2 * Copyright (C) 2001 Momchil Velikov
3 * Portions Copyright (C) 2001 Christoph Hellwig
4 * Copyright (C) 2005 SGI, Christoph Lameter <clameter@sgi.com>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation; either version 2, or (at
9 * your option) any later version.
11 * This program is distributed in the hope that it will be useful, but
12 * WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 #include <linux/errno.h>
22 #include <linux/init.h>
23 #include <linux/kernel.h>
24 #include <linux/module.h>
25 #include <linux/radix-tree.h>
26 #include <linux/percpu.h>
27 #include <linux/slab.h>
28 #include <linux/notifier.h>
29 #include <linux/cpu.h>
30 #include <linux/gfp.h>
31 #include <linux/string.h>
32 #include <linux/bitops.h>
36 #define RADIX_TREE_MAP_SHIFT 6
38 #define RADIX_TREE_MAP_SHIFT 3 /* For more stressful testing */
40 #define RADIX_TREE_TAGS 2
42 #define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT)
43 #define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1)
45 #define RADIX_TREE_TAG_LONGS \
46 ((RADIX_TREE_MAP_SIZE + BITS_PER_LONG - 1) / BITS_PER_LONG)
48 struct radix_tree_node {
50 void *slots[RADIX_TREE_MAP_SIZE];
51 unsigned long tags[RADIX_TREE_TAGS][RADIX_TREE_TAG_LONGS];
54 struct radix_tree_path {
55 struct radix_tree_node *node;
59 #define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long))
60 #define RADIX_TREE_MAX_PATH (RADIX_TREE_INDEX_BITS/RADIX_TREE_MAP_SHIFT + 2)
62 static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH] __read_mostly;
65 * Radix tree node cache.
67 static kmem_cache_t *radix_tree_node_cachep;
70 * Per-cpu pool of preloaded nodes
72 struct radix_tree_preload {
74 struct radix_tree_node *nodes[RADIX_TREE_MAX_PATH];
76 DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
79 * This assumes that the caller has performed appropriate preallocation, and
80 * that the caller has pinned this thread of control to the current CPU.
82 static struct radix_tree_node *
83 radix_tree_node_alloc(struct radix_tree_root *root)
85 struct radix_tree_node *ret;
87 ret = kmem_cache_alloc(radix_tree_node_cachep, root->gfp_mask);
88 if (ret == NULL && !(root->gfp_mask & __GFP_WAIT)) {
89 struct radix_tree_preload *rtp;
91 rtp = &__get_cpu_var(radix_tree_preloads);
93 ret = rtp->nodes[rtp->nr - 1];
94 rtp->nodes[rtp->nr - 1] = NULL;
102 radix_tree_node_free(struct radix_tree_node *node)
104 kmem_cache_free(radix_tree_node_cachep, node);
108 * Load up this CPU's radix_tree_node buffer with sufficient objects to
109 * ensure that the addition of a single element in the tree cannot fail. On
110 * success, return zero, with preemption disabled. On error, return -ENOMEM
111 * with preemption not disabled.
113 int radix_tree_preload(gfp_t gfp_mask)
115 struct radix_tree_preload *rtp;
116 struct radix_tree_node *node;
120 rtp = &__get_cpu_var(radix_tree_preloads);
121 while (rtp->nr < ARRAY_SIZE(rtp->nodes)) {
123 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
127 rtp = &__get_cpu_var(radix_tree_preloads);
128 if (rtp->nr < ARRAY_SIZE(rtp->nodes))
129 rtp->nodes[rtp->nr++] = node;
131 kmem_cache_free(radix_tree_node_cachep, node);
138 static inline void tag_set(struct radix_tree_node *node, int tag, int offset)
140 if (!test_bit(offset, &node->tags[tag][0]))
141 __set_bit(offset, &node->tags[tag][0]);
144 static inline void tag_clear(struct radix_tree_node *node, int tag, int offset)
146 __clear_bit(offset, &node->tags[tag][0]);
149 static inline int tag_get(struct radix_tree_node *node, int tag, int offset)
151 return test_bit(offset, &node->tags[tag][0]);
155 * Return the maximum key which can be store into a
156 * radix tree with height HEIGHT.
158 static inline unsigned long radix_tree_maxindex(unsigned int height)
160 return height_to_maxindex[height];
164 * Extend a radix tree so it can store key @index.
166 static int radix_tree_extend(struct radix_tree_root *root, unsigned long index)
168 struct radix_tree_node *node;
170 char tags[RADIX_TREE_TAGS];
173 /* Figure out what the height should be. */
174 height = root->height + 1;
175 while (index > radix_tree_maxindex(height))
178 if (root->rnode == NULL) {
179 root->height = height;
184 * Prepare the tag status of the top-level node for propagation
185 * into the newly-pushed top-level node(s)
187 for (tag = 0; tag < RADIX_TREE_TAGS; tag++) {
191 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
192 if (root->rnode->tags[tag][idx]) {
200 if (!(node = radix_tree_node_alloc(root)))
203 /* Increase the height. */
204 node->slots[0] = root->rnode;
206 /* Propagate the aggregated tag info into the new root */
207 for (tag = 0; tag < RADIX_TREE_TAGS; tag++) {
209 tag_set(node, tag, 0);
215 } while (height > root->height);
221 * radix_tree_insert - insert into a radix tree
222 * @root: radix tree root
224 * @item: item to insert
226 * Insert an item into the radix tree at position @index.
228 int radix_tree_insert(struct radix_tree_root *root,
229 unsigned long index, void *item)
231 struct radix_tree_node *node = NULL, *slot;
232 unsigned int height, shift;
236 /* Make sure the tree is high enough. */
237 if ((!index && !root->rnode) ||
238 index > radix_tree_maxindex(root->height)) {
239 error = radix_tree_extend(root, index);
245 height = root->height;
246 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
248 offset = 0; /* uninitialised var warning */
251 /* Have to add a child node. */
252 if (!(slot = radix_tree_node_alloc(root)))
255 node->slots[offset] = slot;
261 /* Go a level down */
262 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
264 slot = node->slots[offset];
265 shift -= RADIX_TREE_MAP_SHIFT;
274 node->slots[offset] = item;
275 BUG_ON(tag_get(node, 0, offset));
276 BUG_ON(tag_get(node, 1, offset));
282 EXPORT_SYMBOL(radix_tree_insert);
284 static inline void **__lookup_slot(struct radix_tree_root *root,
287 unsigned int height, shift;
288 struct radix_tree_node **slot;
290 height = root->height;
291 if (index > radix_tree_maxindex(height))
294 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
301 slot = (struct radix_tree_node **)
303 ((index >> shift) & RADIX_TREE_MAP_MASK));
304 shift -= RADIX_TREE_MAP_SHIFT;
308 return (void **)slot;
312 * radix_tree_lookup_slot - lookup a slot in a radix tree
313 * @root: radix tree root
316 * Lookup the slot corresponding to the position @index in the radix tree
317 * @root. This is useful for update-if-exists operations.
319 void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
321 return __lookup_slot(root, index);
323 EXPORT_SYMBOL(radix_tree_lookup_slot);
326 * radix_tree_lookup - perform lookup operation on a radix tree
327 * @root: radix tree root
330 * Lookup the item at the position @index in the radix tree @root.
332 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
336 slot = __lookup_slot(root, index);
337 return slot != NULL ? *slot : NULL;
339 EXPORT_SYMBOL(radix_tree_lookup);
342 * radix_tree_tag_set - set a tag on a radix tree node
343 * @root: radix tree root
347 * Set the search tag corresponging to @index in the radix tree. From
348 * the root all the way down to the leaf node.
350 * Returns the address of the tagged item. Setting a tag on a not-present
353 void *radix_tree_tag_set(struct radix_tree_root *root,
354 unsigned long index, int tag)
356 unsigned int height, shift;
357 struct radix_tree_node *slot;
359 height = root->height;
360 if (index > radix_tree_maxindex(height))
363 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
369 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
370 tag_set(slot, tag, offset);
371 slot = slot->slots[offset];
372 BUG_ON(slot == NULL);
373 shift -= RADIX_TREE_MAP_SHIFT;
379 EXPORT_SYMBOL(radix_tree_tag_set);
382 * radix_tree_tag_clear - clear a tag on a radix tree node
383 * @root: radix tree root
387 * Clear the search tag corresponging to @index in the radix tree. If
388 * this causes the leaf node to have no tags set then clear the tag in the
389 * next-to-leaf node, etc.
391 * Returns the address of the tagged item on success, else NULL. ie:
392 * has the same return value and semantics as radix_tree_lookup().
394 void *radix_tree_tag_clear(struct radix_tree_root *root,
395 unsigned long index, int tag)
397 struct radix_tree_path path[RADIX_TREE_MAX_PATH], *pathp = path;
398 struct radix_tree_node *slot;
399 unsigned int height, shift;
402 height = root->height;
403 if (index > radix_tree_maxindex(height))
406 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
416 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
417 pathp[1].offset = offset;
418 pathp[1].node = slot;
419 slot = slot->slots[offset];
421 shift -= RADIX_TREE_MAP_SHIFT;
432 tag_clear(pathp->node, tag, pathp->offset);
433 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
434 if (pathp->node->tags[tag][idx])
438 } while (pathp->node);
442 EXPORT_SYMBOL(radix_tree_tag_clear);
444 #ifndef __KERNEL__ /* Only the test harness uses this at present */
446 * radix_tree_tag_get - get a tag on a radix tree node
447 * @root: radix tree root
454 * 1: tag present, set
455 * -1: tag present, unset
457 int radix_tree_tag_get(struct radix_tree_root *root,
458 unsigned long index, int tag)
460 unsigned int height, shift;
461 struct radix_tree_node *slot;
462 int saw_unset_tag = 0;
464 height = root->height;
465 if (index > radix_tree_maxindex(height))
468 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
477 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
480 * This is just a debug check. Later, we can bale as soon as
481 * we see an unset tag.
483 if (!tag_get(slot, tag, offset))
486 int ret = tag_get(slot, tag, offset);
488 BUG_ON(ret && saw_unset_tag);
491 slot = slot->slots[offset];
492 shift -= RADIX_TREE_MAP_SHIFT;
496 EXPORT_SYMBOL(radix_tree_tag_get);
500 __lookup(struct radix_tree_root *root, void **results, unsigned long index,
501 unsigned int max_items, unsigned long *next_index)
503 unsigned int nr_found = 0;
504 unsigned int shift, height;
505 struct radix_tree_node *slot;
508 height = root->height;
512 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
515 for ( ; height > 1; height--) {
517 for (i = (index >> shift) & RADIX_TREE_MAP_MASK ;
518 i < RADIX_TREE_MAP_SIZE; i++) {
519 if (slot->slots[i] != NULL)
521 index &= ~((1UL << shift) - 1);
522 index += 1UL << shift;
524 goto out; /* 32-bit wraparound */
526 if (i == RADIX_TREE_MAP_SIZE)
529 shift -= RADIX_TREE_MAP_SHIFT;
530 slot = slot->slots[i];
533 /* Bottom level: grab some items */
534 for (i = index & RADIX_TREE_MAP_MASK; i < RADIX_TREE_MAP_SIZE; i++) {
536 if (slot->slots[i]) {
537 results[nr_found++] = slot->slots[i];
538 if (nr_found == max_items)
548 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
549 * @root: radix tree root
550 * @results: where the results of the lookup are placed
551 * @first_index: start the lookup from this key
552 * @max_items: place up to this many items at *results
554 * Performs an index-ascending scan of the tree for present items. Places
555 * them at *@results and returns the number of items which were placed at
558 * The implementation is naive.
561 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
562 unsigned long first_index, unsigned int max_items)
564 const unsigned long max_index = radix_tree_maxindex(root->height);
565 unsigned long cur_index = first_index;
566 unsigned int ret = 0;
568 while (ret < max_items) {
569 unsigned int nr_found;
570 unsigned long next_index; /* Index of next search */
572 if (cur_index > max_index)
574 nr_found = __lookup(root, results + ret, cur_index,
575 max_items - ret, &next_index);
579 cur_index = next_index;
583 EXPORT_SYMBOL(radix_tree_gang_lookup);
586 * FIXME: the two tag_get()s here should use find_next_bit() instead of
587 * open-coding the search.
590 __lookup_tag(struct radix_tree_root *root, void **results, unsigned long index,
591 unsigned int max_items, unsigned long *next_index, int tag)
593 unsigned int nr_found = 0;
595 unsigned int height = root->height;
596 struct radix_tree_node *slot;
598 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
602 unsigned long i = (index >> shift) & RADIX_TREE_MAP_MASK;
604 for ( ; i < RADIX_TREE_MAP_SIZE; i++) {
605 if (tag_get(slot, tag, i)) {
606 BUG_ON(slot->slots[i] == NULL);
609 index &= ~((1UL << shift) - 1);
610 index += 1UL << shift;
612 goto out; /* 32-bit wraparound */
614 if (i == RADIX_TREE_MAP_SIZE)
617 if (height == 0) { /* Bottom level: grab some items */
618 unsigned long j = index & RADIX_TREE_MAP_MASK;
620 for ( ; j < RADIX_TREE_MAP_SIZE; j++) {
622 if (tag_get(slot, tag, j)) {
623 BUG_ON(slot->slots[j] == NULL);
624 results[nr_found++] = slot->slots[j];
625 if (nr_found == max_items)
630 shift -= RADIX_TREE_MAP_SHIFT;
631 slot = slot->slots[i];
639 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
641 * @root: radix tree root
642 * @results: where the results of the lookup are placed
643 * @first_index: start the lookup from this key
644 * @max_items: place up to this many items at *results
645 * @tag: the tag index
647 * Performs an index-ascending scan of the tree for present items which
648 * have the tag indexed by @tag set. Places the items at *@results and
649 * returns the number of items which were placed at *@results.
652 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
653 unsigned long first_index, unsigned int max_items, int tag)
655 const unsigned long max_index = radix_tree_maxindex(root->height);
656 unsigned long cur_index = first_index;
657 unsigned int ret = 0;
659 while (ret < max_items) {
660 unsigned int nr_found;
661 unsigned long next_index; /* Index of next search */
663 if (cur_index > max_index)
665 nr_found = __lookup_tag(root, results + ret, cur_index,
666 max_items - ret, &next_index, tag);
670 cur_index = next_index;
674 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
677 * radix_tree_delete - delete an item from a radix tree
678 * @root: radix tree root
681 * Remove the item at @index from the radix tree rooted at @root.
683 * Returns the address of the deleted item, or NULL if it was not present.
685 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
687 struct radix_tree_path path[RADIX_TREE_MAX_PATH], *pathp = path;
688 struct radix_tree_path *orig_pathp;
689 struct radix_tree_node *slot;
690 unsigned int height, shift;
692 char tags[RADIX_TREE_TAGS];
695 height = root->height;
696 if (index > radix_tree_maxindex(height))
699 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
703 for ( ; height > 0; height--) {
709 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
710 pathp[1].offset = offset;
711 pathp[1].node = slot;
712 slot = slot->slots[offset];
714 shift -= RADIX_TREE_MAP_SHIFT;
724 * Clear all tags associated with the just-deleted item
726 memset(tags, 0, sizeof(tags));
730 nr_cleared_tags = RADIX_TREE_TAGS;
731 for (tag = 0; tag < RADIX_TREE_TAGS; tag++) {
737 tag_clear(pathp->node, tag, pathp->offset);
739 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
740 if (pathp->node->tags[tag][idx]) {
748 } while (pathp->node && nr_cleared_tags);
750 /* Now free the nodes we do not need anymore */
751 for (pathp = orig_pathp; pathp->node; pathp--) {
752 pathp->node->slots[pathp->offset] = NULL;
753 if (--pathp->node->count)
756 /* Node with zero slots in use so free it */
757 radix_tree_node_free(pathp->node);
764 EXPORT_SYMBOL(radix_tree_delete);
767 * radix_tree_tagged - test whether any items in the tree are tagged
768 * @root: radix tree root
771 int radix_tree_tagged(struct radix_tree_root *root, int tag)
777 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
778 if (root->rnode->tags[tag][idx])
783 EXPORT_SYMBOL(radix_tree_tagged);
786 radix_tree_node_ctor(void *node, kmem_cache_t *cachep, unsigned long flags)
788 memset(node, 0, sizeof(struct radix_tree_node));
791 static __init unsigned long __maxindex(unsigned int height)
793 unsigned int tmp = height * RADIX_TREE_MAP_SHIFT;
794 unsigned long index = (~0UL >> (RADIX_TREE_INDEX_BITS - tmp - 1)) >> 1;
796 if (tmp >= RADIX_TREE_INDEX_BITS)
801 static __init void radix_tree_init_maxindex(void)
805 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
806 height_to_maxindex[i] = __maxindex(i);
809 #ifdef CONFIG_HOTPLUG_CPU
810 static int radix_tree_callback(struct notifier_block *nfb,
811 unsigned long action,
814 int cpu = (long)hcpu;
815 struct radix_tree_preload *rtp;
817 /* Free per-cpu pool of perloaded nodes */
818 if (action == CPU_DEAD) {
819 rtp = &per_cpu(radix_tree_preloads, cpu);
821 kmem_cache_free(radix_tree_node_cachep,
822 rtp->nodes[rtp->nr-1]);
823 rtp->nodes[rtp->nr-1] = NULL;
829 #endif /* CONFIG_HOTPLUG_CPU */
831 void __init radix_tree_init(void)
833 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
834 sizeof(struct radix_tree_node), 0,
835 SLAB_PANIC, radix_tree_node_ctor, NULL);
836 radix_tree_init_maxindex();
837 hotcpu_notifier(radix_tree_callback, 0);