xref: /freebsd/sys/vm/vm_radix.c (revision 87c1627502a5dde91e5284118eec8682b60f27a2)
1 /*
2  * Copyright (c) 2013 EMC Corp.
3  * Copyright (c) 2011 Jeffrey Roberson <jeff@freebsd.org>
4  * Copyright (c) 2008 Mayur Shardul <mayur.shardul@gmail.com>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  */
29 
30 /*
31  * Path-compressed radix trie implementation.
32  * The following code is not generalized into a general purpose library
33  * because there are way too many parameters embedded that should really
34  * be decided by the library consumers.  At the same time, consumers
35  * of this code must achieve highest possible performance.
36  *
37  * The implementation takes into account the following rationale:
38  * - Size of the nodes should be as small as possible but still big enough
39  *   to avoid a large maximum depth for the trie.  This is a balance
40  *   between the necessity to not wire too much physical memory for the nodes
41  *   and the necessity to avoid too much cache pollution during the trie
42  *   operations.
43  * - There is not a huge bias toward the number of lookup operations over
44  *   the number of insert and remove operations.  This basically implies
45  *   that optimizations supposedly helping one operation but hurting the
46  *   other might be carefully evaluated.
47  * - On average not many nodes are expected to be fully populated, hence
48  *   level compression may just complicate things.
49  */
50 
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
53 
54 #include "opt_ddb.h"
55 
56 #include <sys/param.h>
57 #include <sys/systm.h>
58 #include <sys/kernel.h>
59 #include <sys/vmmeter.h>
60 
61 #include <vm/uma.h>
62 #include <vm/vm.h>
63 #include <vm/vm_param.h>
64 #include <vm/vm_page.h>
65 #include <vm/vm_radix.h>
66 
67 #ifdef DDB
68 #include <ddb/ddb.h>
69 #endif
70 
71 /*
72  * These widths should allow the pointers to a node's children to fit within
73  * a single cache line.  The extra levels from a narrow width should not be
74  * a problem thanks to path compression.
75  */
76 #ifdef __LP64__
77 #define	VM_RADIX_WIDTH	4
78 #else
79 #define	VM_RADIX_WIDTH	3
80 #endif
81 
82 #define	VM_RADIX_COUNT	(1 << VM_RADIX_WIDTH)
83 #define	VM_RADIX_MASK	(VM_RADIX_COUNT - 1)
84 #define	VM_RADIX_LIMIT							\
85 	(howmany((sizeof(vm_pindex_t) * NBBY), VM_RADIX_WIDTH) - 1)
86 
87 /* Flag bits stored in node pointers. */
88 #define	VM_RADIX_ISLEAF	0x1
89 #define	VM_RADIX_FLAGS	0x1
90 #define	VM_RADIX_PAD	VM_RADIX_FLAGS
91 
92 /* Returns one unit associated with specified level. */
93 #define	VM_RADIX_UNITLEVEL(lev)						\
94 	((vm_pindex_t)1 << ((VM_RADIX_LIMIT - (lev)) * VM_RADIX_WIDTH))
95 
96 struct vm_radix_node {
97 	vm_pindex_t	 rn_owner;			/* Owner of record. */
98 	uint16_t	 rn_count;			/* Valid children. */
99 	uint16_t	 rn_clev;			/* Current level. */
100 	void		*rn_child[VM_RADIX_COUNT];	/* Child nodes. */
101 };
102 
103 static uma_zone_t vm_radix_node_zone;
104 
105 /*
106  * Allocate a radix node.  Pre-allocation should ensure that the request
107  * will always be satisfied.
108  */
109 static __inline struct vm_radix_node *
110 vm_radix_node_get(vm_pindex_t owner, uint16_t count, uint16_t clevel)
111 {
112 	struct vm_radix_node *rnode;
113 
114 	rnode = uma_zalloc(vm_radix_node_zone, M_NOWAIT);
115 
116 	/*
117 	 * The required number of nodes should already be pre-allocated
118 	 * by vm_radix_prealloc().  However, UMA can hold a few nodes
119 	 * in per-CPU buckets, which will not be accessible by the
120 	 * current CPU.  Thus, the allocation could return NULL when
121 	 * the pre-allocated pool is close to exhaustion.  Anyway,
122 	 * in practice this should never occur because a new node
123 	 * is not always required for insert.  Thus, the pre-allocated
124 	 * pool should have some extra pages that prevent this from
125 	 * becoming a problem.
126 	 */
127 	if (rnode == NULL)
128 		panic("%s: uma_zalloc() returned NULL for a new node",
129 		    __func__);
130 	rnode->rn_owner = owner;
131 	rnode->rn_count = count;
132 	rnode->rn_clev = clevel;
133 	return (rnode);
134 }
135 
136 /*
137  * Free radix node.
138  */
139 static __inline void
140 vm_radix_node_put(struct vm_radix_node *rnode)
141 {
142 
143 	uma_zfree(vm_radix_node_zone, rnode);
144 }
145 
146 /*
147  * Return the position in the array for a given level.
148  */
149 static __inline int
150 vm_radix_slot(vm_pindex_t index, uint16_t level)
151 {
152 
153 	return ((index >> ((VM_RADIX_LIMIT - level) * VM_RADIX_WIDTH)) &
154 	    VM_RADIX_MASK);
155 }
156 
157 /* Trims the key after the specified level. */
158 static __inline vm_pindex_t
159 vm_radix_trimkey(vm_pindex_t index, uint16_t level)
160 {
161 	vm_pindex_t ret;
162 
163 	ret = index;
164 	if (level < VM_RADIX_LIMIT) {
165 		ret >>= (VM_RADIX_LIMIT - level) * VM_RADIX_WIDTH;
166 		ret <<= (VM_RADIX_LIMIT - level) * VM_RADIX_WIDTH;
167 	}
168 	return (ret);
169 }
170 
171 /*
172  * Get the root node for a radix tree.
173  */
174 static __inline struct vm_radix_node *
175 vm_radix_getroot(struct vm_radix *rtree)
176 {
177 
178 	return ((struct vm_radix_node *)rtree->rt_root);
179 }
180 
181 /*
182  * Set the root node for a radix tree.
183  */
184 static __inline void
185 vm_radix_setroot(struct vm_radix *rtree, struct vm_radix_node *rnode)
186 {
187 
188 	rtree->rt_root = (uintptr_t)rnode;
189 }
190 
191 /*
192  * Returns TRUE if the specified radix node is a leaf and FALSE otherwise.
193  */
194 static __inline boolean_t
195 vm_radix_isleaf(struct vm_radix_node *rnode)
196 {
197 
198 	return (((uintptr_t)rnode & VM_RADIX_ISLEAF) != 0);
199 }
200 
201 /*
202  * Returns the associated page extracted from rnode.
203  */
204 static __inline vm_page_t
205 vm_radix_topage(struct vm_radix_node *rnode)
206 {
207 
208 	return ((vm_page_t)((uintptr_t)rnode & ~VM_RADIX_FLAGS));
209 }
210 
211 /*
212  * Adds the page as a child of the provided node.
213  */
214 static __inline void
215 vm_radix_addpage(struct vm_radix_node *rnode, vm_pindex_t index, uint16_t clev,
216     vm_page_t page)
217 {
218 	int slot;
219 
220 	slot = vm_radix_slot(index, clev);
221 	rnode->rn_child[slot] = (void *)((uintptr_t)page | VM_RADIX_ISLEAF);
222 }
223 
224 /*
225  * Returns the slot where two keys differ.
226  * It cannot accept 2 equal keys.
227  */
228 static __inline uint16_t
229 vm_radix_keydiff(vm_pindex_t index1, vm_pindex_t index2)
230 {
231 	uint16_t clev;
232 
233 	KASSERT(index1 != index2, ("%s: passing the same key value %jx",
234 	    __func__, (uintmax_t)index1));
235 
236 	index1 ^= index2;
237 	for (clev = 0; clev <= VM_RADIX_LIMIT ; clev++)
238 		if (vm_radix_slot(index1, clev))
239 			return (clev);
240 	panic("%s: cannot reach this point", __func__);
241 	return (0);
242 }
243 
244 /*
245  * Returns TRUE if it can be determined that key does not belong to the
246  * specified rnode.  Otherwise, returns FALSE.
247  */
248 static __inline boolean_t
249 vm_radix_keybarr(struct vm_radix_node *rnode, vm_pindex_t idx)
250 {
251 
252 	if (rnode->rn_clev > 0) {
253 		idx = vm_radix_trimkey(idx, rnode->rn_clev - 1);
254 		return (idx != rnode->rn_owner);
255 	}
256 	return (FALSE);
257 }
258 
259 /*
260  * Internal helper for vm_radix_reclaim_allnodes().
261  * This function is recursive.
262  */
263 static void
264 vm_radix_reclaim_allnodes_int(struct vm_radix_node *rnode)
265 {
266 	int slot;
267 
268 	KASSERT(rnode->rn_count <= VM_RADIX_COUNT,
269 	    ("vm_radix_reclaim_allnodes_int: bad count in rnode %p", rnode));
270 	for (slot = 0; rnode->rn_count != 0; slot++) {
271 		if (rnode->rn_child[slot] == NULL)
272 			continue;
273 		if (!vm_radix_isleaf(rnode->rn_child[slot]))
274 			vm_radix_reclaim_allnodes_int(rnode->rn_child[slot]);
275 		rnode->rn_child[slot] = NULL;
276 		rnode->rn_count--;
277 	}
278 	vm_radix_node_put(rnode);
279 }
280 
281 #ifdef INVARIANTS
282 /*
283  * Radix node zone destructor.
284  */
285 static void
286 vm_radix_node_zone_dtor(void *mem, int size __unused, void *arg __unused)
287 {
288 	struct vm_radix_node *rnode;
289 	int slot;
290 
291 	rnode = mem;
292 	KASSERT(rnode->rn_count == 0,
293 	    ("vm_radix_node_put: rnode %p has %d children", rnode,
294 	    rnode->rn_count));
295 	for (slot = 0; slot < VM_RADIX_COUNT; slot++)
296 		KASSERT(rnode->rn_child[slot] == NULL,
297 		    ("vm_radix_node_put: rnode %p has a child", rnode));
298 }
299 #endif
300 
301 /*
302  * Radix node zone initializer.
303  */
304 static int
305 vm_radix_node_zone_init(void *mem, int size __unused, int flags __unused)
306 {
307 	struct vm_radix_node *rnode;
308 
309 	rnode = mem;
310 	memset(rnode->rn_child, 0, sizeof(rnode->rn_child));
311 	return (0);
312 }
313 
314 /*
315  * Pre-allocate intermediate nodes from the UMA slab zone.
316  */
317 static void
318 vm_radix_prealloc(void *arg __unused)
319 {
320 	int nodes;
321 
322 	/*
323 	 * Calculate the number of reserved nodes, discounting the pages that
324 	 * are needed to store them.
325 	 */
326 	nodes = ((vm_paddr_t)cnt.v_page_count * PAGE_SIZE) / (PAGE_SIZE +
327 	    sizeof(struct vm_radix_node));
328 	if (!uma_zone_reserve_kva(vm_radix_node_zone, nodes))
329 		panic("%s: unable to create new zone", __func__);
330 	uma_prealloc(vm_radix_node_zone, nodes);
331 }
332 SYSINIT(vm_radix_prealloc, SI_SUB_KMEM, SI_ORDER_SECOND, vm_radix_prealloc,
333     NULL);
334 
335 /*
336  * Initialize the UMA slab zone.
337  * Until vm_radix_prealloc() is called, the zone will be served by the
338  * UMA boot-time pre-allocated pool of pages.
339  */
340 void
341 vm_radix_init(void)
342 {
343 
344 	vm_radix_node_zone = uma_zcreate("RADIX NODE",
345 	    sizeof(struct vm_radix_node), NULL,
346 #ifdef INVARIANTS
347 	    vm_radix_node_zone_dtor,
348 #else
349 	    NULL,
350 #endif
351 	    vm_radix_node_zone_init, NULL, VM_RADIX_PAD, UMA_ZONE_VM |
352 	    UMA_ZONE_NOFREE);
353 }
354 
355 /*
356  * Inserts the key-value pair into the trie.
357  * Panics if the key already exists.
358  */
359 void
360 vm_radix_insert(struct vm_radix *rtree, vm_page_t page)
361 {
362 	vm_pindex_t index, newind;
363 	void **parentp;
364 	struct vm_radix_node *rnode, *tmp;
365 	vm_page_t m;
366 	int slot;
367 	uint16_t clev;
368 
369 	index = page->pindex;
370 
371 	/*
372 	 * The owner of record for root is not really important because it
373 	 * will never be used.
374 	 */
375 	rnode = vm_radix_getroot(rtree);
376 	if (rnode == NULL) {
377 		rtree->rt_root = (uintptr_t)page | VM_RADIX_ISLEAF;
378 		return;
379 	}
380 	parentp = (void **)&rtree->rt_root;
381 	for (;;) {
382 		if (vm_radix_isleaf(rnode)) {
383 			m = vm_radix_topage(rnode);
384 			if (m->pindex == index)
385 				panic("%s: key %jx is already present",
386 				    __func__, (uintmax_t)index);
387 			clev = vm_radix_keydiff(m->pindex, index);
388 			tmp = vm_radix_node_get(vm_radix_trimkey(index,
389 			    clev - 1), 2, clev);
390 			*parentp = tmp;
391 			vm_radix_addpage(tmp, index, clev, page);
392 			vm_radix_addpage(tmp, m->pindex, clev, m);
393 			return;
394 		} else if (vm_radix_keybarr(rnode, index))
395 			break;
396 		slot = vm_radix_slot(index, rnode->rn_clev);
397 		if (rnode->rn_child[slot] == NULL) {
398 			rnode->rn_count++;
399 			vm_radix_addpage(rnode, index, rnode->rn_clev, page);
400 			return;
401 		}
402 		parentp = &rnode->rn_child[slot];
403 		rnode = rnode->rn_child[slot];
404 	}
405 
406 	/*
407 	 * A new node is needed because the right insertion level is reached.
408 	 * Setup the new intermediate node and add the 2 children: the
409 	 * new object and the older edge.
410 	 */
411 	newind = rnode->rn_owner;
412 	clev = vm_radix_keydiff(newind, index);
413 	tmp = vm_radix_node_get(vm_radix_trimkey(index, clev - 1), 2,
414 	    clev);
415 	*parentp = tmp;
416 	vm_radix_addpage(tmp, index, clev, page);
417 	slot = vm_radix_slot(newind, clev);
418 	tmp->rn_child[slot] = rnode;
419 }
420 
421 /*
422  * Returns the value stored at the index.  If the index is not present,
423  * NULL is returned.
424  */
425 vm_page_t
426 vm_radix_lookup(struct vm_radix *rtree, vm_pindex_t index)
427 {
428 	struct vm_radix_node *rnode;
429 	vm_page_t m;
430 	int slot;
431 
432 	rnode = vm_radix_getroot(rtree);
433 	while (rnode != NULL) {
434 		if (vm_radix_isleaf(rnode)) {
435 			m = vm_radix_topage(rnode);
436 			if (m->pindex == index)
437 				return (m);
438 			else
439 				break;
440 		} else if (vm_radix_keybarr(rnode, index))
441 			break;
442 		slot = vm_radix_slot(index, rnode->rn_clev);
443 		rnode = rnode->rn_child[slot];
444 	}
445 	return (NULL);
446 }
447 
448 /*
449  * Look up the nearest entry at a position bigger than or equal to index.
450  */
451 vm_page_t
452 vm_radix_lookup_ge(struct vm_radix *rtree, vm_pindex_t index)
453 {
454 	struct vm_radix_node *stack[VM_RADIX_LIMIT];
455 	vm_pindex_t inc;
456 	vm_page_t m;
457 	struct vm_radix_node *child, *rnode;
458 #ifdef INVARIANTS
459 	int loops = 0;
460 #endif
461 	int slot, tos;
462 
463 	rnode = vm_radix_getroot(rtree);
464 	if (rnode == NULL)
465 		return (NULL);
466 	else if (vm_radix_isleaf(rnode)) {
467 		m = vm_radix_topage(rnode);
468 		if (m->pindex >= index)
469 			return (m);
470 		else
471 			return (NULL);
472 	}
473 	tos = 0;
474 	for (;;) {
475 		/*
476 		 * If the keys differ before the current bisection node,
477 		 * then the search key might rollback to the earliest
478 		 * available bisection node or to the smallest key
479 		 * in the current node (if the owner is bigger than the
480 		 * search key).
481 		 */
482 		if (vm_radix_keybarr(rnode, index)) {
483 			if (index > rnode->rn_owner) {
484 ascend:
485 				KASSERT(++loops < 1000,
486 				    ("vm_radix_lookup_ge: too many loops"));
487 
488 				/*
489 				 * Pop nodes from the stack until either the
490 				 * stack is empty or a node that could have a
491 				 * matching descendant is found.
492 				 */
493 				do {
494 					if (tos == 0)
495 						return (NULL);
496 					rnode = stack[--tos];
497 				} while (vm_radix_slot(index,
498 				    rnode->rn_clev) == (VM_RADIX_COUNT - 1));
499 
500 				/*
501 				 * The following computation cannot overflow
502 				 * because index's slot at the current level
503 				 * is less than VM_RADIX_COUNT - 1.
504 				 */
505 				index = vm_radix_trimkey(index,
506 				    rnode->rn_clev);
507 				index += VM_RADIX_UNITLEVEL(rnode->rn_clev);
508 			} else
509 				index = rnode->rn_owner;
510 			KASSERT(!vm_radix_keybarr(rnode, index),
511 			    ("vm_radix_lookup_ge: keybarr failed"));
512 		}
513 		slot = vm_radix_slot(index, rnode->rn_clev);
514 		child = rnode->rn_child[slot];
515 		if (vm_radix_isleaf(child)) {
516 			m = vm_radix_topage(child);
517 			if (m->pindex >= index)
518 				return (m);
519 		} else if (child != NULL)
520 			goto descend;
521 
522 		/*
523 		 * Look for an available edge or page within the current
524 		 * bisection node.
525 		 */
526                 if (slot < (VM_RADIX_COUNT - 1)) {
527 			inc = VM_RADIX_UNITLEVEL(rnode->rn_clev);
528 			index = vm_radix_trimkey(index, rnode->rn_clev);
529 			do {
530 				index += inc;
531 				slot++;
532 				child = rnode->rn_child[slot];
533 				if (vm_radix_isleaf(child)) {
534 					m = vm_radix_topage(child);
535 					if (m->pindex >= index)
536 						return (m);
537 				} else if (child != NULL)
538 					goto descend;
539 			} while (slot < (VM_RADIX_COUNT - 1));
540 		}
541 		KASSERT(child == NULL || vm_radix_isleaf(child),
542 		    ("vm_radix_lookup_ge: child is radix node"));
543 
544 		/*
545 		 * If a page or edge bigger than the search slot is not found
546 		 * in the current node, ascend to the next higher-level node.
547 		 */
548 		goto ascend;
549 descend:
550 		KASSERT(rnode->rn_clev < VM_RADIX_LIMIT,
551 		    ("vm_radix_lookup_ge: pushing leaf's parent"));
552 		KASSERT(tos < VM_RADIX_LIMIT,
553 		    ("vm_radix_lookup_ge: stack overflow"));
554 		stack[tos++] = rnode;
555 		rnode = child;
556 	}
557 }
558 
559 /*
560  * Look up the nearest entry at a position less than or equal to index.
561  */
562 vm_page_t
563 vm_radix_lookup_le(struct vm_radix *rtree, vm_pindex_t index)
564 {
565 	struct vm_radix_node *stack[VM_RADIX_LIMIT];
566 	vm_pindex_t inc;
567 	vm_page_t m;
568 	struct vm_radix_node *child, *rnode;
569 #ifdef INVARIANTS
570 	int loops = 0;
571 #endif
572 	int slot, tos;
573 
574 	rnode = vm_radix_getroot(rtree);
575 	if (rnode == NULL)
576 		return (NULL);
577 	else if (vm_radix_isleaf(rnode)) {
578 		m = vm_radix_topage(rnode);
579 		if (m->pindex <= index)
580 			return (m);
581 		else
582 			return (NULL);
583 	}
584 	tos = 0;
585 	for (;;) {
586 		/*
587 		 * If the keys differ before the current bisection node,
588 		 * then the search key might rollback to the earliest
589 		 * available bisection node or to the largest key
590 		 * in the current node (if the owner is smaller than the
591 		 * search key).
592 		 */
593 		if (vm_radix_keybarr(rnode, index)) {
594 			if (index > rnode->rn_owner) {
595 				index = rnode->rn_owner + VM_RADIX_COUNT *
596 				    VM_RADIX_UNITLEVEL(rnode->rn_clev);
597 			} else {
598 ascend:
599 				KASSERT(++loops < 1000,
600 				    ("vm_radix_lookup_le: too many loops"));
601 
602 				/*
603 				 * Pop nodes from the stack until either the
604 				 * stack is empty or a node that could have a
605 				 * matching descendant is found.
606 				 */
607 				do {
608 					if (tos == 0)
609 						return (NULL);
610 					rnode = stack[--tos];
611 				} while (vm_radix_slot(index,
612 				    rnode->rn_clev) == 0);
613 
614 				/*
615 				 * The following computation cannot overflow
616 				 * because index's slot at the current level
617 				 * is greater than 0.
618 				 */
619 				index = vm_radix_trimkey(index,
620 				    rnode->rn_clev);
621 			}
622 			index--;
623 			KASSERT(!vm_radix_keybarr(rnode, index),
624 			    ("vm_radix_lookup_le: keybarr failed"));
625 		}
626 		slot = vm_radix_slot(index, rnode->rn_clev);
627 		child = rnode->rn_child[slot];
628 		if (vm_radix_isleaf(child)) {
629 			m = vm_radix_topage(child);
630 			if (m->pindex <= index)
631 				return (m);
632 		} else if (child != NULL)
633 			goto descend;
634 
635 		/*
636 		 * Look for an available edge or page within the current
637 		 * bisection node.
638 		 */
639 		if (slot > 0) {
640 			inc = VM_RADIX_UNITLEVEL(rnode->rn_clev);
641 			index |= inc - 1;
642 			do {
643 				index -= inc;
644 				slot--;
645 				child = rnode->rn_child[slot];
646 				if (vm_radix_isleaf(child)) {
647 					m = vm_radix_topage(child);
648 					if (m->pindex <= index)
649 						return (m);
650 				} else if (child != NULL)
651 					goto descend;
652 			} while (slot > 0);
653 		}
654 		KASSERT(child == NULL || vm_radix_isleaf(child),
655 		    ("vm_radix_lookup_le: child is radix node"));
656 
657 		/*
658 		 * If a page or edge smaller than the search slot is not found
659 		 * in the current node, ascend to the next higher-level node.
660 		 */
661 		goto ascend;
662 descend:
663 		KASSERT(rnode->rn_clev < VM_RADIX_LIMIT,
664 		    ("vm_radix_lookup_le: pushing leaf's parent"));
665 		KASSERT(tos < VM_RADIX_LIMIT,
666 		    ("vm_radix_lookup_le: stack overflow"));
667 		stack[tos++] = rnode;
668 		rnode = child;
669 	}
670 }
671 
672 /*
673  * Remove the specified index from the tree.
674  * Panics if the key is not present.
675  */
676 void
677 vm_radix_remove(struct vm_radix *rtree, vm_pindex_t index)
678 {
679 	struct vm_radix_node *rnode, *parent;
680 	vm_page_t m;
681 	int i, slot;
682 
683 	rnode = vm_radix_getroot(rtree);
684 	if (vm_radix_isleaf(rnode)) {
685 		m = vm_radix_topage(rnode);
686 		if (m->pindex != index)
687 			panic("%s: invalid key found", __func__);
688 		vm_radix_setroot(rtree, NULL);
689 		return;
690 	}
691 	parent = NULL;
692 	for (;;) {
693 		if (rnode == NULL)
694 			panic("vm_radix_remove: impossible to locate the key");
695 		slot = vm_radix_slot(index, rnode->rn_clev);
696 		if (vm_radix_isleaf(rnode->rn_child[slot])) {
697 			m = vm_radix_topage(rnode->rn_child[slot]);
698 			if (m->pindex != index)
699 				panic("%s: invalid key found", __func__);
700 			rnode->rn_child[slot] = NULL;
701 			rnode->rn_count--;
702 			if (rnode->rn_count > 1)
703 				break;
704 			for (i = 0; i < VM_RADIX_COUNT; i++)
705 				if (rnode->rn_child[i] != NULL)
706 					break;
707 			KASSERT(i != VM_RADIX_COUNT,
708 			    ("%s: invalid node configuration", __func__));
709 			if (parent == NULL)
710 				vm_radix_setroot(rtree, rnode->rn_child[i]);
711 			else {
712 				slot = vm_radix_slot(index, parent->rn_clev);
713 				KASSERT(parent->rn_child[slot] == rnode,
714 				    ("%s: invalid child value", __func__));
715 				parent->rn_child[slot] = rnode->rn_child[i];
716 			}
717 			rnode->rn_count--;
718 			rnode->rn_child[i] = NULL;
719 			vm_radix_node_put(rnode);
720 			break;
721 		}
722 		parent = rnode;
723 		rnode = rnode->rn_child[slot];
724 	}
725 }
726 
727 /*
728  * Remove and free all the nodes from the radix tree.
729  * This function is recursive but there is a tight control on it as the
730  * maximum depth of the tree is fixed.
731  */
732 void
733 vm_radix_reclaim_allnodes(struct vm_radix *rtree)
734 {
735 	struct vm_radix_node *root;
736 
737 	root = vm_radix_getroot(rtree);
738 	if (root == NULL)
739 		return;
740 	vm_radix_setroot(rtree, NULL);
741 	if (!vm_radix_isleaf(root))
742 		vm_radix_reclaim_allnodes_int(root);
743 }
744 
745 #ifdef DDB
746 /*
747  * Show details about the given radix node.
748  */
749 DB_SHOW_COMMAND(radixnode, db_show_radixnode)
750 {
751 	struct vm_radix_node *rnode;
752 	int i;
753 
754         if (!have_addr)
755                 return;
756 	rnode = (struct vm_radix_node *)addr;
757 	db_printf("radixnode %p, owner %jx, children count %u, level %u:\n",
758 	    (void *)rnode, (uintmax_t)rnode->rn_owner, rnode->rn_count,
759 	    rnode->rn_clev);
760 	for (i = 0; i < VM_RADIX_COUNT; i++)
761 		if (rnode->rn_child[i] != NULL)
762 			db_printf("slot: %d, val: %p, page: %p, clev: %d\n",
763 			    i, (void *)rnode->rn_child[i],
764 			    vm_radix_isleaf(rnode->rn_child[i]) ?
765 			    vm_radix_topage(rnode->rn_child[i]) : NULL,
766 			    rnode->rn_clev);
767 }
768 #endif /* DDB */
769