Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001 Momchil Velikov | |
3 | * Portions Copyright (C) 2001 Christoph Hellwig | |
cde53535 | 4 | * Copyright (C) 2005 SGI, Christoph Lameter |
7cf9c2c7 | 5 | * Copyright (C) 2006 Nick Piggin |
78c1d784 | 6 | * Copyright (C) 2012 Konstantin Khlebnikov |
1da177e4 LT |
7 | * |
8 | * This program is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU General Public License as | |
10 | * published by the Free Software Foundation; either version 2, or (at | |
11 | * your option) any later version. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, but | |
14 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
16 | * General Public License for more details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public License | |
19 | * along with this program; if not, write to the Free Software | |
20 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
21 | */ | |
22 | ||
23 | #include <linux/errno.h> | |
24 | #include <linux/init.h> | |
25 | #include <linux/kernel.h> | |
8bc3bcc9 | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #include <linux/radix-tree.h> |
28 | #include <linux/percpu.h> | |
29 | #include <linux/slab.h> | |
ce80b067 | 30 | #include <linux/kmemleak.h> |
1da177e4 LT |
31 | #include <linux/notifier.h> |
32 | #include <linux/cpu.h> | |
1da177e4 LT |
33 | #include <linux/string.h> |
34 | #include <linux/bitops.h> | |
7cf9c2c7 | 35 | #include <linux/rcupdate.h> |
92cf2118 | 36 | #include <linux/preempt.h> /* in_interrupt() */ |
1da177e4 LT |
37 | |
38 | ||
26fb1589 JM |
39 | /* |
40 | * The height_to_maxindex array needs to be one deeper than the maximum | |
41 | * path as height 0 holds only 1 entry. | |
42 | */ | |
43 | static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly; | |
1da177e4 LT |
44 | |
45 | /* | |
46 | * Radix tree node cache. | |
47 | */ | |
e18b890b | 48 | static struct kmem_cache *radix_tree_node_cachep; |
1da177e4 | 49 | |
55368052 NP |
50 | /* |
51 | * The radix tree is variable-height, so an insert operation not only has | |
52 | * to build the branch to its corresponding item, it also has to build the | |
53 | * branch to existing items if the size has to be increased (by | |
54 | * radix_tree_extend). | |
55 | * | |
56 | * The worst case is a zero height tree with just a single item at index 0, | |
57 | * and then inserting an item at index ULONG_MAX. This requires 2 new branches | |
58 | * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared. | |
59 | * Hence: | |
60 | */ | |
61 | #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1) | |
62 | ||
1da177e4 LT |
63 | /* |
64 | * Per-cpu pool of preloaded nodes | |
65 | */ | |
66 | struct radix_tree_preload { | |
67 | int nr; | |
9d2a8da0 KS |
68 | /* nodes->private_data points to next preallocated node */ |
69 | struct radix_tree_node *nodes; | |
1da177e4 | 70 | }; |
8cef7d57 | 71 | static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, }; |
1da177e4 | 72 | |
27d20fdd NP |
73 | static inline void *ptr_to_indirect(void *ptr) |
74 | { | |
75 | return (void *)((unsigned long)ptr | RADIX_TREE_INDIRECT_PTR); | |
76 | } | |
77 | ||
78 | static inline void *indirect_to_ptr(void *ptr) | |
79 | { | |
80 | return (void *)((unsigned long)ptr & ~RADIX_TREE_INDIRECT_PTR); | |
81 | } | |
82 | ||
612d6c19 NP |
83 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
84 | { | |
85 | return root->gfp_mask & __GFP_BITS_MASK; | |
86 | } | |
87 | ||
643b52b9 NP |
88 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
89 | int offset) | |
90 | { | |
91 | __set_bit(offset, node->tags[tag]); | |
92 | } | |
93 | ||
94 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
95 | int offset) | |
96 | { | |
97 | __clear_bit(offset, node->tags[tag]); | |
98 | } | |
99 | ||
100 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, | |
101 | int offset) | |
102 | { | |
103 | return test_bit(offset, node->tags[tag]); | |
104 | } | |
105 | ||
106 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) | |
107 | { | |
108 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
109 | } | |
110 | ||
111 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag) | |
112 | { | |
113 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
114 | } | |
115 | ||
116 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
117 | { | |
118 | root->gfp_mask &= __GFP_BITS_MASK; | |
119 | } | |
120 | ||
121 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) | |
122 | { | |
123 | return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); | |
124 | } | |
125 | ||
126 | /* | |
127 | * Returns 1 if any slot in the node has this tag set. | |
128 | * Otherwise returns 0. | |
129 | */ | |
130 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) | |
131 | { | |
132 | int idx; | |
133 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { | |
134 | if (node->tags[tag][idx]) | |
135 | return 1; | |
136 | } | |
137 | return 0; | |
138 | } | |
78c1d784 KK |
139 | |
140 | /** | |
141 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
142 | * | |
143 | * @addr: The address to base the search on | |
144 | * @size: The bitmap size in bits | |
145 | * @offset: The bitnumber to start searching at | |
146 | * | |
147 | * Unrollable variant of find_next_bit() for constant size arrays. | |
148 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
149 | * Returns next bit offset, or size if nothing found. | |
150 | */ | |
151 | static __always_inline unsigned long | |
152 | radix_tree_find_next_bit(const unsigned long *addr, | |
153 | unsigned long size, unsigned long offset) | |
154 | { | |
155 | if (!__builtin_constant_p(size)) | |
156 | return find_next_bit(addr, size, offset); | |
157 | ||
158 | if (offset < size) { | |
159 | unsigned long tmp; | |
160 | ||
161 | addr += offset / BITS_PER_LONG; | |
162 | tmp = *addr >> (offset % BITS_PER_LONG); | |
163 | if (tmp) | |
164 | return __ffs(tmp) + offset; | |
165 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
166 | while (offset < size) { | |
167 | tmp = *++addr; | |
168 | if (tmp) | |
169 | return __ffs(tmp) + offset; | |
170 | offset += BITS_PER_LONG; | |
171 | } | |
172 | } | |
173 | return size; | |
174 | } | |
175 | ||
1da177e4 LT |
176 | /* |
177 | * This assumes that the caller has performed appropriate preallocation, and | |
178 | * that the caller has pinned this thread of control to the current CPU. | |
179 | */ | |
180 | static struct radix_tree_node * | |
181 | radix_tree_node_alloc(struct radix_tree_root *root) | |
182 | { | |
e2848a0e | 183 | struct radix_tree_node *ret = NULL; |
612d6c19 | 184 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 185 | |
5e4c0d97 JK |
186 | /* |
187 | * Preload code isn't irq safe and it doesn't make sence to use | |
188 | * preloading in the interrupt anyway as all the allocations have to | |
189 | * be atomic. So just do normal allocation when in interrupt. | |
190 | */ | |
d0164adc | 191 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
192 | struct radix_tree_preload *rtp; |
193 | ||
58e698af VD |
194 | /* |
195 | * Even if the caller has preloaded, try to allocate from the | |
196 | * cache first for the new node to get accounted. | |
197 | */ | |
198 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
199 | gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN); | |
200 | if (ret) | |
201 | goto out; | |
202 | ||
e2848a0e NP |
203 | /* |
204 | * Provided the caller has preloaded here, we will always | |
205 | * succeed in getting a node here (and never reach | |
206 | * kmem_cache_alloc) | |
207 | */ | |
7c8e0181 | 208 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 209 | if (rtp->nr) { |
9d2a8da0 KS |
210 | ret = rtp->nodes; |
211 | rtp->nodes = ret->private_data; | |
212 | ret->private_data = NULL; | |
1da177e4 LT |
213 | rtp->nr--; |
214 | } | |
ce80b067 CM |
215 | /* |
216 | * Update the allocation stack trace as this is more useful | |
217 | * for debugging. | |
218 | */ | |
219 | kmemleak_update_trace(ret); | |
58e698af | 220 | goto out; |
1da177e4 | 221 | } |
58e698af VD |
222 | ret = kmem_cache_alloc(radix_tree_node_cachep, |
223 | gfp_mask | __GFP_ACCOUNT); | |
224 | out: | |
c0bc9875 | 225 | BUG_ON(radix_tree_is_indirect_ptr(ret)); |
1da177e4 LT |
226 | return ret; |
227 | } | |
228 | ||
7cf9c2c7 NP |
229 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
230 | { | |
231 | struct radix_tree_node *node = | |
232 | container_of(head, struct radix_tree_node, rcu_head); | |
b6dd0865 | 233 | int i; |
643b52b9 NP |
234 | |
235 | /* | |
236 | * must only free zeroed nodes into the slab. radix_tree_shrink | |
237 | * can leave us with a non-NULL entry in the first slot, so clear | |
238 | * that here to make sure. | |
239 | */ | |
b6dd0865 DC |
240 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
241 | tag_clear(node, i, 0); | |
242 | ||
643b52b9 NP |
243 | node->slots[0] = NULL; |
244 | node->count = 0; | |
245 | ||
7cf9c2c7 NP |
246 | kmem_cache_free(radix_tree_node_cachep, node); |
247 | } | |
248 | ||
1da177e4 LT |
249 | static inline void |
250 | radix_tree_node_free(struct radix_tree_node *node) | |
251 | { | |
7cf9c2c7 | 252 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
253 | } |
254 | ||
255 | /* | |
256 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
257 | * ensure that the addition of a single element in the tree cannot fail. On | |
258 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
259 | * with preemption not disabled. | |
b34df792 DH |
260 | * |
261 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 262 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 263 | */ |
5e4c0d97 | 264 | static int __radix_tree_preload(gfp_t gfp_mask) |
1da177e4 LT |
265 | { |
266 | struct radix_tree_preload *rtp; | |
267 | struct radix_tree_node *node; | |
268 | int ret = -ENOMEM; | |
269 | ||
270 | preempt_disable(); | |
7c8e0181 | 271 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 | 272 | while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
1da177e4 | 273 | preempt_enable(); |
488514d1 | 274 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
275 | if (node == NULL) |
276 | goto out; | |
277 | preempt_disable(); | |
7c8e0181 | 278 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 KS |
279 | if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
280 | node->private_data = rtp->nodes; | |
281 | rtp->nodes = node; | |
282 | rtp->nr++; | |
283 | } else { | |
1da177e4 | 284 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 285 | } |
1da177e4 LT |
286 | } |
287 | ret = 0; | |
288 | out: | |
289 | return ret; | |
290 | } | |
5e4c0d97 JK |
291 | |
292 | /* | |
293 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
294 | * ensure that the addition of a single element in the tree cannot fail. On | |
295 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
296 | * with preemption not disabled. | |
297 | * | |
298 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 299 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
300 | */ |
301 | int radix_tree_preload(gfp_t gfp_mask) | |
302 | { | |
303 | /* Warn on non-sensical use... */ | |
d0164adc | 304 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
5e4c0d97 JK |
305 | return __radix_tree_preload(gfp_mask); |
306 | } | |
d7f0923d | 307 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 308 | |
5e4c0d97 JK |
309 | /* |
310 | * The same as above function, except we don't guarantee preloading happens. | |
311 | * We do it, if we decide it helps. On success, return zero with preemption | |
312 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
313 | */ | |
314 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
315 | { | |
d0164adc | 316 | if (gfpflags_allow_blocking(gfp_mask)) |
5e4c0d97 JK |
317 | return __radix_tree_preload(gfp_mask); |
318 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
319 | preempt_disable(); | |
320 | return 0; | |
321 | } | |
322 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
323 | ||
1da177e4 LT |
324 | /* |
325 | * Return the maximum key which can be store into a | |
326 | * radix tree with height HEIGHT. | |
327 | */ | |
328 | static inline unsigned long radix_tree_maxindex(unsigned int height) | |
329 | { | |
330 | return height_to_maxindex[height]; | |
331 | } | |
332 | ||
333 | /* | |
334 | * Extend a radix tree so it can store key @index. | |
335 | */ | |
e6145236 MW |
336 | static int radix_tree_extend(struct radix_tree_root *root, |
337 | unsigned long index, unsigned order) | |
1da177e4 LT |
338 | { |
339 | struct radix_tree_node *node; | |
e2bdb933 | 340 | struct radix_tree_node *slot; |
1da177e4 | 341 | unsigned int height; |
1da177e4 LT |
342 | int tag; |
343 | ||
344 | /* Figure out what the height should be. */ | |
345 | height = root->height + 1; | |
346 | while (index > radix_tree_maxindex(height)) | |
347 | height++; | |
348 | ||
e6145236 | 349 | if ((root->rnode == NULL) && (order == 0)) { |
1da177e4 LT |
350 | root->height = height; |
351 | goto out; | |
352 | } | |
353 | ||
1da177e4 | 354 | do { |
7cf9c2c7 | 355 | unsigned int newheight; |
1da177e4 LT |
356 | if (!(node = radix_tree_node_alloc(root))) |
357 | return -ENOMEM; | |
358 | ||
1da177e4 | 359 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 360 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 361 | if (root_tag_get(root, tag)) |
1da177e4 LT |
362 | tag_set(node, tag, 0); |
363 | } | |
364 | ||
e2bdb933 | 365 | /* Increase the height. */ |
7cf9c2c7 | 366 | newheight = root->height+1; |
449dd698 JW |
367 | BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK); |
368 | node->path = newheight; | |
1da177e4 | 369 | node->count = 1; |
e2bdb933 HD |
370 | node->parent = NULL; |
371 | slot = root->rnode; | |
339e6353 | 372 | if (radix_tree_is_indirect_ptr(slot) && newheight > 1) { |
e2bdb933 HD |
373 | slot = indirect_to_ptr(slot); |
374 | slot->parent = node; | |
339e6353 | 375 | slot = ptr_to_indirect(slot); |
e2bdb933 HD |
376 | } |
377 | node->slots[0] = slot; | |
27d20fdd | 378 | node = ptr_to_indirect(node); |
7cf9c2c7 NP |
379 | rcu_assign_pointer(root->rnode, node); |
380 | root->height = newheight; | |
1da177e4 LT |
381 | } while (height > root->height); |
382 | out: | |
383 | return 0; | |
384 | } | |
385 | ||
386 | /** | |
139e5616 | 387 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
388 | * @root: radix tree root |
389 | * @index: index key | |
e6145236 | 390 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
391 | * @nodep: returns node |
392 | * @slotp: returns slot | |
1da177e4 | 393 | * |
139e5616 JW |
394 | * Create, if necessary, and return the node and slot for an item |
395 | * at position @index in the radix tree @root. | |
396 | * | |
397 | * Until there is more than one item in the tree, no nodes are | |
398 | * allocated and @root->rnode is used as a direct slot instead of | |
399 | * pointing to a node, in which case *@nodep will be NULL. | |
400 | * | |
401 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 402 | */ |
139e5616 | 403 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 MW |
404 | unsigned order, struct radix_tree_node **nodep, |
405 | void ***slotp) | |
1da177e4 | 406 | { |
201b6264 | 407 | struct radix_tree_node *node = NULL, *slot; |
139e5616 | 408 | unsigned int height, shift, offset; |
1da177e4 LT |
409 | int error; |
410 | ||
e6145236 MW |
411 | BUG_ON((0 < order) && (order < RADIX_TREE_MAP_SHIFT)); |
412 | ||
1da177e4 | 413 | /* Make sure the tree is high enough. */ |
612d6c19 | 414 | if (index > radix_tree_maxindex(root->height)) { |
e6145236 | 415 | error = radix_tree_extend(root, index, order); |
1da177e4 LT |
416 | if (error) |
417 | return error; | |
418 | } | |
419 | ||
e6145236 | 420 | slot = root->rnode; |
c0bc9875 | 421 | |
1da177e4 | 422 | height = root->height; |
0070e28d | 423 | shift = height * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
424 | |
425 | offset = 0; /* uninitialised var warning */ | |
e6145236 | 426 | while (shift > order) { |
201b6264 | 427 | if (slot == NULL) { |
1da177e4 | 428 | /* Have to add a child node. */ |
201b6264 | 429 | if (!(slot = radix_tree_node_alloc(root))) |
1da177e4 | 430 | return -ENOMEM; |
449dd698 | 431 | slot->path = height; |
e2bdb933 | 432 | slot->parent = node; |
201b6264 | 433 | if (node) { |
339e6353 MW |
434 | rcu_assign_pointer(node->slots[offset], |
435 | ptr_to_indirect(slot)); | |
1da177e4 | 436 | node->count++; |
449dd698 | 437 | slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT; |
201b6264 | 438 | } else |
339e6353 MW |
439 | rcu_assign_pointer(root->rnode, |
440 | ptr_to_indirect(slot)); | |
e6145236 MW |
441 | } else if (!radix_tree_is_indirect_ptr(slot)) |
442 | break; | |
1da177e4 LT |
443 | |
444 | /* Go a level down */ | |
e6145236 | 445 | height--; |
0070e28d | 446 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 447 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e6145236 | 448 | node = indirect_to_ptr(slot); |
201b6264 | 449 | slot = node->slots[offset]; |
e6145236 MW |
450 | } |
451 | ||
452 | /* Insert pointers to the canonical entry */ | |
453 | if ((shift - order) > 0) { | |
454 | int i, n = 1 << (shift - order); | |
455 | offset = offset & ~(n - 1); | |
456 | slot = ptr_to_indirect(&node->slots[offset]); | |
457 | for (i = 0; i < n; i++) { | |
458 | if (node->slots[offset + i]) | |
459 | return -EEXIST; | |
460 | } | |
461 | ||
462 | for (i = 1; i < n; i++) { | |
463 | rcu_assign_pointer(node->slots[offset + i], slot); | |
464 | node->count++; | |
465 | } | |
612d6c19 | 466 | } |
1da177e4 | 467 | |
139e5616 JW |
468 | if (nodep) |
469 | *nodep = node; | |
470 | if (slotp) | |
471 | *slotp = node ? node->slots + offset : (void **)&root->rnode; | |
472 | return 0; | |
473 | } | |
474 | ||
475 | /** | |
e6145236 | 476 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
477 | * @root: radix tree root |
478 | * @index: index key | |
e6145236 | 479 | * @order: key covers the 2^order indices around index |
139e5616 JW |
480 | * @item: item to insert |
481 | * | |
482 | * Insert an item into the radix tree at position @index. | |
483 | */ | |
e6145236 MW |
484 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
485 | unsigned order, void *item) | |
139e5616 JW |
486 | { |
487 | struct radix_tree_node *node; | |
488 | void **slot; | |
489 | int error; | |
490 | ||
491 | BUG_ON(radix_tree_is_indirect_ptr(item)); | |
492 | ||
e6145236 | 493 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
494 | if (error) |
495 | return error; | |
496 | if (*slot != NULL) | |
1da177e4 | 497 | return -EEXIST; |
139e5616 | 498 | rcu_assign_pointer(*slot, item); |
201b6264 | 499 | |
612d6c19 NP |
500 | if (node) { |
501 | node->count++; | |
139e5616 JW |
502 | BUG_ON(tag_get(node, 0, index & RADIX_TREE_MAP_MASK)); |
503 | BUG_ON(tag_get(node, 1, index & RADIX_TREE_MAP_MASK)); | |
612d6c19 | 504 | } else { |
612d6c19 NP |
505 | BUG_ON(root_tag_get(root, 0)); |
506 | BUG_ON(root_tag_get(root, 1)); | |
507 | } | |
1da177e4 | 508 | |
1da177e4 LT |
509 | return 0; |
510 | } | |
e6145236 | 511 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 512 | |
139e5616 JW |
513 | /** |
514 | * __radix_tree_lookup - lookup an item in a radix tree | |
515 | * @root: radix tree root | |
516 | * @index: index key | |
517 | * @nodep: returns node | |
518 | * @slotp: returns slot | |
519 | * | |
520 | * Lookup and return the item at position @index in the radix | |
521 | * tree @root. | |
522 | * | |
523 | * Until there is more than one item in the tree, no nodes are | |
524 | * allocated and @root->rnode is used as a direct slot instead of | |
525 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 526 | */ |
139e5616 JW |
527 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
528 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 529 | { |
139e5616 | 530 | struct radix_tree_node *node, *parent; |
1da177e4 | 531 | unsigned int height, shift; |
139e5616 | 532 | void **slot; |
612d6c19 | 533 | |
2676a58c | 534 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 | 535 | if (node == NULL) |
1da177e4 LT |
536 | return NULL; |
537 | ||
c0bc9875 | 538 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
539 | if (index > 0) |
540 | return NULL; | |
139e5616 JW |
541 | |
542 | if (nodep) | |
543 | *nodep = NULL; | |
544 | if (slotp) | |
545 | *slotp = (void **)&root->rnode; | |
546 | return node; | |
7cf9c2c7 | 547 | } |
27d20fdd | 548 | node = indirect_to_ptr(node); |
7cf9c2c7 | 549 | |
449dd698 | 550 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
551 | if (index > radix_tree_maxindex(height)) |
552 | return NULL; | |
612d6c19 | 553 | |
1da177e4 | 554 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 555 | |
7cf9c2c7 | 556 | do { |
139e5616 JW |
557 | parent = node; |
558 | slot = node->slots + ((index >> shift) & RADIX_TREE_MAP_MASK); | |
2676a58c | 559 | node = rcu_dereference_raw(*slot); |
7cf9c2c7 | 560 | if (node == NULL) |
1da177e4 | 561 | return NULL; |
e6145236 MW |
562 | if (!radix_tree_is_indirect_ptr(node)) |
563 | break; | |
339e6353 | 564 | node = indirect_to_ptr(node); |
1da177e4 | 565 | |
1da177e4 LT |
566 | shift -= RADIX_TREE_MAP_SHIFT; |
567 | height--; | |
7cf9c2c7 | 568 | } while (height > 0); |
1da177e4 | 569 | |
139e5616 JW |
570 | if (nodep) |
571 | *nodep = parent; | |
572 | if (slotp) | |
573 | *slotp = slot; | |
574 | return node; | |
b72b71c6 HS |
575 | } |
576 | ||
577 | /** | |
578 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
579 | * @root: radix tree root | |
580 | * @index: index key | |
581 | * | |
582 | * Returns: the slot corresponding to the position @index in the | |
583 | * radix tree @root. This is useful for update-if-exists operations. | |
584 | * | |
585 | * This function can be called under rcu_read_lock iff the slot is not | |
586 | * modified by radix_tree_replace_slot, otherwise it must be called | |
587 | * exclusive from other writers. Any dereference of the slot must be done | |
588 | * using radix_tree_deref_slot. | |
589 | */ | |
590 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
591 | { | |
139e5616 JW |
592 | void **slot; |
593 | ||
594 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
595 | return NULL; | |
596 | return slot; | |
a4331366 | 597 | } |
a4331366 HR |
598 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
599 | ||
600 | /** | |
601 | * radix_tree_lookup - perform lookup operation on a radix tree | |
602 | * @root: radix tree root | |
603 | * @index: index key | |
604 | * | |
605 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
606 | * |
607 | * This function can be called under rcu_read_lock, however the caller | |
608 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
609 | * them safely). No RCU barriers are required to access or modify the | |
610 | * returned item, however. | |
a4331366 HR |
611 | */ |
612 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
613 | { | |
139e5616 | 614 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
615 | } |
616 | EXPORT_SYMBOL(radix_tree_lookup); | |
617 | ||
618 | /** | |
619 | * radix_tree_tag_set - set a tag on a radix tree node | |
620 | * @root: radix tree root | |
621 | * @index: index key | |
622 | * @tag: tag index | |
623 | * | |
daff89f3 JC |
624 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
625 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
626 | * the root all the way down to the leaf node. |
627 | * | |
628 | * Returns the address of the tagged item. Setting a tag on a not-present | |
629 | * item is a bug. | |
630 | */ | |
631 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 632 | unsigned long index, unsigned int tag) |
1da177e4 LT |
633 | { |
634 | unsigned int height, shift; | |
201b6264 | 635 | struct radix_tree_node *slot; |
1da177e4 LT |
636 | |
637 | height = root->height; | |
4c91c364 | 638 | BUG_ON(index > radix_tree_maxindex(height)); |
1da177e4 | 639 | |
27d20fdd | 640 | slot = indirect_to_ptr(root->rnode); |
612d6c19 | 641 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
642 | |
643 | while (height > 0) { | |
644 | int offset; | |
645 | ||
646 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
d5274261 NP |
647 | if (!tag_get(slot, tag, offset)) |
648 | tag_set(slot, tag, offset); | |
201b6264 CL |
649 | slot = slot->slots[offset]; |
650 | BUG_ON(slot == NULL); | |
e6145236 MW |
651 | if (!radix_tree_is_indirect_ptr(slot)) |
652 | break; | |
339e6353 | 653 | slot = indirect_to_ptr(slot); |
1da177e4 LT |
654 | shift -= RADIX_TREE_MAP_SHIFT; |
655 | height--; | |
656 | } | |
657 | ||
612d6c19 NP |
658 | /* set the root's tag bit */ |
659 | if (slot && !root_tag_get(root, tag)) | |
660 | root_tag_set(root, tag); | |
661 | ||
201b6264 | 662 | return slot; |
1da177e4 LT |
663 | } |
664 | EXPORT_SYMBOL(radix_tree_tag_set); | |
665 | ||
666 | /** | |
667 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
668 | * @root: radix tree root | |
669 | * @index: index key | |
670 | * @tag: tag index | |
671 | * | |
daff89f3 JC |
672 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
673 | * corresponding to @index in the radix tree. If | |
1da177e4 LT |
674 | * this causes the leaf node to have no tags set then clear the tag in the |
675 | * next-to-leaf node, etc. | |
676 | * | |
677 | * Returns the address of the tagged item on success, else NULL. ie: | |
678 | * has the same return value and semantics as radix_tree_lookup(). | |
679 | */ | |
680 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 681 | unsigned long index, unsigned int tag) |
1da177e4 | 682 | { |
e2bdb933 | 683 | struct radix_tree_node *node = NULL; |
612d6c19 | 684 | struct radix_tree_node *slot = NULL; |
1da177e4 | 685 | unsigned int height, shift; |
e2bdb933 | 686 | int uninitialized_var(offset); |
1da177e4 LT |
687 | |
688 | height = root->height; | |
689 | if (index > radix_tree_maxindex(height)) | |
690 | goto out; | |
691 | ||
e2bdb933 | 692 | shift = height * RADIX_TREE_MAP_SHIFT; |
339e6353 | 693 | slot = root->rnode; |
1da177e4 | 694 | |
e2bdb933 | 695 | while (shift) { |
201b6264 | 696 | if (slot == NULL) |
1da177e4 | 697 | goto out; |
e6145236 MW |
698 | if (!radix_tree_is_indirect_ptr(slot)) |
699 | break; | |
339e6353 | 700 | slot = indirect_to_ptr(slot); |
1da177e4 | 701 | |
e2bdb933 | 702 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 703 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e2bdb933 | 704 | node = slot; |
201b6264 | 705 | slot = slot->slots[offset]; |
1da177e4 LT |
706 | } |
707 | ||
612d6c19 | 708 | if (slot == NULL) |
1da177e4 LT |
709 | goto out; |
710 | ||
e2bdb933 HD |
711 | while (node) { |
712 | if (!tag_get(node, tag, offset)) | |
d5274261 | 713 | goto out; |
e2bdb933 HD |
714 | tag_clear(node, tag, offset); |
715 | if (any_tag_set(node, tag)) | |
6e954b9e | 716 | goto out; |
e2bdb933 HD |
717 | |
718 | index >>= RADIX_TREE_MAP_SHIFT; | |
719 | offset = index & RADIX_TREE_MAP_MASK; | |
720 | node = node->parent; | |
612d6c19 NP |
721 | } |
722 | ||
723 | /* clear the root's tag bit */ | |
724 | if (root_tag_get(root, tag)) | |
725 | root_tag_clear(root, tag); | |
726 | ||
1da177e4 | 727 | out: |
612d6c19 | 728 | return slot; |
1da177e4 LT |
729 | } |
730 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
731 | ||
1da177e4 | 732 | /** |
32605a18 MT |
733 | * radix_tree_tag_get - get a tag on a radix tree node |
734 | * @root: radix tree root | |
735 | * @index: index key | |
daff89f3 | 736 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 737 | * |
32605a18 | 738 | * Return values: |
1da177e4 | 739 | * |
612d6c19 NP |
740 | * 0: tag not present or not set |
741 | * 1: tag set | |
ce82653d DH |
742 | * |
743 | * Note that the return value of this function may not be relied on, even if | |
744 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
745 | * from concurrency. | |
1da177e4 LT |
746 | */ |
747 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 748 | unsigned long index, unsigned int tag) |
1da177e4 LT |
749 | { |
750 | unsigned int height, shift; | |
7cf9c2c7 | 751 | struct radix_tree_node *node; |
1da177e4 | 752 | |
612d6c19 NP |
753 | /* check the root's tag bit */ |
754 | if (!root_tag_get(root, tag)) | |
755 | return 0; | |
756 | ||
2676a58c | 757 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
758 | if (node == NULL) |
759 | return 0; | |
760 | ||
c0bc9875 | 761 | if (!radix_tree_is_indirect_ptr(node)) |
7cf9c2c7 | 762 | return (index == 0); |
27d20fdd | 763 | node = indirect_to_ptr(node); |
7cf9c2c7 | 764 | |
449dd698 | 765 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
766 | if (index > radix_tree_maxindex(height)) |
767 | return 0; | |
612d6c19 | 768 | |
1da177e4 | 769 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
770 | |
771 | for ( ; ; ) { | |
772 | int offset; | |
773 | ||
7cf9c2c7 | 774 | if (node == NULL) |
1da177e4 | 775 | return 0; |
339e6353 | 776 | node = indirect_to_ptr(node); |
1da177e4 LT |
777 | |
778 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
7cf9c2c7 | 779 | if (!tag_get(node, tag, offset)) |
3fa36acb | 780 | return 0; |
ce82653d | 781 | if (height == 1) |
3fa36acb | 782 | return 1; |
2676a58c | 783 | node = rcu_dereference_raw(node->slots[offset]); |
e6145236 MW |
784 | if (!radix_tree_is_indirect_ptr(node)) |
785 | return 1; | |
1da177e4 LT |
786 | shift -= RADIX_TREE_MAP_SHIFT; |
787 | height--; | |
788 | } | |
789 | } | |
790 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 791 | |
78c1d784 KK |
792 | /** |
793 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
794 | * | |
795 | * @root: radix tree root | |
796 | * @iter: iterator state | |
797 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
798 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
799 | */ | |
800 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
801 | struct radix_tree_iter *iter, unsigned flags) | |
802 | { | |
803 | unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
804 | struct radix_tree_node *rnode, *node; | |
449dd698 | 805 | unsigned long index, offset, height; |
78c1d784 KK |
806 | |
807 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
808 | return NULL; | |
809 | ||
810 | /* | |
811 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
812 | * during iterating; it can be zero only at the beginning. | |
813 | * And we cannot overflow iter->next_index in a single step, | |
814 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
815 | * |
816 | * This condition also used by radix_tree_next_slot() to stop | |
817 | * contiguous iterating, and forbid swithing to the next chunk. | |
78c1d784 KK |
818 | */ |
819 | index = iter->next_index; | |
820 | if (!index && iter->index) | |
821 | return NULL; | |
822 | ||
823 | rnode = rcu_dereference_raw(root->rnode); | |
824 | if (radix_tree_is_indirect_ptr(rnode)) { | |
825 | rnode = indirect_to_ptr(rnode); | |
826 | } else if (rnode && !index) { | |
827 | /* Single-slot tree */ | |
828 | iter->index = 0; | |
829 | iter->next_index = 1; | |
830 | iter->tags = 1; | |
831 | return (void **)&root->rnode; | |
832 | } else | |
833 | return NULL; | |
834 | ||
835 | restart: | |
449dd698 JW |
836 | height = rnode->path & RADIX_TREE_HEIGHT_MASK; |
837 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
78c1d784 KK |
838 | offset = index >> shift; |
839 | ||
840 | /* Index outside of the tree */ | |
841 | if (offset >= RADIX_TREE_MAP_SIZE) | |
842 | return NULL; | |
843 | ||
844 | node = rnode; | |
845 | while (1) { | |
e6145236 | 846 | struct radix_tree_node *slot; |
78c1d784 KK |
847 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
848 | !test_bit(offset, node->tags[tag]) : | |
849 | !node->slots[offset]) { | |
850 | /* Hole detected */ | |
851 | if (flags & RADIX_TREE_ITER_CONTIG) | |
852 | return NULL; | |
853 | ||
854 | if (flags & RADIX_TREE_ITER_TAGGED) | |
855 | offset = radix_tree_find_next_bit( | |
856 | node->tags[tag], | |
857 | RADIX_TREE_MAP_SIZE, | |
858 | offset + 1); | |
859 | else | |
860 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
861 | if (node->slots[offset]) | |
862 | break; | |
863 | } | |
864 | index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1); | |
865 | index += offset << shift; | |
866 | /* Overflow after ~0UL */ | |
867 | if (!index) | |
868 | return NULL; | |
869 | if (offset == RADIX_TREE_MAP_SIZE) | |
870 | goto restart; | |
871 | } | |
872 | ||
873 | /* This is leaf-node */ | |
874 | if (!shift) | |
875 | break; | |
876 | ||
e6145236 MW |
877 | slot = rcu_dereference_raw(node->slots[offset]); |
878 | if (slot == NULL) | |
78c1d784 | 879 | goto restart; |
e6145236 MW |
880 | if (!radix_tree_is_indirect_ptr(slot)) |
881 | break; | |
882 | node = indirect_to_ptr(slot); | |
78c1d784 KK |
883 | shift -= RADIX_TREE_MAP_SHIFT; |
884 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
885 | } | |
886 | ||
887 | /* Update the iterator state */ | |
888 | iter->index = index; | |
889 | iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1; | |
890 | ||
891 | /* Construct iter->tags bit-mask from node->tags[tag] array */ | |
892 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
893 | unsigned tag_long, tag_bit; | |
894 | ||
895 | tag_long = offset / BITS_PER_LONG; | |
896 | tag_bit = offset % BITS_PER_LONG; | |
897 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
898 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
899 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
900 | /* Pick tags from next element */ | |
901 | if (tag_bit) | |
902 | iter->tags |= node->tags[tag][tag_long + 1] << | |
903 | (BITS_PER_LONG - tag_bit); | |
904 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
905 | iter->next_index = index + BITS_PER_LONG; | |
906 | } | |
907 | } | |
908 | ||
909 | return node->slots + offset; | |
910 | } | |
911 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
912 | ||
ebf8aa44 JK |
913 | /** |
914 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
915 | * tag if item has another tag set | |
916 | * @root: radix tree root | |
917 | * @first_indexp: pointer to a starting index of a range to scan | |
918 | * @last_index: last index of a range to scan | |
919 | * @nr_to_tag: maximum number items to tag | |
920 | * @iftag: tag index to test | |
921 | * @settag: tag index to set if tested tag is set | |
922 | * | |
923 | * This function scans range of radix tree from first_index to last_index | |
924 | * (inclusive). For each item in the range if iftag is set, the function sets | |
925 | * also settag. The function stops either after tagging nr_to_tag items or | |
926 | * after reaching last_index. | |
927 | * | |
144dcfc0 DC |
928 | * The tags must be set from the leaf level only and propagated back up the |
929 | * path to the root. We must do this so that we resolve the full path before | |
930 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
931 | * we can get to the leaf node and find that the index that has the iftag | |
932 | * set is outside the range we are scanning. This reults in dangling tags and | |
933 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
934 | * | |
ebf8aa44 JK |
935 | * The function returns number of leaves where the tag was set and sets |
936 | * *first_indexp to the first unscanned index. | |
d5ed3a4a JK |
937 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
938 | * be prepared to handle that. | |
ebf8aa44 JK |
939 | */ |
940 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
941 | unsigned long *first_indexp, unsigned long last_index, | |
942 | unsigned long nr_to_tag, | |
943 | unsigned int iftag, unsigned int settag) | |
944 | { | |
144dcfc0 | 945 | unsigned int height = root->height; |
e2bdb933 | 946 | struct radix_tree_node *node = NULL; |
144dcfc0 DC |
947 | struct radix_tree_node *slot; |
948 | unsigned int shift; | |
949 | unsigned long tagged = 0; | |
950 | unsigned long index = *first_indexp; | |
ebf8aa44 JK |
951 | |
952 | last_index = min(last_index, radix_tree_maxindex(height)); | |
953 | if (index > last_index) | |
954 | return 0; | |
955 | if (!nr_to_tag) | |
956 | return 0; | |
957 | if (!root_tag_get(root, iftag)) { | |
958 | *first_indexp = last_index + 1; | |
959 | return 0; | |
960 | } | |
961 | if (height == 0) { | |
962 | *first_indexp = last_index + 1; | |
963 | root_tag_set(root, settag); | |
964 | return 1; | |
965 | } | |
966 | ||
967 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
27d20fdd | 968 | slot = indirect_to_ptr(root->rnode); |
ebf8aa44 JK |
969 | |
970 | for (;;) { | |
e2bdb933 | 971 | unsigned long upindex; |
ebf8aa44 JK |
972 | int offset; |
973 | ||
974 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
975 | if (!slot->slots[offset]) | |
976 | goto next; | |
977 | if (!tag_get(slot, iftag, offset)) | |
978 | goto next; | |
e2bdb933 | 979 | if (shift) { |
e2bdb933 | 980 | node = slot; |
144dcfc0 | 981 | slot = slot->slots[offset]; |
e6145236 MW |
982 | if (radix_tree_is_indirect_ptr(slot)) { |
983 | slot = indirect_to_ptr(slot); | |
984 | shift -= RADIX_TREE_MAP_SHIFT; | |
985 | continue; | |
986 | } else { | |
987 | slot = node; | |
988 | node = node->parent; | |
989 | } | |
144dcfc0 DC |
990 | } |
991 | ||
992 | /* tag the leaf */ | |
e6145236 | 993 | tagged += 1 << shift; |
ebf8aa44 | 994 | tag_set(slot, settag, offset); |
144dcfc0 DC |
995 | |
996 | /* walk back up the path tagging interior nodes */ | |
e2bdb933 HD |
997 | upindex = index; |
998 | while (node) { | |
999 | upindex >>= RADIX_TREE_MAP_SHIFT; | |
1000 | offset = upindex & RADIX_TREE_MAP_MASK; | |
1001 | ||
144dcfc0 | 1002 | /* stop if we find a node with the tag already set */ |
e2bdb933 | 1003 | if (tag_get(node, settag, offset)) |
144dcfc0 | 1004 | break; |
e2bdb933 HD |
1005 | tag_set(node, settag, offset); |
1006 | node = node->parent; | |
ebf8aa44 | 1007 | } |
144dcfc0 | 1008 | |
e2bdb933 HD |
1009 | /* |
1010 | * Small optimization: now clear that node pointer. | |
1011 | * Since all of this slot's ancestors now have the tag set | |
1012 | * from setting it above, we have no further need to walk | |
1013 | * back up the tree setting tags, until we update slot to | |
1014 | * point to another radix_tree_node. | |
1015 | */ | |
1016 | node = NULL; | |
1017 | ||
ebf8aa44 JK |
1018 | next: |
1019 | /* Go to next item at level determined by 'shift' */ | |
1020 | index = ((index >> shift) + 1) << shift; | |
d5ed3a4a JK |
1021 | /* Overflow can happen when last_index is ~0UL... */ |
1022 | if (index > last_index || !index) | |
ebf8aa44 JK |
1023 | break; |
1024 | if (tagged >= nr_to_tag) | |
1025 | break; | |
1026 | while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) { | |
1027 | /* | |
1028 | * We've fully scanned this node. Go up. Because | |
1029 | * last_index is guaranteed to be in the tree, what | |
1030 | * we do below cannot wander astray. | |
1031 | */ | |
e2bdb933 | 1032 | slot = slot->parent; |
ebf8aa44 JK |
1033 | shift += RADIX_TREE_MAP_SHIFT; |
1034 | } | |
1035 | } | |
1036 | /* | |
ac15ee69 TO |
1037 | * We need not to tag the root tag if there is no tag which is set with |
1038 | * settag within the range from *first_indexp to last_index. | |
ebf8aa44 | 1039 | */ |
ac15ee69 TO |
1040 | if (tagged > 0) |
1041 | root_tag_set(root, settag); | |
ebf8aa44 JK |
1042 | *first_indexp = index; |
1043 | ||
1044 | return tagged; | |
1045 | } | |
1046 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
1047 | ||
1da177e4 LT |
1048 | /** |
1049 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1050 | * @root: radix tree root | |
1051 | * @results: where the results of the lookup are placed | |
1052 | * @first_index: start the lookup from this key | |
1053 | * @max_items: place up to this many items at *results | |
1054 | * | |
1055 | * Performs an index-ascending scan of the tree for present items. Places | |
1056 | * them at *@results and returns the number of items which were placed at | |
1057 | * *@results. | |
1058 | * | |
1059 | * The implementation is naive. | |
7cf9c2c7 NP |
1060 | * |
1061 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1062 | * rcu_read_lock. In this case, rather than the returned results being | |
1063 | * an atomic snapshot of the tree at a single point in time, the semantics | |
1064 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups | |
1065 | * have been issued in individual locks, and results stored in 'results'. | |
1da177e4 LT |
1066 | */ |
1067 | unsigned int | |
1068 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1069 | unsigned long first_index, unsigned int max_items) | |
1070 | { | |
cebbd29e KK |
1071 | struct radix_tree_iter iter; |
1072 | void **slot; | |
1073 | unsigned int ret = 0; | |
7cf9c2c7 | 1074 | |
cebbd29e | 1075 | if (unlikely(!max_items)) |
7cf9c2c7 | 1076 | return 0; |
1da177e4 | 1077 | |
cebbd29e | 1078 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1079 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1080 | if (!results[ret]) |
1081 | continue; | |
46437f9a MW |
1082 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1083 | slot = radix_tree_iter_retry(&iter); | |
1084 | continue; | |
1085 | } | |
cebbd29e | 1086 | if (++ret == max_items) |
1da177e4 | 1087 | break; |
1da177e4 | 1088 | } |
7cf9c2c7 | 1089 | |
1da177e4 LT |
1090 | return ret; |
1091 | } | |
1092 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1093 | ||
47feff2c NP |
1094 | /** |
1095 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1096 | * @root: radix tree root | |
1097 | * @results: where the results of the lookup are placed | |
6328650b | 1098 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1099 | * @first_index: start the lookup from this key |
1100 | * @max_items: place up to this many items at *results | |
1101 | * | |
1102 | * Performs an index-ascending scan of the tree for present items. Places | |
1103 | * their slots at *@results and returns the number of items which were | |
1104 | * placed at *@results. | |
1105 | * | |
1106 | * The implementation is naive. | |
1107 | * | |
1108 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1109 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1110 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1111 | */ | |
1112 | unsigned int | |
6328650b HD |
1113 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1114 | void ***results, unsigned long *indices, | |
47feff2c NP |
1115 | unsigned long first_index, unsigned int max_items) |
1116 | { | |
cebbd29e KK |
1117 | struct radix_tree_iter iter; |
1118 | void **slot; | |
1119 | unsigned int ret = 0; | |
47feff2c | 1120 | |
cebbd29e | 1121 | if (unlikely(!max_items)) |
47feff2c NP |
1122 | return 0; |
1123 | ||
cebbd29e KK |
1124 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1125 | results[ret] = slot; | |
6328650b | 1126 | if (indices) |
cebbd29e KK |
1127 | indices[ret] = iter.index; |
1128 | if (++ret == max_items) | |
47feff2c | 1129 | break; |
47feff2c NP |
1130 | } |
1131 | ||
1132 | return ret; | |
1133 | } | |
1134 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1135 | ||
1da177e4 LT |
1136 | /** |
1137 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1138 | * based on a tag | |
1139 | * @root: radix tree root | |
1140 | * @results: where the results of the lookup are placed | |
1141 | * @first_index: start the lookup from this key | |
1142 | * @max_items: place up to this many items at *results | |
daff89f3 | 1143 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1144 | * |
1145 | * Performs an index-ascending scan of the tree for present items which | |
1146 | * have the tag indexed by @tag set. Places the items at *@results and | |
1147 | * returns the number of items which were placed at *@results. | |
1148 | */ | |
1149 | unsigned int | |
1150 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1151 | unsigned long first_index, unsigned int max_items, |
1152 | unsigned int tag) | |
1da177e4 | 1153 | { |
cebbd29e KK |
1154 | struct radix_tree_iter iter; |
1155 | void **slot; | |
1156 | unsigned int ret = 0; | |
612d6c19 | 1157 | |
cebbd29e | 1158 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1159 | return 0; |
1160 | ||
cebbd29e | 1161 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1162 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1163 | if (!results[ret]) |
1164 | continue; | |
46437f9a MW |
1165 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1166 | slot = radix_tree_iter_retry(&iter); | |
1167 | continue; | |
1168 | } | |
cebbd29e | 1169 | if (++ret == max_items) |
1da177e4 | 1170 | break; |
1da177e4 | 1171 | } |
7cf9c2c7 | 1172 | |
1da177e4 LT |
1173 | return ret; |
1174 | } | |
1175 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1176 | ||
47feff2c NP |
1177 | /** |
1178 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1179 | * radix tree based on a tag | |
1180 | * @root: radix tree root | |
1181 | * @results: where the results of the lookup are placed | |
1182 | * @first_index: start the lookup from this key | |
1183 | * @max_items: place up to this many items at *results | |
1184 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1185 | * | |
1186 | * Performs an index-ascending scan of the tree for present items which | |
1187 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1188 | * returns the number of slots which were placed at *@results. | |
1189 | */ | |
1190 | unsigned int | |
1191 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1192 | unsigned long first_index, unsigned int max_items, | |
1193 | unsigned int tag) | |
1194 | { | |
cebbd29e KK |
1195 | struct radix_tree_iter iter; |
1196 | void **slot; | |
1197 | unsigned int ret = 0; | |
47feff2c | 1198 | |
cebbd29e | 1199 | if (unlikely(!max_items)) |
47feff2c NP |
1200 | return 0; |
1201 | ||
cebbd29e KK |
1202 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1203 | results[ret] = slot; | |
1204 | if (++ret == max_items) | |
47feff2c | 1205 | break; |
47feff2c NP |
1206 | } |
1207 | ||
1208 | return ret; | |
1209 | } | |
1210 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1211 | ||
e504f3fd HD |
1212 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
1213 | #include <linux/sched.h> /* for cond_resched() */ | |
1214 | ||
1215 | /* | |
1216 | * This linear search is at present only useful to shmem_unuse_inode(). | |
1217 | */ | |
1218 | static unsigned long __locate(struct radix_tree_node *slot, void *item, | |
1219 | unsigned long index, unsigned long *found_index) | |
1220 | { | |
1221 | unsigned int shift, height; | |
1222 | unsigned long i; | |
1223 | ||
449dd698 | 1224 | height = slot->path & RADIX_TREE_HEIGHT_MASK; |
e504f3fd HD |
1225 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1226 | ||
1227 | for ( ; height > 1; height--) { | |
1228 | i = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1229 | for (;;) { | |
1230 | if (slot->slots[i] != NULL) | |
1231 | break; | |
1232 | index &= ~((1UL << shift) - 1); | |
1233 | index += 1UL << shift; | |
1234 | if (index == 0) | |
1235 | goto out; /* 32-bit wraparound */ | |
1236 | i++; | |
1237 | if (i == RADIX_TREE_MAP_SIZE) | |
1238 | goto out; | |
1239 | } | |
1240 | ||
e504f3fd HD |
1241 | slot = rcu_dereference_raw(slot->slots[i]); |
1242 | if (slot == NULL) | |
1243 | goto out; | |
e6145236 MW |
1244 | if (!radix_tree_is_indirect_ptr(slot)) { |
1245 | if (slot == item) { | |
1246 | *found_index = index + i; | |
1247 | index = 0; | |
1248 | } else { | |
1249 | index += shift; | |
1250 | } | |
1251 | goto out; | |
1252 | } | |
339e6353 | 1253 | slot = indirect_to_ptr(slot); |
e6145236 | 1254 | shift -= RADIX_TREE_MAP_SHIFT; |
e504f3fd HD |
1255 | } |
1256 | ||
1257 | /* Bottom level: check items */ | |
1258 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
1259 | if (slot->slots[i] == item) { | |
1260 | *found_index = index + i; | |
1261 | index = 0; | |
1262 | goto out; | |
1263 | } | |
1264 | } | |
1265 | index += RADIX_TREE_MAP_SIZE; | |
1266 | out: | |
1267 | return index; | |
1268 | } | |
1269 | ||
1270 | /** | |
1271 | * radix_tree_locate_item - search through radix tree for item | |
1272 | * @root: radix tree root | |
1273 | * @item: item to be found | |
1274 | * | |
1275 | * Returns index where item was found, or -1 if not found. | |
1276 | * Caller must hold no lock (since this time-consuming function needs | |
1277 | * to be preemptible), and must check afterwards if item is still there. | |
1278 | */ | |
1279 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1280 | { | |
1281 | struct radix_tree_node *node; | |
1282 | unsigned long max_index; | |
1283 | unsigned long cur_index = 0; | |
1284 | unsigned long found_index = -1; | |
1285 | ||
1286 | do { | |
1287 | rcu_read_lock(); | |
1288 | node = rcu_dereference_raw(root->rnode); | |
1289 | if (!radix_tree_is_indirect_ptr(node)) { | |
1290 | rcu_read_unlock(); | |
1291 | if (node == item) | |
1292 | found_index = 0; | |
1293 | break; | |
1294 | } | |
1295 | ||
1296 | node = indirect_to_ptr(node); | |
449dd698 JW |
1297 | max_index = radix_tree_maxindex(node->path & |
1298 | RADIX_TREE_HEIGHT_MASK); | |
5f30fc94 HD |
1299 | if (cur_index > max_index) { |
1300 | rcu_read_unlock(); | |
e504f3fd | 1301 | break; |
5f30fc94 | 1302 | } |
e504f3fd HD |
1303 | |
1304 | cur_index = __locate(node, item, cur_index, &found_index); | |
1305 | rcu_read_unlock(); | |
1306 | cond_resched(); | |
1307 | } while (cur_index != 0 && cur_index <= max_index); | |
1308 | ||
1309 | return found_index; | |
1310 | } | |
1311 | #else | |
1312 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1313 | { | |
1314 | return -1; | |
1315 | } | |
1316 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ | |
47feff2c | 1317 | |
a5f51c96 NP |
1318 | /** |
1319 | * radix_tree_shrink - shrink height of a radix tree to minimal | |
1320 | * @root radix tree root | |
1321 | */ | |
1322 | static inline void radix_tree_shrink(struct radix_tree_root *root) | |
1323 | { | |
1324 | /* try to shrink tree height */ | |
c0bc9875 | 1325 | while (root->height > 0) { |
a5f51c96 | 1326 | struct radix_tree_node *to_free = root->rnode; |
e2bdb933 | 1327 | struct radix_tree_node *slot; |
a5f51c96 | 1328 | |
c0bc9875 | 1329 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
27d20fdd | 1330 | to_free = indirect_to_ptr(to_free); |
c0bc9875 NP |
1331 | |
1332 | /* | |
1333 | * The candidate node has more than one child, or its child | |
e6145236 MW |
1334 | * is not at the leftmost slot, or it is a multiorder entry, |
1335 | * we cannot shrink. | |
c0bc9875 NP |
1336 | */ |
1337 | if (to_free->count != 1) | |
1338 | break; | |
339e6353 MW |
1339 | slot = to_free->slots[0]; |
1340 | if (!slot) | |
c0bc9875 NP |
1341 | break; |
1342 | ||
7cf9c2c7 NP |
1343 | /* |
1344 | * We don't need rcu_assign_pointer(), since we are simply | |
27d20fdd NP |
1345 | * moving the node from one part of the tree to another: if it |
1346 | * was safe to dereference the old pointer to it | |
7cf9c2c7 | 1347 | * (to_free->slots[0]), it will be safe to dereference the new |
27d20fdd | 1348 | * one (root->rnode) as far as dependent read barriers go. |
7cf9c2c7 | 1349 | */ |
e2bdb933 | 1350 | if (root->height > 1) { |
e6145236 MW |
1351 | if (!radix_tree_is_indirect_ptr(slot)) |
1352 | break; | |
1353 | ||
339e6353 | 1354 | slot = indirect_to_ptr(slot); |
e2bdb933 HD |
1355 | slot->parent = NULL; |
1356 | slot = ptr_to_indirect(slot); | |
1357 | } | |
1358 | root->rnode = slot; | |
a5f51c96 | 1359 | root->height--; |
27d20fdd NP |
1360 | |
1361 | /* | |
1362 | * We have a dilemma here. The node's slot[0] must not be | |
1363 | * NULLed in case there are concurrent lookups expecting to | |
1364 | * find the item. However if this was a bottom-level node, | |
1365 | * then it may be subject to the slot pointer being visible | |
1366 | * to callers dereferencing it. If item corresponding to | |
1367 | * slot[0] is subsequently deleted, these callers would expect | |
1368 | * their slot to become empty sooner or later. | |
1369 | * | |
1370 | * For example, lockless pagecache will look up a slot, deref | |
1371 | * the page pointer, and if the page is 0 refcount it means it | |
1372 | * was concurrently deleted from pagecache so try the deref | |
1373 | * again. Fortunately there is already a requirement for logic | |
1374 | * to retry the entire slot lookup -- the indirect pointer | |
1375 | * problem (replacing direct root node with an indirect pointer | |
1376 | * also results in a stale slot). So tag the slot as indirect | |
1377 | * to force callers to retry. | |
1378 | */ | |
1379 | if (root->height == 0) | |
1380 | *((unsigned long *)&to_free->slots[0]) |= | |
1381 | RADIX_TREE_INDIRECT_PTR; | |
1382 | ||
a5f51c96 NP |
1383 | radix_tree_node_free(to_free); |
1384 | } | |
1385 | } | |
1386 | ||
139e5616 JW |
1387 | /** |
1388 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1389 | * @root: radix tree root | |
139e5616 JW |
1390 | * @node: node containing @index |
1391 | * | |
1392 | * After clearing the slot at @index in @node from radix tree | |
1393 | * rooted at @root, call this function to attempt freeing the | |
1394 | * node and shrinking the tree. | |
1395 | * | |
1396 | * Returns %true if @node was freed, %false otherwise. | |
1397 | */ | |
449dd698 | 1398 | bool __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1399 | struct radix_tree_node *node) |
1400 | { | |
1401 | bool deleted = false; | |
1402 | ||
1403 | do { | |
1404 | struct radix_tree_node *parent; | |
1405 | ||
1406 | if (node->count) { | |
1407 | if (node == indirect_to_ptr(root->rnode)) { | |
1408 | radix_tree_shrink(root); | |
1409 | if (root->height == 0) | |
1410 | deleted = true; | |
1411 | } | |
1412 | return deleted; | |
1413 | } | |
1414 | ||
1415 | parent = node->parent; | |
1416 | if (parent) { | |
449dd698 | 1417 | unsigned int offset; |
139e5616 | 1418 | |
449dd698 JW |
1419 | offset = node->path >> RADIX_TREE_HEIGHT_SHIFT; |
1420 | parent->slots[offset] = NULL; | |
139e5616 JW |
1421 | parent->count--; |
1422 | } else { | |
1423 | root_tag_clear_all(root); | |
1424 | root->height = 0; | |
1425 | root->rnode = NULL; | |
1426 | } | |
1427 | ||
1428 | radix_tree_node_free(node); | |
1429 | deleted = true; | |
1430 | ||
1431 | node = parent; | |
1432 | } while (node); | |
1433 | ||
1434 | return deleted; | |
1435 | } | |
1436 | ||
1da177e4 | 1437 | /** |
53c59f26 | 1438 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1439 | * @root: radix tree root |
1440 | * @index: index key | |
53c59f26 | 1441 | * @item: expected item |
1da177e4 | 1442 | * |
53c59f26 | 1443 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1444 | * |
53c59f26 JW |
1445 | * Returns the address of the deleted item, or NULL if it was not present |
1446 | * or the entry at the given @index was not @item. | |
1da177e4 | 1447 | */ |
53c59f26 JW |
1448 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1449 | unsigned long index, void *item) | |
1da177e4 | 1450 | { |
139e5616 | 1451 | struct radix_tree_node *node; |
e6145236 | 1452 | unsigned int offset, i; |
139e5616 JW |
1453 | void **slot; |
1454 | void *entry; | |
d5274261 | 1455 | int tag; |
1da177e4 | 1456 | |
139e5616 JW |
1457 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1458 | if (!entry) | |
1459 | return NULL; | |
1da177e4 | 1460 | |
139e5616 JW |
1461 | if (item && entry != item) |
1462 | return NULL; | |
1463 | ||
1464 | if (!node) { | |
612d6c19 NP |
1465 | root_tag_clear_all(root); |
1466 | root->rnode = NULL; | |
139e5616 | 1467 | return entry; |
612d6c19 | 1468 | } |
1da177e4 | 1469 | |
139e5616 | 1470 | offset = index & RADIX_TREE_MAP_MASK; |
53c59f26 | 1471 | |
1da177e4 | 1472 | /* |
e2bdb933 HD |
1473 | * Clear all tags associated with the item to be deleted. |
1474 | * This way of doing it would be inefficient, but seldom is any set. | |
1da177e4 | 1475 | */ |
daff89f3 | 1476 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
e2bdb933 | 1477 | if (tag_get(node, tag, offset)) |
612d6c19 | 1478 | radix_tree_tag_clear(root, index, tag); |
d5274261 | 1479 | } |
1da177e4 | 1480 | |
e6145236 MW |
1481 | /* Delete any sibling slots pointing to this slot */ |
1482 | for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) { | |
1483 | if (node->slots[offset + i] != ptr_to_indirect(slot)) | |
1484 | break; | |
1485 | node->slots[offset + i] = NULL; | |
1486 | node->count--; | |
1487 | } | |
139e5616 JW |
1488 | node->slots[offset] = NULL; |
1489 | node->count--; | |
e2bdb933 | 1490 | |
449dd698 | 1491 | __radix_tree_delete_node(root, node); |
612d6c19 | 1492 | |
139e5616 | 1493 | return entry; |
1da177e4 | 1494 | } |
53c59f26 JW |
1495 | EXPORT_SYMBOL(radix_tree_delete_item); |
1496 | ||
1497 | /** | |
1498 | * radix_tree_delete - delete an item from a radix tree | |
1499 | * @root: radix tree root | |
1500 | * @index: index key | |
1501 | * | |
1502 | * Remove the item at @index from the radix tree rooted at @root. | |
1503 | * | |
1504 | * Returns the address of the deleted item, or NULL if it was not present. | |
1505 | */ | |
1506 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1507 | { | |
1508 | return radix_tree_delete_item(root, index, NULL); | |
1509 | } | |
1da177e4 LT |
1510 | EXPORT_SYMBOL(radix_tree_delete); |
1511 | ||
1512 | /** | |
1513 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1514 | * @root: radix tree root | |
1515 | * @tag: tag to test | |
1516 | */ | |
daff89f3 | 1517 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1518 | { |
612d6c19 | 1519 | return root_tag_get(root, tag); |
1da177e4 LT |
1520 | } |
1521 | EXPORT_SYMBOL(radix_tree_tagged); | |
1522 | ||
1523 | static void | |
449dd698 | 1524 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1525 | { |
449dd698 JW |
1526 | struct radix_tree_node *node = arg; |
1527 | ||
1528 | memset(node, 0, sizeof(*node)); | |
1529 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1530 | } |
1531 | ||
1532 | static __init unsigned long __maxindex(unsigned int height) | |
1533 | { | |
430d275a PL |
1534 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
1535 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1536 | ||
1537 | if (shift < 0) | |
1538 | return ~0UL; | |
1539 | if (shift >= BITS_PER_LONG) | |
1540 | return 0UL; | |
1541 | return ~0UL >> shift; | |
1da177e4 LT |
1542 | } |
1543 | ||
1544 | static __init void radix_tree_init_maxindex(void) | |
1545 | { | |
1546 | unsigned int i; | |
1547 | ||
1548 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1549 | height_to_maxindex[i] = __maxindex(i); | |
1550 | } | |
1551 | ||
1da177e4 LT |
1552 | static int radix_tree_callback(struct notifier_block *nfb, |
1553 | unsigned long action, | |
1554 | void *hcpu) | |
1555 | { | |
1556 | int cpu = (long)hcpu; | |
1557 | struct radix_tree_preload *rtp; | |
9d2a8da0 | 1558 | struct radix_tree_node *node; |
1da177e4 LT |
1559 | |
1560 | /* Free per-cpu pool of perloaded nodes */ | |
8bb78442 | 1561 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
1562 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1563 | while (rtp->nr) { | |
9d2a8da0 KS |
1564 | node = rtp->nodes; |
1565 | rtp->nodes = node->private_data; | |
1566 | kmem_cache_free(radix_tree_node_cachep, node); | |
1567 | rtp->nr--; | |
1da177e4 LT |
1568 | } |
1569 | } | |
1570 | return NOTIFY_OK; | |
1571 | } | |
1da177e4 LT |
1572 | |
1573 | void __init radix_tree_init(void) | |
1574 | { | |
1575 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1576 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1577 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1578 | radix_tree_node_ctor); | |
1da177e4 LT |
1579 | radix_tree_init_maxindex(); |
1580 | hotcpu_notifier(radix_tree_callback, 0); | |
1581 | } |