xref: /netbsd-src/sys/dev/raidframe/rf_dagutils.c (revision 08c81a9c2dc8c7300e893321eb65c0925d60871c)
1 /*	$NetBSD: rf_dagutils.c,v 1.15 2002/09/14 17:53:59 oster Exp $	*/
2 /*
3  * Copyright (c) 1995 Carnegie-Mellon University.
4  * All rights reserved.
5  *
6  * Authors: Mark Holland, William V. Courtright II, Jim Zelenka
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  * rf_dagutils.c -- utility routines for manipulating dags
32  *
33  *****************************************************************************/
34 
35 #include <sys/cdefs.h>
36 __KERNEL_RCSID(0, "$NetBSD: rf_dagutils.c,v 1.15 2002/09/14 17:53:59 oster Exp $");
37 
38 #include <dev/raidframe/raidframevar.h>
39 
40 #include "rf_archs.h"
41 #include "rf_threadstuff.h"
42 #include "rf_raid.h"
43 #include "rf_dag.h"
44 #include "rf_dagutils.h"
45 #include "rf_dagfuncs.h"
46 #include "rf_general.h"
47 #include "rf_freelist.h"
48 #include "rf_map.h"
49 #include "rf_shutdown.h"
50 
51 #define SNUM_DIFF(_a_,_b_) (((_a_)>(_b_))?((_a_)-(_b_)):((_b_)-(_a_)))
52 
53 RF_RedFuncs_t rf_xorFuncs = {
54 	rf_RegularXorFunc, "Reg Xr",
55 rf_SimpleXorFunc, "Simple Xr"};
56 
57 RF_RedFuncs_t rf_xorRecoveryFuncs = {
58 	rf_RecoveryXorFunc, "Recovery Xr",
59 rf_RecoveryXorFunc, "Recovery Xr"};
60 
61 #if RF_DEBUG_VALIDATE_DAG
62 static void rf_RecurPrintDAG(RF_DagNode_t *, int, int);
63 static void rf_PrintDAG(RF_DagHeader_t *);
64 static int rf_ValidateBranch(RF_DagNode_t *, int *, int *,
65 			     RF_DagNode_t **, int);
66 static void rf_ValidateBranchVisitedBits(RF_DagNode_t *, int, int);
67 static void rf_ValidateVisitedBits(RF_DagHeader_t *);
68 #endif /* RF_DEBUG_VALIDATE_DAG */
69 
70 /******************************************************************************
71  *
72  * InitNode - initialize a dag node
73  *
74  * the size of the propList array is always the same as that of the
75  * successors array.
76  *
77  *****************************************************************************/
78 void
79 rf_InitNode(
80     RF_DagNode_t * node,
81     RF_NodeStatus_t initstatus,
82     int commit,
83     int (*doFunc) (RF_DagNode_t * node),
84     int (*undoFunc) (RF_DagNode_t * node),
85     int (*wakeFunc) (RF_DagNode_t * node, int status),
86     int nSucc,
87     int nAnte,
88     int nParam,
89     int nResult,
90     RF_DagHeader_t * hdr,
91     char *name,
92     RF_AllocListElem_t * alist)
93 {
94 	void  **ptrs;
95 	int     nptrs;
96 
97 	if (nAnte > RF_MAX_ANTECEDENTS)
98 		RF_PANIC();
99 	node->status = initstatus;
100 	node->commitNode = commit;
101 	node->doFunc = doFunc;
102 	node->undoFunc = undoFunc;
103 	node->wakeFunc = wakeFunc;
104 	node->numParams = nParam;
105 	node->numResults = nResult;
106 	node->numAntecedents = nAnte;
107 	node->numAntDone = 0;
108 	node->next = NULL;
109 	node->numSuccedents = nSucc;
110 	node->name = name;
111 	node->dagHdr = hdr;
112 	node->visited = 0;
113 
114 	/* allocate all the pointers with one call to malloc */
115 	nptrs = nSucc + nAnte + nResult + nSucc;
116 
117 	if (nptrs <= RF_DAG_PTRCACHESIZE) {
118 		/*
119 	         * The dag_ptrs field of the node is basically some scribble
120 	         * space to be used here. We could get rid of it, and always
121 	         * allocate the range of pointers, but that's expensive. So,
122 	         * we pick a "common case" size for the pointer cache. Hopefully,
123 	         * we'll find that:
124 	         * (1) Generally, nptrs doesn't exceed RF_DAG_PTRCACHESIZE by
125 	         *     only a little bit (least efficient case)
126 	         * (2) Generally, ntprs isn't a lot less than RF_DAG_PTRCACHESIZE
127 	         *     (wasted memory)
128 	         */
129 		ptrs = (void **) node->dag_ptrs;
130 	} else {
131 		RF_CallocAndAdd(ptrs, nptrs, sizeof(void *), (void **), alist);
132 	}
133 	node->succedents = (nSucc) ? (RF_DagNode_t **) ptrs : NULL;
134 	node->antecedents = (nAnte) ? (RF_DagNode_t **) (ptrs + nSucc) : NULL;
135 	node->results = (nResult) ? (void **) (ptrs + nSucc + nAnte) : NULL;
136 	node->propList = (nSucc) ? (RF_PropHeader_t **) (ptrs + nSucc + nAnte + nResult) : NULL;
137 
138 	if (nParam) {
139 		if (nParam <= RF_DAG_PARAMCACHESIZE) {
140 			node->params = (RF_DagParam_t *) node->dag_params;
141 		} else {
142 			RF_CallocAndAdd(node->params, nParam, sizeof(RF_DagParam_t), (RF_DagParam_t *), alist);
143 		}
144 	} else {
145 		node->params = NULL;
146 	}
147 }
148 
149 
150 
151 /******************************************************************************
152  *
153  * allocation and deallocation routines
154  *
155  *****************************************************************************/
156 
157 void
158 rf_FreeDAG(dag_h)
159 	RF_DagHeader_t *dag_h;
160 {
161 	RF_AccessStripeMapHeader_t *asmap, *t_asmap;
162 	RF_DagHeader_t *nextDag;
163 
164 	while (dag_h) {
165 		nextDag = dag_h->next;
166 		rf_FreeAllocList(dag_h->allocList);
167 		for (asmap = dag_h->asmList; asmap;) {
168 			t_asmap = asmap;
169 			asmap = asmap->next;
170 			rf_FreeAccessStripeMap(t_asmap);
171 		}
172 		rf_FreeDAGHeader(dag_h);
173 		dag_h = nextDag;
174 	}
175 }
176 
177 RF_PropHeader_t *
178 rf_MakePropListEntry(
179     RF_DagHeader_t * dag_h,
180     int resultNum,
181     int paramNum,
182     RF_PropHeader_t * next,
183     RF_AllocListElem_t * allocList)
184 {
185 	RF_PropHeader_t *p;
186 
187 	RF_CallocAndAdd(p, 1, sizeof(RF_PropHeader_t),
188 	    (RF_PropHeader_t *), allocList);
189 	p->resultNum = resultNum;
190 	p->paramNum = paramNum;
191 	p->next = next;
192 	return (p);
193 }
194 
195 static RF_FreeList_t *rf_dagh_freelist;
196 
197 #define RF_MAX_FREE_DAGH 128
198 #define RF_DAGH_INC       16
199 #define RF_DAGH_INITIAL   32
200 
201 static void rf_ShutdownDAGs(void *);
202 static void
203 rf_ShutdownDAGs(ignored)
204 	void   *ignored;
205 {
206 	RF_FREELIST_DESTROY(rf_dagh_freelist, next, (RF_DagHeader_t *));
207 }
208 
209 int
210 rf_ConfigureDAGs(listp)
211 	RF_ShutdownList_t **listp;
212 {
213 	int     rc;
214 
215 	RF_FREELIST_CREATE(rf_dagh_freelist, RF_MAX_FREE_DAGH,
216 	    RF_DAGH_INC, sizeof(RF_DagHeader_t));
217 	if (rf_dagh_freelist == NULL)
218 		return (ENOMEM);
219 	rc = rf_ShutdownCreate(listp, rf_ShutdownDAGs, NULL);
220 	if (rc) {
221 		rf_print_unable_to_add_shutdown(__FILE__, __LINE__, rc);
222 		rf_ShutdownDAGs(NULL);
223 		return (rc);
224 	}
225 	RF_FREELIST_PRIME(rf_dagh_freelist, RF_DAGH_INITIAL, next,
226 	    (RF_DagHeader_t *));
227 	return (0);
228 }
229 
230 RF_DagHeader_t *
231 rf_AllocDAGHeader()
232 {
233 	RF_DagHeader_t *dh;
234 
235 	RF_FREELIST_GET(rf_dagh_freelist, dh, next, (RF_DagHeader_t *));
236 	if (dh) {
237 		memset((char *) dh, 0, sizeof(RF_DagHeader_t));
238 	}
239 	return (dh);
240 }
241 
242 void
243 rf_FreeDAGHeader(RF_DagHeader_t * dh)
244 {
245 	RF_FREELIST_FREE(rf_dagh_freelist, dh, next);
246 }
247 /* allocates a buffer big enough to hold the data described by pda */
248 void   *
249 rf_AllocBuffer(
250     RF_Raid_t * raidPtr,
251     RF_DagHeader_t * dag_h,
252     RF_PhysDiskAddr_t * pda,
253     RF_AllocListElem_t * allocList)
254 {
255 	char   *p;
256 
257 	RF_MallocAndAdd(p, pda->numSector << raidPtr->logBytesPerSector,
258 	    (char *), allocList);
259 	return ((void *) p);
260 }
261 #if RF_DEBUG_VALIDATE_DAG
262 /******************************************************************************
263  *
264  * debug routines
265  *
266  *****************************************************************************/
267 
268 char   *
269 rf_NodeStatusString(RF_DagNode_t * node)
270 {
271 	switch (node->status) {
272 		case rf_wait:return ("wait");
273 	case rf_fired:
274 		return ("fired");
275 	case rf_good:
276 		return ("good");
277 	case rf_bad:
278 		return ("bad");
279 	default:
280 		return ("?");
281 	}
282 }
283 
284 void
285 rf_PrintNodeInfoString(RF_DagNode_t * node)
286 {
287 	RF_PhysDiskAddr_t *pda;
288 	int     (*df) (RF_DagNode_t *) = node->doFunc;
289 	int     i, lk, unlk;
290 	void   *bufPtr;
291 
292 	if ((df == rf_DiskReadFunc) || (df == rf_DiskWriteFunc)
293 	    || (df == rf_DiskReadMirrorIdleFunc)
294 	    || (df == rf_DiskReadMirrorPartitionFunc)) {
295 		pda = (RF_PhysDiskAddr_t *) node->params[0].p;
296 		bufPtr = (void *) node->params[1].p;
297 		lk = RF_EXTRACT_LOCK_FLAG(node->params[3].v);
298 		unlk = RF_EXTRACT_UNLOCK_FLAG(node->params[3].v);
299 		RF_ASSERT(!(lk && unlk));
300 		printf("r %d c %d offs %ld nsect %d buf 0x%lx %s\n", pda->row, pda->col,
301 		    (long) pda->startSector, (int) pda->numSector, (long) bufPtr,
302 		    (lk) ? "LOCK" : ((unlk) ? "UNLK" : " "));
303 		return;
304 	}
305 	if (df == rf_DiskUnlockFunc) {
306 		pda = (RF_PhysDiskAddr_t *) node->params[0].p;
307 		lk = RF_EXTRACT_LOCK_FLAG(node->params[3].v);
308 		unlk = RF_EXTRACT_UNLOCK_FLAG(node->params[3].v);
309 		RF_ASSERT(!(lk && unlk));
310 		printf("r %d c %d %s\n", pda->row, pda->col,
311 		    (lk) ? "LOCK" : ((unlk) ? "UNLK" : "nop"));
312 		return;
313 	}
314 	if ((df == rf_SimpleXorFunc) || (df == rf_RegularXorFunc)
315 	    || (df == rf_RecoveryXorFunc)) {
316 		printf("result buf 0x%lx\n", (long) node->results[0]);
317 		for (i = 0; i < node->numParams - 1; i += 2) {
318 			pda = (RF_PhysDiskAddr_t *) node->params[i].p;
319 			bufPtr = (RF_PhysDiskAddr_t *) node->params[i + 1].p;
320 			printf("    buf 0x%lx r%d c%d offs %ld nsect %d\n",
321 			    (long) bufPtr, pda->row, pda->col,
322 			    (long) pda->startSector, (int) pda->numSector);
323 		}
324 		return;
325 	}
326 #if RF_INCLUDE_PARITYLOGGING > 0
327 	if (df == rf_ParityLogOverwriteFunc || df == rf_ParityLogUpdateFunc) {
328 		for (i = 0; i < node->numParams - 1; i += 2) {
329 			pda = (RF_PhysDiskAddr_t *) node->params[i].p;
330 			bufPtr = (RF_PhysDiskAddr_t *) node->params[i + 1].p;
331 			printf(" r%d c%d offs %ld nsect %d buf 0x%lx\n",
332 			    pda->row, pda->col, (long) pda->startSector,
333 			    (int) pda->numSector, (long) bufPtr);
334 		}
335 		return;
336 	}
337 #endif				/* RF_INCLUDE_PARITYLOGGING > 0 */
338 
339 	if ((df == rf_TerminateFunc) || (df == rf_NullNodeFunc)) {
340 		printf("\n");
341 		return;
342 	}
343 	printf("?\n");
344 }
345 
346 static void
347 rf_RecurPrintDAG(node, depth, unvisited)
348 	RF_DagNode_t *node;
349 	int     depth;
350 	int     unvisited;
351 {
352 	char   *anttype;
353 	int     i;
354 
355 	node->visited = (unvisited) ? 0 : 1;
356 	printf("(%d) %d C%d %s: %s,s%d %d/%d,a%d/%d,p%d,r%d S{", depth,
357 	    node->nodeNum, node->commitNode, node->name, rf_NodeStatusString(node),
358 	    node->numSuccedents, node->numSuccFired, node->numSuccDone,
359 	    node->numAntecedents, node->numAntDone, node->numParams, node->numResults);
360 	for (i = 0; i < node->numSuccedents; i++) {
361 		printf("%d%s", node->succedents[i]->nodeNum,
362 		    ((i == node->numSuccedents - 1) ? "\0" : " "));
363 	}
364 	printf("} A{");
365 	for (i = 0; i < node->numAntecedents; i++) {
366 		switch (node->antType[i]) {
367 		case rf_trueData:
368 			anttype = "T";
369 			break;
370 		case rf_antiData:
371 			anttype = "A";
372 			break;
373 		case rf_outputData:
374 			anttype = "O";
375 			break;
376 		case rf_control:
377 			anttype = "C";
378 			break;
379 		default:
380 			anttype = "?";
381 			break;
382 		}
383 		printf("%d(%s)%s", node->antecedents[i]->nodeNum, anttype, (i == node->numAntecedents - 1) ? "\0" : " ");
384 	}
385 	printf("}; ");
386 	rf_PrintNodeInfoString(node);
387 	for (i = 0; i < node->numSuccedents; i++) {
388 		if (node->succedents[i]->visited == unvisited)
389 			rf_RecurPrintDAG(node->succedents[i], depth + 1, unvisited);
390 	}
391 }
392 
393 static void
394 rf_PrintDAG(dag_h)
395 	RF_DagHeader_t *dag_h;
396 {
397 	int     unvisited, i;
398 	char   *status;
399 
400 	/* set dag status */
401 	switch (dag_h->status) {
402 	case rf_enable:
403 		status = "enable";
404 		break;
405 	case rf_rollForward:
406 		status = "rollForward";
407 		break;
408 	case rf_rollBackward:
409 		status = "rollBackward";
410 		break;
411 	default:
412 		status = "illegal!";
413 		break;
414 	}
415 	/* find out if visited bits are currently set or clear */
416 	unvisited = dag_h->succedents[0]->visited;
417 
418 	printf("DAG type:  %s\n", dag_h->creator);
419 	printf("format is (depth) num commit type: status,nSucc nSuccFired/nSuccDone,nAnte/nAnteDone,nParam,nResult S{x} A{x(type)};  info\n");
420 	printf("(0) %d Hdr: %s, s%d, (commit %d/%d) S{", dag_h->nodeNum,
421 	    status, dag_h->numSuccedents, dag_h->numCommitNodes, dag_h->numCommits);
422 	for (i = 0; i < dag_h->numSuccedents; i++) {
423 		printf("%d%s", dag_h->succedents[i]->nodeNum,
424 		    ((i == dag_h->numSuccedents - 1) ? "\0" : " "));
425 	}
426 	printf("};\n");
427 	for (i = 0; i < dag_h->numSuccedents; i++) {
428 		if (dag_h->succedents[i]->visited == unvisited)
429 			rf_RecurPrintDAG(dag_h->succedents[i], 1, unvisited);
430 	}
431 }
432 /* assigns node numbers */
433 int
434 rf_AssignNodeNums(RF_DagHeader_t * dag_h)
435 {
436 	int     unvisited, i, nnum;
437 	RF_DagNode_t *node;
438 
439 	nnum = 0;
440 	unvisited = dag_h->succedents[0]->visited;
441 
442 	dag_h->nodeNum = nnum++;
443 	for (i = 0; i < dag_h->numSuccedents; i++) {
444 		node = dag_h->succedents[i];
445 		if (node->visited == unvisited) {
446 			nnum = rf_RecurAssignNodeNums(dag_h->succedents[i], nnum, unvisited);
447 		}
448 	}
449 	return (nnum);
450 }
451 
452 int
453 rf_RecurAssignNodeNums(node, num, unvisited)
454 	RF_DagNode_t *node;
455 	int     num;
456 	int     unvisited;
457 {
458 	int     i;
459 
460 	node->visited = (unvisited) ? 0 : 1;
461 
462 	node->nodeNum = num++;
463 	for (i = 0; i < node->numSuccedents; i++) {
464 		if (node->succedents[i]->visited == unvisited) {
465 			num = rf_RecurAssignNodeNums(node->succedents[i], num, unvisited);
466 		}
467 	}
468 	return (num);
469 }
470 /* set the header pointers in each node to "newptr" */
471 void
472 rf_ResetDAGHeaderPointers(dag_h, newptr)
473 	RF_DagHeader_t *dag_h;
474 	RF_DagHeader_t *newptr;
475 {
476 	int     i;
477 	for (i = 0; i < dag_h->numSuccedents; i++)
478 		if (dag_h->succedents[i]->dagHdr != newptr)
479 			rf_RecurResetDAGHeaderPointers(dag_h->succedents[i], newptr);
480 }
481 
482 void
483 rf_RecurResetDAGHeaderPointers(node, newptr)
484 	RF_DagNode_t *node;
485 	RF_DagHeader_t *newptr;
486 {
487 	int     i;
488 	node->dagHdr = newptr;
489 	for (i = 0; i < node->numSuccedents; i++)
490 		if (node->succedents[i]->dagHdr != newptr)
491 			rf_RecurResetDAGHeaderPointers(node->succedents[i], newptr);
492 }
493 
494 
495 void
496 rf_PrintDAGList(RF_DagHeader_t * dag_h)
497 {
498 	int     i = 0;
499 
500 	for (; dag_h; dag_h = dag_h->next) {
501 		rf_AssignNodeNums(dag_h);
502 		printf("\n\nDAG %d IN LIST:\n", i++);
503 		rf_PrintDAG(dag_h);
504 	}
505 }
506 
507 static int
508 rf_ValidateBranch(node, scount, acount, nodes, unvisited)
509 	RF_DagNode_t *node;
510 	int    *scount;
511 	int    *acount;
512 	RF_DagNode_t **nodes;
513 	int     unvisited;
514 {
515 	int     i, retcode = 0;
516 
517 	/* construct an array of node pointers indexed by node num */
518 	node->visited = (unvisited) ? 0 : 1;
519 	nodes[node->nodeNum] = node;
520 
521 	if (node->next != NULL) {
522 		printf("INVALID DAG: next pointer in node is not NULL\n");
523 		retcode = 1;
524 	}
525 	if (node->status != rf_wait) {
526 		printf("INVALID DAG: Node status is not wait\n");
527 		retcode = 1;
528 	}
529 	if (node->numAntDone != 0) {
530 		printf("INVALID DAG: numAntDone is not zero\n");
531 		retcode = 1;
532 	}
533 	if (node->doFunc == rf_TerminateFunc) {
534 		if (node->numSuccedents != 0) {
535 			printf("INVALID DAG: Terminator node has succedents\n");
536 			retcode = 1;
537 		}
538 	} else {
539 		if (node->numSuccedents == 0) {
540 			printf("INVALID DAG: Non-terminator node has no succedents\n");
541 			retcode = 1;
542 		}
543 	}
544 	for (i = 0; i < node->numSuccedents; i++) {
545 		if (!node->succedents[i]) {
546 			printf("INVALID DAG: succedent %d of node %s is NULL\n", i, node->name);
547 			retcode = 1;
548 		}
549 		scount[node->succedents[i]->nodeNum]++;
550 	}
551 	for (i = 0; i < node->numAntecedents; i++) {
552 		if (!node->antecedents[i]) {
553 			printf("INVALID DAG: antecedent %d of node %s is NULL\n", i, node->name);
554 			retcode = 1;
555 		}
556 		acount[node->antecedents[i]->nodeNum]++;
557 	}
558 	for (i = 0; i < node->numSuccedents; i++) {
559 		if (node->succedents[i]->visited == unvisited) {
560 			if (rf_ValidateBranch(node->succedents[i], scount,
561 				acount, nodes, unvisited)) {
562 				retcode = 1;
563 			}
564 		}
565 	}
566 	return (retcode);
567 }
568 
569 static void
570 rf_ValidateBranchVisitedBits(node, unvisited, rl)
571 	RF_DagNode_t *node;
572 	int     unvisited;
573 	int     rl;
574 {
575 	int     i;
576 
577 	RF_ASSERT(node->visited == unvisited);
578 	for (i = 0; i < node->numSuccedents; i++) {
579 		if (node->succedents[i] == NULL) {
580 			printf("node=%lx node->succedents[%d] is NULL\n", (long) node, i);
581 			RF_ASSERT(0);
582 		}
583 		rf_ValidateBranchVisitedBits(node->succedents[i], unvisited, rl + 1);
584 	}
585 }
586 /* NOTE:  never call this on a big dag, because it is exponential
587  * in execution time
588  */
589 static void
590 rf_ValidateVisitedBits(dag)
591 	RF_DagHeader_t *dag;
592 {
593 	int     i, unvisited;
594 
595 	unvisited = dag->succedents[0]->visited;
596 
597 	for (i = 0; i < dag->numSuccedents; i++) {
598 		if (dag->succedents[i] == NULL) {
599 			printf("dag=%lx dag->succedents[%d] is NULL\n", (long) dag, i);
600 			RF_ASSERT(0);
601 		}
602 		rf_ValidateBranchVisitedBits(dag->succedents[i], unvisited, 0);
603 	}
604 }
605 /* validate a DAG.  _at entry_ verify that:
606  *   -- numNodesCompleted is zero
607  *   -- node queue is null
608  *   -- dag status is rf_enable
609  *   -- next pointer is null on every node
610  *   -- all nodes have status wait
611  *   -- numAntDone is zero in all nodes
612  *   -- terminator node has zero successors
613  *   -- no other node besides terminator has zero successors
614  *   -- no successor or antecedent pointer in a node is NULL
615  *   -- number of times that each node appears as a successor of another node
616  *      is equal to the antecedent count on that node
617  *   -- number of times that each node appears as an antecedent of another node
618  *      is equal to the succedent count on that node
619  *   -- what else?
620  */
621 int
622 rf_ValidateDAG(dag_h)
623 	RF_DagHeader_t *dag_h;
624 {
625 	int     i, nodecount;
626 	int    *scount, *acount;/* per-node successor and antecedent counts */
627 	RF_DagNode_t **nodes;	/* array of ptrs to nodes in dag */
628 	int     retcode = 0;
629 	int     unvisited;
630 	int     commitNodeCount = 0;
631 
632 	if (rf_validateVisitedDebug)
633 		rf_ValidateVisitedBits(dag_h);
634 
635 	if (dag_h->numNodesCompleted != 0) {
636 		printf("INVALID DAG: num nodes completed is %d, should be 0\n", dag_h->numNodesCompleted);
637 		retcode = 1;
638 		goto validate_dag_bad;
639 	}
640 	if (dag_h->status != rf_enable) {
641 		printf("INVALID DAG: not enabled\n");
642 		retcode = 1;
643 		goto validate_dag_bad;
644 	}
645 	if (dag_h->numCommits != 0) {
646 		printf("INVALID DAG: numCommits != 0 (%d)\n", dag_h->numCommits);
647 		retcode = 1;
648 		goto validate_dag_bad;
649 	}
650 	if (dag_h->numSuccedents != 1) {
651 		/* currently, all dags must have only one succedent */
652 		printf("INVALID DAG: numSuccedents !1 (%d)\n", dag_h->numSuccedents);
653 		retcode = 1;
654 		goto validate_dag_bad;
655 	}
656 	nodecount = rf_AssignNodeNums(dag_h);
657 
658 	unvisited = dag_h->succedents[0]->visited;
659 
660 	RF_Calloc(scount, nodecount, sizeof(int), (int *));
661 	RF_Calloc(acount, nodecount, sizeof(int), (int *));
662 	RF_Calloc(nodes, nodecount, sizeof(RF_DagNode_t *), (RF_DagNode_t **));
663 	for (i = 0; i < dag_h->numSuccedents; i++) {
664 		if ((dag_h->succedents[i]->visited == unvisited)
665 		    && rf_ValidateBranch(dag_h->succedents[i], scount,
666 			acount, nodes, unvisited)) {
667 			retcode = 1;
668 		}
669 	}
670 	/* start at 1 to skip the header node */
671 	for (i = 1; i < nodecount; i++) {
672 		if (nodes[i]->commitNode)
673 			commitNodeCount++;
674 		if (nodes[i]->doFunc == NULL) {
675 			printf("INVALID DAG: node %s has an undefined doFunc\n", nodes[i]->name);
676 			retcode = 1;
677 			goto validate_dag_out;
678 		}
679 		if (nodes[i]->undoFunc == NULL) {
680 			printf("INVALID DAG: node %s has an undefined doFunc\n", nodes[i]->name);
681 			retcode = 1;
682 			goto validate_dag_out;
683 		}
684 		if (nodes[i]->numAntecedents != scount[nodes[i]->nodeNum]) {
685 			printf("INVALID DAG: node %s has %d antecedents but appears as a succedent %d times\n",
686 			    nodes[i]->name, nodes[i]->numAntecedents, scount[nodes[i]->nodeNum]);
687 			retcode = 1;
688 			goto validate_dag_out;
689 		}
690 		if (nodes[i]->numSuccedents != acount[nodes[i]->nodeNum]) {
691 			printf("INVALID DAG: node %s has %d succedents but appears as an antecedent %d times\n",
692 			    nodes[i]->name, nodes[i]->numSuccedents, acount[nodes[i]->nodeNum]);
693 			retcode = 1;
694 			goto validate_dag_out;
695 		}
696 	}
697 
698 	if (dag_h->numCommitNodes != commitNodeCount) {
699 		printf("INVALID DAG: incorrect commit node count.  hdr->numCommitNodes (%d) found (%d) commit nodes in graph\n",
700 		    dag_h->numCommitNodes, commitNodeCount);
701 		retcode = 1;
702 		goto validate_dag_out;
703 	}
704 validate_dag_out:
705 	RF_Free(scount, nodecount * sizeof(int));
706 	RF_Free(acount, nodecount * sizeof(int));
707 	RF_Free(nodes, nodecount * sizeof(RF_DagNode_t *));
708 	if (retcode)
709 		rf_PrintDAGList(dag_h);
710 
711 	if (rf_validateVisitedDebug)
712 		rf_ValidateVisitedBits(dag_h);
713 
714 	return (retcode);
715 
716 validate_dag_bad:
717 	rf_PrintDAGList(dag_h);
718 	return (retcode);
719 }
720 
721 #endif /* RF_DEBUG_VALIDATE_DAG */
722 
723 /******************************************************************************
724  *
725  * misc construction routines
726  *
727  *****************************************************************************/
728 
729 void
730 rf_redirect_asm(
731     RF_Raid_t * raidPtr,
732     RF_AccessStripeMap_t * asmap)
733 {
734 	int     ds = (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) ? 1 : 0;
735 	int     row = asmap->physInfo->row;
736 	int     fcol = raidPtr->reconControl[row]->fcol;
737 	int     srow = raidPtr->reconControl[row]->spareRow;
738 	int     scol = raidPtr->reconControl[row]->spareCol;
739 	RF_PhysDiskAddr_t *pda;
740 
741 	RF_ASSERT(raidPtr->status[row] == rf_rs_reconstructing);
742 	for (pda = asmap->physInfo; pda; pda = pda->next) {
743 		if (pda->col == fcol) {
744 			if (rf_dagDebug) {
745 				if (!rf_CheckRUReconstructed(raidPtr->reconControl[row]->reconMap,
746 					pda->startSector)) {
747 					RF_PANIC();
748 				}
749 			}
750 			/* printf("Remapped data for large write\n"); */
751 			if (ds) {
752 				raidPtr->Layout.map->MapSector(raidPtr, pda->raidAddress,
753 				    &pda->row, &pda->col, &pda->startSector, RF_REMAP);
754 			} else {
755 				pda->row = srow;
756 				pda->col = scol;
757 			}
758 		}
759 	}
760 	for (pda = asmap->parityInfo; pda; pda = pda->next) {
761 		if (pda->col == fcol) {
762 			if (rf_dagDebug) {
763 				if (!rf_CheckRUReconstructed(raidPtr->reconControl[row]->reconMap, pda->startSector)) {
764 					RF_PANIC();
765 				}
766 			}
767 		}
768 		if (ds) {
769 			(raidPtr->Layout.map->MapParity) (raidPtr, pda->raidAddress, &pda->row, &pda->col, &pda->startSector, RF_REMAP);
770 		} else {
771 			pda->row = srow;
772 			pda->col = scol;
773 		}
774 	}
775 }
776 
777 
778 /* this routine allocates read buffers and generates stripe maps for the
779  * regions of the array from the start of the stripe to the start of the
780  * access, and from the end of the access to the end of the stripe.  It also
781  * computes and returns the number of DAG nodes needed to read all this data.
782  * Note that this routine does the wrong thing if the access is fully
783  * contained within one stripe unit, so we RF_ASSERT against this case at the
784  * start.
785  */
786 void
787 rf_MapUnaccessedPortionOfStripe(
788     RF_Raid_t * raidPtr,
789     RF_RaidLayout_t * layoutPtr,/* in: layout information */
790     RF_AccessStripeMap_t * asmap,	/* in: access stripe map */
791     RF_DagHeader_t * dag_h,	/* in: header of the dag to create */
792     RF_AccessStripeMapHeader_t ** new_asm_h,	/* in: ptr to array of 2
793 						 * headers, to be filled in */
794     int *nRodNodes,		/* out: num nodes to be generated to read
795 				 * unaccessed data */
796     char **sosBuffer,		/* out: pointers to newly allocated buffer */
797     char **eosBuffer,
798     RF_AllocListElem_t * allocList)
799 {
800 	RF_RaidAddr_t sosRaidAddress, eosRaidAddress;
801 	RF_SectorNum_t sosNumSector, eosNumSector;
802 
803 	RF_ASSERT(asmap->numStripeUnitsAccessed > (layoutPtr->numDataCol / 2));
804 	/* generate an access map for the region of the array from start of
805 	 * stripe to start of access */
806 	new_asm_h[0] = new_asm_h[1] = NULL;
807 	*nRodNodes = 0;
808 	if (!rf_RaidAddressStripeAligned(layoutPtr, asmap->raidAddress)) {
809 		sosRaidAddress = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, asmap->raidAddress);
810 		sosNumSector = asmap->raidAddress - sosRaidAddress;
811 		RF_MallocAndAdd(*sosBuffer, rf_RaidAddressToByte(raidPtr, sosNumSector), (char *), allocList);
812 		new_asm_h[0] = rf_MapAccess(raidPtr, sosRaidAddress, sosNumSector, *sosBuffer, RF_DONT_REMAP);
813 		new_asm_h[0]->next = dag_h->asmList;
814 		dag_h->asmList = new_asm_h[0];
815 		*nRodNodes += new_asm_h[0]->stripeMap->numStripeUnitsAccessed;
816 
817 		RF_ASSERT(new_asm_h[0]->stripeMap->next == NULL);
818 		/* we're totally within one stripe here */
819 		if (asmap->flags & RF_ASM_REDIR_LARGE_WRITE)
820 			rf_redirect_asm(raidPtr, new_asm_h[0]->stripeMap);
821 	}
822 	/* generate an access map for the region of the array from end of
823 	 * access to end of stripe */
824 	if (!rf_RaidAddressStripeAligned(layoutPtr, asmap->endRaidAddress)) {
825 		eosRaidAddress = asmap->endRaidAddress;
826 		eosNumSector = rf_RaidAddressOfNextStripeBoundary(layoutPtr, eosRaidAddress) - eosRaidAddress;
827 		RF_MallocAndAdd(*eosBuffer, rf_RaidAddressToByte(raidPtr, eosNumSector), (char *), allocList);
828 		new_asm_h[1] = rf_MapAccess(raidPtr, eosRaidAddress, eosNumSector, *eosBuffer, RF_DONT_REMAP);
829 		new_asm_h[1]->next = dag_h->asmList;
830 		dag_h->asmList = new_asm_h[1];
831 		*nRodNodes += new_asm_h[1]->stripeMap->numStripeUnitsAccessed;
832 
833 		RF_ASSERT(new_asm_h[1]->stripeMap->next == NULL);
834 		/* we're totally within one stripe here */
835 		if (asmap->flags & RF_ASM_REDIR_LARGE_WRITE)
836 			rf_redirect_asm(raidPtr, new_asm_h[1]->stripeMap);
837 	}
838 }
839 
840 
841 
842 /* returns non-zero if the indicated ranges of stripe unit offsets overlap */
843 int
844 rf_PDAOverlap(
845     RF_RaidLayout_t * layoutPtr,
846     RF_PhysDiskAddr_t * src,
847     RF_PhysDiskAddr_t * dest)
848 {
849 	RF_SectorNum_t soffs = rf_StripeUnitOffset(layoutPtr, src->startSector);
850 	RF_SectorNum_t doffs = rf_StripeUnitOffset(layoutPtr, dest->startSector);
851 	/* use -1 to be sure we stay within SU */
852 	RF_SectorNum_t send = rf_StripeUnitOffset(layoutPtr, src->startSector + src->numSector - 1);
853 	RF_SectorNum_t dend = rf_StripeUnitOffset(layoutPtr, dest->startSector + dest->numSector - 1);
854 	return ((RF_MAX(soffs, doffs) <= RF_MIN(send, dend)) ? 1 : 0);
855 }
856 
857 
858 /* GenerateFailedAccessASMs
859  *
860  * this routine figures out what portion of the stripe needs to be read
861  * to effect the degraded read or write operation.  It's primary function
862  * is to identify everything required to recover the data, and then
863  * eliminate anything that is already being accessed by the user.
864  *
865  * The main result is two new ASMs, one for the region from the start of the
866  * stripe to the start of the access, and one for the region from the end of
867  * the access to the end of the stripe.  These ASMs describe everything that
868  * needs to be read to effect the degraded access.  Other results are:
869  *    nXorBufs -- the total number of buffers that need to be XORed together to
870  *                recover the lost data,
871  *    rpBufPtr -- ptr to a newly-allocated buffer to hold the parity.  If NULL
872  *                at entry, not allocated.
873  *    overlappingPDAs --
874  *                describes which of the non-failed PDAs in the user access
875  *                overlap data that needs to be read to effect recovery.
876  *                overlappingPDAs[i]==1 if and only if, neglecting the failed
877  *                PDA, the ith pda in the input asm overlaps data that needs
878  *                to be read for recovery.
879  */
880  /* in: asm - ASM for the actual access, one stripe only */
881  /* in: failedPDA - which component of the access has failed */
882  /* in: dag_h - header of the DAG we're going to create */
883  /* out: new_asm_h - the two new ASMs */
884  /* out: nXorBufs - the total number of xor bufs required */
885  /* out: rpBufPtr - a buffer for the parity read */
886 void
887 rf_GenerateFailedAccessASMs(
888     RF_Raid_t * raidPtr,
889     RF_AccessStripeMap_t * asmap,
890     RF_PhysDiskAddr_t * failedPDA,
891     RF_DagHeader_t * dag_h,
892     RF_AccessStripeMapHeader_t ** new_asm_h,
893     int *nXorBufs,
894     char **rpBufPtr,
895     char *overlappingPDAs,
896     RF_AllocListElem_t * allocList)
897 {
898 	RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
899 
900 	/* s=start, e=end, s=stripe, a=access, f=failed, su=stripe unit */
901 	RF_RaidAddr_t sosAddr, sosEndAddr, eosStartAddr, eosAddr;
902 
903 	RF_SectorCount_t numSect[2], numParitySect;
904 	RF_PhysDiskAddr_t *pda;
905 	char   *rdBuf, *bufP;
906 	int     foundit, i;
907 
908 	bufP = NULL;
909 	foundit = 0;
910 	/* first compute the following raid addresses: start of stripe,
911 	 * (sosAddr) MIN(start of access, start of failed SU),   (sosEndAddr)
912 	 * MAX(end of access, end of failed SU),       (eosStartAddr) end of
913 	 * stripe (i.e. start of next stripe)   (eosAddr) */
914 	sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, asmap->raidAddress);
915 	sosEndAddr = RF_MIN(asmap->raidAddress, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, failedPDA->raidAddress));
916 	eosStartAddr = RF_MAX(asmap->endRaidAddress, rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, failedPDA->raidAddress));
917 	eosAddr = rf_RaidAddressOfNextStripeBoundary(layoutPtr, asmap->raidAddress);
918 
919 	/* now generate access stripe maps for each of the above regions of
920 	 * the stripe.  Use a dummy (NULL) buf ptr for now */
921 
922 	new_asm_h[0] = (sosAddr != sosEndAddr) ? rf_MapAccess(raidPtr, sosAddr, sosEndAddr - sosAddr, NULL, RF_DONT_REMAP) : NULL;
923 	new_asm_h[1] = (eosStartAddr != eosAddr) ? rf_MapAccess(raidPtr, eosStartAddr, eosAddr - eosStartAddr, NULL, RF_DONT_REMAP) : NULL;
924 
925 	/* walk through the PDAs and range-restrict each SU to the region of
926 	 * the SU touched on the failed PDA.  also compute total data buffer
927 	 * space requirements in this step.  Ignore the parity for now. */
928 
929 	numSect[0] = numSect[1] = 0;
930 	if (new_asm_h[0]) {
931 		new_asm_h[0]->next = dag_h->asmList;
932 		dag_h->asmList = new_asm_h[0];
933 		for (pda = new_asm_h[0]->stripeMap->physInfo; pda; pda = pda->next) {
934 			rf_RangeRestrictPDA(raidPtr, failedPDA, pda, RF_RESTRICT_NOBUFFER, 0);
935 			numSect[0] += pda->numSector;
936 		}
937 	}
938 	if (new_asm_h[1]) {
939 		new_asm_h[1]->next = dag_h->asmList;
940 		dag_h->asmList = new_asm_h[1];
941 		for (pda = new_asm_h[1]->stripeMap->physInfo; pda; pda = pda->next) {
942 			rf_RangeRestrictPDA(raidPtr, failedPDA, pda, RF_RESTRICT_NOBUFFER, 0);
943 			numSect[1] += pda->numSector;
944 		}
945 	}
946 	numParitySect = failedPDA->numSector;
947 
948 	/* allocate buffer space for the data & parity we have to read to
949 	 * recover from the failure */
950 
951 	if (numSect[0] + numSect[1] + ((rpBufPtr) ? numParitySect : 0)) {	/* don't allocate parity
952 										 * buf if not needed */
953 		RF_MallocAndAdd(rdBuf, rf_RaidAddressToByte(raidPtr, numSect[0] + numSect[1] + numParitySect), (char *), allocList);
954 		bufP = rdBuf;
955 		if (rf_degDagDebug)
956 			printf("Newly allocated buffer (%d bytes) is 0x%lx\n",
957 			    (int) rf_RaidAddressToByte(raidPtr, numSect[0] + numSect[1] + numParitySect), (unsigned long) bufP);
958 	}
959 	/* now walk through the pdas one last time and assign buffer pointers
960 	 * (ugh!).  Again, ignore the parity.  also, count nodes to find out
961 	 * how many bufs need to be xored together */
962 	(*nXorBufs) = 1;	/* in read case, 1 is for parity.  In write
963 				 * case, 1 is for failed data */
964 	if (new_asm_h[0]) {
965 		for (pda = new_asm_h[0]->stripeMap->physInfo; pda; pda = pda->next) {
966 			pda->bufPtr = bufP;
967 			bufP += rf_RaidAddressToByte(raidPtr, pda->numSector);
968 		}
969 		*nXorBufs += new_asm_h[0]->stripeMap->numStripeUnitsAccessed;
970 	}
971 	if (new_asm_h[1]) {
972 		for (pda = new_asm_h[1]->stripeMap->physInfo; pda; pda = pda->next) {
973 			pda->bufPtr = bufP;
974 			bufP += rf_RaidAddressToByte(raidPtr, pda->numSector);
975 		}
976 		(*nXorBufs) += new_asm_h[1]->stripeMap->numStripeUnitsAccessed;
977 	}
978 	if (rpBufPtr)
979 		*rpBufPtr = bufP;	/* the rest of the buffer is for
980 					 * parity */
981 
982 	/* the last step is to figure out how many more distinct buffers need
983 	 * to get xor'd to produce the missing unit.  there's one for each
984 	 * user-data read node that overlaps the portion of the failed unit
985 	 * being accessed */
986 
987 	for (foundit = i = 0, pda = asmap->physInfo; pda; i++, pda = pda->next) {
988 		if (pda == failedPDA) {
989 			i--;
990 			foundit = 1;
991 			continue;
992 		}
993 		if (rf_PDAOverlap(layoutPtr, pda, failedPDA)) {
994 			overlappingPDAs[i] = 1;
995 			(*nXorBufs)++;
996 		}
997 	}
998 	if (!foundit) {
999 		RF_ERRORMSG("GenerateFailedAccessASMs: did not find failedPDA in asm list\n");
1000 		RF_ASSERT(0);
1001 	}
1002 	if (rf_degDagDebug) {
1003 		if (new_asm_h[0]) {
1004 			printf("First asm:\n");
1005 			rf_PrintFullAccessStripeMap(new_asm_h[0], 1);
1006 		}
1007 		if (new_asm_h[1]) {
1008 			printf("Second asm:\n");
1009 			rf_PrintFullAccessStripeMap(new_asm_h[1], 1);
1010 		}
1011 	}
1012 }
1013 
1014 
1015 /* adjusts the offset and number of sectors in the destination pda so that
1016  * it covers at most the region of the SU covered by the source PDA.  This
1017  * is exclusively a restriction:  the number of sectors indicated by the
1018  * target PDA can only shrink.
1019  *
1020  * For example:  s = sectors within SU indicated by source PDA
1021  *               d = sectors within SU indicated by dest PDA
1022  *               r = results, stored in dest PDA
1023  *
1024  * |--------------- one stripe unit ---------------------|
1025  * |           sssssssssssssssssssssssssssssssss         |
1026  * |    ddddddddddddddddddddddddddddddddddddddddddddd    |
1027  * |           rrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrr         |
1028  *
1029  * Another example:
1030  *
1031  * |--------------- one stripe unit ---------------------|
1032  * |           sssssssssssssssssssssssssssssssss         |
1033  * |    ddddddddddddddddddddddd                          |
1034  * |           rrrrrrrrrrrrrrrr                          |
1035  *
1036  */
1037 void
1038 rf_RangeRestrictPDA(
1039     RF_Raid_t * raidPtr,
1040     RF_PhysDiskAddr_t * src,
1041     RF_PhysDiskAddr_t * dest,
1042     int dobuffer,
1043     int doraidaddr)
1044 {
1045 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
1046 	RF_SectorNum_t soffs = rf_StripeUnitOffset(layoutPtr, src->startSector);
1047 	RF_SectorNum_t doffs = rf_StripeUnitOffset(layoutPtr, dest->startSector);
1048 	RF_SectorNum_t send = rf_StripeUnitOffset(layoutPtr, src->startSector + src->numSector - 1);	/* use -1 to be sure we
1049 													 * stay within SU */
1050 	RF_SectorNum_t dend = rf_StripeUnitOffset(layoutPtr, dest->startSector + dest->numSector - 1);
1051 	RF_SectorNum_t subAddr = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, dest->startSector);	/* stripe unit boundary */
1052 
1053 	dest->startSector = subAddr + RF_MAX(soffs, doffs);
1054 	dest->numSector = subAddr + RF_MIN(send, dend) + 1 - dest->startSector;
1055 
1056 	if (dobuffer)
1057 		dest->bufPtr += (soffs > doffs) ? rf_RaidAddressToByte(raidPtr, soffs - doffs) : 0;
1058 	if (doraidaddr) {
1059 		dest->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, dest->raidAddress) +
1060 		    rf_StripeUnitOffset(layoutPtr, dest->startSector);
1061 	}
1062 }
1063 
1064 #if (RF_INCLUDE_CHAINDECLUSTER > 0)
1065 
1066 /*
1067  * Want the highest of these primes to be the largest one
1068  * less than the max expected number of columns (won't hurt
1069  * to be too small or too large, but won't be optimal, either)
1070  * --jimz
1071  */
1072 #define NLOWPRIMES 8
1073 static int lowprimes[NLOWPRIMES] = {2, 3, 5, 7, 11, 13, 17, 19};
1074 /*****************************************************************************
1075  * compute the workload shift factor.  (chained declustering)
1076  *
1077  * return nonzero if access should shift to secondary, otherwise,
1078  * access is to primary
1079  *****************************************************************************/
1080 int
1081 rf_compute_workload_shift(
1082     RF_Raid_t * raidPtr,
1083     RF_PhysDiskAddr_t * pda)
1084 {
1085 	/*
1086          * variables:
1087          *  d   = column of disk containing primary
1088          *  f   = column of failed disk
1089          *  n   = number of disks in array
1090          *  sd  = "shift distance" (number of columns that d is to the right of f)
1091          *  row = row of array the access is in
1092          *  v   = numerator of redirection ratio
1093          *  k   = denominator of redirection ratio
1094          */
1095 	RF_RowCol_t d, f, sd, row, n;
1096 	int     k, v, ret, i;
1097 
1098 	row = pda->row;
1099 	n = raidPtr->numCol;
1100 
1101 	/* assign column of primary copy to d */
1102 	d = pda->col;
1103 
1104 	/* assign column of dead disk to f */
1105 	for (f = 0; ((!RF_DEAD_DISK(raidPtr->Disks[row][f].status)) && (f < n)); f++);
1106 
1107 	RF_ASSERT(f < n);
1108 	RF_ASSERT(f != d);
1109 
1110 	sd = (f > d) ? (n + d - f) : (d - f);
1111 	RF_ASSERT(sd < n);
1112 
1113 	/*
1114          * v of every k accesses should be redirected
1115          *
1116          * v/k := (n-1-sd)/(n-1)
1117          */
1118 	v = (n - 1 - sd);
1119 	k = (n - 1);
1120 
1121 #if 1
1122 	/*
1123          * XXX
1124          * Is this worth it?
1125          *
1126          * Now reduce the fraction, by repeatedly factoring
1127          * out primes (just like they teach in elementary school!)
1128          */
1129 	for (i = 0; i < NLOWPRIMES; i++) {
1130 		if (lowprimes[i] > v)
1131 			break;
1132 		while (((v % lowprimes[i]) == 0) && ((k % lowprimes[i]) == 0)) {
1133 			v /= lowprimes[i];
1134 			k /= lowprimes[i];
1135 		}
1136 	}
1137 #endif
1138 
1139 	raidPtr->hist_diskreq[row][d]++;
1140 	if (raidPtr->hist_diskreq[row][d] > v) {
1141 		ret = 0;	/* do not redirect */
1142 	} else {
1143 		ret = 1;	/* redirect */
1144 	}
1145 
1146 #if 0
1147 	printf("d=%d f=%d sd=%d v=%d k=%d ret=%d h=%d\n", d, f, sd, v, k, ret,
1148 	    raidPtr->hist_diskreq[row][d]);
1149 #endif
1150 
1151 	if (raidPtr->hist_diskreq[row][d] >= k) {
1152 		/* reset counter */
1153 		raidPtr->hist_diskreq[row][d] = 0;
1154 	}
1155 	return (ret);
1156 }
1157 #endif /* (RF_INCLUDE_CHAINDECLUSTER > 0) */
1158 
1159 /*
1160  * Disk selection routines
1161  */
1162 
1163 /*
1164  * Selects the disk with the shortest queue from a mirror pair.
1165  * Both the disk I/Os queued in RAIDframe as well as those at the physical
1166  * disk are counted as members of the "queue"
1167  */
1168 void
1169 rf_SelectMirrorDiskIdle(RF_DagNode_t * node)
1170 {
1171 	RF_Raid_t *raidPtr = (RF_Raid_t *) node->dagHdr->raidPtr;
1172 	RF_RowCol_t rowData, colData, rowMirror, colMirror;
1173 	int     dataQueueLength, mirrorQueueLength, usemirror;
1174 	RF_PhysDiskAddr_t *data_pda = (RF_PhysDiskAddr_t *) node->params[0].p;
1175 	RF_PhysDiskAddr_t *mirror_pda = (RF_PhysDiskAddr_t *) node->params[4].p;
1176 	RF_PhysDiskAddr_t *tmp_pda;
1177 	RF_RaidDisk_t **disks = raidPtr->Disks;
1178 	RF_DiskQueue_t **dqs = raidPtr->Queues, *dataQueue, *mirrorQueue;
1179 
1180 	/* return the [row col] of the disk with the shortest queue */
1181 	rowData = data_pda->row;
1182 	colData = data_pda->col;
1183 	rowMirror = mirror_pda->row;
1184 	colMirror = mirror_pda->col;
1185 	dataQueue = &(dqs[rowData][colData]);
1186 	mirrorQueue = &(dqs[rowMirror][colMirror]);
1187 
1188 #ifdef RF_LOCK_QUEUES_TO_READ_LEN
1189 	RF_LOCK_QUEUE_MUTEX(dataQueue, "SelectMirrorDiskIdle");
1190 #endif				/* RF_LOCK_QUEUES_TO_READ_LEN */
1191 	dataQueueLength = dataQueue->queueLength + dataQueue->numOutstanding;
1192 #ifdef RF_LOCK_QUEUES_TO_READ_LEN
1193 	RF_UNLOCK_QUEUE_MUTEX(dataQueue, "SelectMirrorDiskIdle");
1194 	RF_LOCK_QUEUE_MUTEX(mirrorQueue, "SelectMirrorDiskIdle");
1195 #endif				/* RF_LOCK_QUEUES_TO_READ_LEN */
1196 	mirrorQueueLength = mirrorQueue->queueLength + mirrorQueue->numOutstanding;
1197 #ifdef RF_LOCK_QUEUES_TO_READ_LEN
1198 	RF_UNLOCK_QUEUE_MUTEX(mirrorQueue, "SelectMirrorDiskIdle");
1199 #endif				/* RF_LOCK_QUEUES_TO_READ_LEN */
1200 
1201 	usemirror = 0;
1202 	if (RF_DEAD_DISK(disks[rowMirror][colMirror].status)) {
1203 		usemirror = 0;
1204 	} else
1205 		if (RF_DEAD_DISK(disks[rowData][colData].status)) {
1206 			usemirror = 1;
1207 		} else
1208 			if (raidPtr->parity_good == RF_RAID_DIRTY) {
1209 				/* Trust only the main disk */
1210 				usemirror = 0;
1211 			} else
1212 				if (dataQueueLength < mirrorQueueLength) {
1213 					usemirror = 0;
1214 				} else
1215 					if (mirrorQueueLength < dataQueueLength) {
1216 						usemirror = 1;
1217 					} else {
1218 						/* queues are equal length. attempt
1219 						 * cleverness. */
1220 						if (SNUM_DIFF(dataQueue->last_deq_sector, data_pda->startSector)
1221 						    <= SNUM_DIFF(mirrorQueue->last_deq_sector, mirror_pda->startSector)) {
1222 							usemirror = 0;
1223 						} else {
1224 							usemirror = 1;
1225 						}
1226 					}
1227 
1228 	if (usemirror) {
1229 		/* use mirror (parity) disk, swap params 0 & 4 */
1230 		tmp_pda = data_pda;
1231 		node->params[0].p = mirror_pda;
1232 		node->params[4].p = tmp_pda;
1233 	} else {
1234 		/* use data disk, leave param 0 unchanged */
1235 	}
1236 	/* printf("dataQueueLength %d, mirrorQueueLength
1237 	 * %d\n",dataQueueLength, mirrorQueueLength); */
1238 }
1239 /*
1240  * Do simple partitioning. This assumes that
1241  * the data and parity disks are laid out identically.
1242  */
1243 void
1244 rf_SelectMirrorDiskPartition(RF_DagNode_t * node)
1245 {
1246 	RF_Raid_t *raidPtr = (RF_Raid_t *) node->dagHdr->raidPtr;
1247 	RF_RowCol_t rowData, colData, rowMirror, colMirror;
1248 	RF_PhysDiskAddr_t *data_pda = (RF_PhysDiskAddr_t *) node->params[0].p;
1249 	RF_PhysDiskAddr_t *mirror_pda = (RF_PhysDiskAddr_t *) node->params[4].p;
1250 	RF_PhysDiskAddr_t *tmp_pda;
1251 	RF_RaidDisk_t **disks = raidPtr->Disks;
1252 	RF_DiskQueue_t **dqs = raidPtr->Queues, *dataQueue, *mirrorQueue;
1253 	int     usemirror;
1254 
1255 	/* return the [row col] of the disk with the shortest queue */
1256 	rowData = data_pda->row;
1257 	colData = data_pda->col;
1258 	rowMirror = mirror_pda->row;
1259 	colMirror = mirror_pda->col;
1260 	dataQueue = &(dqs[rowData][colData]);
1261 	mirrorQueue = &(dqs[rowMirror][colMirror]);
1262 
1263 	usemirror = 0;
1264 	if (RF_DEAD_DISK(disks[rowMirror][colMirror].status)) {
1265 		usemirror = 0;
1266 	} else
1267 		if (RF_DEAD_DISK(disks[rowData][colData].status)) {
1268 			usemirror = 1;
1269 		} else
1270 			if (raidPtr->parity_good == RF_RAID_DIRTY) {
1271 				/* Trust only the main disk */
1272 				usemirror = 0;
1273 			} else
1274 				if (data_pda->startSector <
1275 				    (disks[rowData][colData].numBlocks / 2)) {
1276 					usemirror = 0;
1277 				} else {
1278 					usemirror = 1;
1279 				}
1280 
1281 	if (usemirror) {
1282 		/* use mirror (parity) disk, swap params 0 & 4 */
1283 		tmp_pda = data_pda;
1284 		node->params[0].p = mirror_pda;
1285 		node->params[4].p = tmp_pda;
1286 	} else {
1287 		/* use data disk, leave param 0 unchanged */
1288 	}
1289 }
1290