedents);
219 		RF_ASSERT(node->numSuccFired <= node->numSuccedents);
220 		RF_ASSERT(node->numSuccFired <= node->numSuccDone);
221 		if ((node->status == rf_good) && (node->numSuccDone == node->numSuccedents))
222 			ready = RF_TRUE;
223 		else
224 			ready = RF_FALSE;
225 		break;
226 	default:
227 		printf("Execution engine found illegal DAG status in NodeReady\n");
228 		RF_PANIC();
229 		break;
230 	}
231 
232 	return (ready);
233 }
234 
235 
236 
237 /* user context and dag-exec-thread context:
238  * Fire a node.  The node's status field determines which function, do or undo,
239  * to be fired.
240  * This routine assumes that the node's status field has alread been set to
241  * "fired" or "recover" to indicate the direction of execution.
242  */
243 static void
244 FireNode(RF_DagNode_t * node)
245 {
246 	int     tid;
247 
248 	switch (node->status) {
249 	case rf_fired:
250 		/* fire the do function of a node */
251 		if (rf_engineDebug) {
252 			rf_get_threadid(tid);
253 			printf("[%d] Firing node 0x%lx (%s)\n", tid, (unsigned long) node, node->name);
254 		}
255 		if (node->flags & RF_DAGNODE_FLAG_YIELD) {
256 #if defined(__NetBSD__) && defined(_KERNEL)
257 			/* thread_block(); */
258 			/* printf("Need to block the thread here...\n");  */
259 			/* XXX thread_block is actually mentioned in
260 			 * /usr/include/vm/vm_extern.h */
261 #else
262 			thread_block();
263 #endif
264 		}
265 		(*(node->doFunc)) (node);
266 		break;
267 	case rf_recover:
268 		/* fire the undo function of a node */
269 		if (rf_engineDebug || 1) {
270 			rf_get_threadid(tid);
271 			printf("[%d] Firing (undo) node 0x%lx (%s)\n", tid, (unsigned long) node, node->name);
272 		}
273 		if (node->flags & RF_DAGNODE_FLAG_YIELD)
274 #if defined(__NetBSD__) && defined(_KERNEL)
275 			/* thread_block(); */
276 			/* printf("Need to block the thread here...\n"); */
277 			/* XXX thread_block is actually mentioned in
278 			 * /usr/include/vm/vm_extern.h */
279 #else
280 			thread_block();
281 #endif
282 		(*(node->undoFunc)) (node);
283 		break;
284 	default:
285 		RF_PANIC();
286 		break;
287 	}
288 }
289 
290 
291 
292 /* user context:
293  * Attempt to fire each node in a linear array.
294  * The entire list is fired atomically.
295  */
296 static void
297 FireNodeArray(
298     int numNodes,
299     RF_DagNode_t ** nodeList)
300 {
301 	RF_DagStatus_t dstat;
302 	RF_DagNode_t *node;
303 	int     i, j;
304 
305 	/* first, mark all nodes which are ready to be fired */
306 	for (i = 0; i < numNodes; i++) {
307 		node = nodeList[i];
308 		dstat = node->dagHdr->status;
309 		RF_ASSERT((node->status == rf_wait) || (node->status == rf_good));
310 		if (NodeReady(node)) {
311 			if ((dstat == rf_enable) || (dstat == rf_rollForward)) {
312 				RF_ASSERT(node->status == rf_wait);
313 				if (node->commitNode)
314 					node->dagHdr->numCommits++;
315 				node->status = rf_fired;
316 				for (j = 0; j < node->numAntecedents; j++)
317 					node->antecedents[j]->numSuccFired++;
318 			} else {
319 				RF_ASSERT(dstat == rf_rollBackward);
320 				RF_ASSERT(node->status == rf_good);
321 				RF_ASSERT(node->commitNode == RF_FALSE);	/* only one commit node
322 										 * per graph */
323 				node->status = rf_recover;
324 			}
325 		}
326 	}
327 	/* now, fire the nodes */
328 	for (i = 0; i < numNodes; i++) {
329 		if ((nodeList[i]->status == rf_fired) || (nodeList[i]->status == rf_recover))
330 			FireNode(nodeList[i]);
331 	}
332 }
333 
334 
335 /* user context:
336  * Attempt to fire each node in a linked list.
337  * The entire list is fired atomically.
338  */
339 static void
340 FireNodeList(RF_DagNode_t * nodeList)
341 {
342 	RF_DagNode_t *node, *next;
343 	RF_DagStatus_t dstat;
344 	int     j;
345 
346 	if (nodeList) {
347 		/* first, mark all nodes which are ready to be fired */
348 		for (node = nodeList; node; node = next) {
349 			next = node->next;
350 			dstat = node->dagHdr->status;
351 			RF_ASSERT((node->status == rf_wait) || (node->status == rf_good));
352 			if (NodeReady(node)) {
353 				if ((dstat == rf_enable) || (dstat == rf_rollForward)) {
354 					RF_ASSERT(node->status == rf_wait);
355 					if (node->commitNode)
356 						node->dagHdr->numCommits++;
357 					node->status = rf_fired;
358 					for (j = 0; j < node->numAntecedents; j++)
359 						node->antecedents[j]->numSuccFired++;
360 				} else {
361 					RF_ASSERT(dstat == rf_rollBackward);
362 					RF_ASSERT(node->status == rf_good);
363 					RF_ASSERT(node->commitNode == RF_FALSE);	/* only one commit node
364 											 * per graph */
365 					node->status = rf_recover;
366 				}
367 			}
368 		}
369 		/* now, fire the nodes */
370 		for (node = nodeList; node; node = next) {
371 			next = node->next;
372 			if ((node->status == rf_fired) || (node->status == rf_recover))
373 				FireNode(node);
374 		}
375 	}
376 }
377 /* interrupt context:
378  * for each succedent
379  *    propagate required results from node to succedent
380  *    increment succedent's numAntDone
381  *    place newly-enable nodes on node queue for firing
382  *
383  * To save context switches, we don't place NIL nodes on the node queue,
384  * but rather just process them as if they had fired.  Note that NIL nodes
385  * that are the direct successors of the header will actually get fired by
386  * DispatchDAG, which is fine because no context switches are involved.
387  *
388  * Important:  when running at user level, this can be called by any
389  * disk thread, and so the increment and check of the antecedent count
390  * must be locked.  I used the node queue mutex and locked down the
391  * entire function, but this is certainly overkill.
392  */
393 static void
394 PropagateResults(
395     RF_DagNode_t * node,
396     int context)
397 {
398 	RF_DagNode_t *s, *a;
399 	RF_Raid_t *raidPtr;
400 	int     tid, i, ks;
401 	RF_DagNode_t *finishlist = NULL;	/* a list of NIL nodes to be
402 						 * finished */
403 	RF_DagNode_t *skiplist = NULL;	/* list of nodes with failed truedata
404 					 * antecedents */
405 	RF_DagNode_t *firelist = NULL;	/* a list of nodes to be fired */
406 	RF_DagNode_t *q = NULL, *qh = NULL, *next;
407 	int     j, skipNode;
408 
409 	rf_get_threadid(tid);
410 
411 	raidPtr = node->dagHdr->raidPtr;
412 
413 	DO_LOCK(raidPtr);
414 
415 	/* debug - validate fire counts */
416 	for (i = 0; i < node->numAntecedents; i++) {
417 		a = *(node->antecedents + i);
418 		RF_ASSERT(a->numSuccFired >= a->numSuccDone);
419 		RF_ASSERT(a->numSuccFired <= a->numSuccedents);
420 		a->numSuccDone++;
421 	}
422 
423 	switch (node->dagHdr->status) {
424 	case rf_enable:
425 	case rf_rollForward:
426 		for (i = 0; i < node->numSuccedents; i++) {
427 			s = *(node->succedents + i);
428 			RF_ASSERT(s->status == rf_wait);
429 			(s->numAntDone)++;
430 			if (s->numAntDone == s->numAntecedents) {
431 				/* look for NIL nodes */
432 				if (s->doFunc == rf_NullNodeFunc) {
433 					/* don't fire NIL nodes, just process
434 					 * them */
435 					s->next = finishlist;
436 					finishlist = s;
437 				} else {
438 					/* look to see if the node is to be
439 					 * skipped */
440 					skipNode = RF_FALSE;
441 					for (j = 0; j < s->numAntecedents; j++)
442 						if ((s->antType[j] == rf_trueData) && (s->antecedents[j]->status == rf_bad))
443 							skipNode = RF_TRUE;
444 					if (skipNode) {
445 						/* this node has one or more
446 						 * failed true data
447 						 * dependencies, so skip it */
448 						s->next = skiplist;
449 						skiplist = s;
450 					} else
451 						/* add s to list of nodes (q)
452 						 * to execute */
453 						if (context != RF_INTR_CONTEXT) {
454 							/* we only have to
455 							 * enqueue if we're at
456 							 * intr context */
457 							s->next = firelist;	/* put node on a list to
458 										 * be fired after we
459 										 * unlock */
460 							firelist = s;
461 						} else {	/* enqueue the node for
462 								 * the dag exec thread
463 								 * to fire */
464 							RF_ASSERT(NodeReady(s));
465 							if (q) {
466 								q->next = s;
467 								q = s;
468 							} else {
469 								qh = q = s;
470 								qh->next = NULL;
471 							}
472 						}
473 				}
474 			}
475 		}
476 
477 		if (q) {
478 			/* xfer our local list of nodes to the node queue */
479 			q->next = raidPtr->node_queue;
480 			raidPtr->node_queue = qh;
481 			DO_SIGNAL(raidPtr);
482 		}
483 		DO_UNLOCK(raidPtr);
484 
485 		for (; skiplist; skiplist = next) {
486 			next = skiplist->next;
487 			skiplist->status = rf_skipped;
488 			for (i = 0; i < skiplist->numAntecedents; i++) {
489 				skiplist->antecedents[i]->numSuccFired++;
490 			}
491 			if (skiplist->commitNode) {
492 				skiplist->dagHdr->numCommits++;
493 			}
494 			rf_FinishNode(skiplist, context);
495 		}
496 		for (; finishlist; finishlist = next) {
497 			/* NIL nodes: no need to fire them */
498 			next = finishlist->next;
499 			finishlist->status = rf_good;
500 			for (i = 0; i < finishlist->numAntecedents; i++) {
501 				finishlist->antecedents[i]->numSuccFired++;
502 			}
503 			if (finishlist->commitNode)
504 				finishlist->dagHdr->numCommits++;
505 			/*
506 		         * Okay, here we're calling rf_FinishNode() on nodes that
507 		         * have the null function as their work proc. Such a node
508 		         * could be the terminal node in a DAG. If so, it will
509 		         * cause the DAG to complete, which will in turn free
510 		         * memory used by the DAG, which includes the node in
511 		         * question. Thus, we must avoid referencing the node
512 		         * at all after calling rf_FinishNode() on it.
513 		         */
514 			rf_FinishNode(finishlist, context);	/* recursive call */
515 		}
516 		/* fire all nodes in firelist */
517 		FireNodeList(firelist);
518 		break;
519 
520 	case rf_rollBackward:
521 		for (i = 0; i < node->numAntecedents; i++) {
522 			a = *(node->antecedents + i);
523 			RF_ASSERT(a->status == rf_good);
524 			RF_ASSERT(a->numSuccDone <= a->numSuccedents);
525 			RF_ASSERT(a->numSuccDone <= a->numSuccFired);
526 
527 			if (a->numSuccDone == a->numSuccFired) {
528 				if (a->undoFunc == rf_NullNodeFunc) {
529 					/* don't fire NIL nodes, just process
530 					 * them */
531 					a->next = finishlist;
532 					finishlist = a;
533 				} else {
534 					if (context != RF_INTR_CONTEXT) {
535 						/* we only have to enqueue if
536 						 * we're at intr context */
537 						a->next = firelist;	/* put node on a list to
538 									 * be fired after we
539 									 * unlock */
540 						firelist = a;
541 					} else {	/* enqueue the node for
542 							 * the dag exec thread
543 							 * to fire */
544 						RF_ASSERT(NodeReady(a));
545 						if (q) {
546 							q->next = a;
547 							q = a;
548 						} else {
549 							qh = q = a;
550 							qh->next = NULL;
551 						}
552 					}
553 				}
554 			}
555 		}
556 		if (q) {
557 			/* xfer our local list of nodes to the node queue */
558 			q->next = raidPtr->node_queue;
559 			raidPtr->node_queue = qh;
560 			DO_SIGNAL(raidPtr);
561 		}
562 		DO_UNLOCK(raidPtr);
563 		for (; finishlist; finishlist = next) {	/* NIL nodes: no need to
564 							 * fire them */
565 			next = finishlist->next;
566 			finishlist->status = rf_good;
567 			/*
568 		         * Okay, here we're calling rf_FinishNode() on nodes that
569 		         * have the null function as their work proc. Such a node
570 		         * could be the first node in a DAG. If so, it will
571 		         * cause the DAG to complete, which will in turn free
572 		         * memory used by the DAG, which includes the node in
573 		         * question. Thus, we must avoid referencing the node
574 		         * at all after calling rf_FinishNode() on it.
575 		         */
576 			rf_FinishNode(finishlist, context);	/* recursive call */
577 		}
578 		/* fire all nodes in firelist */
579 		FireNodeList(firelist);
580 
581 		break;
582 	default:
583 		printf("Engine found illegal DAG status in PropagateResults()\n");
584 		RF_PANIC();
585 		break;
586 	}
587 }
588 
589 
590 
591 /*
592  * Process a fired node which has completed
593  */
594 static void
595 ProcessNode(
596     RF_DagNode_t * node,
597     int context)
598 {
599 	RF_Raid_t *raidPtr;
600 	int     tid;
601 
602 	raidPtr = node->dagHdr->raidPtr;
603 
604 	switch (node->status) {
605 	case rf_good:
606 		/* normal case, don't need to do anything */
607 		break;
608 	case rf_bad:
609 		if ((node->dagHdr->numCommits > 0) || (node->dagHdr->numCommitNodes == 0)) {
610 			node->dagHdr->status = rf_rollForward;	/* crossed commit
611 								 * barrier */
612 			if (rf_engineDebug || 1) {
613 				rf_get_threadid(tid);
614 				printf("[%d] node (%s) returned fail, rolling forward\n", tid, node->name);
615 			}
616 		} else {
617 			node->dagHdr->status = rf_rollBackward;	/* never reached commit
618 								 * barrier */
619 			if (rf_engineDebug || 1) {
620 				rf_get_threadid(tid);
621 				printf("[%d] node (%s) returned fail, rolling backward\n", tid, node->name);
622 			}
623 		}
624 		break;
625 	case rf_undone:
626 		/* normal rollBackward case, don't need to do anything */
627 		break;
628 	case rf_panic:
629 		/* an undo node failed!!! */
630 		printf("UNDO of a node failed!!!/n");
631 		break;
632 	default:
633 		printf("node finished execution with an illegal status!!!\n");
634 		RF_PANIC();
635 		break;
636 	}
637 
638 	/* enqueue node's succedents (antecedents if rollBackward) for
639 	 * execution */
640 	PropagateResults(node, context);
641 }
642 
643 
644 
645 /* user context or dag-exec-thread context:
646  * This is the first step in post-processing a newly-completed node.
647  * This routine is called by each node execution function to mark the node
648  * as complete and fire off any successors that have been enabled.
649  */
650 int
651 rf_FinishNode(
652     RF_DagNode_t * node,
653     int context)
654 {
655 	/* as far as I can tell, retcode is not used -wvcii */
656 	int     retcode = RF_FALSE;
657 	node->dagHdr->numNodesCompleted++;
658 	ProcessNode(node, context);
659 
660 	return (retcode);
661 }
662 
663 
664 /* user context:
665  * submit dag for execution, return non-zero if we have to wait for completion.
666  * if and only if we return non-zero, we'll cause cbFunc to get invoked with
667  * cbArg when the DAG has completed.
668  *
669  * for now we always return 1.  If the DAG does not cause any I/O, then the callback
670  * may get invoked before DispatchDAG returns.  There's code in state 5 of ContinueRaidAccess
671  * to handle this.
672  *
673  * All we do here is fire the direct successors of the header node.  The
674  * DAG execution thread does the rest of the dag processing.
675  */
676 int
677 rf_DispatchDAG(
678     RF_DagHeader_t * dag,
679     void (*cbFunc) (void *),
680     void *cbArg)
681 {
682 	RF_Raid_t *raidPtr;
683 	int     tid;
684 
685 	raidPtr = dag->raidPtr;
686 	if (dag->tracerec) {
687 		RF_ETIMER_START(dag->tracerec->timer);
688 	}
689 	if (rf_engineDebug || rf_validateDAGDebug) {
690 		if (rf_ValidateDAG(dag))
691 			RF_PANIC();
692 	}
693 	if (rf_engineDebug) {
694 		rf_get_threadid(tid);
695 		printf("[%d] Entering DispatchDAG\n", tid);
696 	}
697 	raidPtr->dags_in_flight++;	/* debug only:  blow off proper
698 					 * locking */
699 	dag->cbFunc = cbFunc;
700 	dag->cbArg = cbArg;
701 	dag->numNodesCompleted = 0;
702 	dag->status = rf_enable;
703 	FireNodeArray(dag->numSuccedents, dag->succedents);
704 	return (1);
705 }
706 /* dedicated kernel thread:
707  * the thread that handles all DAG node firing.
708  * To minimize locking and unlocking, we grab a copy of the entire node queue and then set the
709  * node queue to NULL before doing any firing of nodes.  This way we only have to release the
710  * lock once.  Of course, it's probably rare that there's more than one node in the queue at
711  * any one time, but it sometimes happens.
712  *
713  * In the kernel, this thread runs at spl0 and is not swappable.  I copied these
714  * characteristics from the aio_completion_thread.
715  */
716 
717 static void
718 DAGExecutionThread(RF_ThreadArg_t arg)
719 {
720 	RF_DagNode_t *nd, *local_nq, *term_nq, *fire_nq;
721 	RF_Raid_t *raidPtr;
722 	int     ks, tid;
723 	int     s;
724 
725 	raidPtr = (RF_Raid_t *) arg;
726 
727 	rf_assign_threadid();
728 	if (rf_engineDebug) {
729 		rf_get_threadid(tid);
730 		printf("[%d] Engine thread is running\n", tid);
731 	}
732 #ifndef __NetBSD__
733 	thread = current_thread();
734 	thread_swappable(thread, RF_FALSE);
735 	thread->priority = thread->sched_pri = BASEPRI_SYSTEM;
736 	s = spl0();
737 #endif
738 	/* XXX what to put here XXX */
739 
740 	s = splbio();
741 
742 	RF_THREADGROUP_RUNNING(&raidPtr->engine_tg);
743 
744 	DO_LOCK(raidPtr);
745 	while (!raidPtr->shutdown_engine) {
746 
747 		while (raidPtr->node_queue != NULL) {
748 			local_nq = raidPtr->node_queue;
749 			fire_nq = NULL;
750 			term_nq = NULL;
751 			raidPtr->node_queue = NULL;
752 			DO_UNLOCK(raidPtr);
753 
754 			/* first, strip out the terminal nodes */
755 			while (local_nq) {
756 				nd = local_nq;
757 				local_nq = local_nq->next;
758 				switch (nd->dagHdr->status) {
759 				case rf_enable:
760 				case rf_rollForward:
761 					if (nd->numSuccedents == 0) {
762 						/* end of the dag, add to
763 						 * callback list */
764 						nd->next = term_nq;
765 						term_nq = nd;
766 					} else {
767 						/* not the end, add to the
768 						 * fire queue */
769 						nd->next = fire_nq;
770 						fire_nq = nd;
771 					}
772 					break;
773 				case rf_rollBackward:
774 					if (nd->numAntecedents == 0) {
775 						/* end of the dag, add to the
776 						 * callback list */
777 						nd->next = term_nq;
778 						term_nq = nd;
779 					} else {
780 						/* not the end, add to the
781 						 * fire queue */
782 						nd->next = fire_nq;
783 						fire_nq = nd;
784 					}
785 					break;
786 				default:
787 					RF_PANIC();
788 					break;
789 				}
790 			}
791 
792 			/* execute callback of dags which have reached the
793 			 * terminal node */
794 			while (term_nq) {
795 				nd = term_nq;
796 				term_nq = term_nq->next;
797 				nd->next = NULL;
798 				(nd->dagHdr->cbFunc) (nd->dagHdr->cbArg);
799 				raidPtr->dags_in_flight--;	/* debug only */
800 			}
801 
802 			/* fire remaining nodes */
803 			FireNodeList(fire_nq);
804 
805 			DO_LOCK(raidPtr);
806 		}
807 		while (!raidPtr->shutdown_engine && raidPtr->node_queue == NULL)
808 			DO_WAIT(raidPtr);
809 	}
810 	DO_UNLOCK(raidPtr);
811 
812 	RF_THREADGROUP_DONE(&raidPtr->engine_tg);
813 #ifdef __NetBSD__
814 	splx(s);
815 	kthread_exit(0);
816 #else
817 	splx(s);
818 	thread_terminate(thread);
819 	thread_halt_self();
820 #endif
821 }
822