xref: /freebsd/sys/vm/vm_radix.c (revision 8ef24a0d4b28fe230e20637f56869cc4148cd2ca)
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 << ((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.
107  */
108 static __inline struct vm_radix_node *
109 vm_radix_node_get(vm_pindex_t owner, uint16_t count, uint16_t clevel)
110 {
111 	struct vm_radix_node *rnode;
112 
113 	rnode = uma_zalloc(vm_radix_node_zone, M_NOWAIT | M_ZERO);
114 	if (rnode == NULL)
115 		return (NULL);
116 	rnode->rn_owner = owner;
117 	rnode->rn_count = count;
118 	rnode->rn_clev = clevel;
119 	return (rnode);
120 }
121 
122 /*
123  * Free radix node.
124  */
125 static __inline void
126 vm_radix_node_put(struct vm_radix_node *rnode)
127 {
128 
129 	uma_zfree(vm_radix_node_zone, rnode);
130 }
131 
132 /*
133  * Return the position in the array for a given level.
134  */
135 static __inline int
136 vm_radix_slot(vm_pindex_t index, uint16_t level)
137 {
138 
139 	return ((index >> (level * VM_RADIX_WIDTH)) & VM_RADIX_MASK);
140 }
141 
142 /* Trims the key after the specified level. */
143 static __inline vm_pindex_t
144 vm_radix_trimkey(vm_pindex_t index, uint16_t level)
145 {
146 	vm_pindex_t ret;
147 
148 	ret = index;
149 	if (level > 0) {
150 		ret >>= level * VM_RADIX_WIDTH;
151 		ret <<= level * VM_RADIX_WIDTH;
152 	}
153 	return (ret);
154 }
155 
156 /*
157  * Get the root node for a radix tree.
158  */
159 static __inline struct vm_radix_node *
160 vm_radix_getroot(struct vm_radix *rtree)
161 {
162 
163 	return ((struct vm_radix_node *)rtree->rt_root);
164 }
165 
166 /*
167  * Set the root node for a radix tree.
168  */
169 static __inline void
170 vm_radix_setroot(struct vm_radix *rtree, struct vm_radix_node *rnode)
171 {
172 
173 	rtree->rt_root = (uintptr_t)rnode;
174 }
175 
176 /*
177  * Returns TRUE if the specified radix node is a leaf and FALSE otherwise.
178  */
179 static __inline boolean_t
180 vm_radix_isleaf(struct vm_radix_node *rnode)
181 {
182 
183 	return (((uintptr_t)rnode & VM_RADIX_ISLEAF) != 0);
184 }
185 
186 /*
187  * Returns the associated page extracted from rnode.
188  */
189 static __inline vm_page_t
190 vm_radix_topage(struct vm_radix_node *rnode)
191 {
192 
193 	return ((vm_page_t)((uintptr_t)rnode & ~VM_RADIX_FLAGS));
194 }
195 
196 /*
197  * Adds the page as a child of the provided node.
198  */
199 static __inline void
200 vm_radix_addpage(struct vm_radix_node *rnode, vm_pindex_t index, uint16_t clev,
201     vm_page_t page)
202 {
203 	int slot;
204 
205 	slot = vm_radix_slot(index, clev);
206 	rnode->rn_child[slot] = (void *)((uintptr_t)page | VM_RADIX_ISLEAF);
207 }
208 
209 /*
210  * Returns the slot where two keys differ.
211  * It cannot accept 2 equal keys.
212  */
213 static __inline uint16_t
214 vm_radix_keydiff(vm_pindex_t index1, vm_pindex_t index2)
215 {
216 	uint16_t clev;
217 
218 	KASSERT(index1 != index2, ("%s: passing the same key value %jx",
219 	    __func__, (uintmax_t)index1));
220 
221 	index1 ^= index2;
222 	for (clev = VM_RADIX_LIMIT;; clev--)
223 		if (vm_radix_slot(index1, clev) != 0)
224 			return (clev);
225 }
226 
227 /*
228  * Returns TRUE if it can be determined that key does not belong to the
229  * specified rnode.  Otherwise, returns FALSE.
230  */
231 static __inline boolean_t
232 vm_radix_keybarr(struct vm_radix_node *rnode, vm_pindex_t idx)
233 {
234 
235 	if (rnode->rn_clev < VM_RADIX_LIMIT) {
236 		idx = vm_radix_trimkey(idx, rnode->rn_clev + 1);
237 		return (idx != rnode->rn_owner);
238 	}
239 	return (FALSE);
240 }
241 
242 /*
243  * Internal helper for vm_radix_reclaim_allnodes().
244  * This function is recursive.
245  */
246 static void
247 vm_radix_reclaim_allnodes_int(struct vm_radix_node *rnode)
248 {
249 	int slot;
250 
251 	KASSERT(rnode->rn_count <= VM_RADIX_COUNT,
252 	    ("vm_radix_reclaim_allnodes_int: bad count in rnode %p", rnode));
253 	for (slot = 0; rnode->rn_count != 0; slot++) {
254 		if (rnode->rn_child[slot] == NULL)
255 			continue;
256 		if (!vm_radix_isleaf(rnode->rn_child[slot]))
257 			vm_radix_reclaim_allnodes_int(rnode->rn_child[slot]);
258 		rnode->rn_child[slot] = NULL;
259 		rnode->rn_count--;
260 	}
261 	vm_radix_node_put(rnode);
262 }
263 
264 #ifdef INVARIANTS
265 /*
266  * Radix node zone destructor.
267  */
268 static void
269 vm_radix_node_zone_dtor(void *mem, int size __unused, void *arg __unused)
270 {
271 	struct vm_radix_node *rnode;
272 	int slot;
273 
274 	rnode = mem;
275 	KASSERT(rnode->rn_count == 0,
276 	    ("vm_radix_node_put: rnode %p has %d children", rnode,
277 	    rnode->rn_count));
278 	for (slot = 0; slot < VM_RADIX_COUNT; slot++)
279 		KASSERT(rnode->rn_child[slot] == NULL,
280 		    ("vm_radix_node_put: rnode %p has a child", rnode));
281 }
282 #endif
283 
284 #ifndef UMA_MD_SMALL_ALLOC
285 /*
286  * Reserve the KVA necessary to satisfy the node allocation.
287  * This is mandatory in architectures not supporting direct
288  * mapping as they will need otherwise to carve into the kernel maps for
289  * every node allocation, resulting into deadlocks for consumers already
290  * working with kernel maps.
291  */
292 static void
293 vm_radix_reserve_kva(void *arg __unused)
294 {
295 
296 	/*
297 	 * Calculate the number of reserved nodes, discounting the pages that
298 	 * are needed to store them.
299 	 */
300 	if (!uma_zone_reserve_kva(vm_radix_node_zone,
301 	    ((vm_paddr_t)vm_cnt.v_page_count * PAGE_SIZE) / (PAGE_SIZE +
302 	    sizeof(struct vm_radix_node))))
303 		panic("%s: unable to reserve KVA", __func__);
304 }
305 SYSINIT(vm_radix_reserve_kva, SI_SUB_KMEM, SI_ORDER_THIRD,
306     vm_radix_reserve_kva, NULL);
307 #endif
308 
309 /*
310  * Initialize the UMA slab zone.
311  * Until vm_radix_prealloc() is called, the zone will be served by the
312  * UMA boot-time pre-allocated pool of pages.
313  */
314 void
315 vm_radix_init(void)
316 {
317 
318 	vm_radix_node_zone = uma_zcreate("RADIX NODE",
319 	    sizeof(struct vm_radix_node), NULL,
320 #ifdef INVARIANTS
321 	    vm_radix_node_zone_dtor,
322 #else
323 	    NULL,
324 #endif
325 	    NULL, NULL, VM_RADIX_PAD, UMA_ZONE_VM);
326 }
327 
328 /*
329  * Inserts the key-value pair into the trie.
330  * Panics if the key already exists.
331  */
332 int
333 vm_radix_insert(struct vm_radix *rtree, vm_page_t page)
334 {
335 	vm_pindex_t index, newind;
336 	void **parentp;
337 	struct vm_radix_node *rnode, *tmp;
338 	vm_page_t m;
339 	int slot;
340 	uint16_t clev;
341 
342 	index = page->pindex;
343 
344 restart:
345 
346 	/*
347 	 * The owner of record for root is not really important because it
348 	 * will never be used.
349 	 */
350 	rnode = vm_radix_getroot(rtree);
351 	if (rnode == NULL) {
352 		rtree->rt_root = (uintptr_t)page | VM_RADIX_ISLEAF;
353 		return (0);
354 	}
355 	parentp = (void **)&rtree->rt_root;
356 	for (;;) {
357 		if (vm_radix_isleaf(rnode)) {
358 			m = vm_radix_topage(rnode);
359 			if (m->pindex == index)
360 				panic("%s: key %jx is already present",
361 				    __func__, (uintmax_t)index);
362 			clev = vm_radix_keydiff(m->pindex, index);
363 
364 			/*
365 			 * During node allocation the trie that is being
366 			 * walked can be modified because of recursing radix
367 			 * trie operations.
368 			 * If this is the case, the recursing functions signal
369 			 * such situation and the insert operation must
370 			 * start from scratch again.
371 			 * The freed radix node will then be in the UMA
372 			 * caches very likely to avoid the same situation
373 			 * to happen.
374 			 */
375 			rtree->rt_flags |= RT_INSERT_INPROG;
376 			tmp = vm_radix_node_get(vm_radix_trimkey(index,
377 			    clev + 1), 2, clev);
378 			rtree->rt_flags &= ~RT_INSERT_INPROG;
379 			if (tmp == NULL) {
380 				rtree->rt_flags &= ~RT_TRIE_MODIFIED;
381 				return (ENOMEM);
382 			}
383 			if ((rtree->rt_flags & RT_TRIE_MODIFIED) != 0) {
384 				rtree->rt_flags &= ~RT_TRIE_MODIFIED;
385 				tmp->rn_count = 0;
386 				vm_radix_node_put(tmp);
387 				goto restart;
388 			}
389 			*parentp = tmp;
390 			vm_radix_addpage(tmp, index, clev, page);
391 			vm_radix_addpage(tmp, m->pindex, clev, m);
392 			return (0);
393 		} else if (vm_radix_keybarr(rnode, index))
394 			break;
395 		slot = vm_radix_slot(index, rnode->rn_clev);
396 		if (rnode->rn_child[slot] == NULL) {
397 			rnode->rn_count++;
398 			vm_radix_addpage(rnode, index, rnode->rn_clev, page);
399 			return (0);
400 		}
401 		parentp = &rnode->rn_child[slot];
402 		rnode = rnode->rn_child[slot];
403 	}
404 
405 	/*
406 	 * A new node is needed because the right insertion level is reached.
407 	 * Setup the new intermediate node and add the 2 children: the
408 	 * new object and the older edge.
409 	 */
410 	newind = rnode->rn_owner;
411 	clev = vm_radix_keydiff(newind, index);
412 
413 	/* See the comments above. */
414 	rtree->rt_flags |= RT_INSERT_INPROG;
415 	tmp = vm_radix_node_get(vm_radix_trimkey(index, clev + 1), 2, clev);
416 	rtree->rt_flags &= ~RT_INSERT_INPROG;
417 	if (tmp == NULL) {
418 		rtree->rt_flags &= ~RT_TRIE_MODIFIED;
419 		return (ENOMEM);
420 	}
421 	if ((rtree->rt_flags & RT_TRIE_MODIFIED) != 0) {
422 		rtree->rt_flags &= ~RT_TRIE_MODIFIED;
423 		tmp->rn_count = 0;
424 		vm_radix_node_put(tmp);
425 		goto restart;
426 	}
427 	*parentp = tmp;
428 	vm_radix_addpage(tmp, index, clev, page);
429 	slot = vm_radix_slot(newind, clev);
430 	tmp->rn_child[slot] = rnode;
431 	return (0);
432 }
433 
434 /*
435  * Returns TRUE if the specified radix tree contains a single leaf and FALSE
436  * otherwise.
437  */
438 boolean_t
439 vm_radix_is_singleton(struct vm_radix *rtree)
440 {
441 	struct vm_radix_node *rnode;
442 
443 	rnode = vm_radix_getroot(rtree);
444 	if (rnode == NULL)
445 		return (FALSE);
446 	return (vm_radix_isleaf(rnode));
447 }
448 
449 /*
450  * Returns the value stored at the index.  If the index is not present,
451  * NULL is returned.
452  */
453 vm_page_t
454 vm_radix_lookup(struct vm_radix *rtree, vm_pindex_t index)
455 {
456 	struct vm_radix_node *rnode;
457 	vm_page_t m;
458 	int slot;
459 
460 	rnode = vm_radix_getroot(rtree);
461 	while (rnode != NULL) {
462 		if (vm_radix_isleaf(rnode)) {
463 			m = vm_radix_topage(rnode);
464 			if (m->pindex == index)
465 				return (m);
466 			else
467 				break;
468 		} else if (vm_radix_keybarr(rnode, index))
469 			break;
470 		slot = vm_radix_slot(index, rnode->rn_clev);
471 		rnode = rnode->rn_child[slot];
472 	}
473 	return (NULL);
474 }
475 
476 /*
477  * Look up the nearest entry at a position bigger than or equal to index.
478  */
479 vm_page_t
480 vm_radix_lookup_ge(struct vm_radix *rtree, vm_pindex_t index)
481 {
482 	struct vm_radix_node *stack[VM_RADIX_LIMIT];
483 	vm_pindex_t inc;
484 	vm_page_t m;
485 	struct vm_radix_node *child, *rnode;
486 #ifdef INVARIANTS
487 	int loops = 0;
488 #endif
489 	int slot, tos;
490 
491 	rnode = vm_radix_getroot(rtree);
492 	if (rnode == NULL)
493 		return (NULL);
494 	else if (vm_radix_isleaf(rnode)) {
495 		m = vm_radix_topage(rnode);
496 		if (m->pindex >= index)
497 			return (m);
498 		else
499 			return (NULL);
500 	}
501 	tos = 0;
502 	for (;;) {
503 		/*
504 		 * If the keys differ before the current bisection node,
505 		 * then the search key might rollback to the earliest
506 		 * available bisection node or to the smallest key
507 		 * in the current node (if the owner is bigger than the
508 		 * search key).
509 		 */
510 		if (vm_radix_keybarr(rnode, index)) {
511 			if (index > rnode->rn_owner) {
512 ascend:
513 				KASSERT(++loops < 1000,
514 				    ("vm_radix_lookup_ge: too many loops"));
515 
516 				/*
517 				 * Pop nodes from the stack until either the
518 				 * stack is empty or a node that could have a
519 				 * matching descendant is found.
520 				 */
521 				do {
522 					if (tos == 0)
523 						return (NULL);
524 					rnode = stack[--tos];
525 				} while (vm_radix_slot(index,
526 				    rnode->rn_clev) == (VM_RADIX_COUNT - 1));
527 
528 				/*
529 				 * The following computation cannot overflow
530 				 * because index's slot at the current level
531 				 * is less than VM_RADIX_COUNT - 1.
532 				 */
533 				index = vm_radix_trimkey(index,
534 				    rnode->rn_clev);
535 				index += VM_RADIX_UNITLEVEL(rnode->rn_clev);
536 			} else
537 				index = rnode->rn_owner;
538 			KASSERT(!vm_radix_keybarr(rnode, index),
539 			    ("vm_radix_lookup_ge: keybarr failed"));
540 		}
541 		slot = vm_radix_slot(index, rnode->rn_clev);
542 		child = rnode->rn_child[slot];
543 		if (vm_radix_isleaf(child)) {
544 			m = vm_radix_topage(child);
545 			if (m->pindex >= index)
546 				return (m);
547 		} else if (child != NULL)
548 			goto descend;
549 
550 		/*
551 		 * Look for an available edge or page within the current
552 		 * bisection node.
553 		 */
554                 if (slot < (VM_RADIX_COUNT - 1)) {
555 			inc = VM_RADIX_UNITLEVEL(rnode->rn_clev);
556 			index = vm_radix_trimkey(index, rnode->rn_clev);
557 			do {
558 				index += inc;
559 				slot++;
560 				child = rnode->rn_child[slot];
561 				if (vm_radix_isleaf(child)) {
562 					m = vm_radix_topage(child);
563 					if (m->pindex >= index)
564 						return (m);
565 				} else if (child != NULL)
566 					goto descend;
567 			} while (slot < (VM_RADIX_COUNT - 1));
568 		}
569 		KASSERT(child == NULL || vm_radix_isleaf(child),
570 		    ("vm_radix_lookup_ge: child is radix node"));
571 
572 		/*
573 		 * If a page or edge bigger than the search slot is not found
574 		 * in the current node, ascend to the next higher-level node.
575 		 */
576 		goto ascend;
577 descend:
578 		KASSERT(rnode->rn_clev > 0,
579 		    ("vm_radix_lookup_ge: pushing leaf's parent"));
580 		KASSERT(tos < VM_RADIX_LIMIT,
581 		    ("vm_radix_lookup_ge: stack overflow"));
582 		stack[tos++] = rnode;
583 		rnode = child;
584 	}
585 }
586 
587 /*
588  * Look up the nearest entry at a position less than or equal to index.
589  */
590 vm_page_t
591 vm_radix_lookup_le(struct vm_radix *rtree, vm_pindex_t index)
592 {
593 	struct vm_radix_node *stack[VM_RADIX_LIMIT];
594 	vm_pindex_t inc;
595 	vm_page_t m;
596 	struct vm_radix_node *child, *rnode;
597 #ifdef INVARIANTS
598 	int loops = 0;
599 #endif
600 	int slot, tos;
601 
602 	rnode = vm_radix_getroot(rtree);
603 	if (rnode == NULL)
604 		return (NULL);
605 	else if (vm_radix_isleaf(rnode)) {
606 		m = vm_radix_topage(rnode);
607 		if (m->pindex <= index)
608 			return (m);
609 		else
610 			return (NULL);
611 	}
612 	tos = 0;
613 	for (;;) {
614 		/*
615 		 * If the keys differ before the current bisection node,
616 		 * then the search key might rollback to the earliest
617 		 * available bisection node or to the largest key
618 		 * in the current node (if the owner is smaller than the
619 		 * search key).
620 		 */
621 		if (vm_radix_keybarr(rnode, index)) {
622 			if (index > rnode->rn_owner) {
623 				index = rnode->rn_owner + VM_RADIX_COUNT *
624 				    VM_RADIX_UNITLEVEL(rnode->rn_clev);
625 			} else {
626 ascend:
627 				KASSERT(++loops < 1000,
628 				    ("vm_radix_lookup_le: too many loops"));
629 
630 				/*
631 				 * Pop nodes from the stack until either the
632 				 * stack is empty or a node that could have a
633 				 * matching descendant is found.
634 				 */
635 				do {
636 					if (tos == 0)
637 						return (NULL);
638 					rnode = stack[--tos];
639 				} while (vm_radix_slot(index,
640 				    rnode->rn_clev) == 0);
641 
642 				/*
643 				 * The following computation cannot overflow
644 				 * because index's slot at the current level
645 				 * is greater than 0.
646 				 */
647 				index = vm_radix_trimkey(index,
648 				    rnode->rn_clev);
649 			}
650 			index--;
651 			KASSERT(!vm_radix_keybarr(rnode, index),
652 			    ("vm_radix_lookup_le: keybarr failed"));
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 page or edge smaller than the search slot is not found
687 		 * in the current node, ascend to the next higher-level node.
688 		 */
689 		goto ascend;
690 descend:
691 		KASSERT(rnode->rn_clev > 0,
692 		    ("vm_radix_lookup_le: pushing leaf's parent"));
693 		KASSERT(tos < VM_RADIX_LIMIT,
694 		    ("vm_radix_lookup_le: stack overflow"));
695 		stack[tos++] = rnode;
696 		rnode = child;
697 	}
698 }
699 
700 /*
701  * Remove the specified index from the tree.
702  * Panics if the key is not present.
703  */
704 void
705 vm_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 	/*
712 	 * Detect if a page is going to be removed from a trie which is
713 	 * already undergoing another trie operation.
714 	 * Right now this is only possible for vm_radix_remove() recursing
715 	 * into vm_radix_insert().
716 	 * If this is the case, the caller must be notified about this
717 	 * situation.  It will also takecare to update the RT_TRIE_MODIFIED
718 	 * accordingly.
719 	 * The RT_TRIE_MODIFIED bit is set here because the remove operation
720 	 * will always succeed.
721 	 */
722 	if ((rtree->rt_flags & RT_INSERT_INPROG) != 0)
723 		rtree->rt_flags |= RT_TRIE_MODIFIED;
724 
725 	rnode = vm_radix_getroot(rtree);
726 	if (vm_radix_isleaf(rnode)) {
727 		m = vm_radix_topage(rnode);
728 		if (m->pindex != index)
729 			panic("%s: invalid key found", __func__);
730 		vm_radix_setroot(rtree, NULL);
731 		return;
732 	}
733 	parent = NULL;
734 	for (;;) {
735 		if (rnode == NULL)
736 			panic("vm_radix_remove: impossible to locate the key");
737 		slot = vm_radix_slot(index, rnode->rn_clev);
738 		if (vm_radix_isleaf(rnode->rn_child[slot])) {
739 			m = vm_radix_topage(rnode->rn_child[slot]);
740 			if (m->pindex != index)
741 				panic("%s: invalid key found", __func__);
742 			rnode->rn_child[slot] = NULL;
743 			rnode->rn_count--;
744 			if (rnode->rn_count > 1)
745 				break;
746 			for (i = 0; i < VM_RADIX_COUNT; i++)
747 				if (rnode->rn_child[i] != NULL)
748 					break;
749 			KASSERT(i != VM_RADIX_COUNT,
750 			    ("%s: invalid node configuration", __func__));
751 			if (parent == NULL)
752 				vm_radix_setroot(rtree, rnode->rn_child[i]);
753 			else {
754 				slot = vm_radix_slot(index, parent->rn_clev);
755 				KASSERT(parent->rn_child[slot] == rnode,
756 				    ("%s: invalid child value", __func__));
757 				parent->rn_child[slot] = rnode->rn_child[i];
758 			}
759 			rnode->rn_count--;
760 			rnode->rn_child[i] = NULL;
761 			vm_radix_node_put(rnode);
762 			break;
763 		}
764 		parent = rnode;
765 		rnode = rnode->rn_child[slot];
766 	}
767 }
768 
769 /*
770  * Remove and free all the nodes from the radix tree.
771  * This function is recursive but there is a tight control on it as the
772  * maximum depth of the tree is fixed.
773  */
774 void
775 vm_radix_reclaim_allnodes(struct vm_radix *rtree)
776 {
777 	struct vm_radix_node *root;
778 
779 	KASSERT((rtree->rt_flags & RT_INSERT_INPROG) == 0,
780 	    ("vm_radix_reclaim_allnodes: unexpected trie recursion"));
781 
782 	root = vm_radix_getroot(rtree);
783 	if (root == NULL)
784 		return;
785 	vm_radix_setroot(rtree, NULL);
786 	if (!vm_radix_isleaf(root))
787 		vm_radix_reclaim_allnodes_int(root);
788 }
789 
790 /*
791  * Replace an existing page in the trie with another one.
792  * Panics if there is not an old page in the trie at the new page's index.
793  */
794 vm_page_t
795 vm_radix_replace(struct vm_radix *rtree, vm_page_t newpage)
796 {
797 	struct vm_radix_node *rnode;
798 	vm_page_t m;
799 	vm_pindex_t index;
800 	int slot;
801 
802 	index = newpage->pindex;
803 	rnode = vm_radix_getroot(rtree);
804 	if (rnode == NULL)
805 		panic("%s: replacing page on an empty trie", __func__);
806 	if (vm_radix_isleaf(rnode)) {
807 		m = vm_radix_topage(rnode);
808 		if (m->pindex != index)
809 			panic("%s: original replacing root key not found",
810 			    __func__);
811 		rtree->rt_root = (uintptr_t)newpage | VM_RADIX_ISLEAF;
812 		return (m);
813 	}
814 	for (;;) {
815 		slot = vm_radix_slot(index, rnode->rn_clev);
816 		if (vm_radix_isleaf(rnode->rn_child[slot])) {
817 			m = vm_radix_topage(rnode->rn_child[slot]);
818 			if (m->pindex == index) {
819 				rnode->rn_child[slot] =
820 				    (void *)((uintptr_t)newpage |
821 				    VM_RADIX_ISLEAF);
822 				return (m);
823 			} else
824 				break;
825 		} else if (rnode->rn_child[slot] == NULL ||
826 		    vm_radix_keybarr(rnode->rn_child[slot], index))
827 			break;
828 		rnode = rnode->rn_child[slot];
829 	}
830 	panic("%s: original replacing page not found", __func__);
831 }
832 
833 #ifdef DDB
834 /*
835  * Show details about the given radix node.
836  */
837 DB_SHOW_COMMAND(radixnode, db_show_radixnode)
838 {
839 	struct vm_radix_node *rnode;
840 	int i;
841 
842         if (!have_addr)
843                 return;
844 	rnode = (struct vm_radix_node *)addr;
845 	db_printf("radixnode %p, owner %jx, children count %u, level %u:\n",
846 	    (void *)rnode, (uintmax_t)rnode->rn_owner, rnode->rn_count,
847 	    rnode->rn_clev);
848 	for (i = 0; i < VM_RADIX_COUNT; i++)
849 		if (rnode->rn_child[i] != NULL)
850 			db_printf("slot: %d, val: %p, page: %p, clev: %d\n",
851 			    i, (void *)rnode->rn_child[i],
852 			    vm_radix_isleaf(rnode->rn_child[i]) ?
853 			    vm_radix_topage(rnode->rn_child[i]) : NULL,
854 			    rnode->rn_clev);
855 }
856 #endif /* DDB */
857