xref: /netbsd/sys/dev/raidframe/rf_engine.c (revision 9272c734)
1 /*	$NetBSD: rf_engine.c,v 1.53 2019/10/10 03:43:59 christos 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 1.53 2019/10/10 03:43:59 christos Exp $");
59 
60 #include <sys/errno.h>
61 
62 #include "rf_threadstuff.h"
63 #include "rf_dag.h"
64 #include "rf_engine.h"
65 #include "rf_etimer.h"
66 #include "rf_general.h"
67 #include "rf_dagutils.h"
68 #include "rf_shutdown.h"
69 #include "rf_raid.h"
70 #include "rf_kintf.h"
71 #include "rf_paritymap.h"
72 
73 static void rf_ShutdownEngine(void *);
74 static void DAGExecutionThread(RF_ThreadArg_t arg);
75 static void rf_RaidIOThread(RF_ThreadArg_t arg);
76 
77 /* synchronization primitives for this file.  DO_WAIT should be enclosed in a while loop. */
78 
79 #define DO_LOCK(_r_) \
80 	rf_lock_mutex2((_r_)->node_queue_mutex)
81 
82 #define DO_UNLOCK(_r_) \
83 	rf_unlock_mutex2((_r_)->node_queue_mutex)
84 
85 #define	DO_WAIT(_r_) \
86 	rf_wait_cond2((_r_)->node_queue_cv, (_r_)->node_queue_mutex)
87 
88 #define	DO_SIGNAL(_r_) \
89 	rf_broadcast_cond2((_r_)->node_queue_cv)	/* XXX rf_signal_cond2? */
90 
91 static void
rf_ShutdownEngine(void * arg)92 rf_ShutdownEngine(void *arg)
93 {
94 	RF_Raid_t *raidPtr;
95 
96 	raidPtr = (RF_Raid_t *) arg;
97 
98 	/* Tell the rf_RaidIOThread to shutdown */
99 	rf_lock_mutex2(raidPtr->iodone_lock);
100 
101 	raidPtr->shutdown_raidio = 1;
102 	rf_signal_cond2(raidPtr->iodone_cv);
103 
104 	/* ...and wait for it to tell us it has finished */
105 	while (raidPtr->shutdown_raidio)
106 		rf_wait_cond2(raidPtr->iodone_cv, raidPtr->iodone_lock);
107 
108 	rf_unlock_mutex2(raidPtr->iodone_lock);
109 
110  	/* Now shut down the DAG execution engine. */
111  	DO_LOCK(raidPtr);
112   	raidPtr->shutdown_engine = 1;
113   	DO_SIGNAL(raidPtr);
114 
115 	/* ...and wait for it to tell us it has finished */
116 	while (raidPtr->shutdown_engine)
117 		DO_WAIT(raidPtr);
118 
119  	DO_UNLOCK(raidPtr);
120 
121 	rf_destroy_mutex2(raidPtr->node_queue_mutex);
122 	rf_destroy_cond2(raidPtr->node_queue_cv);
123 
124 	rf_destroy_mutex2(raidPtr->iodone_lock);
125 	rf_destroy_cond2(raidPtr->iodone_cv);
126 }
127 
128 int
rf_ConfigureEngine(RF_ShutdownList_t ** listp,RF_Raid_t * raidPtr,RF_Config_t * cfgPtr)129 rf_ConfigureEngine(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
130 		   RF_Config_t *cfgPtr)
131 {
132 
133 	/*
134 	 * Initialise iodone for the IO thread.
135 	 */
136 	TAILQ_INIT(&(raidPtr->iodone));
137 	rf_init_mutex2(raidPtr->iodone_lock, IPL_VM);
138 	rf_init_cond2(raidPtr->iodone_cv, "raidiow");
139 
140 	rf_init_mutex2(raidPtr->node_queue_mutex, IPL_VM);
141 	rf_init_cond2(raidPtr->node_queue_cv, "rfnodeq");
142 	raidPtr->node_queue = NULL;
143 	raidPtr->dags_in_flight = 0;
144 
145 	/* we create the execution thread only once per system boot. no need
146 	 * to check return code b/c the kernel panics if it can't create the
147 	 * thread. */
148 #if RF_DEBUG_ENGINE
149 	if (rf_engineDebug) {
150 		printf("raid%d: Creating engine thread\n", raidPtr->raidid);
151 	}
152 #endif
153 	if (RF_CREATE_ENGINE_THREAD(raidPtr->engine_thread,
154 				    DAGExecutionThread, raidPtr,
155 				    "raid%d", raidPtr->raidid)) {
156 		printf("raid%d: Unable to create engine thread\n",
157 		       raidPtr->raidid);
158 		return (ENOMEM);
159 	}
160 	if (RF_CREATE_ENGINE_THREAD(raidPtr->engine_helper_thread,
161 				    rf_RaidIOThread, raidPtr,
162 				    "raidio%d", raidPtr->raidid)) {
163 		printf("raid%d: Unable to create raidio thread\n",
164 		       raidPtr->raidid);
165 		return (ENOMEM);
166 	}
167 #if RF_DEBUG_ENGINE
168 	if (rf_engineDebug) {
169 		printf("raid%d: Created engine thread\n", raidPtr->raidid);
170 	}
171 #endif
172 
173 	/* engine thread is now running and waiting for work */
174 #if RF_DEBUG_ENGINE
175 	if (rf_engineDebug) {
176 		printf("raid%d: Engine thread running and waiting for events\n", raidPtr->raidid);
177 	}
178 #endif
179 	rf_ShutdownCreate(listp, rf_ShutdownEngine, raidPtr);
180 
181 	return (0);
182 }
183 
184 #if 0
185 static int
186 BranchDone(RF_DagNode_t *node)
187 {
188 	int     i;
189 
190 	/* return true if forward execution is completed for a node and its
191 	 * succedents */
192 	switch (node->status) {
193 	case rf_wait:
194 		/* should never be called in this state */
195 		RF_PANIC();
196 		break;
197 	case rf_fired:
198 		/* node is currently executing, so we're not done */
199 		return (RF_FALSE);
200 	case rf_good:
201 		/* for each succedent recursively check branch */
202 		for (i = 0; i < node->numSuccedents; i++)
203 			if (!BranchDone(node->succedents[i]))
204 				return RF_FALSE;
205 		return RF_TRUE;	/* node and all succedent branches aren't in
206 				 * fired state */
207 	case rf_bad:
208 		/* succedents can't fire */
209 		return (RF_TRUE);
210 	case rf_recover:
211 		/* should never be called in this state */
212 		RF_PANIC();
213 		break;
214 	case rf_undone:
215 	case rf_panic:
216 		/* XXX need to fix this case */
217 		/* for now, assume that we're done */
218 		return (RF_TRUE);
219 	default:
220 		/* illegal node status */
221 		RF_PANIC();
222 		break;
223 	}
224 }
225 #endif
226 
227 static int
NodeReady(RF_DagNode_t * node)228 NodeReady(RF_DagNode_t *node)
229 {
230 	int     ready;
231 
232 	ready = RF_FALSE;
233 
234 	switch (node->dagHdr->status) {
235 	case rf_enable:
236 	case rf_rollForward:
237 		if ((node->status == rf_wait) &&
238 		    (node->numAntecedents == node->numAntDone))
239 			ready = RF_TRUE;
240 		break;
241 	case rf_rollBackward:
242 		RF_ASSERT(node->numSuccDone <= node->numSuccedents);
243 		RF_ASSERT(node->numSuccFired <= node->numSuccedents);
244 		RF_ASSERT(node->numSuccFired <= node->numSuccDone);
245 		if ((node->status == rf_good) &&
246 		    (node->numSuccDone == node->numSuccedents))
247 			ready = RF_TRUE;
248 		break;
249 	default:
250 		printf("Execution engine found illegal DAG status in NodeReady\n");
251 		RF_PANIC();
252 		break;
253 	}
254 
255 	return (ready);
256 }
257 
258 
259 
260 /* user context and dag-exec-thread context: Fire a node.  The node's
261  * status field determines which function, do or undo, to be fired.
262  * This routine assumes that the node's status field has alread been
263  * set to "fired" or "recover" to indicate the direction of execution.
264  */
265 static void
FireNode(RF_DagNode_t * node)266 FireNode(RF_DagNode_t *node)
267 {
268 	switch (node->status) {
269 	case rf_fired:
270 		/* fire the do function of a node */
271 #if RF_DEBUG_ENGINE
272 		if (rf_engineDebug) {
273 			printf("raid%d: Firing node 0x%lx (%s)\n",
274 			       node->dagHdr->raidPtr->raidid,
275 			       (unsigned long) node, node->name);
276 		}
277 #endif
278 		if (node->flags & RF_DAGNODE_FLAG_YIELD) {
279 #if defined(__NetBSD__) && defined(_KERNEL)
280 			/* thread_block(); */
281 			/* printf("Need to block the thread here...\n");  */
282 			/* XXX thread_block is actually mentioned in
283 			 * /usr/include/vm/vm_extern.h */
284 #else
285 			thread_block();
286 #endif
287 		}
288 		(*(node->doFunc)) (node);
289 		break;
290 	case rf_recover:
291 		/* fire the undo function of a node */
292 #if RF_DEBUG_ENGINE
293 		if (rf_engineDebug) {
294 			printf("raid%d: Firing (undo) node 0x%lx (%s)\n",
295 			       node->dagHdr->raidPtr->raidid,
296 			       (unsigned long) node, node->name);
297 		}
298 #endif
299 		if (node->flags & RF_DAGNODE_FLAG_YIELD)
300 #if defined(__NetBSD__) && defined(_KERNEL)
301 			/* thread_block(); */
302 			/* printf("Need to block the thread here...\n"); */
303 			/* XXX thread_block is actually mentioned in
304 			 * /usr/include/vm/vm_extern.h */
305 #else
306 			thread_block();
307 #endif
308 		(*(node->undoFunc)) (node);
309 		break;
310 	default:
311 		RF_PANIC();
312 		break;
313 	}
314 }
315 
316 
317 
318 /* user context:
319  * Attempt to fire each node in a linear array.
320  * The entire list is fired atomically.
321  */
322 static void
FireNodeArray(int numNodes,RF_DagNode_t ** nodeList)323 FireNodeArray(int numNodes, RF_DagNode_t **nodeList)
324 {
325 	RF_DagStatus_t dstat;
326 	RF_DagNode_t *node;
327 	int     i, j;
328 
329 	/* first, mark all nodes which are ready to be fired */
330 	for (i = 0; i < numNodes; i++) {
331 		node = nodeList[i];
332 		dstat = node->dagHdr->status;
333 		RF_ASSERT((node->status == rf_wait) ||
334 			  (node->status == rf_good));
335 		if (NodeReady(node)) {
336 			if ((dstat == rf_enable) ||
337 			    (dstat == rf_rollForward)) {
338 				RF_ASSERT(node->status == rf_wait);
339 				if (node->commitNode)
340 					node->dagHdr->numCommits++;
341 				node->status = rf_fired;
342 				for (j = 0; j < node->numAntecedents; j++)
343 					node->antecedents[j]->numSuccFired++;
344 			} else {
345 				RF_ASSERT(dstat == rf_rollBackward);
346 				RF_ASSERT(node->status == rf_good);
347 				/* only one commit node per graph */
348 				RF_ASSERT(node->commitNode == RF_FALSE);
349 				node->status = rf_recover;
350 			}
351 		}
352 	}
353 	/* now, fire the nodes */
354 	for (i = 0; i < numNodes; i++) {
355 		if ((nodeList[i]->status == rf_fired) ||
356 		    (nodeList[i]->status == rf_recover))
357 			FireNode(nodeList[i]);
358 	}
359 }
360 
361 
362 /* user context:
363  * Attempt to fire each node in a linked list.
364  * The entire list is fired atomically.
365  */
366 static void
FireNodeList(RF_DagNode_t * nodeList)367 FireNodeList(RF_DagNode_t *nodeList)
368 {
369 	RF_DagNode_t *node, *next;
370 	RF_DagStatus_t dstat;
371 	int     j;
372 
373 	if (nodeList) {
374 		/* first, mark all nodes which are ready to be fired */
375 		for (node = nodeList; node; node = next) {
376 			next = node->next;
377 			dstat = node->dagHdr->status;
378 			RF_ASSERT((node->status == rf_wait) ||
379 				  (node->status == rf_good));
380 			if (NodeReady(node)) {
381 				if ((dstat == rf_enable) ||
382 				    (dstat == rf_rollForward)) {
383 					RF_ASSERT(node->status == rf_wait);
384 					if (node->commitNode)
385 						node->dagHdr->numCommits++;
386 					node->status = rf_fired;
387 					for (j = 0; j < node->numAntecedents; j++)
388 						node->antecedents[j]->numSuccFired++;
389 				} else {
390 					RF_ASSERT(dstat == rf_rollBackward);
391 					RF_ASSERT(node->status == rf_good);
392 					/* only one commit node per graph */
393 					RF_ASSERT(node->commitNode == RF_FALSE);
394 					node->status = rf_recover;
395 				}
396 			}
397 		}
398 		/* now, fire the nodes */
399 		for (node = nodeList; node; node = next) {
400 			next = node->next;
401 			if ((node->status == rf_fired) ||
402 			    (node->status == rf_recover))
403 				FireNode(node);
404 		}
405 	}
406 }
407 /* interrupt context:
408  * for each succedent
409  *    propagate required results from node to succedent
410  *    increment succedent's numAntDone
411  *    place newly-enable nodes on node queue for firing
412  *
413  * To save context switches, we don't place NIL nodes on the node queue,
414  * but rather just process them as if they had fired.  Note that NIL nodes
415  * that are the direct successors of the header will actually get fired by
416  * DispatchDAG, which is fine because no context switches are involved.
417  *
418  * Important:  when running at user level, this can be called by any
419  * disk thread, and so the increment and check of the antecedent count
420  * must be locked.  I used the node queue mutex and locked down the
421  * entire function, but this is certainly overkill.
422  */
423 static void
PropagateResults(RF_DagNode_t * node,int context)424 PropagateResults(RF_DagNode_t *node, int context)
425 {
426 	RF_DagNode_t *s, *a;
427 	RF_Raid_t *raidPtr;
428 	int     i;
429 	RF_DagNode_t *finishlist = NULL;	/* a list of NIL nodes to be
430 						 * finished */
431 	RF_DagNode_t *skiplist = NULL;	/* list of nodes with failed truedata
432 					 * antecedents */
433 	RF_DagNode_t *firelist = NULL;	/* a list of nodes to be fired */
434 	RF_DagNode_t *q = NULL, *qh = NULL, *next;
435 	int     j, skipNode;
436 
437 	raidPtr = node->dagHdr->raidPtr;
438 
439 	DO_LOCK(raidPtr);
440 
441 	/* debug - validate fire counts */
442 	for (i = 0; i < node->numAntecedents; i++) {
443 		a = *(node->antecedents + i);
444 		RF_ASSERT(a->numSuccFired >= a->numSuccDone);
445 		RF_ASSERT(a->numSuccFired <= a->numSuccedents);
446 		a->numSuccDone++;
447 	}
448 
449 	switch (node->dagHdr->status) {
450 	case rf_enable:
451 	case rf_rollForward:
452 		for (i = 0; i < node->numSuccedents; i++) {
453 			s = *(node->succedents + i);
454 			RF_ASSERT(s->status == rf_wait);
455 			(s->numAntDone)++;
456 			if (s->numAntDone == s->numAntecedents) {
457 				/* look for NIL nodes */
458 				if (s->doFunc == rf_NullNodeFunc) {
459 					/* don't fire NIL nodes, just process
460 					 * them */
461 					s->next = finishlist;
462 					finishlist = s;
463 				} else {
464 					/* look to see if the node is to be
465 					 * skipped */
466 					skipNode = RF_FALSE;
467 					for (j = 0; j < s->numAntecedents; j++)
468 						if ((s->antType[j] == rf_trueData) && (s->antecedents[j]->status == rf_bad))
469 							skipNode = RF_TRUE;
470 					if (skipNode) {
471 						/* this node has one or more
472 						 * failed true data
473 						 * dependencies, so skip it */
474 						s->next = skiplist;
475 						skiplist = s;
476 					} else
477 						/* add s to list of nodes (q)
478 						 * to execute */
479 						if (context != RF_INTR_CONTEXT) {
480 							/* we only have to
481 							 * enqueue if we're at
482 							 * intr context */
483 							/* put node on
484                                                            a list to
485                                                            be fired
486                                                            after we
487                                                            unlock */
488 							s->next = firelist;
489 							firelist = s;
490 						} else {
491 							/* enqueue the
492 							   node for
493 							   the dag
494 							   exec thread
495 							   to fire */
496 							RF_ASSERT(NodeReady(s));
497 							if (q) {
498 								q->next = s;
499 								q = s;
500 							} else {
501 								qh = q = s;
502 								qh->next = NULL;
503 							}
504 						}
505 				}
506 			}
507 		}
508 
509 		if (q) {
510 			/* xfer our local list of nodes to the node queue */
511 			q->next = raidPtr->node_queue;
512 			raidPtr->node_queue = qh;
513 			DO_SIGNAL(raidPtr);
514 		}
515 		DO_UNLOCK(raidPtr);
516 
517 		for (; skiplist; skiplist = next) {
518 			next = skiplist->next;
519 			skiplist->status = rf_skipped;
520 			for (i = 0; i < skiplist->numAntecedents; i++) {
521 				skiplist->antecedents[i]->numSuccFired++;
522 			}
523 			if (skiplist->commitNode) {
524 				skiplist->dagHdr->numCommits++;
525 			}
526 			rf_FinishNode(skiplist, context);
527 		}
528 		for (; finishlist; finishlist = next) {
529 			/* NIL nodes: no need to fire them */
530 			next = finishlist->next;
531 			finishlist->status = rf_good;
532 			for (i = 0; i < finishlist->numAntecedents; i++) {
533 				finishlist->antecedents[i]->numSuccFired++;
534 			}
535 			if (finishlist->commitNode)
536 				finishlist->dagHdr->numCommits++;
537 			/*
538 		         * Okay, here we're calling rf_FinishNode() on
539 		         * nodes that have the null function as their
540 		         * work proc. Such a node could be the
541 		         * terminal node in a DAG. If so, it will
542 		         * cause the DAG to complete, which will in
543 		         * turn free memory used by the DAG, which
544 		         * includes the node in question. Thus, we
545 		         * must avoid referencing the node at all
546 		         * after calling rf_FinishNode() on it.  */
547 			rf_FinishNode(finishlist, context);	/* recursive call */
548 		}
549 		/* fire all nodes in firelist */
550 		FireNodeList(firelist);
551 		break;
552 
553 	case rf_rollBackward:
554 		for (i = 0; i < node->numAntecedents; i++) {
555 			a = *(node->antecedents + i);
556 			RF_ASSERT(a->status == rf_good);
557 			RF_ASSERT(a->numSuccDone <= a->numSuccedents);
558 			RF_ASSERT(a->numSuccDone <= a->numSuccFired);
559 
560 			if (a->numSuccDone == a->numSuccFired) {
561 				if (a->undoFunc == rf_NullNodeFunc) {
562 					/* don't fire NIL nodes, just process
563 					 * them */
564 					a->next = finishlist;
565 					finishlist = a;
566 				} else {
567 					if (context != RF_INTR_CONTEXT) {
568 						/* we only have to enqueue if
569 						 * we're at intr context */
570 						/* put node on a list to be
571 						   fired after we unlock */
572 						a->next = firelist;
573 
574 						firelist = a;
575 					} else {
576 						/* enqueue the node for the
577 						   dag exec thread to fire */
578 						RF_ASSERT(NodeReady(a));
579 						if (q) {
580 							q->next = a;
581 							q = a;
582 						} else {
583 							qh = q = a;
584 							qh->next = NULL;
585 						}
586 					}
587 				}
588 			}
589 		}
590 		if (q) {
591 			/* xfer our local list of nodes to the node queue */
592 			q->next = raidPtr->node_queue;
593 			raidPtr->node_queue = qh;
594 			DO_SIGNAL(raidPtr);
595 		}
596 		DO_UNLOCK(raidPtr);
597 		for (; finishlist; finishlist = next) {
598 			/* NIL nodes: no need to fire them */
599 			next = finishlist->next;
600 			finishlist->status = rf_good;
601 			/*
602 		         * Okay, here we're calling rf_FinishNode() on
603 		         * nodes that have the null function as their
604 		         * work proc. Such a node could be the first
605 		         * node in a DAG. If so, it will cause the DAG
606 		         * to complete, which will in turn free memory
607 		         * used by the DAG, which includes the node in
608 		         * question. Thus, we must avoid referencing
609 		         * the node at all after calling
610 		         * rf_FinishNode() on it.  */
611 			rf_FinishNode(finishlist, context);	/* recursive call */
612 		}
613 		/* fire all nodes in firelist */
614 		FireNodeList(firelist);
615 
616 		break;
617 	default:
618 		printf("Engine found illegal DAG status in PropagateResults()\n");
619 		RF_PANIC();
620 		break;
621 	}
622 }
623 
624 
625 
626 /*
627  * Process a fired node which has completed
628  */
629 static void
ProcessNode(RF_DagNode_t * node,int context)630 ProcessNode(RF_DagNode_t *node, int context)
631 {
632 #if RF_DEBUG_ENGINE
633 	RF_Raid_t *raidPtr;
634 
635 	raidPtr = node->dagHdr->raidPtr;
636 #endif
637 
638 	switch (node->status) {
639 	case rf_good:
640 		/* normal case, don't need to do anything */
641 		break;
642 	case rf_bad:
643 		if ((node->dagHdr->numCommits > 0) ||
644 		    (node->dagHdr->numCommitNodes == 0)) {
645 			/* crossed commit barrier */
646 			node->dagHdr->status = rf_rollForward;
647 #if RF_DEBUG_ENGINE
648 			if (rf_engineDebug) {
649 				printf("raid%d: node (%s) returned fail, rolling forward\n", raidPtr->raidid, node->name);
650 			}
651 #endif
652 		} else {
653 			/* never reached commit barrier */
654 			node->dagHdr->status = rf_rollBackward;
655 #if RF_DEBUG_ENGINE
656 			if (rf_engineDebug) {
657 				printf("raid%d: node (%s) returned fail, rolling backward\n", raidPtr->raidid, node->name);
658 			}
659 #endif
660 		}
661 		break;
662 	case rf_undone:
663 		/* normal rollBackward case, don't need to do anything */
664 		break;
665 	case rf_panic:
666 		/* an undo node failed!!! */
667 		printf("UNDO of a node failed!!!\n");
668 		break;
669 	default:
670 		printf("node finished execution with an illegal status!!!\n");
671 		RF_PANIC();
672 		break;
673 	}
674 
675 	/* enqueue node's succedents (antecedents if rollBackward) for
676 	 * execution */
677 	PropagateResults(node, context);
678 }
679 
680 
681 
682 /* user context or dag-exec-thread context:
683  * This is the first step in post-processing a newly-completed node.
684  * This routine is called by each node execution function to mark the node
685  * as complete and fire off any successors that have been enabled.
686  */
687 void
rf_FinishNode(RF_DagNode_t * node,int context)688 rf_FinishNode(RF_DagNode_t *node, int context)
689 {
690 	node->dagHdr->numNodesCompleted++;
691 	ProcessNode(node, context);
692 }
693 
694 
695 /* user context: submit dag for execution, return non-zero if we have
696  * to wait for completion.  if and only if we return non-zero, we'll
697  * cause cbFunc to get invoked with cbArg when the DAG has completed.
698  *
699  * for now we always return 1.  If the DAG does not cause any I/O,
700  * then the callback may get invoked before DispatchDAG returns.
701  * There's code in state 5 of ContinueRaidAccess to handle this.
702  *
703  * All we do here is fire the direct successors of the header node.
704  * The DAG execution thread does the rest of the dag processing.  */
705 int
rf_DispatchDAG(RF_DagHeader_t * dag,void (* cbFunc)(void *),void * cbArg)706 rf_DispatchDAG(RF_DagHeader_t *dag, void (*cbFunc) (void *),
707 	       void *cbArg)
708 {
709 	RF_Raid_t *raidPtr;
710 
711 	raidPtr = dag->raidPtr;
712 #if RF_ACC_TRACE > 0
713 	if (dag->tracerec) {
714 		RF_ETIMER_START(dag->tracerec->timer);
715 	}
716 #endif
717 #if DEBUG
718 #if RF_DEBUG_VALIDATE_DAG
719 	if (rf_engineDebug || rf_validateDAGDebug) {
720 		if (rf_ValidateDAG(dag))
721 			RF_PANIC();
722 	}
723 #endif
724 #endif
725 #if RF_DEBUG_ENGINE
726 	if (rf_engineDebug) {
727 		printf("raid%d: Entering DispatchDAG\n", raidPtr->raidid);
728 	}
729 #endif
730 	raidPtr->dags_in_flight++;	/* debug only:  blow off proper
731 					 * locking */
732 	dag->cbFunc = cbFunc;
733 	dag->cbArg = cbArg;
734 	dag->numNodesCompleted = 0;
735 	dag->status = rf_enable;
736 	FireNodeArray(dag->numSuccedents, dag->succedents);
737 	return (1);
738 }
739 /* dedicated kernel thread: the thread that handles all DAG node
740  * firing.  To minimize locking and unlocking, we grab a copy of the
741  * entire node queue and then set the node queue to NULL before doing
742  * any firing of nodes.  This way we only have to release the lock
743  * once.  Of course, it's probably rare that there's more than one
744  * node in the queue at any one time, but it sometimes happens.
745  */
746 
747 static void
DAGExecutionThread(RF_ThreadArg_t arg)748 DAGExecutionThread(RF_ThreadArg_t arg)
749 {
750 	RF_DagNode_t *nd, *local_nq, *term_nq, *fire_nq;
751 	RF_Raid_t *raidPtr;
752 
753 	raidPtr = (RF_Raid_t *) arg;
754 
755 #if RF_DEBUG_ENGINE
756 	if (rf_engineDebug) {
757 		printf("raid%d: Engine thread is running\n", raidPtr->raidid);
758 	}
759 #endif
760 
761 	DO_LOCK(raidPtr);
762 	while (!raidPtr->shutdown_engine) {
763 
764 		while (raidPtr->node_queue != NULL) {
765 			local_nq = raidPtr->node_queue;
766 			fire_nq = NULL;
767 			term_nq = NULL;
768 			raidPtr->node_queue = NULL;
769 			DO_UNLOCK(raidPtr);
770 
771 			/* first, strip out the terminal nodes */
772 			while (local_nq) {
773 				nd = local_nq;
774 				local_nq = local_nq->next;
775 				switch (nd->dagHdr->status) {
776 				case rf_enable:
777 				case rf_rollForward:
778 					if (nd->numSuccedents == 0) {
779 						/* end of the dag, add to
780 						 * callback list */
781 						nd->next = term_nq;
782 						term_nq = nd;
783 					} else {
784 						/* not the end, add to the
785 						 * fire queue */
786 						nd->next = fire_nq;
787 						fire_nq = nd;
788 					}
789 					break;
790 				case rf_rollBackward:
791 					if (nd->numAntecedents == 0) {
792 						/* end of the dag, add to the
793 						 * callback list */
794 						nd->next = term_nq;
795 						term_nq = nd;
796 					} else {
797 						/* not the end, add to the
798 						 * fire queue */
799 						nd->next = fire_nq;
800 						fire_nq = nd;
801 					}
802 					break;
803 				default:
804 					RF_PANIC();
805 					break;
806 				}
807 			}
808 
809 			/* execute callback of dags which have reached the
810 			 * terminal node */
811 			while (term_nq) {
812 				nd = term_nq;
813 				term_nq = term_nq->next;
814 				nd->next = NULL;
815 				(nd->dagHdr->cbFunc) (nd->dagHdr->cbArg);
816 				raidPtr->dags_in_flight--;	/* debug only */
817 			}
818 
819 			/* fire remaining nodes */
820 			FireNodeList(fire_nq);
821 
822 			DO_LOCK(raidPtr);
823 		}
824 		while (!raidPtr->shutdown_engine &&
825 		       raidPtr->node_queue == NULL) {
826 			DO_WAIT(raidPtr);
827 		}
828 	}
829 
830 	/* Let rf_ShutdownEngine know that we're done... */
831 	raidPtr->shutdown_engine = 0;
832 	DO_SIGNAL(raidPtr);
833 
834 	DO_UNLOCK(raidPtr);
835 
836 	kthread_exit(0);
837 }
838 
839 /*
840  * rf_RaidIOThread() -- When I/O to a component begins, raidstrategy()
841  * puts the I/O on a buffer queue, and then signals raidPtr->iodone.  If
842  * necessary, this function calls raidstart() to initiate the I/O.
843  * When I/O to a component completes, KernelWakeupFunc() puts the
844  * completed request onto raidPtr->iodone TAILQ.  This function looks
845  * after requests on that queue by calling rf_DiskIOComplete() for the
846  * request, and by calling any required CompleteFunc for the request.
847  */
848 
849 static void
rf_RaidIOThread(RF_ThreadArg_t arg)850 rf_RaidIOThread(RF_ThreadArg_t arg)
851 {
852 	RF_Raid_t *raidPtr;
853 	RF_DiskQueueData_t *req;
854 
855 	raidPtr = (RF_Raid_t *) arg;
856 
857 	rf_lock_mutex2(raidPtr->iodone_lock);
858 
859 	while (!raidPtr->shutdown_raidio) {
860 		/* if there is nothing to do, then snooze. */
861 		if (TAILQ_EMPTY(&(raidPtr->iodone)) &&
862 		    rf_buf_queue_check(raidPtr)) {
863 			rf_wait_cond2(raidPtr->iodone_cv, raidPtr->iodone_lock);
864 		}
865 
866 		/* Check for deferred parity-map-related work. */
867 		if (raidPtr->parity_map != NULL) {
868 			rf_unlock_mutex2(raidPtr->iodone_lock);
869 			rf_paritymap_checkwork(raidPtr->parity_map);
870 			rf_lock_mutex2(raidPtr->iodone_lock);
871 		}
872 
873 		/* See what I/Os, if any, have arrived */
874 		while ((req = TAILQ_FIRST(&(raidPtr->iodone))) != NULL) {
875 			TAILQ_REMOVE(&(raidPtr->iodone), req, iodone_entries);
876 			rf_unlock_mutex2(raidPtr->iodone_lock);
877 			rf_DiskIOComplete(req->queue, req, req->error);
878 			(req->CompleteFunc) (req->argument, req->error);
879 			rf_lock_mutex2(raidPtr->iodone_lock);
880 		}
881 
882 		/* process any pending outgoing IO */
883 		rf_unlock_mutex2(raidPtr->iodone_lock);
884 		raidstart(raidPtr);
885 		rf_lock_mutex2(raidPtr->iodone_lock);
886 
887 	}
888 
889 	/* Let rf_ShutdownEngine know that we're done... */
890 	raidPtr->shutdown_raidio = 0;
891 	rf_signal_cond2(raidPtr->iodone_cv);
892 
893 	rf_unlock_mutex2(raidPtr->iodone_lock);
894 
895 	kthread_exit(0);
896 }
897