vm_radix.c revision 250018
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: head/sys/vm/vm_radix.c 250018 2013-04-28 08:29:00Z alc $");
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
96struct 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
103static 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 */
109static __inline struct vm_radix_node *
110vm_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 */
139static __inline void
140vm_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 */
149static __inline int
150vm_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. */
158static __inline vm_pindex_t
159vm_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 */
174static __inline struct vm_radix_node *
175vm_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 */
184static __inline void
185vm_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 */
194static __inline boolean_t
195vm_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 */
204static __inline vm_page_t
205vm_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 */
214static __inline void
215vm_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 */
228static __inline uint16_t
229vm_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 */
248static __inline boolean_t
249vm_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 * Adjusts the idx key to the first upper level available, based on a valid
261 * initial level and map of available levels.
262 * Returns a value bigger than 0 to signal that there are not valid levels
263 * available.
264 */
265static __inline int
266vm_radix_addlev(vm_pindex_t *idx, boolean_t *levels, uint16_t ilev)
267{
268
269	for (; levels[ilev] == FALSE ||
270	    vm_radix_slot(*idx, ilev) == (VM_RADIX_COUNT - 1); ilev--)
271		if (ilev == 0)
272			return (1);
273
274	/*
275	 * The following computation cannot overflow because *idx's slot at
276	 * ilev is less than VM_RADIX_COUNT - 1.
277	 */
278	*idx = vm_radix_trimkey(*idx, ilev);
279	*idx += VM_RADIX_UNITLEVEL(ilev);
280	return (0);
281}
282
283/*
284 * Adjusts the idx key to the first lower level available, based on a valid
285 * initial level and map of available levels.
286 * Returns a value bigger than 0 to signal that there are not valid levels
287 * available.
288 */
289static __inline int
290vm_radix_declev(vm_pindex_t *idx, boolean_t *levels, uint16_t ilev)
291{
292
293	for (; levels[ilev] == FALSE ||
294	    vm_radix_slot(*idx, ilev) == 0; ilev--)
295		if (ilev == 0)
296			return (1);
297
298	/*
299	 * The following computation cannot overflow because *idx's slot at
300	 * ilev is greater than 0.
301	 */
302	*idx = vm_radix_trimkey(*idx, ilev);
303	*idx -= 1;
304	return (0);
305}
306
307/*
308 * Internal helper for vm_radix_reclaim_allnodes().
309 * This function is recursive.
310 */
311static void
312vm_radix_reclaim_allnodes_int(struct vm_radix_node *rnode)
313{
314	int slot;
315
316	KASSERT(rnode->rn_count <= VM_RADIX_COUNT,
317	    ("vm_radix_reclaim_allnodes_int: bad count in rnode %p", rnode));
318	for (slot = 0; rnode->rn_count != 0; slot++) {
319		if (rnode->rn_child[slot] == NULL)
320			continue;
321		if (!vm_radix_isleaf(rnode->rn_child[slot]))
322			vm_radix_reclaim_allnodes_int(rnode->rn_child[slot]);
323		rnode->rn_child[slot] = NULL;
324		rnode->rn_count--;
325	}
326	vm_radix_node_put(rnode);
327}
328
329#ifdef INVARIANTS
330/*
331 * Radix node zone destructor.
332 */
333static void
334vm_radix_node_zone_dtor(void *mem, int size __unused, void *arg __unused)
335{
336	struct vm_radix_node *rnode;
337	int slot;
338
339	rnode = mem;
340	KASSERT(rnode->rn_count == 0,
341	    ("vm_radix_node_put: rnode %p has %d children", rnode,
342	    rnode->rn_count));
343	for (slot = 0; slot < VM_RADIX_COUNT; slot++)
344		KASSERT(rnode->rn_child[slot] == NULL,
345		    ("vm_radix_node_put: rnode %p has a child", rnode));
346}
347#endif
348
349/*
350 * Radix node zone initializer.
351 */
352static int
353vm_radix_node_zone_init(void *mem, int size __unused, int flags __unused)
354{
355	struct vm_radix_node *rnode;
356
357	rnode = mem;
358	memset(rnode->rn_child, 0, sizeof(rnode->rn_child));
359	return (0);
360}
361
362/*
363 * Pre-allocate intermediate nodes from the UMA slab zone.
364 */
365static void
366vm_radix_prealloc(void *arg __unused)
367{
368	int nodes;
369
370	/*
371	 * Calculate the number of reserved nodes, discounting the pages that
372	 * are needed to store them.
373	 */
374	nodes = ((vm_paddr_t)cnt.v_page_count * PAGE_SIZE) / (PAGE_SIZE +
375	    sizeof(struct vm_radix_node));
376	if (!uma_zone_reserve_kva(vm_radix_node_zone, nodes))
377		panic("%s: unable to create new zone", __func__);
378	uma_prealloc(vm_radix_node_zone, nodes);
379}
380SYSINIT(vm_radix_prealloc, SI_SUB_KMEM, SI_ORDER_SECOND, vm_radix_prealloc,
381    NULL);
382
383/*
384 * Initialize the UMA slab zone.
385 * Until vm_radix_prealloc() is called, the zone will be served by the
386 * UMA boot-time pre-allocated pool of pages.
387 */
388void
389vm_radix_init(void)
390{
391
392	vm_radix_node_zone = uma_zcreate("RADIX NODE",
393	    sizeof(struct vm_radix_node), NULL,
394#ifdef INVARIANTS
395	    vm_radix_node_zone_dtor,
396#else
397	    NULL,
398#endif
399	    vm_radix_node_zone_init, NULL, VM_RADIX_PAD, UMA_ZONE_VM |
400	    UMA_ZONE_NOFREE);
401}
402
403/*
404 * Inserts the key-value pair into the trie.
405 * Panics if the key already exists.
406 */
407void
408vm_radix_insert(struct vm_radix *rtree, vm_page_t page)
409{
410	vm_pindex_t index, newind;
411	void **parentp;
412	struct vm_radix_node *rnode, *tmp;
413	vm_page_t m;
414	int slot;
415	uint16_t clev;
416
417	index = page->pindex;
418
419	/*
420	 * The owner of record for root is not really important because it
421	 * will never be used.
422	 */
423	rnode = vm_radix_getroot(rtree);
424	if (rnode == NULL) {
425		rtree->rt_root = (uintptr_t)page | VM_RADIX_ISLEAF;
426		return;
427	}
428	parentp = (void **)&rtree->rt_root;
429	for (;;) {
430		if (vm_radix_isleaf(rnode)) {
431			m = vm_radix_topage(rnode);
432			if (m->pindex == index)
433				panic("%s: key %jx is already present",
434				    __func__, (uintmax_t)index);
435			clev = vm_radix_keydiff(m->pindex, index);
436			tmp = vm_radix_node_get(vm_radix_trimkey(index,
437			    clev - 1), 2, clev);
438			*parentp = tmp;
439			vm_radix_addpage(tmp, index, clev, page);
440			vm_radix_addpage(tmp, m->pindex, clev, m);
441			return;
442		} else if (vm_radix_keybarr(rnode, index))
443			break;
444		slot = vm_radix_slot(index, rnode->rn_clev);
445		if (rnode->rn_child[slot] == NULL) {
446			rnode->rn_count++;
447			vm_radix_addpage(rnode, index, rnode->rn_clev, page);
448			return;
449		}
450		parentp = &rnode->rn_child[slot];
451		rnode = rnode->rn_child[slot];
452	}
453
454	/*
455	 * A new node is needed because the right insertion level is reached.
456	 * Setup the new intermediate node and add the 2 children: the
457	 * new object and the older edge.
458	 */
459	newind = rnode->rn_owner;
460	clev = vm_radix_keydiff(newind, index);
461	tmp = vm_radix_node_get(vm_radix_trimkey(index, clev - 1), 2,
462	    clev);
463	*parentp = tmp;
464	vm_radix_addpage(tmp, index, clev, page);
465	slot = vm_radix_slot(newind, clev);
466	tmp->rn_child[slot] = rnode;
467}
468
469/*
470 * Returns the value stored at the index.  If the index is not present,
471 * NULL is returned.
472 */
473vm_page_t
474vm_radix_lookup(struct vm_radix *rtree, vm_pindex_t index)
475{
476	struct vm_radix_node *rnode;
477	vm_page_t m;
478	int slot;
479
480	rnode = vm_radix_getroot(rtree);
481	while (rnode != NULL) {
482		if (vm_radix_isleaf(rnode)) {
483			m = vm_radix_topage(rnode);
484			if (m->pindex == index)
485				return (m);
486			else
487				break;
488		} else if (vm_radix_keybarr(rnode, index))
489			break;
490		slot = vm_radix_slot(index, rnode->rn_clev);
491		rnode = rnode->rn_child[slot];
492	}
493	return (NULL);
494}
495
496/*
497 * Look up the nearest entry at a position bigger than or equal to index.
498 */
499vm_page_t
500vm_radix_lookup_ge(struct vm_radix *rtree, vm_pindex_t index)
501{
502	vm_pindex_t inc;
503	vm_page_t m;
504	struct vm_radix_node *child, *rnode;
505	int slot;
506	uint16_t difflev;
507	boolean_t maplevels[VM_RADIX_LIMIT + 1];
508#ifdef INVARIANTS
509	int loops = 0;
510#endif
511
512	rnode = vm_radix_getroot(rtree);
513	if (rnode == NULL)
514		return (NULL);
515	else if (vm_radix_isleaf(rnode)) {
516		m = vm_radix_topage(rnode);
517		if (m->pindex >= index)
518			return (m);
519		else
520			return (NULL);
521	}
522restart:
523	KASSERT(++loops < 1000, ("%s: too many loops", __func__));
524	for (difflev = 0; difflev < (VM_RADIX_LIMIT + 1); difflev++)
525		maplevels[difflev] = FALSE;
526	for (;;) {
527		maplevels[rnode->rn_clev] = TRUE;
528
529		/*
530		 * If the keys differ before the current bisection node,
531		 * then the search key might rollback to the earliest
532		 * available bisection node or to the smallest key
533		 * in the current node (if the owner is bigger than the
534		 * search key).
535		 * The maplevels array records any node has been seen
536		 * at a given level.  This aids the search for a valid
537		 * bisection node.
538		 */
539		if (vm_radix_keybarr(rnode, index)) {
540			if (index > rnode->rn_owner) {
541				difflev = vm_radix_keydiff(index,
542				    rnode->rn_owner);
543				if (vm_radix_addlev(&index, maplevels,
544				    difflev) > 0)
545					break;
546				rnode = vm_radix_getroot(rtree);
547				goto restart;
548			} else
549				index = rnode->rn_owner;
550		}
551		slot = vm_radix_slot(index, rnode->rn_clev);
552		child = rnode->rn_child[slot];
553		if (vm_radix_isleaf(child)) {
554			m = vm_radix_topage(child);
555			if (m->pindex >= index)
556				return (m);
557		} else if (child != NULL)
558			goto descend;
559
560		/*
561		 * Look for an available edge or page within the current
562		 * bisection node.
563		 */
564                if (slot < (VM_RADIX_COUNT - 1)) {
565			inc = VM_RADIX_UNITLEVEL(rnode->rn_clev);
566			index = vm_radix_trimkey(index, rnode->rn_clev);
567			do {
568				index += inc;
569				slot++;
570				child = rnode->rn_child[slot];
571				if (vm_radix_isleaf(child)) {
572					m = vm_radix_topage(child);
573					if (m->pindex >= index)
574						return (m);
575				} else if (child != NULL)
576					goto descend;
577			} while (slot < (VM_RADIX_COUNT - 1));
578		}
579		KASSERT(child == NULL || vm_radix_isleaf(child),
580		    ("vm_radix_lookup_ge: child is radix node"));
581
582		/*
583		 * If a valid page or edge bigger than the search slot is
584		 * found in the traversal, skip to the next higher-level key.
585		 */
586		if (rnode->rn_clev == 0 || vm_radix_addlev(&index, maplevels,
587		    rnode->rn_clev - 1) > 0)
588			break;
589		rnode = vm_radix_getroot(rtree);
590		goto restart;
591descend:
592		rnode = child;
593	}
594	return (NULL);
595}
596
597/*
598 * Look up the nearest entry at a position less than or equal to index.
599 */
600vm_page_t
601vm_radix_lookup_le(struct vm_radix *rtree, vm_pindex_t index)
602{
603	vm_pindex_t inc;
604	vm_page_t m;
605	struct vm_radix_node *child, *rnode;
606	int slot;
607	uint16_t difflev;
608	boolean_t maplevels[VM_RADIX_LIMIT + 1];
609#ifdef INVARIANTS
610	int loops = 0;
611#endif
612
613	rnode = vm_radix_getroot(rtree);
614	if (rnode == NULL)
615		return (NULL);
616	else if (vm_radix_isleaf(rnode)) {
617		m = vm_radix_topage(rnode);
618		if (m->pindex <= index)
619			return (m);
620		else
621			return (NULL);
622	}
623restart:
624	KASSERT(++loops < 1000, ("%s: too many loops", __func__));
625	for (difflev = 0; difflev < (VM_RADIX_LIMIT + 1); difflev++)
626		maplevels[difflev] = FALSE;
627	for (;;) {
628		maplevels[rnode->rn_clev] = TRUE;
629
630		/*
631		 * If the keys differ before the current bisection node,
632		 * then the search key might rollback to the earliest
633		 * available bisection node or to the largest key
634		 * in the current node (if the owner is smaller than the
635		 * search key).
636		 * The maplevels array records any node has been seen
637		 * at a given level.  This aids the search for a valid
638		 * bisection node.
639		 */
640		if (vm_radix_keybarr(rnode, index)) {
641			if (index > rnode->rn_owner) {
642				index = rnode->rn_owner + VM_RADIX_COUNT *
643				    VM_RADIX_UNITLEVEL(rnode->rn_clev) - 1;
644			} else {
645				difflev = vm_radix_keydiff(index,
646				    rnode->rn_owner);
647				if (vm_radix_declev(&index, maplevels,
648				    difflev) > 0)
649					break;
650				rnode = vm_radix_getroot(rtree);
651				goto restart;
652			}
653		}
654		slot = vm_radix_slot(index, rnode->rn_clev);
655		child = rnode->rn_child[slot];
656		if (vm_radix_isleaf(child)) {
657			m = vm_radix_topage(child);
658			if (m->pindex <= index)
659				return (m);
660		} else if (child != NULL)
661			goto descend;
662
663		/*
664		 * Look for an available edge or page within the current
665		 * bisection node.
666		 */
667		if (slot > 0) {
668			inc = VM_RADIX_UNITLEVEL(rnode->rn_clev);
669			index |= inc - 1;
670			do {
671				index -= inc;
672				slot--;
673				child = rnode->rn_child[slot];
674				if (vm_radix_isleaf(child)) {
675					m = vm_radix_topage(child);
676					if (m->pindex <= index)
677						return (m);
678				} else if (child != NULL)
679					goto descend;
680			} while (slot > 0);
681		}
682		KASSERT(child == NULL || vm_radix_isleaf(child),
683		    ("vm_radix_lookup_le: child is radix node"));
684
685		/*
686		 * If a valid page or edge smaller than the search slot is
687		 * found in the traversal, skip to the next higher-level key.
688		 */
689		if (rnode->rn_clev == 0 || vm_radix_declev(&index, maplevels,
690		    rnode->rn_clev - 1) > 0)
691			break;
692		rnode = vm_radix_getroot(rtree);
693		goto restart;
694descend:
695		rnode = child;
696	}
697	return (NULL);
698}
699
700/*
701 * Remove the specified index from the tree.
702 * Panics if the key is not present.
703 */
704void
705vm_radix_remove(struct vm_radix *rtree, vm_pindex_t index)
706{
707	struct vm_radix_node *rnode, *parent;
708	vm_page_t m;
709	int i, slot;
710
711	rnode = vm_radix_getroot(rtree);
712	if (vm_radix_isleaf(rnode)) {
713		m = vm_radix_topage(rnode);
714		if (m->pindex != index)
715			panic("%s: invalid key found", __func__);
716		vm_radix_setroot(rtree, NULL);
717		return;
718	}
719	parent = NULL;
720	for (;;) {
721		if (rnode == NULL)
722			panic("vm_radix_remove: impossible to locate the key");
723		slot = vm_radix_slot(index, rnode->rn_clev);
724		if (vm_radix_isleaf(rnode->rn_child[slot])) {
725			m = vm_radix_topage(rnode->rn_child[slot]);
726			if (m->pindex != index)
727				panic("%s: invalid key found", __func__);
728			rnode->rn_child[slot] = NULL;
729			rnode->rn_count--;
730			if (rnode->rn_count > 1)
731				break;
732			for (i = 0; i < VM_RADIX_COUNT; i++)
733				if (rnode->rn_child[i] != NULL)
734					break;
735			KASSERT(i != VM_RADIX_COUNT,
736			    ("%s: invalid node configuration", __func__));
737			if (parent == NULL)
738				vm_radix_setroot(rtree, rnode->rn_child[i]);
739			else {
740				slot = vm_radix_slot(index, parent->rn_clev);
741				KASSERT(parent->rn_child[slot] == rnode,
742				    ("%s: invalid child value", __func__));
743				parent->rn_child[slot] = rnode->rn_child[i];
744			}
745			rnode->rn_count--;
746			rnode->rn_child[i] = NULL;
747			vm_radix_node_put(rnode);
748			break;
749		}
750		parent = rnode;
751		rnode = rnode->rn_child[slot];
752	}
753}
754
755/*
756 * Remove and free all the nodes from the radix tree.
757 * This function is recursive but there is a tight control on it as the
758 * maximum depth of the tree is fixed.
759 */
760void
761vm_radix_reclaim_allnodes(struct vm_radix *rtree)
762{
763	struct vm_radix_node *root;
764
765	root = vm_radix_getroot(rtree);
766	if (root == NULL)
767		return;
768	vm_radix_setroot(rtree, NULL);
769	if (!vm_radix_isleaf(root))
770		vm_radix_reclaim_allnodes_int(root);
771}
772
773#ifdef DDB
774/*
775 * Show details about the given radix node.
776 */
777DB_SHOW_COMMAND(radixnode, db_show_radixnode)
778{
779	struct vm_radix_node *rnode;
780	int i;
781
782        if (!have_addr)
783                return;
784	rnode = (struct vm_radix_node *)addr;
785	db_printf("radixnode %p, owner %jx, children count %u, level %u:\n",
786	    (void *)rnode, (uintmax_t)rnode->rn_owner, rnode->rn_count,
787	    rnode->rn_clev);
788	for (i = 0; i < VM_RADIX_COUNT; i++)
789		if (rnode->rn_child[i] != NULL)
790			db_printf("slot: %d, val: %p, page: %p, clev: %d\n",
791			    i, (void *)rnode->rn_child[i],
792			    vm_radix_isleaf(rnode->rn_child[i]) ?
793			    vm_radix_topage(rnode->rn_child[i]) : NULL,
794			    rnode->rn_clev);
795}
796#endif /* DDB */
797