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 * Returns 1 if any slot in the node has this tag set.
156 * Otherwise returns 0.
158 static inline int any_tag_set(struct radix_tree_node *node, int tag)
161 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
162 if (node->tags[tag][idx])
169 * Return the maximum key which can be store into a
170 * radix tree with height HEIGHT.
172 static inline unsigned long radix_tree_maxindex(unsigned int height)
174 return height_to_maxindex[height];
178 * Extend a radix tree so it can store key @index.
180 static int radix_tree_extend(struct radix_tree_root *root, unsigned long index)
182 struct radix_tree_node *node;
184 char tags[RADIX_TREE_TAGS];
187 /* Figure out what the height should be. */
188 height = root->height + 1;
189 while (index > radix_tree_maxindex(height))
192 if (root->rnode == NULL) {
193 root->height = height;
198 * Prepare the tag status of the top-level node for propagation
199 * into the newly-pushed top-level node(s)
201 for (tag = 0; tag < RADIX_TREE_TAGS; tag++) {
203 if (any_tag_set(root->rnode, tag))
208 if (!(node = radix_tree_node_alloc(root)))
211 /* Increase the height. */
212 node->slots[0] = root->rnode;
214 /* Propagate the aggregated tag info into the new root */
215 for (tag = 0; tag < RADIX_TREE_TAGS; tag++) {
217 tag_set(node, tag, 0);
223 } while (height > root->height);
229 * radix_tree_insert - insert into a radix tree
230 * @root: radix tree root
232 * @item: item to insert
234 * Insert an item into the radix tree at position @index.
236 int radix_tree_insert(struct radix_tree_root *root,
237 unsigned long index, void *item)
239 struct radix_tree_node *node = NULL, *slot;
240 unsigned int height, shift;
244 /* Make sure the tree is high enough. */
245 if ((!index && !root->rnode) ||
246 index > radix_tree_maxindex(root->height)) {
247 error = radix_tree_extend(root, index);
253 height = root->height;
254 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
256 offset = 0; /* uninitialised var warning */
259 /* Have to add a child node. */
260 if (!(slot = radix_tree_node_alloc(root)))
263 node->slots[offset] = slot;
269 /* Go a level down */
270 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
272 slot = node->slots[offset];
273 shift -= RADIX_TREE_MAP_SHIFT;
282 node->slots[offset] = item;
283 BUG_ON(tag_get(node, 0, offset));
284 BUG_ON(tag_get(node, 1, offset));
290 EXPORT_SYMBOL(radix_tree_insert);
292 static inline void **__lookup_slot(struct radix_tree_root *root,
295 unsigned int height, shift;
296 struct radix_tree_node **slot;
298 height = root->height;
299 if (index > radix_tree_maxindex(height))
302 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
309 slot = (struct radix_tree_node **)
311 ((index >> shift) & RADIX_TREE_MAP_MASK));
312 shift -= RADIX_TREE_MAP_SHIFT;
316 return (void **)slot;
320 * radix_tree_lookup_slot - lookup a slot in a radix tree
321 * @root: radix tree root
324 * Lookup the slot corresponding to the position @index in the radix tree
325 * @root. This is useful for update-if-exists operations.
327 void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
329 return __lookup_slot(root, index);
331 EXPORT_SYMBOL(radix_tree_lookup_slot);
334 * radix_tree_lookup - perform lookup operation on a radix tree
335 * @root: radix tree root
338 * Lookup the item at the position @index in the radix tree @root.
340 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
344 slot = __lookup_slot(root, index);
345 return slot != NULL ? *slot : NULL;
347 EXPORT_SYMBOL(radix_tree_lookup);
350 * radix_tree_tag_set - set a tag on a radix tree node
351 * @root: radix tree root
355 * Set the search tag corresponging to @index in the radix tree. From
356 * the root all the way down to the leaf node.
358 * Returns the address of the tagged item. Setting a tag on a not-present
361 void *radix_tree_tag_set(struct radix_tree_root *root,
362 unsigned long index, int tag)
364 unsigned int height, shift;
365 struct radix_tree_node *slot;
367 height = root->height;
368 if (index > radix_tree_maxindex(height))
371 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
377 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
378 tag_set(slot, tag, offset);
379 slot = slot->slots[offset];
380 BUG_ON(slot == NULL);
381 shift -= RADIX_TREE_MAP_SHIFT;
387 EXPORT_SYMBOL(radix_tree_tag_set);
390 * radix_tree_tag_clear - clear a tag on a radix tree node
391 * @root: radix tree root
395 * Clear the search tag corresponging to @index in the radix tree. If
396 * this causes the leaf node to have no tags set then clear the tag in the
397 * next-to-leaf node, etc.
399 * Returns the address of the tagged item on success, else NULL. ie:
400 * has the same return value and semantics as radix_tree_lookup().
402 void *radix_tree_tag_clear(struct radix_tree_root *root,
403 unsigned long index, int tag)
405 struct radix_tree_path path[RADIX_TREE_MAX_PATH], *pathp = path;
406 struct radix_tree_node *slot;
407 unsigned int height, shift;
410 height = root->height;
411 if (index > radix_tree_maxindex(height))
414 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
424 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
425 pathp[1].offset = offset;
426 pathp[1].node = slot;
427 slot = slot->slots[offset];
429 shift -= RADIX_TREE_MAP_SHIFT;
438 tag_clear(pathp->node, tag, pathp->offset);
439 if (any_tag_set(pathp->node, tag))
442 } while (pathp->node);
446 EXPORT_SYMBOL(radix_tree_tag_clear);
448 #ifndef __KERNEL__ /* Only the test harness uses this at present */
450 * radix_tree_tag_get - get a tag on a radix tree node
451 * @root: radix tree root
458 * 1: tag present, set
459 * -1: tag present, unset
461 int radix_tree_tag_get(struct radix_tree_root *root,
462 unsigned long index, int tag)
464 unsigned int height, shift;
465 struct radix_tree_node *slot;
466 int saw_unset_tag = 0;
468 height = root->height;
469 if (index > radix_tree_maxindex(height))
472 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
481 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
484 * This is just a debug check. Later, we can bale as soon as
485 * we see an unset tag.
487 if (!tag_get(slot, tag, offset))
490 int ret = tag_get(slot, tag, offset);
492 BUG_ON(ret && saw_unset_tag);
495 slot = slot->slots[offset];
496 shift -= RADIX_TREE_MAP_SHIFT;
500 EXPORT_SYMBOL(radix_tree_tag_get);
504 __lookup(struct radix_tree_root *root, void **results, unsigned long index,
505 unsigned int max_items, unsigned long *next_index)
507 unsigned int nr_found = 0;
508 unsigned int shift, height;
509 struct radix_tree_node *slot;
512 height = root->height;
516 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
519 for ( ; height > 1; height--) {
521 for (i = (index >> shift) & RADIX_TREE_MAP_MASK ;
522 i < RADIX_TREE_MAP_SIZE; i++) {
523 if (slot->slots[i] != NULL)
525 index &= ~((1UL << shift) - 1);
526 index += 1UL << shift;
528 goto out; /* 32-bit wraparound */
530 if (i == RADIX_TREE_MAP_SIZE)
533 shift -= RADIX_TREE_MAP_SHIFT;
534 slot = slot->slots[i];
537 /* Bottom level: grab some items */
538 for (i = index & RADIX_TREE_MAP_MASK; i < RADIX_TREE_MAP_SIZE; i++) {
540 if (slot->slots[i]) {
541 results[nr_found++] = slot->slots[i];
542 if (nr_found == max_items)
552 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
553 * @root: radix tree root
554 * @results: where the results of the lookup are placed
555 * @first_index: start the lookup from this key
556 * @max_items: place up to this many items at *results
558 * Performs an index-ascending scan of the tree for present items. Places
559 * them at *@results and returns the number of items which were placed at
562 * The implementation is naive.
565 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
566 unsigned long first_index, unsigned int max_items)
568 const unsigned long max_index = radix_tree_maxindex(root->height);
569 unsigned long cur_index = first_index;
570 unsigned int ret = 0;
572 while (ret < max_items) {
573 unsigned int nr_found;
574 unsigned long next_index; /* Index of next search */
576 if (cur_index > max_index)
578 nr_found = __lookup(root, results + ret, cur_index,
579 max_items - ret, &next_index);
583 cur_index = next_index;
587 EXPORT_SYMBOL(radix_tree_gang_lookup);
590 * FIXME: the two tag_get()s here should use find_next_bit() instead of
591 * open-coding the search.
594 __lookup_tag(struct radix_tree_root *root, void **results, unsigned long index,
595 unsigned int max_items, unsigned long *next_index, int tag)
597 unsigned int nr_found = 0;
599 unsigned int height = root->height;
600 struct radix_tree_node *slot;
602 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
606 unsigned long i = (index >> shift) & RADIX_TREE_MAP_MASK;
608 for ( ; i < RADIX_TREE_MAP_SIZE; i++) {
609 if (tag_get(slot, tag, i)) {
610 BUG_ON(slot->slots[i] == NULL);
613 index &= ~((1UL << shift) - 1);
614 index += 1UL << shift;
616 goto out; /* 32-bit wraparound */
618 if (i == RADIX_TREE_MAP_SIZE)
621 if (height == 0) { /* Bottom level: grab some items */
622 unsigned long j = index & RADIX_TREE_MAP_MASK;
624 for ( ; j < RADIX_TREE_MAP_SIZE; j++) {
626 if (tag_get(slot, tag, j)) {
627 BUG_ON(slot->slots[j] == NULL);
628 results[nr_found++] = slot->slots[j];
629 if (nr_found == max_items)
634 shift -= RADIX_TREE_MAP_SHIFT;
635 slot = slot->slots[i];
643 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
645 * @root: radix tree root
646 * @results: where the results of the lookup are placed
647 * @first_index: start the lookup from this key
648 * @max_items: place up to this many items at *results
649 * @tag: the tag index
651 * Performs an index-ascending scan of the tree for present items which
652 * have the tag indexed by @tag set. Places the items at *@results and
653 * returns the number of items which were placed at *@results.
656 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
657 unsigned long first_index, unsigned int max_items, int tag)
659 const unsigned long max_index = radix_tree_maxindex(root->height);
660 unsigned long cur_index = first_index;
661 unsigned int ret = 0;
663 while (ret < max_items) {
664 unsigned int nr_found;
665 unsigned long next_index; /* Index of next search */
667 if (cur_index > max_index)
669 nr_found = __lookup_tag(root, results + ret, cur_index,
670 max_items - ret, &next_index, tag);
674 cur_index = next_index;
678 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
681 * radix_tree_delete - delete an item from a radix tree
682 * @root: radix tree root
685 * Remove the item at @index from the radix tree rooted at @root.
687 * Returns the address of the deleted item, or NULL if it was not present.
689 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
691 struct radix_tree_path path[RADIX_TREE_MAX_PATH], *pathp = path;
692 struct radix_tree_path *orig_pathp;
693 struct radix_tree_node *slot;
694 unsigned int height, shift;
696 char tags[RADIX_TREE_TAGS];
699 height = root->height;
700 if (index > radix_tree_maxindex(height))
703 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
707 for ( ; height > 0; height--) {
713 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
714 pathp[1].offset = offset;
715 pathp[1].node = slot;
716 slot = slot->slots[offset];
718 shift -= RADIX_TREE_MAP_SHIFT;
728 * Clear all tags associated with the just-deleted item
730 memset(tags, 0, sizeof(tags));
734 nr_cleared_tags = RADIX_TREE_TAGS;
735 for (tag = 0; tag < RADIX_TREE_TAGS; tag++) {
739 tag_clear(pathp->node, tag, pathp->offset);
741 if (any_tag_set(pathp->node, tag)) {
747 } while (pathp->node && nr_cleared_tags);
749 /* Now free the nodes we do not need anymore */
750 for (pathp = orig_pathp; pathp->node; pathp--) {
751 pathp->node->slots[pathp->offset] = NULL;
752 if (--pathp->node->count)
755 /* Node with zero slots in use so free it */
756 radix_tree_node_free(pathp->node);
763 EXPORT_SYMBOL(radix_tree_delete);
766 * radix_tree_tagged - test whether any items in the tree are tagged
767 * @root: radix tree root
770 int radix_tree_tagged(struct radix_tree_root *root, int tag)
772 struct radix_tree_node *rnode;
776 return any_tag_set(rnode, tag);
778 EXPORT_SYMBOL(radix_tree_tagged);
781 radix_tree_node_ctor(void *node, kmem_cache_t *cachep, unsigned long flags)
783 memset(node, 0, sizeof(struct radix_tree_node));
786 static __init unsigned long __maxindex(unsigned int height)
788 unsigned int tmp = height * RADIX_TREE_MAP_SHIFT;
789 unsigned long index = (~0UL >> (RADIX_TREE_INDEX_BITS - tmp - 1)) >> 1;
791 if (tmp >= RADIX_TREE_INDEX_BITS)
796 static __init void radix_tree_init_maxindex(void)
800 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
801 height_to_maxindex[i] = __maxindex(i);
804 #ifdef CONFIG_HOTPLUG_CPU
805 static int radix_tree_callback(struct notifier_block *nfb,
806 unsigned long action,
809 int cpu = (long)hcpu;
810 struct radix_tree_preload *rtp;
812 /* Free per-cpu pool of perloaded nodes */
813 if (action == CPU_DEAD) {
814 rtp = &per_cpu(radix_tree_preloads, cpu);
816 kmem_cache_free(radix_tree_node_cachep,
817 rtp->nodes[rtp->nr-1]);
818 rtp->nodes[rtp->nr-1] = NULL;
824 #endif /* CONFIG_HOTPLUG_CPU */
826 void __init radix_tree_init(void)
828 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
829 sizeof(struct radix_tree_node), 0,
830 SLAB_PANIC, radix_tree_node_ctor, NULL);
831 radix_tree_init_maxindex();
832 hotcpu_notifier(radix_tree_callback, 0);