Commit | Line | Data |
---|---|---|
54a611b6 LH |
1 | // SPDX-License-Identifier: GPL-2.0+ |
2 | /* | |
3 | * Maple Tree implementation | |
4 | * Copyright (c) 2018-2022 Oracle Corporation | |
5 | * Authors: Liam R. Howlett <Liam.Howlett@oracle.com> | |
6 | * Matthew Wilcox <willy@infradead.org> | |
7 | */ | |
8 | ||
9 | /* | |
10 | * DOC: Interesting implementation details of the Maple Tree | |
11 | * | |
12 | * Each node type has a number of slots for entries and a number of slots for | |
13 | * pivots. In the case of dense nodes, the pivots are implied by the position | |
14 | * and are simply the slot index + the minimum of the node. | |
15 | * | |
16 | * In regular B-Tree terms, pivots are called keys. The term pivot is used to | |
17 | * indicate that the tree is specifying ranges, Pivots may appear in the | |
18 | * subtree with an entry attached to the value where as keys are unique to a | |
19 | * specific position of a B-tree. Pivot values are inclusive of the slot with | |
20 | * the same index. | |
21 | * | |
22 | * | |
23 | * The following illustrates the layout of a range64 nodes slots and pivots. | |
24 | * | |
25 | * | |
26 | * Slots -> | 0 | 1 | 2 | ... | 12 | 13 | 14 | 15 | | |
27 | * ┬ ┬ ┬ ┬ ┬ ┬ ┬ ┬ ┬ | |
28 | * │ │ │ │ │ │ │ │ └─ Implied maximum | |
29 | * │ │ │ │ │ │ │ └─ Pivot 14 | |
30 | * │ │ │ │ │ │ └─ Pivot 13 | |
31 | * │ │ │ │ │ └─ Pivot 12 | |
32 | * │ │ │ │ └─ Pivot 11 | |
33 | * │ │ │ └─ Pivot 2 | |
34 | * │ │ └─ Pivot 1 | |
35 | * │ └─ Pivot 0 | |
36 | * └─ Implied minimum | |
37 | * | |
38 | * Slot contents: | |
39 | * Internal (non-leaf) nodes contain pointers to other nodes. | |
40 | * Leaf nodes contain entries. | |
41 | * | |
42 | * The location of interest is often referred to as an offset. All offsets have | |
43 | * a slot, but the last offset has an implied pivot from the node above (or | |
44 | * UINT_MAX for the root node. | |
45 | * | |
46 | * Ranges complicate certain write activities. When modifying any of | |
47 | * the B-tree variants, it is known that one entry will either be added or | |
48 | * deleted. When modifying the Maple Tree, one store operation may overwrite | |
49 | * the entire data set, or one half of the tree, or the middle half of the tree. | |
50 | * | |
51 | */ | |
52 | ||
53 | ||
54 | #include <linux/maple_tree.h> | |
55 | #include <linux/xarray.h> | |
56 | #include <linux/types.h> | |
57 | #include <linux/export.h> | |
58 | #include <linux/slab.h> | |
59 | #include <linux/limits.h> | |
60 | #include <asm/barrier.h> | |
61 | ||
62 | #define CREATE_TRACE_POINTS | |
63 | #include <trace/events/maple_tree.h> | |
64 | ||
65 | #define MA_ROOT_PARENT 1 | |
66 | ||
67 | /* | |
68 | * Maple state flags | |
69 | * * MA_STATE_BULK - Bulk insert mode | |
70 | * * MA_STATE_REBALANCE - Indicate a rebalance during bulk insert | |
71 | * * MA_STATE_PREALLOC - Preallocated nodes, WARN_ON allocation | |
72 | */ | |
73 | #define MA_STATE_BULK 1 | |
74 | #define MA_STATE_REBALANCE 2 | |
75 | #define MA_STATE_PREALLOC 4 | |
76 | ||
77 | #define ma_parent_ptr(x) ((struct maple_pnode *)(x)) | |
78 | #define ma_mnode_ptr(x) ((struct maple_node *)(x)) | |
79 | #define ma_enode_ptr(x) ((struct maple_enode *)(x)) | |
80 | static struct kmem_cache *maple_node_cache; | |
81 | ||
82 | #ifdef CONFIG_DEBUG_MAPLE_TREE | |
83 | static const unsigned long mt_max[] = { | |
84 | [maple_dense] = MAPLE_NODE_SLOTS, | |
85 | [maple_leaf_64] = ULONG_MAX, | |
86 | [maple_range_64] = ULONG_MAX, | |
87 | [maple_arange_64] = ULONG_MAX, | |
88 | }; | |
89 | #define mt_node_max(x) mt_max[mte_node_type(x)] | |
90 | #endif | |
91 | ||
92 | static const unsigned char mt_slots[] = { | |
93 | [maple_dense] = MAPLE_NODE_SLOTS, | |
94 | [maple_leaf_64] = MAPLE_RANGE64_SLOTS, | |
95 | [maple_range_64] = MAPLE_RANGE64_SLOTS, | |
96 | [maple_arange_64] = MAPLE_ARANGE64_SLOTS, | |
97 | }; | |
98 | #define mt_slot_count(x) mt_slots[mte_node_type(x)] | |
99 | ||
100 | static const unsigned char mt_pivots[] = { | |
101 | [maple_dense] = 0, | |
102 | [maple_leaf_64] = MAPLE_RANGE64_SLOTS - 1, | |
103 | [maple_range_64] = MAPLE_RANGE64_SLOTS - 1, | |
104 | [maple_arange_64] = MAPLE_ARANGE64_SLOTS - 1, | |
105 | }; | |
106 | #define mt_pivot_count(x) mt_pivots[mte_node_type(x)] | |
107 | ||
108 | static const unsigned char mt_min_slots[] = { | |
109 | [maple_dense] = MAPLE_NODE_SLOTS / 2, | |
110 | [maple_leaf_64] = (MAPLE_RANGE64_SLOTS / 2) - 2, | |
111 | [maple_range_64] = (MAPLE_RANGE64_SLOTS / 2) - 2, | |
112 | [maple_arange_64] = (MAPLE_ARANGE64_SLOTS / 2) - 1, | |
113 | }; | |
114 | #define mt_min_slot_count(x) mt_min_slots[mte_node_type(x)] | |
115 | ||
116 | #define MAPLE_BIG_NODE_SLOTS (MAPLE_RANGE64_SLOTS * 2 + 2) | |
117 | #define MAPLE_BIG_NODE_GAPS (MAPLE_ARANGE64_SLOTS * 2 + 1) | |
118 | ||
119 | struct maple_big_node { | |
120 | struct maple_pnode *parent; | |
121 | unsigned long pivot[MAPLE_BIG_NODE_SLOTS - 1]; | |
122 | union { | |
123 | struct maple_enode *slot[MAPLE_BIG_NODE_SLOTS]; | |
124 | struct { | |
125 | unsigned long padding[MAPLE_BIG_NODE_GAPS]; | |
126 | unsigned long gap[MAPLE_BIG_NODE_GAPS]; | |
127 | }; | |
128 | }; | |
129 | unsigned char b_end; | |
130 | enum maple_type type; | |
131 | }; | |
132 | ||
133 | /* | |
134 | * The maple_subtree_state is used to build a tree to replace a segment of an | |
135 | * existing tree in a more atomic way. Any walkers of the older tree will hit a | |
136 | * dead node and restart on updates. | |
137 | */ | |
138 | struct maple_subtree_state { | |
139 | struct ma_state *orig_l; /* Original left side of subtree */ | |
140 | struct ma_state *orig_r; /* Original right side of subtree */ | |
141 | struct ma_state *l; /* New left side of subtree */ | |
142 | struct ma_state *m; /* New middle of subtree (rare) */ | |
143 | struct ma_state *r; /* New right side of subtree */ | |
144 | struct ma_topiary *free; /* nodes to be freed */ | |
145 | struct ma_topiary *destroy; /* Nodes to be destroyed (walked and freed) */ | |
146 | struct maple_big_node *bn; | |
147 | }; | |
148 | ||
44081c77 AB |
149 | #ifdef CONFIG_KASAN_STACK |
150 | /* Prevent mas_wr_bnode() from exceeding the stack frame limit */ | |
151 | #define noinline_for_kasan noinline_for_stack | |
152 | #else | |
153 | #define noinline_for_kasan inline | |
154 | #endif | |
155 | ||
54a611b6 LH |
156 | /* Functions */ |
157 | static inline struct maple_node *mt_alloc_one(gfp_t gfp) | |
158 | { | |
541e06b7 | 159 | return kmem_cache_alloc(maple_node_cache, gfp); |
54a611b6 LH |
160 | } |
161 | ||
162 | static inline int mt_alloc_bulk(gfp_t gfp, size_t size, void **nodes) | |
163 | { | |
541e06b7 | 164 | return kmem_cache_alloc_bulk(maple_node_cache, gfp, size, nodes); |
54a611b6 LH |
165 | } |
166 | ||
167 | static inline void mt_free_bulk(size_t size, void __rcu **nodes) | |
168 | { | |
169 | kmem_cache_free_bulk(maple_node_cache, size, (void **)nodes); | |
170 | } | |
171 | ||
172 | static void mt_free_rcu(struct rcu_head *head) | |
173 | { | |
174 | struct maple_node *node = container_of(head, struct maple_node, rcu); | |
175 | ||
176 | kmem_cache_free(maple_node_cache, node); | |
177 | } | |
178 | ||
179 | /* | |
180 | * ma_free_rcu() - Use rcu callback to free a maple node | |
181 | * @node: The node to free | |
182 | * | |
183 | * The maple tree uses the parent pointer to indicate this node is no longer in | |
184 | * use and will be freed. | |
185 | */ | |
186 | static void ma_free_rcu(struct maple_node *node) | |
187 | { | |
c13af03d | 188 | WARN_ON(node->parent != ma_parent_ptr(node)); |
54a611b6 LH |
189 | call_rcu(&node->rcu, mt_free_rcu); |
190 | } | |
191 | ||
54a611b6 LH |
192 | static void mas_set_height(struct ma_state *mas) |
193 | { | |
194 | unsigned int new_flags = mas->tree->ma_flags; | |
195 | ||
196 | new_flags &= ~MT_FLAGS_HEIGHT_MASK; | |
197 | BUG_ON(mas->depth > MAPLE_HEIGHT_MAX); | |
198 | new_flags |= mas->depth << MT_FLAGS_HEIGHT_OFFSET; | |
199 | mas->tree->ma_flags = new_flags; | |
200 | } | |
201 | ||
202 | static unsigned int mas_mt_height(struct ma_state *mas) | |
203 | { | |
204 | return mt_height(mas->tree); | |
205 | } | |
206 | ||
207 | static inline enum maple_type mte_node_type(const struct maple_enode *entry) | |
208 | { | |
209 | return ((unsigned long)entry >> MAPLE_NODE_TYPE_SHIFT) & | |
210 | MAPLE_NODE_TYPE_MASK; | |
211 | } | |
212 | ||
213 | static inline bool ma_is_dense(const enum maple_type type) | |
214 | { | |
215 | return type < maple_leaf_64; | |
216 | } | |
217 | ||
218 | static inline bool ma_is_leaf(const enum maple_type type) | |
219 | { | |
220 | return type < maple_range_64; | |
221 | } | |
222 | ||
223 | static inline bool mte_is_leaf(const struct maple_enode *entry) | |
224 | { | |
225 | return ma_is_leaf(mte_node_type(entry)); | |
226 | } | |
227 | ||
228 | /* | |
229 | * We also reserve values with the bottom two bits set to '10' which are | |
230 | * below 4096 | |
231 | */ | |
232 | static inline bool mt_is_reserved(const void *entry) | |
233 | { | |
234 | return ((unsigned long)entry < MAPLE_RESERVED_RANGE) && | |
235 | xa_is_internal(entry); | |
236 | } | |
237 | ||
238 | static inline void mas_set_err(struct ma_state *mas, long err) | |
239 | { | |
240 | mas->node = MA_ERROR(err); | |
241 | } | |
242 | ||
243 | static inline bool mas_is_ptr(struct ma_state *mas) | |
244 | { | |
245 | return mas->node == MAS_ROOT; | |
246 | } | |
247 | ||
248 | static inline bool mas_is_start(struct ma_state *mas) | |
249 | { | |
250 | return mas->node == MAS_START; | |
251 | } | |
252 | ||
253 | bool mas_is_err(struct ma_state *mas) | |
254 | { | |
255 | return xa_is_err(mas->node); | |
256 | } | |
257 | ||
258 | static inline bool mas_searchable(struct ma_state *mas) | |
259 | { | |
260 | if (mas_is_none(mas)) | |
261 | return false; | |
262 | ||
263 | if (mas_is_ptr(mas)) | |
264 | return false; | |
265 | ||
266 | return true; | |
267 | } | |
268 | ||
269 | static inline struct maple_node *mte_to_node(const struct maple_enode *entry) | |
270 | { | |
271 | return (struct maple_node *)((unsigned long)entry & ~MAPLE_NODE_MASK); | |
272 | } | |
273 | ||
274 | /* | |
275 | * mte_to_mat() - Convert a maple encoded node to a maple topiary node. | |
276 | * @entry: The maple encoded node | |
277 | * | |
278 | * Return: a maple topiary pointer | |
279 | */ | |
280 | static inline struct maple_topiary *mte_to_mat(const struct maple_enode *entry) | |
281 | { | |
282 | return (struct maple_topiary *) | |
283 | ((unsigned long)entry & ~MAPLE_NODE_MASK); | |
284 | } | |
285 | ||
286 | /* | |
287 | * mas_mn() - Get the maple state node. | |
288 | * @mas: The maple state | |
289 | * | |
290 | * Return: the maple node (not encoded - bare pointer). | |
291 | */ | |
292 | static inline struct maple_node *mas_mn(const struct ma_state *mas) | |
293 | { | |
294 | return mte_to_node(mas->node); | |
295 | } | |
296 | ||
297 | /* | |
298 | * mte_set_node_dead() - Set a maple encoded node as dead. | |
299 | * @mn: The maple encoded node. | |
300 | */ | |
301 | static inline void mte_set_node_dead(struct maple_enode *mn) | |
302 | { | |
303 | mte_to_node(mn)->parent = ma_parent_ptr(mte_to_node(mn)); | |
304 | smp_wmb(); /* Needed for RCU */ | |
305 | } | |
306 | ||
307 | /* Bit 1 indicates the root is a node */ | |
308 | #define MAPLE_ROOT_NODE 0x02 | |
309 | /* maple_type stored bit 3-6 */ | |
310 | #define MAPLE_ENODE_TYPE_SHIFT 0x03 | |
311 | /* Bit 2 means a NULL somewhere below */ | |
312 | #define MAPLE_ENODE_NULL 0x04 | |
313 | ||
314 | static inline struct maple_enode *mt_mk_node(const struct maple_node *node, | |
315 | enum maple_type type) | |
316 | { | |
317 | return (void *)((unsigned long)node | | |
318 | (type << MAPLE_ENODE_TYPE_SHIFT) | MAPLE_ENODE_NULL); | |
319 | } | |
320 | ||
321 | static inline void *mte_mk_root(const struct maple_enode *node) | |
322 | { | |
323 | return (void *)((unsigned long)node | MAPLE_ROOT_NODE); | |
324 | } | |
325 | ||
326 | static inline void *mte_safe_root(const struct maple_enode *node) | |
327 | { | |
328 | return (void *)((unsigned long)node & ~MAPLE_ROOT_NODE); | |
329 | } | |
330 | ||
6e7ba8b5 | 331 | static inline void *mte_set_full(const struct maple_enode *node) |
54a611b6 | 332 | { |
6e7ba8b5 | 333 | return (void *)((unsigned long)node & ~MAPLE_ENODE_NULL); |
54a611b6 LH |
334 | } |
335 | ||
6e7ba8b5 | 336 | static inline void *mte_clear_full(const struct maple_enode *node) |
54a611b6 | 337 | { |
6e7ba8b5 LH |
338 | return (void *)((unsigned long)node | MAPLE_ENODE_NULL); |
339 | } | |
340 | ||
341 | static inline bool mte_has_null(const struct maple_enode *node) | |
342 | { | |
343 | return (unsigned long)node & MAPLE_ENODE_NULL; | |
54a611b6 LH |
344 | } |
345 | ||
346 | static inline bool ma_is_root(struct maple_node *node) | |
347 | { | |
348 | return ((unsigned long)node->parent & MA_ROOT_PARENT); | |
349 | } | |
350 | ||
351 | static inline bool mte_is_root(const struct maple_enode *node) | |
352 | { | |
353 | return ma_is_root(mte_to_node(node)); | |
354 | } | |
355 | ||
356 | static inline bool mas_is_root_limits(const struct ma_state *mas) | |
357 | { | |
358 | return !mas->min && mas->max == ULONG_MAX; | |
359 | } | |
360 | ||
361 | static inline bool mt_is_alloc(struct maple_tree *mt) | |
362 | { | |
363 | return (mt->ma_flags & MT_FLAGS_ALLOC_RANGE); | |
364 | } | |
365 | ||
366 | /* | |
367 | * The Parent Pointer | |
368 | * Excluding root, the parent pointer is 256B aligned like all other tree nodes. | |
369 | * When storing a 32 or 64 bit values, the offset can fit into 5 bits. The 16 | |
370 | * bit values need an extra bit to store the offset. This extra bit comes from | |
371 | * a reuse of the last bit in the node type. This is possible by using bit 1 to | |
372 | * indicate if bit 2 is part of the type or the slot. | |
373 | * | |
374 | * Note types: | |
375 | * 0x??1 = Root | |
376 | * 0x?00 = 16 bit nodes | |
377 | * 0x010 = 32 bit nodes | |
378 | * 0x110 = 64 bit nodes | |
379 | * | |
380 | * Slot size and alignment | |
381 | * 0b??1 : Root | |
382 | * 0b?00 : 16 bit values, type in 0-1, slot in 2-7 | |
383 | * 0b010 : 32 bit values, type in 0-2, slot in 3-7 | |
384 | * 0b110 : 64 bit values, type in 0-2, slot in 3-7 | |
385 | */ | |
386 | ||
387 | #define MAPLE_PARENT_ROOT 0x01 | |
388 | ||
389 | #define MAPLE_PARENT_SLOT_SHIFT 0x03 | |
390 | #define MAPLE_PARENT_SLOT_MASK 0xF8 | |
391 | ||
392 | #define MAPLE_PARENT_16B_SLOT_SHIFT 0x02 | |
393 | #define MAPLE_PARENT_16B_SLOT_MASK 0xFC | |
394 | ||
395 | #define MAPLE_PARENT_RANGE64 0x06 | |
396 | #define MAPLE_PARENT_RANGE32 0x04 | |
397 | #define MAPLE_PARENT_NOT_RANGE16 0x02 | |
398 | ||
399 | /* | |
400 | * mte_parent_shift() - Get the parent shift for the slot storage. | |
401 | * @parent: The parent pointer cast as an unsigned long | |
402 | * Return: The shift into that pointer to the star to of the slot | |
403 | */ | |
404 | static inline unsigned long mte_parent_shift(unsigned long parent) | |
405 | { | |
406 | /* Note bit 1 == 0 means 16B */ | |
407 | if (likely(parent & MAPLE_PARENT_NOT_RANGE16)) | |
408 | return MAPLE_PARENT_SLOT_SHIFT; | |
409 | ||
410 | return MAPLE_PARENT_16B_SLOT_SHIFT; | |
411 | } | |
412 | ||
413 | /* | |
414 | * mte_parent_slot_mask() - Get the slot mask for the parent. | |
415 | * @parent: The parent pointer cast as an unsigned long. | |
416 | * Return: The slot mask for that parent. | |
417 | */ | |
418 | static inline unsigned long mte_parent_slot_mask(unsigned long parent) | |
419 | { | |
420 | /* Note bit 1 == 0 means 16B */ | |
421 | if (likely(parent & MAPLE_PARENT_NOT_RANGE16)) | |
422 | return MAPLE_PARENT_SLOT_MASK; | |
423 | ||
424 | return MAPLE_PARENT_16B_SLOT_MASK; | |
425 | } | |
426 | ||
427 | /* | |
428 | * mas_parent_enum() - Return the maple_type of the parent from the stored | |
429 | * parent type. | |
430 | * @mas: The maple state | |
431 | * @node: The maple_enode to extract the parent's enum | |
432 | * Return: The node->parent maple_type | |
433 | */ | |
434 | static inline | |
435 | enum maple_type mte_parent_enum(struct maple_enode *p_enode, | |
436 | struct maple_tree *mt) | |
437 | { | |
438 | unsigned long p_type; | |
439 | ||
440 | p_type = (unsigned long)p_enode; | |
441 | if (p_type & MAPLE_PARENT_ROOT) | |
442 | return 0; /* Validated in the caller. */ | |
443 | ||
444 | p_type &= MAPLE_NODE_MASK; | |
445 | p_type = p_type & ~(MAPLE_PARENT_ROOT | mte_parent_slot_mask(p_type)); | |
446 | ||
447 | switch (p_type) { | |
448 | case MAPLE_PARENT_RANGE64: /* or MAPLE_PARENT_ARANGE64 */ | |
449 | if (mt_is_alloc(mt)) | |
450 | return maple_arange_64; | |
451 | return maple_range_64; | |
452 | } | |
453 | ||
454 | return 0; | |
455 | } | |
456 | ||
457 | static inline | |
458 | enum maple_type mas_parent_enum(struct ma_state *mas, struct maple_enode *enode) | |
459 | { | |
460 | return mte_parent_enum(ma_enode_ptr(mte_to_node(enode)->parent), mas->tree); | |
461 | } | |
462 | ||
463 | /* | |
464 | * mte_set_parent() - Set the parent node and encode the slot | |
465 | * @enode: The encoded maple node. | |
466 | * @parent: The encoded maple node that is the parent of @enode. | |
467 | * @slot: The slot that @enode resides in @parent. | |
468 | * | |
469 | * Slot number is encoded in the enode->parent bit 3-6 or 2-6, depending on the | |
470 | * parent type. | |
471 | */ | |
472 | static inline | |
473 | void mte_set_parent(struct maple_enode *enode, const struct maple_enode *parent, | |
474 | unsigned char slot) | |
475 | { | |
831978e3 | 476 | unsigned long val = (unsigned long)parent; |
54a611b6 LH |
477 | unsigned long shift; |
478 | unsigned long type; | |
479 | enum maple_type p_type = mte_node_type(parent); | |
480 | ||
481 | BUG_ON(p_type == maple_dense); | |
482 | BUG_ON(p_type == maple_leaf_64); | |
483 | ||
484 | switch (p_type) { | |
485 | case maple_range_64: | |
486 | case maple_arange_64: | |
487 | shift = MAPLE_PARENT_SLOT_SHIFT; | |
488 | type = MAPLE_PARENT_RANGE64; | |
489 | break; | |
490 | default: | |
491 | case maple_dense: | |
492 | case maple_leaf_64: | |
493 | shift = type = 0; | |
494 | break; | |
495 | } | |
496 | ||
497 | val &= ~MAPLE_NODE_MASK; /* Clear all node metadata in parent */ | |
498 | val |= (slot << shift) | type; | |
499 | mte_to_node(enode)->parent = ma_parent_ptr(val); | |
500 | } | |
501 | ||
502 | /* | |
503 | * mte_parent_slot() - get the parent slot of @enode. | |
504 | * @enode: The encoded maple node. | |
505 | * | |
506 | * Return: The slot in the parent node where @enode resides. | |
507 | */ | |
508 | static inline unsigned int mte_parent_slot(const struct maple_enode *enode) | |
509 | { | |
831978e3 | 510 | unsigned long val = (unsigned long)mte_to_node(enode)->parent; |
54a611b6 | 511 | |
84fd3e1e | 512 | if (val & MA_ROOT_PARENT) |
54a611b6 LH |
513 | return 0; |
514 | ||
515 | /* | |
516 | * Okay to use MAPLE_PARENT_16B_SLOT_MASK as the last bit will be lost | |
517 | * by shift if the parent shift is MAPLE_PARENT_SLOT_SHIFT | |
518 | */ | |
519 | return (val & MAPLE_PARENT_16B_SLOT_MASK) >> mte_parent_shift(val); | |
520 | } | |
521 | ||
522 | /* | |
523 | * mte_parent() - Get the parent of @node. | |
524 | * @node: The encoded maple node. | |
525 | * | |
526 | * Return: The parent maple node. | |
527 | */ | |
528 | static inline struct maple_node *mte_parent(const struct maple_enode *enode) | |
529 | { | |
530 | return (void *)((unsigned long) | |
531 | (mte_to_node(enode)->parent) & ~MAPLE_NODE_MASK); | |
532 | } | |
533 | ||
534 | /* | |
535 | * ma_dead_node() - check if the @enode is dead. | |
536 | * @enode: The encoded maple node | |
537 | * | |
538 | * Return: true if dead, false otherwise. | |
539 | */ | |
540 | static inline bool ma_dead_node(const struct maple_node *node) | |
541 | { | |
0a2b18d9 | 542 | struct maple_node *parent; |
54a611b6 | 543 | |
0a2b18d9 LH |
544 | /* Do not reorder reads from the node prior to the parent check */ |
545 | smp_rmb(); | |
546 | parent = (void *)((unsigned long) node->parent & ~MAPLE_NODE_MASK); | |
54a611b6 LH |
547 | return (parent == node); |
548 | } | |
39d0bd86 | 549 | |
54a611b6 LH |
550 | /* |
551 | * mte_dead_node() - check if the @enode is dead. | |
552 | * @enode: The encoded maple node | |
553 | * | |
554 | * Return: true if dead, false otherwise. | |
555 | */ | |
556 | static inline bool mte_dead_node(const struct maple_enode *enode) | |
557 | { | |
558 | struct maple_node *parent, *node; | |
559 | ||
560 | node = mte_to_node(enode); | |
0a2b18d9 LH |
561 | /* Do not reorder reads from the node prior to the parent check */ |
562 | smp_rmb(); | |
54a611b6 LH |
563 | parent = mte_parent(enode); |
564 | return (parent == node); | |
565 | } | |
566 | ||
567 | /* | |
568 | * mas_allocated() - Get the number of nodes allocated in a maple state. | |
569 | * @mas: The maple state | |
570 | * | |
571 | * The ma_state alloc member is overloaded to hold a pointer to the first | |
572 | * allocated node or to the number of requested nodes to allocate. If bit 0 is | |
573 | * set, then the alloc contains the number of requested nodes. If there is an | |
574 | * allocated node, then the total allocated nodes is in that node. | |
575 | * | |
576 | * Return: The total number of nodes allocated | |
577 | */ | |
578 | static inline unsigned long mas_allocated(const struct ma_state *mas) | |
579 | { | |
580 | if (!mas->alloc || ((unsigned long)mas->alloc & 0x1)) | |
581 | return 0; | |
582 | ||
583 | return mas->alloc->total; | |
584 | } | |
585 | ||
586 | /* | |
587 | * mas_set_alloc_req() - Set the requested number of allocations. | |
588 | * @mas: the maple state | |
589 | * @count: the number of allocations. | |
590 | * | |
591 | * The requested number of allocations is either in the first allocated node, | |
592 | * located in @mas->alloc->request_count, or directly in @mas->alloc if there is | |
593 | * no allocated node. Set the request either in the node or do the necessary | |
594 | * encoding to store in @mas->alloc directly. | |
595 | */ | |
596 | static inline void mas_set_alloc_req(struct ma_state *mas, unsigned long count) | |
597 | { | |
598 | if (!mas->alloc || ((unsigned long)mas->alloc & 0x1)) { | |
599 | if (!count) | |
600 | mas->alloc = NULL; | |
601 | else | |
602 | mas->alloc = (struct maple_alloc *)(((count) << 1U) | 1U); | |
603 | return; | |
604 | } | |
605 | ||
606 | mas->alloc->request_count = count; | |
607 | } | |
608 | ||
609 | /* | |
610 | * mas_alloc_req() - get the requested number of allocations. | |
611 | * @mas: The maple state | |
612 | * | |
613 | * The alloc count is either stored directly in @mas, or in | |
614 | * @mas->alloc->request_count if there is at least one node allocated. Decode | |
615 | * the request count if it's stored directly in @mas->alloc. | |
616 | * | |
617 | * Return: The allocation request count. | |
618 | */ | |
619 | static inline unsigned int mas_alloc_req(const struct ma_state *mas) | |
620 | { | |
621 | if ((unsigned long)mas->alloc & 0x1) | |
622 | return (unsigned long)(mas->alloc) >> 1; | |
623 | else if (mas->alloc) | |
624 | return mas->alloc->request_count; | |
625 | return 0; | |
626 | } | |
627 | ||
628 | /* | |
629 | * ma_pivots() - Get a pointer to the maple node pivots. | |
630 | * @node - the maple node | |
631 | * @type - the node type | |
632 | * | |
39d0bd86 LH |
633 | * In the event of a dead node, this array may be %NULL |
634 | * | |
54a611b6 LH |
635 | * Return: A pointer to the maple node pivots |
636 | */ | |
637 | static inline unsigned long *ma_pivots(struct maple_node *node, | |
638 | enum maple_type type) | |
639 | { | |
640 | switch (type) { | |
641 | case maple_arange_64: | |
642 | return node->ma64.pivot; | |
643 | case maple_range_64: | |
644 | case maple_leaf_64: | |
645 | return node->mr64.pivot; | |
646 | case maple_dense: | |
647 | return NULL; | |
648 | } | |
649 | return NULL; | |
650 | } | |
651 | ||
652 | /* | |
653 | * ma_gaps() - Get a pointer to the maple node gaps. | |
654 | * @node - the maple node | |
655 | * @type - the node type | |
656 | * | |
657 | * Return: A pointer to the maple node gaps | |
658 | */ | |
659 | static inline unsigned long *ma_gaps(struct maple_node *node, | |
660 | enum maple_type type) | |
661 | { | |
662 | switch (type) { | |
663 | case maple_arange_64: | |
664 | return node->ma64.gap; | |
665 | case maple_range_64: | |
666 | case maple_leaf_64: | |
667 | case maple_dense: | |
668 | return NULL; | |
669 | } | |
670 | return NULL; | |
671 | } | |
672 | ||
673 | /* | |
674 | * mte_pivot() - Get the pivot at @piv of the maple encoded node. | |
675 | * @mn: The maple encoded node. | |
676 | * @piv: The pivot. | |
677 | * | |
678 | * Return: the pivot at @piv of @mn. | |
679 | */ | |
680 | static inline unsigned long mte_pivot(const struct maple_enode *mn, | |
681 | unsigned char piv) | |
682 | { | |
683 | struct maple_node *node = mte_to_node(mn); | |
ab6ef70a | 684 | enum maple_type type = mte_node_type(mn); |
54a611b6 | 685 | |
ab6ef70a | 686 | if (piv >= mt_pivots[type]) { |
54a611b6 LH |
687 | WARN_ON(1); |
688 | return 0; | |
689 | } | |
ab6ef70a | 690 | switch (type) { |
54a611b6 LH |
691 | case maple_arange_64: |
692 | return node->ma64.pivot[piv]; | |
693 | case maple_range_64: | |
694 | case maple_leaf_64: | |
695 | return node->mr64.pivot[piv]; | |
696 | case maple_dense: | |
697 | return 0; | |
698 | } | |
699 | return 0; | |
700 | } | |
701 | ||
702 | /* | |
703 | * mas_safe_pivot() - get the pivot at @piv or mas->max. | |
704 | * @mas: The maple state | |
705 | * @pivots: The pointer to the maple node pivots | |
706 | * @piv: The pivot to fetch | |
707 | * @type: The maple node type | |
708 | * | |
709 | * Return: The pivot at @piv within the limit of the @pivots array, @mas->max | |
710 | * otherwise. | |
711 | */ | |
712 | static inline unsigned long | |
713 | mas_safe_pivot(const struct ma_state *mas, unsigned long *pivots, | |
714 | unsigned char piv, enum maple_type type) | |
715 | { | |
716 | if (piv >= mt_pivots[type]) | |
717 | return mas->max; | |
718 | ||
719 | return pivots[piv]; | |
720 | } | |
721 | ||
722 | /* | |
723 | * mas_safe_min() - Return the minimum for a given offset. | |
724 | * @mas: The maple state | |
725 | * @pivots: The pointer to the maple node pivots | |
726 | * @offset: The offset into the pivot array | |
727 | * | |
728 | * Return: The minimum range value that is contained in @offset. | |
729 | */ | |
730 | static inline unsigned long | |
731 | mas_safe_min(struct ma_state *mas, unsigned long *pivots, unsigned char offset) | |
732 | { | |
733 | if (likely(offset)) | |
734 | return pivots[offset - 1] + 1; | |
735 | ||
736 | return mas->min; | |
737 | } | |
738 | ||
739 | /* | |
740 | * mas_logical_pivot() - Get the logical pivot of a given offset. | |
741 | * @mas: The maple state | |
742 | * @pivots: The pointer to the maple node pivots | |
743 | * @offset: The offset into the pivot array | |
744 | * @type: The maple node type | |
745 | * | |
746 | * When there is no value at a pivot (beyond the end of the data), then the | |
747 | * pivot is actually @mas->max. | |
748 | * | |
749 | * Return: the logical pivot of a given @offset. | |
750 | */ | |
751 | static inline unsigned long | |
752 | mas_logical_pivot(struct ma_state *mas, unsigned long *pivots, | |
753 | unsigned char offset, enum maple_type type) | |
754 | { | |
755 | unsigned long lpiv = mas_safe_pivot(mas, pivots, offset, type); | |
756 | ||
757 | if (likely(lpiv)) | |
758 | return lpiv; | |
759 | ||
760 | if (likely(offset)) | |
761 | return mas->max; | |
762 | ||
763 | return lpiv; | |
764 | } | |
765 | ||
766 | /* | |
767 | * mte_set_pivot() - Set a pivot to a value in an encoded maple node. | |
768 | * @mn: The encoded maple node | |
769 | * @piv: The pivot offset | |
770 | * @val: The value of the pivot | |
771 | */ | |
772 | static inline void mte_set_pivot(struct maple_enode *mn, unsigned char piv, | |
773 | unsigned long val) | |
774 | { | |
775 | struct maple_node *node = mte_to_node(mn); | |
776 | enum maple_type type = mte_node_type(mn); | |
777 | ||
778 | BUG_ON(piv >= mt_pivots[type]); | |
779 | switch (type) { | |
780 | default: | |
781 | case maple_range_64: | |
782 | case maple_leaf_64: | |
783 | node->mr64.pivot[piv] = val; | |
784 | break; | |
785 | case maple_arange_64: | |
786 | node->ma64.pivot[piv] = val; | |
787 | break; | |
788 | case maple_dense: | |
789 | break; | |
790 | } | |
791 | ||
792 | } | |
793 | ||
794 | /* | |
795 | * ma_slots() - Get a pointer to the maple node slots. | |
796 | * @mn: The maple node | |
797 | * @mt: The maple node type | |
798 | * | |
799 | * Return: A pointer to the maple node slots | |
800 | */ | |
801 | static inline void __rcu **ma_slots(struct maple_node *mn, enum maple_type mt) | |
802 | { | |
803 | switch (mt) { | |
804 | default: | |
805 | case maple_arange_64: | |
806 | return mn->ma64.slot; | |
807 | case maple_range_64: | |
808 | case maple_leaf_64: | |
809 | return mn->mr64.slot; | |
810 | case maple_dense: | |
811 | return mn->slot; | |
812 | } | |
813 | } | |
814 | ||
815 | static inline bool mt_locked(const struct maple_tree *mt) | |
816 | { | |
817 | return mt_external_lock(mt) ? mt_lock_is_held(mt) : | |
818 | lockdep_is_held(&mt->ma_lock); | |
819 | } | |
820 | ||
821 | static inline void *mt_slot(const struct maple_tree *mt, | |
822 | void __rcu **slots, unsigned char offset) | |
823 | { | |
824 | return rcu_dereference_check(slots[offset], mt_locked(mt)); | |
825 | } | |
826 | ||
790e1fa8 LH |
827 | static inline void *mt_slot_locked(struct maple_tree *mt, void __rcu **slots, |
828 | unsigned char offset) | |
829 | { | |
830 | return rcu_dereference_protected(slots[offset], mt_locked(mt)); | |
831 | } | |
54a611b6 LH |
832 | /* |
833 | * mas_slot_locked() - Get the slot value when holding the maple tree lock. | |
834 | * @mas: The maple state | |
835 | * @slots: The pointer to the slots | |
836 | * @offset: The offset into the slots array to fetch | |
837 | * | |
838 | * Return: The entry stored in @slots at the @offset. | |
839 | */ | |
840 | static inline void *mas_slot_locked(struct ma_state *mas, void __rcu **slots, | |
841 | unsigned char offset) | |
842 | { | |
790e1fa8 | 843 | return mt_slot_locked(mas->tree, slots, offset); |
54a611b6 LH |
844 | } |
845 | ||
846 | /* | |
847 | * mas_slot() - Get the slot value when not holding the maple tree lock. | |
848 | * @mas: The maple state | |
849 | * @slots: The pointer to the slots | |
850 | * @offset: The offset into the slots array to fetch | |
851 | * | |
852 | * Return: The entry stored in @slots at the @offset | |
853 | */ | |
854 | static inline void *mas_slot(struct ma_state *mas, void __rcu **slots, | |
855 | unsigned char offset) | |
856 | { | |
857 | return mt_slot(mas->tree, slots, offset); | |
858 | } | |
859 | ||
860 | /* | |
861 | * mas_root() - Get the maple tree root. | |
862 | * @mas: The maple state. | |
863 | * | |
864 | * Return: The pointer to the root of the tree | |
865 | */ | |
866 | static inline void *mas_root(struct ma_state *mas) | |
867 | { | |
868 | return rcu_dereference_check(mas->tree->ma_root, mt_locked(mas->tree)); | |
869 | } | |
870 | ||
871 | static inline void *mt_root_locked(struct maple_tree *mt) | |
872 | { | |
873 | return rcu_dereference_protected(mt->ma_root, mt_locked(mt)); | |
874 | } | |
875 | ||
876 | /* | |
877 | * mas_root_locked() - Get the maple tree root when holding the maple tree lock. | |
878 | * @mas: The maple state. | |
879 | * | |
880 | * Return: The pointer to the root of the tree | |
881 | */ | |
882 | static inline void *mas_root_locked(struct ma_state *mas) | |
883 | { | |
884 | return mt_root_locked(mas->tree); | |
885 | } | |
886 | ||
887 | static inline struct maple_metadata *ma_meta(struct maple_node *mn, | |
888 | enum maple_type mt) | |
889 | { | |
890 | switch (mt) { | |
891 | case maple_arange_64: | |
892 | return &mn->ma64.meta; | |
893 | default: | |
894 | return &mn->mr64.meta; | |
895 | } | |
896 | } | |
897 | ||
898 | /* | |
899 | * ma_set_meta() - Set the metadata information of a node. | |
900 | * @mn: The maple node | |
901 | * @mt: The maple node type | |
902 | * @offset: The offset of the highest sub-gap in this node. | |
903 | * @end: The end of the data in this node. | |
904 | */ | |
905 | static inline void ma_set_meta(struct maple_node *mn, enum maple_type mt, | |
906 | unsigned char offset, unsigned char end) | |
907 | { | |
908 | struct maple_metadata *meta = ma_meta(mn, mt); | |
909 | ||
910 | meta->gap = offset; | |
911 | meta->end = end; | |
912 | } | |
913 | ||
2e5b4921 | 914 | /* |
790e1fa8 LH |
915 | * mt_clear_meta() - clear the metadata information of a node, if it exists |
916 | * @mt: The maple tree | |
2e5b4921 | 917 | * @mn: The maple node |
790e1fa8 | 918 | * @type: The maple node type |
2e5b4921 LH |
919 | * @offset: The offset of the highest sub-gap in this node. |
920 | * @end: The end of the data in this node. | |
921 | */ | |
790e1fa8 LH |
922 | static inline void mt_clear_meta(struct maple_tree *mt, struct maple_node *mn, |
923 | enum maple_type type) | |
2e5b4921 LH |
924 | { |
925 | struct maple_metadata *meta; | |
926 | unsigned long *pivots; | |
927 | void __rcu **slots; | |
928 | void *next; | |
929 | ||
790e1fa8 | 930 | switch (type) { |
2e5b4921 LH |
931 | case maple_range_64: |
932 | pivots = mn->mr64.pivot; | |
933 | if (unlikely(pivots[MAPLE_RANGE64_SLOTS - 2])) { | |
934 | slots = mn->mr64.slot; | |
790e1fa8 LH |
935 | next = mt_slot_locked(mt, slots, |
936 | MAPLE_RANGE64_SLOTS - 1); | |
937 | if (unlikely((mte_to_node(next) && | |
938 | mte_node_type(next)))) | |
939 | return; /* no metadata, could be node */ | |
2e5b4921 LH |
940 | } |
941 | fallthrough; | |
942 | case maple_arange_64: | |
790e1fa8 | 943 | meta = ma_meta(mn, type); |
2e5b4921 LH |
944 | break; |
945 | default: | |
946 | return; | |
947 | } | |
948 | ||
949 | meta->gap = 0; | |
950 | meta->end = 0; | |
951 | } | |
952 | ||
54a611b6 LH |
953 | /* |
954 | * ma_meta_end() - Get the data end of a node from the metadata | |
955 | * @mn: The maple node | |
956 | * @mt: The maple node type | |
957 | */ | |
958 | static inline unsigned char ma_meta_end(struct maple_node *mn, | |
959 | enum maple_type mt) | |
960 | { | |
961 | struct maple_metadata *meta = ma_meta(mn, mt); | |
962 | ||
963 | return meta->end; | |
964 | } | |
965 | ||
966 | /* | |
967 | * ma_meta_gap() - Get the largest gap location of a node from the metadata | |
968 | * @mn: The maple node | |
969 | * @mt: The maple node type | |
970 | */ | |
971 | static inline unsigned char ma_meta_gap(struct maple_node *mn, | |
972 | enum maple_type mt) | |
973 | { | |
974 | BUG_ON(mt != maple_arange_64); | |
975 | ||
976 | return mn->ma64.meta.gap; | |
977 | } | |
978 | ||
979 | /* | |
980 | * ma_set_meta_gap() - Set the largest gap location in a nodes metadata | |
981 | * @mn: The maple node | |
982 | * @mn: The maple node type | |
983 | * @offset: The location of the largest gap. | |
984 | */ | |
985 | static inline void ma_set_meta_gap(struct maple_node *mn, enum maple_type mt, | |
986 | unsigned char offset) | |
987 | { | |
988 | ||
989 | struct maple_metadata *meta = ma_meta(mn, mt); | |
990 | ||
991 | meta->gap = offset; | |
992 | } | |
993 | ||
994 | /* | |
995 | * mat_add() - Add a @dead_enode to the ma_topiary of a list of dead nodes. | |
996 | * @mat - the ma_topiary, a linked list of dead nodes. | |
997 | * @dead_enode - the node to be marked as dead and added to the tail of the list | |
998 | * | |
999 | * Add the @dead_enode to the linked list in @mat. | |
1000 | */ | |
1001 | static inline void mat_add(struct ma_topiary *mat, | |
1002 | struct maple_enode *dead_enode) | |
1003 | { | |
1004 | mte_set_node_dead(dead_enode); | |
1005 | mte_to_mat(dead_enode)->next = NULL; | |
1006 | if (!mat->tail) { | |
1007 | mat->tail = mat->head = dead_enode; | |
1008 | return; | |
1009 | } | |
1010 | ||
1011 | mte_to_mat(mat->tail)->next = dead_enode; | |
1012 | mat->tail = dead_enode; | |
1013 | } | |
1014 | ||
1015 | static void mte_destroy_walk(struct maple_enode *, struct maple_tree *); | |
1016 | static inline void mas_free(struct ma_state *mas, struct maple_enode *used); | |
1017 | ||
1018 | /* | |
1019 | * mas_mat_free() - Free all nodes in a dead list. | |
1020 | * @mas - the maple state | |
1021 | * @mat - the ma_topiary linked list of dead nodes to free. | |
1022 | * | |
1023 | * Free walk a dead list. | |
1024 | */ | |
1025 | static void mas_mat_free(struct ma_state *mas, struct ma_topiary *mat) | |
1026 | { | |
1027 | struct maple_enode *next; | |
1028 | ||
1029 | while (mat->head) { | |
1030 | next = mte_to_mat(mat->head)->next; | |
1031 | mas_free(mas, mat->head); | |
1032 | mat->head = next; | |
1033 | } | |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * mas_mat_destroy() - Free all nodes and subtrees in a dead list. | |
1038 | * @mas - the maple state | |
1039 | * @mat - the ma_topiary linked list of dead nodes to free. | |
1040 | * | |
1041 | * Destroy walk a dead list. | |
1042 | */ | |
1043 | static void mas_mat_destroy(struct ma_state *mas, struct ma_topiary *mat) | |
1044 | { | |
1045 | struct maple_enode *next; | |
1046 | ||
1047 | while (mat->head) { | |
1048 | next = mte_to_mat(mat->head)->next; | |
1049 | mte_destroy_walk(mat->head, mat->mtree); | |
1050 | mat->head = next; | |
1051 | } | |
1052 | } | |
1053 | /* | |
1054 | * mas_descend() - Descend into the slot stored in the ma_state. | |
1055 | * @mas - the maple state. | |
1056 | * | |
1057 | * Note: Not RCU safe, only use in write side or debug code. | |
1058 | */ | |
1059 | static inline void mas_descend(struct ma_state *mas) | |
1060 | { | |
1061 | enum maple_type type; | |
1062 | unsigned long *pivots; | |
1063 | struct maple_node *node; | |
1064 | void __rcu **slots; | |
1065 | ||
1066 | node = mas_mn(mas); | |
1067 | type = mte_node_type(mas->node); | |
1068 | pivots = ma_pivots(node, type); | |
1069 | slots = ma_slots(node, type); | |
1070 | ||
1071 | if (mas->offset) | |
1072 | mas->min = pivots[mas->offset - 1] + 1; | |
1073 | mas->max = mas_safe_pivot(mas, pivots, mas->offset, type); | |
1074 | mas->node = mas_slot(mas, slots, mas->offset); | |
1075 | } | |
1076 | ||
1077 | /* | |
1078 | * mte_set_gap() - Set a maple node gap. | |
1079 | * @mn: The encoded maple node | |
1080 | * @gap: The offset of the gap to set | |
1081 | * @val: The gap value | |
1082 | */ | |
1083 | static inline void mte_set_gap(const struct maple_enode *mn, | |
1084 | unsigned char gap, unsigned long val) | |
1085 | { | |
1086 | switch (mte_node_type(mn)) { | |
1087 | default: | |
1088 | break; | |
1089 | case maple_arange_64: | |
1090 | mte_to_node(mn)->ma64.gap[gap] = val; | |
1091 | break; | |
1092 | } | |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * mas_ascend() - Walk up a level of the tree. | |
1097 | * @mas: The maple state | |
1098 | * | |
1099 | * Sets the @mas->max and @mas->min to the correct values when walking up. This | |
1100 | * may cause several levels of walking up to find the correct min and max. | |
1101 | * May find a dead node which will cause a premature return. | |
1102 | * Return: 1 on dead node, 0 otherwise | |
1103 | */ | |
1104 | static int mas_ascend(struct ma_state *mas) | |
1105 | { | |
1106 | struct maple_enode *p_enode; /* parent enode. */ | |
1107 | struct maple_enode *a_enode; /* ancestor enode. */ | |
1108 | struct maple_node *a_node; /* ancestor node. */ | |
1109 | struct maple_node *p_node; /* parent node. */ | |
1110 | unsigned char a_slot; | |
1111 | enum maple_type a_type; | |
1112 | unsigned long min, max; | |
1113 | unsigned long *pivots; | |
1114 | unsigned char offset; | |
1115 | bool set_max = false, set_min = false; | |
1116 | ||
1117 | a_node = mas_mn(mas); | |
1118 | if (ma_is_root(a_node)) { | |
1119 | mas->offset = 0; | |
1120 | return 0; | |
1121 | } | |
1122 | ||
1123 | p_node = mte_parent(mas->node); | |
1124 | if (unlikely(a_node == p_node)) | |
1125 | return 1; | |
1126 | a_type = mas_parent_enum(mas, mas->node); | |
1127 | offset = mte_parent_slot(mas->node); | |
1128 | a_enode = mt_mk_node(p_node, a_type); | |
1129 | ||
1130 | /* Check to make sure all parent information is still accurate */ | |
1131 | if (p_node != mte_parent(mas->node)) | |
1132 | return 1; | |
1133 | ||
1134 | mas->node = a_enode; | |
1135 | mas->offset = offset; | |
1136 | ||
1137 | if (mte_is_root(a_enode)) { | |
1138 | mas->max = ULONG_MAX; | |
1139 | mas->min = 0; | |
1140 | return 0; | |
1141 | } | |
1142 | ||
1143 | min = 0; | |
1144 | max = ULONG_MAX; | |
1145 | do { | |
1146 | p_enode = a_enode; | |
1147 | a_type = mas_parent_enum(mas, p_enode); | |
1148 | a_node = mte_parent(p_enode); | |
1149 | a_slot = mte_parent_slot(p_enode); | |
54a611b6 | 1150 | a_enode = mt_mk_node(a_node, a_type); |
39d0bd86 LH |
1151 | pivots = ma_pivots(a_node, a_type); |
1152 | ||
1153 | if (unlikely(ma_dead_node(a_node))) | |
1154 | return 1; | |
54a611b6 LH |
1155 | |
1156 | if (!set_min && a_slot) { | |
1157 | set_min = true; | |
1158 | min = pivots[a_slot - 1] + 1; | |
1159 | } | |
1160 | ||
1161 | if (!set_max && a_slot < mt_pivots[a_type]) { | |
1162 | set_max = true; | |
1163 | max = pivots[a_slot]; | |
1164 | } | |
1165 | ||
1166 | if (unlikely(ma_dead_node(a_node))) | |
1167 | return 1; | |
1168 | ||
1169 | if (unlikely(ma_is_root(a_node))) | |
1170 | break; | |
1171 | ||
1172 | } while (!set_min || !set_max); | |
1173 | ||
1174 | mas->max = max; | |
1175 | mas->min = min; | |
1176 | return 0; | |
1177 | } | |
1178 | ||
1179 | /* | |
1180 | * mas_pop_node() - Get a previously allocated maple node from the maple state. | |
1181 | * @mas: The maple state | |
1182 | * | |
1183 | * Return: A pointer to a maple node. | |
1184 | */ | |
1185 | static inline struct maple_node *mas_pop_node(struct ma_state *mas) | |
1186 | { | |
1187 | struct maple_alloc *ret, *node = mas->alloc; | |
1188 | unsigned long total = mas_allocated(mas); | |
541e06b7 | 1189 | unsigned int req = mas_alloc_req(mas); |
54a611b6 LH |
1190 | |
1191 | /* nothing or a request pending. */ | |
541e06b7 | 1192 | if (WARN_ON(!total)) |
54a611b6 LH |
1193 | return NULL; |
1194 | ||
1195 | if (total == 1) { | |
1196 | /* single allocation in this ma_state */ | |
1197 | mas->alloc = NULL; | |
1198 | ret = node; | |
1199 | goto single_node; | |
1200 | } | |
1201 | ||
541e06b7 | 1202 | if (node->node_count == 1) { |
54a611b6 LH |
1203 | /* Single allocation in this node. */ |
1204 | mas->alloc = node->slot[0]; | |
54a611b6 LH |
1205 | mas->alloc->total = node->total - 1; |
1206 | ret = node; | |
1207 | goto new_head; | |
1208 | } | |
54a611b6 | 1209 | node->total--; |
541e06b7 LH |
1210 | ret = node->slot[--node->node_count]; |
1211 | node->slot[node->node_count] = NULL; | |
54a611b6 LH |
1212 | |
1213 | single_node: | |
1214 | new_head: | |
541e06b7 LH |
1215 | if (req) { |
1216 | req++; | |
1217 | mas_set_alloc_req(mas, req); | |
54a611b6 | 1218 | } |
541e06b7 LH |
1219 | |
1220 | memset(ret, 0, sizeof(*ret)); | |
54a611b6 LH |
1221 | return (struct maple_node *)ret; |
1222 | } | |
1223 | ||
1224 | /* | |
1225 | * mas_push_node() - Push a node back on the maple state allocation. | |
1226 | * @mas: The maple state | |
1227 | * @used: The used maple node | |
1228 | * | |
1229 | * Stores the maple node back into @mas->alloc for reuse. Updates allocated and | |
1230 | * requested node count as necessary. | |
1231 | */ | |
1232 | static inline void mas_push_node(struct ma_state *mas, struct maple_node *used) | |
1233 | { | |
1234 | struct maple_alloc *reuse = (struct maple_alloc *)used; | |
1235 | struct maple_alloc *head = mas->alloc; | |
1236 | unsigned long count; | |
1237 | unsigned int requested = mas_alloc_req(mas); | |
1238 | ||
54a611b6 LH |
1239 | count = mas_allocated(mas); |
1240 | ||
541e06b7 LH |
1241 | reuse->request_count = 0; |
1242 | reuse->node_count = 0; | |
1243 | if (count && (head->node_count < MAPLE_ALLOC_SLOTS)) { | |
1244 | head->slot[head->node_count++] = reuse; | |
54a611b6 LH |
1245 | head->total++; |
1246 | goto done; | |
1247 | } | |
1248 | ||
1249 | reuse->total = 1; | |
1250 | if ((head) && !((unsigned long)head & 0x1)) { | |
54a611b6 | 1251 | reuse->slot[0] = head; |
541e06b7 | 1252 | reuse->node_count = 1; |
54a611b6 LH |
1253 | reuse->total += head->total; |
1254 | } | |
1255 | ||
1256 | mas->alloc = reuse; | |
1257 | done: | |
1258 | if (requested > 1) | |
1259 | mas_set_alloc_req(mas, requested - 1); | |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * mas_alloc_nodes() - Allocate nodes into a maple state | |
1264 | * @mas: The maple state | |
1265 | * @gfp: The GFP Flags | |
1266 | */ | |
1267 | static inline void mas_alloc_nodes(struct ma_state *mas, gfp_t gfp) | |
1268 | { | |
1269 | struct maple_alloc *node; | |
54a611b6 | 1270 | unsigned long allocated = mas_allocated(mas); |
54a611b6 LH |
1271 | unsigned int requested = mas_alloc_req(mas); |
1272 | unsigned int count; | |
1273 | void **slots = NULL; | |
1274 | unsigned int max_req = 0; | |
1275 | ||
1276 | if (!requested) | |
1277 | return; | |
1278 | ||
1279 | mas_set_alloc_req(mas, 0); | |
1280 | if (mas->mas_flags & MA_STATE_PREALLOC) { | |
1281 | if (allocated) | |
1282 | return; | |
1283 | WARN_ON(!allocated); | |
1284 | } | |
1285 | ||
541e06b7 | 1286 | if (!allocated || mas->alloc->node_count == MAPLE_ALLOC_SLOTS) { |
54a611b6 LH |
1287 | node = (struct maple_alloc *)mt_alloc_one(gfp); |
1288 | if (!node) | |
1289 | goto nomem_one; | |
1290 | ||
541e06b7 | 1291 | if (allocated) { |
54a611b6 | 1292 | node->slot[0] = mas->alloc; |
541e06b7 LH |
1293 | node->node_count = 1; |
1294 | } else { | |
1295 | node->node_count = 0; | |
1296 | } | |
54a611b6 | 1297 | |
54a611b6 | 1298 | mas->alloc = node; |
541e06b7 | 1299 | node->total = ++allocated; |
54a611b6 LH |
1300 | requested--; |
1301 | } | |
1302 | ||
1303 | node = mas->alloc; | |
541e06b7 | 1304 | node->request_count = 0; |
54a611b6 | 1305 | while (requested) { |
1f5f12ec PZ |
1306 | max_req = MAPLE_ALLOC_SLOTS - node->node_count; |
1307 | slots = (void **)&node->slot[node->node_count]; | |
54a611b6 LH |
1308 | max_req = min(requested, max_req); |
1309 | count = mt_alloc_bulk(gfp, max_req, slots); | |
1310 | if (!count) | |
1311 | goto nomem_bulk; | |
1312 | ||
1f5f12ec PZ |
1313 | if (node->node_count == 0) { |
1314 | node->slot[0]->node_count = 0; | |
1315 | node->slot[0]->request_count = 0; | |
1316 | } | |
1317 | ||
54a611b6 | 1318 | node->node_count += count; |
541e06b7 | 1319 | allocated += count; |
c61b3a2b | 1320 | node = node->slot[0]; |
54a611b6 LH |
1321 | requested -= count; |
1322 | } | |
541e06b7 | 1323 | mas->alloc->total = allocated; |
54a611b6 LH |
1324 | return; |
1325 | ||
1326 | nomem_bulk: | |
1327 | /* Clean up potential freed allocations on bulk failure */ | |
1328 | memset(slots, 0, max_req * sizeof(unsigned long)); | |
1329 | nomem_one: | |
1330 | mas_set_alloc_req(mas, requested); | |
1331 | if (mas->alloc && !(((unsigned long)mas->alloc & 0x1))) | |
541e06b7 | 1332 | mas->alloc->total = allocated; |
54a611b6 | 1333 | mas_set_err(mas, -ENOMEM); |
54a611b6 LH |
1334 | } |
1335 | ||
1336 | /* | |
1337 | * mas_free() - Free an encoded maple node | |
1338 | * @mas: The maple state | |
1339 | * @used: The encoded maple node to free. | |
1340 | * | |
1341 | * Uses rcu free if necessary, pushes @used back on the maple state allocations | |
1342 | * otherwise. | |
1343 | */ | |
1344 | static inline void mas_free(struct ma_state *mas, struct maple_enode *used) | |
1345 | { | |
1346 | struct maple_node *tmp = mte_to_node(used); | |
1347 | ||
1348 | if (mt_in_rcu(mas->tree)) | |
1349 | ma_free_rcu(tmp); | |
1350 | else | |
1351 | mas_push_node(mas, tmp); | |
1352 | } | |
1353 | ||
1354 | /* | |
1355 | * mas_node_count() - Check if enough nodes are allocated and request more if | |
1356 | * there is not enough nodes. | |
1357 | * @mas: The maple state | |
1358 | * @count: The number of nodes needed | |
1359 | * @gfp: the gfp flags | |
1360 | */ | |
1361 | static void mas_node_count_gfp(struct ma_state *mas, int count, gfp_t gfp) | |
1362 | { | |
1363 | unsigned long allocated = mas_allocated(mas); | |
1364 | ||
1365 | if (allocated < count) { | |
1366 | mas_set_alloc_req(mas, count - allocated); | |
1367 | mas_alloc_nodes(mas, gfp); | |
1368 | } | |
1369 | } | |
1370 | ||
1371 | /* | |
1372 | * mas_node_count() - Check if enough nodes are allocated and request more if | |
1373 | * there is not enough nodes. | |
1374 | * @mas: The maple state | |
1375 | * @count: The number of nodes needed | |
1376 | * | |
1377 | * Note: Uses GFP_NOWAIT | __GFP_NOWARN for gfp flags. | |
1378 | */ | |
1379 | static void mas_node_count(struct ma_state *mas, int count) | |
1380 | { | |
1381 | return mas_node_count_gfp(mas, count, GFP_NOWAIT | __GFP_NOWARN); | |
1382 | } | |
1383 | ||
1384 | /* | |
1385 | * mas_start() - Sets up maple state for operations. | |
1386 | * @mas: The maple state. | |
1387 | * | |
46b34584 | 1388 | * If mas->node == MAS_START, then set the min, max and depth to |
54a611b6 LH |
1389 | * defaults. |
1390 | * | |
1391 | * Return: | |
1392 | * - If mas->node is an error or not MAS_START, return NULL. | |
1393 | * - If it's an empty tree: NULL & mas->node == MAS_NONE | |
1394 | * - If it's a single entry: The entry & mas->node == MAS_ROOT | |
1395 | * - If it's a tree: NULL & mas->node == safe root node. | |
1396 | */ | |
1397 | static inline struct maple_enode *mas_start(struct ma_state *mas) | |
1398 | { | |
1399 | if (likely(mas_is_start(mas))) { | |
1400 | struct maple_enode *root; | |
1401 | ||
54a611b6 LH |
1402 | mas->min = 0; |
1403 | mas->max = ULONG_MAX; | |
1404 | mas->depth = 0; | |
54a611b6 | 1405 | |
a7b92d59 | 1406 | retry: |
54a611b6 LH |
1407 | root = mas_root(mas); |
1408 | /* Tree with nodes */ | |
1409 | if (likely(xa_is_node(root))) { | |
9bbba563 | 1410 | mas->depth = 1; |
54a611b6 | 1411 | mas->node = mte_safe_root(root); |
46b34584 | 1412 | mas->offset = 0; |
a7b92d59 LH |
1413 | if (mte_dead_node(mas->node)) |
1414 | goto retry; | |
1415 | ||
54a611b6 LH |
1416 | return NULL; |
1417 | } | |
1418 | ||
1419 | /* empty tree */ | |
1420 | if (unlikely(!root)) { | |
46b34584 | 1421 | mas->node = MAS_NONE; |
54a611b6 LH |
1422 | mas->offset = MAPLE_NODE_SLOTS; |
1423 | return NULL; | |
1424 | } | |
1425 | ||
1426 | /* Single entry tree */ | |
1427 | mas->node = MAS_ROOT; | |
1428 | mas->offset = MAPLE_NODE_SLOTS; | |
1429 | ||
1430 | /* Single entry tree. */ | |
1431 | if (mas->index > 0) | |
1432 | return NULL; | |
1433 | ||
1434 | return root; | |
1435 | } | |
1436 | ||
1437 | return NULL; | |
1438 | } | |
1439 | ||
1440 | /* | |
1441 | * ma_data_end() - Find the end of the data in a node. | |
1442 | * @node: The maple node | |
1443 | * @type: The maple node type | |
1444 | * @pivots: The array of pivots in the node | |
1445 | * @max: The maximum value in the node | |
1446 | * | |
1447 | * Uses metadata to find the end of the data when possible. | |
1448 | * Return: The zero indexed last slot with data (may be null). | |
1449 | */ | |
1450 | static inline unsigned char ma_data_end(struct maple_node *node, | |
1451 | enum maple_type type, | |
1452 | unsigned long *pivots, | |
1453 | unsigned long max) | |
1454 | { | |
1455 | unsigned char offset; | |
1456 | ||
39d0bd86 LH |
1457 | if (!pivots) |
1458 | return 0; | |
1459 | ||
54a611b6 LH |
1460 | if (type == maple_arange_64) |
1461 | return ma_meta_end(node, type); | |
1462 | ||
1463 | offset = mt_pivots[type] - 1; | |
1464 | if (likely(!pivots[offset])) | |
1465 | return ma_meta_end(node, type); | |
1466 | ||
1467 | if (likely(pivots[offset] == max)) | |
1468 | return offset; | |
1469 | ||
1470 | return mt_pivots[type]; | |
1471 | } | |
1472 | ||
1473 | /* | |
1474 | * mas_data_end() - Find the end of the data (slot). | |
1475 | * @mas: the maple state | |
1476 | * | |
1477 | * This method is optimized to check the metadata of a node if the node type | |
1478 | * supports data end metadata. | |
1479 | * | |
1480 | * Return: The zero indexed last slot with data (may be null). | |
1481 | */ | |
1482 | static inline unsigned char mas_data_end(struct ma_state *mas) | |
1483 | { | |
1484 | enum maple_type type; | |
1485 | struct maple_node *node; | |
1486 | unsigned char offset; | |
1487 | unsigned long *pivots; | |
1488 | ||
1489 | type = mte_node_type(mas->node); | |
1490 | node = mas_mn(mas); | |
1491 | if (type == maple_arange_64) | |
1492 | return ma_meta_end(node, type); | |
1493 | ||
1494 | pivots = ma_pivots(node, type); | |
39d0bd86 LH |
1495 | if (unlikely(ma_dead_node(node))) |
1496 | return 0; | |
1497 | ||
54a611b6 LH |
1498 | offset = mt_pivots[type] - 1; |
1499 | if (likely(!pivots[offset])) | |
1500 | return ma_meta_end(node, type); | |
1501 | ||
1502 | if (likely(pivots[offset] == mas->max)) | |
1503 | return offset; | |
1504 | ||
1505 | return mt_pivots[type]; | |
1506 | } | |
1507 | ||
1508 | /* | |
1509 | * mas_leaf_max_gap() - Returns the largest gap in a leaf node | |
1510 | * @mas - the maple state | |
1511 | * | |
1512 | * Return: The maximum gap in the leaf. | |
1513 | */ | |
1514 | static unsigned long mas_leaf_max_gap(struct ma_state *mas) | |
1515 | { | |
1516 | enum maple_type mt; | |
1517 | unsigned long pstart, gap, max_gap; | |
1518 | struct maple_node *mn; | |
1519 | unsigned long *pivots; | |
1520 | void __rcu **slots; | |
1521 | unsigned char i; | |
1522 | unsigned char max_piv; | |
1523 | ||
1524 | mt = mte_node_type(mas->node); | |
1525 | mn = mas_mn(mas); | |
1526 | slots = ma_slots(mn, mt); | |
1527 | max_gap = 0; | |
1528 | if (unlikely(ma_is_dense(mt))) { | |
1529 | gap = 0; | |
1530 | for (i = 0; i < mt_slots[mt]; i++) { | |
1531 | if (slots[i]) { | |
1532 | if (gap > max_gap) | |
1533 | max_gap = gap; | |
1534 | gap = 0; | |
1535 | } else { | |
1536 | gap++; | |
1537 | } | |
1538 | } | |
1539 | if (gap > max_gap) | |
1540 | max_gap = gap; | |
1541 | return max_gap; | |
1542 | } | |
1543 | ||
1544 | /* | |
1545 | * Check the first implied pivot optimizes the loop below and slot 1 may | |
1546 | * be skipped if there is a gap in slot 0. | |
1547 | */ | |
1548 | pivots = ma_pivots(mn, mt); | |
1549 | if (likely(!slots[0])) { | |
1550 | max_gap = pivots[0] - mas->min + 1; | |
1551 | i = 2; | |
1552 | } else { | |
1553 | i = 1; | |
1554 | } | |
1555 | ||
1556 | /* reduce max_piv as the special case is checked before the loop */ | |
1557 | max_piv = ma_data_end(mn, mt, pivots, mas->max) - 1; | |
1558 | /* | |
1559 | * Check end implied pivot which can only be a gap on the right most | |
1560 | * node. | |
1561 | */ | |
1562 | if (unlikely(mas->max == ULONG_MAX) && !slots[max_piv + 1]) { | |
1563 | gap = ULONG_MAX - pivots[max_piv]; | |
1564 | if (gap > max_gap) | |
1565 | max_gap = gap; | |
1566 | } | |
1567 | ||
1568 | for (; i <= max_piv; i++) { | |
1569 | /* data == no gap. */ | |
1570 | if (likely(slots[i])) | |
1571 | continue; | |
1572 | ||
1573 | pstart = pivots[i - 1]; | |
1574 | gap = pivots[i] - pstart; | |
1575 | if (gap > max_gap) | |
1576 | max_gap = gap; | |
1577 | ||
1578 | /* There cannot be two gaps in a row. */ | |
1579 | i++; | |
1580 | } | |
1581 | return max_gap; | |
1582 | } | |
1583 | ||
1584 | /* | |
1585 | * ma_max_gap() - Get the maximum gap in a maple node (non-leaf) | |
1586 | * @node: The maple node | |
1587 | * @gaps: The pointer to the gaps | |
1588 | * @mt: The maple node type | |
1589 | * @*off: Pointer to store the offset location of the gap. | |
1590 | * | |
1591 | * Uses the metadata data end to scan backwards across set gaps. | |
1592 | * | |
1593 | * Return: The maximum gap value | |
1594 | */ | |
1595 | static inline unsigned long | |
1596 | ma_max_gap(struct maple_node *node, unsigned long *gaps, enum maple_type mt, | |
1597 | unsigned char *off) | |
1598 | { | |
1599 | unsigned char offset, i; | |
1600 | unsigned long max_gap = 0; | |
1601 | ||
1602 | i = offset = ma_meta_end(node, mt); | |
1603 | do { | |
1604 | if (gaps[i] > max_gap) { | |
1605 | max_gap = gaps[i]; | |
1606 | offset = i; | |
1607 | } | |
1608 | } while (i--); | |
1609 | ||
1610 | *off = offset; | |
1611 | return max_gap; | |
1612 | } | |
1613 | ||
1614 | /* | |
1615 | * mas_max_gap() - find the largest gap in a non-leaf node and set the slot. | |
1616 | * @mas: The maple state. | |
1617 | * | |
1618 | * If the metadata gap is set to MAPLE_ARANGE64_META_MAX, there is no gap. | |
1619 | * | |
1620 | * Return: The gap value. | |
1621 | */ | |
1622 | static inline unsigned long mas_max_gap(struct ma_state *mas) | |
1623 | { | |
1624 | unsigned long *gaps; | |
1625 | unsigned char offset; | |
1626 | enum maple_type mt; | |
1627 | struct maple_node *node; | |
1628 | ||
1629 | mt = mte_node_type(mas->node); | |
1630 | if (ma_is_leaf(mt)) | |
1631 | return mas_leaf_max_gap(mas); | |
1632 | ||
1633 | node = mas_mn(mas); | |
1634 | offset = ma_meta_gap(node, mt); | |
1635 | if (offset == MAPLE_ARANGE64_META_MAX) | |
1636 | return 0; | |
1637 | ||
1638 | gaps = ma_gaps(node, mt); | |
1639 | return gaps[offset]; | |
1640 | } | |
1641 | ||
1642 | /* | |
1643 | * mas_parent_gap() - Set the parent gap and any gaps above, as needed | |
1644 | * @mas: The maple state | |
1645 | * @offset: The gap offset in the parent to set | |
1646 | * @new: The new gap value. | |
1647 | * | |
1648 | * Set the parent gap then continue to set the gap upwards, using the metadata | |
1649 | * of the parent to see if it is necessary to check the node above. | |
1650 | */ | |
1651 | static inline void mas_parent_gap(struct ma_state *mas, unsigned char offset, | |
1652 | unsigned long new) | |
1653 | { | |
1654 | unsigned long meta_gap = 0; | |
1655 | struct maple_node *pnode; | |
1656 | struct maple_enode *penode; | |
1657 | unsigned long *pgaps; | |
1658 | unsigned char meta_offset; | |
1659 | enum maple_type pmt; | |
1660 | ||
1661 | pnode = mte_parent(mas->node); | |
1662 | pmt = mas_parent_enum(mas, mas->node); | |
1663 | penode = mt_mk_node(pnode, pmt); | |
1664 | pgaps = ma_gaps(pnode, pmt); | |
1665 | ||
1666 | ascend: | |
1667 | meta_offset = ma_meta_gap(pnode, pmt); | |
1668 | if (meta_offset == MAPLE_ARANGE64_META_MAX) | |
1669 | meta_gap = 0; | |
1670 | else | |
1671 | meta_gap = pgaps[meta_offset]; | |
1672 | ||
1673 | pgaps[offset] = new; | |
1674 | ||
1675 | if (meta_gap == new) | |
1676 | return; | |
1677 | ||
1678 | if (offset != meta_offset) { | |
1679 | if (meta_gap > new) | |
1680 | return; | |
1681 | ||
1682 | ma_set_meta_gap(pnode, pmt, offset); | |
1683 | } else if (new < meta_gap) { | |
1684 | meta_offset = 15; | |
1685 | new = ma_max_gap(pnode, pgaps, pmt, &meta_offset); | |
1686 | ma_set_meta_gap(pnode, pmt, meta_offset); | |
1687 | } | |
1688 | ||
1689 | if (ma_is_root(pnode)) | |
1690 | return; | |
1691 | ||
1692 | /* Go to the parent node. */ | |
1693 | pnode = mte_parent(penode); | |
1694 | pmt = mas_parent_enum(mas, penode); | |
1695 | pgaps = ma_gaps(pnode, pmt); | |
1696 | offset = mte_parent_slot(penode); | |
1697 | penode = mt_mk_node(pnode, pmt); | |
1698 | goto ascend; | |
1699 | } | |
1700 | ||
1701 | /* | |
1702 | * mas_update_gap() - Update a nodes gaps and propagate up if necessary. | |
1703 | * @mas - the maple state. | |
1704 | */ | |
1705 | static inline void mas_update_gap(struct ma_state *mas) | |
1706 | { | |
1707 | unsigned char pslot; | |
1708 | unsigned long p_gap; | |
1709 | unsigned long max_gap; | |
1710 | ||
1711 | if (!mt_is_alloc(mas->tree)) | |
1712 | return; | |
1713 | ||
1714 | if (mte_is_root(mas->node)) | |
1715 | return; | |
1716 | ||
1717 | max_gap = mas_max_gap(mas); | |
1718 | ||
1719 | pslot = mte_parent_slot(mas->node); | |
1720 | p_gap = ma_gaps(mte_parent(mas->node), | |
1721 | mas_parent_enum(mas, mas->node))[pslot]; | |
1722 | ||
1723 | if (p_gap != max_gap) | |
1724 | mas_parent_gap(mas, pslot, max_gap); | |
1725 | } | |
1726 | ||
1727 | /* | |
1728 | * mas_adopt_children() - Set the parent pointer of all nodes in @parent to | |
1729 | * @parent with the slot encoded. | |
1730 | * @mas - the maple state (for the tree) | |
1731 | * @parent - the maple encoded node containing the children. | |
1732 | */ | |
1733 | static inline void mas_adopt_children(struct ma_state *mas, | |
1734 | struct maple_enode *parent) | |
1735 | { | |
1736 | enum maple_type type = mte_node_type(parent); | |
1737 | struct maple_node *node = mas_mn(mas); | |
1738 | void __rcu **slots = ma_slots(node, type); | |
1739 | unsigned long *pivots = ma_pivots(node, type); | |
1740 | struct maple_enode *child; | |
1741 | unsigned char offset; | |
1742 | ||
1743 | offset = ma_data_end(node, type, pivots, mas->max); | |
1744 | do { | |
1745 | child = mas_slot_locked(mas, slots, offset); | |
1746 | mte_set_parent(child, parent, offset); | |
1747 | } while (offset--); | |
1748 | } | |
1749 | ||
1750 | /* | |
1751 | * mas_replace() - Replace a maple node in the tree with mas->node. Uses the | |
1752 | * parent encoding to locate the maple node in the tree. | |
1753 | * @mas - the ma_state to use for operations. | |
1754 | * @advanced - boolean to adopt the child nodes and free the old node (false) or | |
1755 | * leave the node (true) and handle the adoption and free elsewhere. | |
1756 | */ | |
1757 | static inline void mas_replace(struct ma_state *mas, bool advanced) | |
1758 | __must_hold(mas->tree->lock) | |
1759 | { | |
1760 | struct maple_node *mn = mas_mn(mas); | |
1761 | struct maple_enode *old_enode; | |
1762 | unsigned char offset = 0; | |
1763 | void __rcu **slots = NULL; | |
1764 | ||
1765 | if (ma_is_root(mn)) { | |
1766 | old_enode = mas_root_locked(mas); | |
1767 | } else { | |
1768 | offset = mte_parent_slot(mas->node); | |
1769 | slots = ma_slots(mte_parent(mas->node), | |
1770 | mas_parent_enum(mas, mas->node)); | |
1771 | old_enode = mas_slot_locked(mas, slots, offset); | |
1772 | } | |
1773 | ||
1774 | if (!advanced && !mte_is_leaf(mas->node)) | |
1775 | mas_adopt_children(mas, mas->node); | |
1776 | ||
1777 | if (mte_is_root(mas->node)) { | |
1778 | mn->parent = ma_parent_ptr( | |
1779 | ((unsigned long)mas->tree | MA_ROOT_PARENT)); | |
1780 | rcu_assign_pointer(mas->tree->ma_root, mte_mk_root(mas->node)); | |
1781 | mas_set_height(mas); | |
1782 | } else { | |
1783 | rcu_assign_pointer(slots[offset], mas->node); | |
1784 | } | |
1785 | ||
c13af03d LH |
1786 | if (!advanced) { |
1787 | mte_set_node_dead(old_enode); | |
54a611b6 | 1788 | mas_free(mas, old_enode); |
c13af03d | 1789 | } |
54a611b6 LH |
1790 | } |
1791 | ||
1792 | /* | |
1793 | * mas_new_child() - Find the new child of a node. | |
1794 | * @mas: the maple state | |
1795 | * @child: the maple state to store the child. | |
1796 | */ | |
1797 | static inline bool mas_new_child(struct ma_state *mas, struct ma_state *child) | |
1798 | __must_hold(mas->tree->lock) | |
1799 | { | |
1800 | enum maple_type mt; | |
1801 | unsigned char offset; | |
1802 | unsigned char end; | |
1803 | unsigned long *pivots; | |
1804 | struct maple_enode *entry; | |
1805 | struct maple_node *node; | |
1806 | void __rcu **slots; | |
1807 | ||
1808 | mt = mte_node_type(mas->node); | |
1809 | node = mas_mn(mas); | |
1810 | slots = ma_slots(node, mt); | |
1811 | pivots = ma_pivots(node, mt); | |
1812 | end = ma_data_end(node, mt, pivots, mas->max); | |
1813 | for (offset = mas->offset; offset <= end; offset++) { | |
1814 | entry = mas_slot_locked(mas, slots, offset); | |
1815 | if (mte_parent(entry) == node) { | |
1816 | *child = *mas; | |
1817 | mas->offset = offset + 1; | |
1818 | child->offset = offset; | |
1819 | mas_descend(child); | |
1820 | child->offset = 0; | |
1821 | return true; | |
1822 | } | |
1823 | } | |
1824 | return false; | |
1825 | } | |
1826 | ||
1827 | /* | |
1828 | * mab_shift_right() - Shift the data in mab right. Note, does not clean out the | |
1829 | * old data or set b_node->b_end. | |
1830 | * @b_node: the maple_big_node | |
1831 | * @shift: the shift count | |
1832 | */ | |
1833 | static inline void mab_shift_right(struct maple_big_node *b_node, | |
1834 | unsigned char shift) | |
1835 | { | |
1836 | unsigned long size = b_node->b_end * sizeof(unsigned long); | |
1837 | ||
1838 | memmove(b_node->pivot + shift, b_node->pivot, size); | |
1839 | memmove(b_node->slot + shift, b_node->slot, size); | |
1840 | if (b_node->type == maple_arange_64) | |
1841 | memmove(b_node->gap + shift, b_node->gap, size); | |
1842 | } | |
1843 | ||
1844 | /* | |
1845 | * mab_middle_node() - Check if a middle node is needed (unlikely) | |
1846 | * @b_node: the maple_big_node that contains the data. | |
1847 | * @size: the amount of data in the b_node | |
1848 | * @split: the potential split location | |
1849 | * @slot_count: the size that can be stored in a single node being considered. | |
1850 | * | |
1851 | * Return: true if a middle node is required. | |
1852 | */ | |
1853 | static inline bool mab_middle_node(struct maple_big_node *b_node, int split, | |
1854 | unsigned char slot_count) | |
1855 | { | |
1856 | unsigned char size = b_node->b_end; | |
1857 | ||
1858 | if (size >= 2 * slot_count) | |
1859 | return true; | |
1860 | ||
1861 | if (!b_node->slot[split] && (size >= 2 * slot_count - 1)) | |
1862 | return true; | |
1863 | ||
1864 | return false; | |
1865 | } | |
1866 | ||
1867 | /* | |
1868 | * mab_no_null_split() - ensure the split doesn't fall on a NULL | |
1869 | * @b_node: the maple_big_node with the data | |
1870 | * @split: the suggested split location | |
1871 | * @slot_count: the number of slots in the node being considered. | |
1872 | * | |
1873 | * Return: the split location. | |
1874 | */ | |
1875 | static inline int mab_no_null_split(struct maple_big_node *b_node, | |
1876 | unsigned char split, unsigned char slot_count) | |
1877 | { | |
1878 | if (!b_node->slot[split]) { | |
1879 | /* | |
1880 | * If the split is less than the max slot && the right side will | |
1881 | * still be sufficient, then increment the split on NULL. | |
1882 | */ | |
1883 | if ((split < slot_count - 1) && | |
1884 | (b_node->b_end - split) > (mt_min_slots[b_node->type])) | |
1885 | split++; | |
1886 | else | |
1887 | split--; | |
1888 | } | |
1889 | return split; | |
1890 | } | |
1891 | ||
1892 | /* | |
1893 | * mab_calc_split() - Calculate the split location and if there needs to be two | |
1894 | * splits. | |
1895 | * @bn: The maple_big_node with the data | |
1896 | * @mid_split: The second split, if required. 0 otherwise. | |
1897 | * | |
1898 | * Return: The first split location. The middle split is set in @mid_split. | |
1899 | */ | |
1900 | static inline int mab_calc_split(struct ma_state *mas, | |
1901 | struct maple_big_node *bn, unsigned char *mid_split, unsigned long min) | |
1902 | { | |
1903 | unsigned char b_end = bn->b_end; | |
1904 | int split = b_end / 2; /* Assume equal split. */ | |
1905 | unsigned char slot_min, slot_count = mt_slots[bn->type]; | |
1906 | ||
1907 | /* | |
1908 | * To support gap tracking, all NULL entries are kept together and a node cannot | |
1909 | * end on a NULL entry, with the exception of the left-most leaf. The | |
1910 | * limitation means that the split of a node must be checked for this condition | |
1911 | * and be able to put more data in one direction or the other. | |
1912 | */ | |
1913 | if (unlikely((mas->mas_flags & MA_STATE_BULK))) { | |
1914 | *mid_split = 0; | |
1915 | split = b_end - mt_min_slots[bn->type]; | |
1916 | ||
1917 | if (!ma_is_leaf(bn->type)) | |
1918 | return split; | |
1919 | ||
1920 | mas->mas_flags |= MA_STATE_REBALANCE; | |
1921 | if (!bn->slot[split]) | |
1922 | split--; | |
1923 | return split; | |
1924 | } | |
1925 | ||
1926 | /* | |
1927 | * Although extremely rare, it is possible to enter what is known as the 3-way | |
1928 | * split scenario. The 3-way split comes about by means of a store of a range | |
1929 | * that overwrites the end and beginning of two full nodes. The result is a set | |
1930 | * of entries that cannot be stored in 2 nodes. Sometimes, these two nodes can | |
1931 | * also be located in different parent nodes which are also full. This can | |
1932 | * carry upwards all the way to the root in the worst case. | |
1933 | */ | |
1934 | if (unlikely(mab_middle_node(bn, split, slot_count))) { | |
1935 | split = b_end / 3; | |
1936 | *mid_split = split * 2; | |
1937 | } else { | |
1938 | slot_min = mt_min_slots[bn->type]; | |
1939 | ||
1940 | *mid_split = 0; | |
1941 | /* | |
1942 | * Avoid having a range less than the slot count unless it | |
1943 | * causes one node to be deficient. | |
1944 | * NOTE: mt_min_slots is 1 based, b_end and split are zero. | |
1945 | */ | |
1946 | while (((bn->pivot[split] - min) < slot_count - 1) && | |
1947 | (split < slot_count - 1) && (b_end - split > slot_min)) | |
1948 | split++; | |
1949 | } | |
1950 | ||
1951 | /* Avoid ending a node on a NULL entry */ | |
1952 | split = mab_no_null_split(bn, split, slot_count); | |
54a611b6 | 1953 | |
e11cb683 VY |
1954 | if (unlikely(*mid_split)) |
1955 | *mid_split = mab_no_null_split(bn, *mid_split, slot_count); | |
54a611b6 LH |
1956 | |
1957 | return split; | |
1958 | } | |
1959 | ||
1960 | /* | |
1961 | * mas_mab_cp() - Copy data from a maple state inclusively to a maple_big_node | |
1962 | * and set @b_node->b_end to the next free slot. | |
1963 | * @mas: The maple state | |
1964 | * @mas_start: The starting slot to copy | |
1965 | * @mas_end: The end slot to copy (inclusively) | |
1966 | * @b_node: The maple_big_node to place the data | |
1967 | * @mab_start: The starting location in maple_big_node to store the data. | |
1968 | */ | |
1969 | static inline void mas_mab_cp(struct ma_state *mas, unsigned char mas_start, | |
1970 | unsigned char mas_end, struct maple_big_node *b_node, | |
1971 | unsigned char mab_start) | |
1972 | { | |
1973 | enum maple_type mt; | |
1974 | struct maple_node *node; | |
1975 | void __rcu **slots; | |
1976 | unsigned long *pivots, *gaps; | |
1977 | int i = mas_start, j = mab_start; | |
1978 | unsigned char piv_end; | |
1979 | ||
1980 | node = mas_mn(mas); | |
1981 | mt = mte_node_type(mas->node); | |
1982 | pivots = ma_pivots(node, mt); | |
1983 | if (!i) { | |
1984 | b_node->pivot[j] = pivots[i++]; | |
1985 | if (unlikely(i > mas_end)) | |
1986 | goto complete; | |
1987 | j++; | |
1988 | } | |
1989 | ||
1990 | piv_end = min(mas_end, mt_pivots[mt]); | |
1991 | for (; i < piv_end; i++, j++) { | |
1992 | b_node->pivot[j] = pivots[i]; | |
1993 | if (unlikely(!b_node->pivot[j])) | |
1994 | break; | |
1995 | ||
1996 | if (unlikely(mas->max == b_node->pivot[j])) | |
1997 | goto complete; | |
1998 | } | |
1999 | ||
2000 | if (likely(i <= mas_end)) | |
2001 | b_node->pivot[j] = mas_safe_pivot(mas, pivots, i, mt); | |
2002 | ||
2003 | complete: | |
2004 | b_node->b_end = ++j; | |
2005 | j -= mab_start; | |
2006 | slots = ma_slots(node, mt); | |
2007 | memcpy(b_node->slot + mab_start, slots + mas_start, sizeof(void *) * j); | |
2008 | if (!ma_is_leaf(mt) && mt_is_alloc(mas->tree)) { | |
2009 | gaps = ma_gaps(node, mt); | |
2010 | memcpy(b_node->gap + mab_start, gaps + mas_start, | |
2011 | sizeof(unsigned long) * j); | |
2012 | } | |
2013 | } | |
2014 | ||
2015 | /* | |
2016 | * mas_leaf_set_meta() - Set the metadata of a leaf if possible. | |
2017 | * @mas: The maple state | |
2018 | * @node: The maple node | |
2019 | * @pivots: pointer to the maple node pivots | |
2020 | * @mt: The maple type | |
2021 | * @end: The assumed end | |
2022 | * | |
2023 | * Note, end may be incremented within this function but not modified at the | |
2024 | * source. This is fine since the metadata is the last thing to be stored in a | |
2025 | * node during a write. | |
2026 | */ | |
2027 | static inline void mas_leaf_set_meta(struct ma_state *mas, | |
2028 | struct maple_node *node, unsigned long *pivots, | |
2029 | enum maple_type mt, unsigned char end) | |
2030 | { | |
2031 | /* There is no room for metadata already */ | |
2032 | if (mt_pivots[mt] <= end) | |
2033 | return; | |
2034 | ||
2035 | if (pivots[end] && pivots[end] < mas->max) | |
2036 | end++; | |
2037 | ||
2038 | if (end < mt_slots[mt] - 1) | |
2039 | ma_set_meta(node, mt, 0, end); | |
2040 | } | |
2041 | ||
2042 | /* | |
2043 | * mab_mas_cp() - Copy data from maple_big_node to a maple encoded node. | |
2044 | * @b_node: the maple_big_node that has the data | |
2045 | * @mab_start: the start location in @b_node. | |
2046 | * @mab_end: The end location in @b_node (inclusively) | |
2047 | * @mas: The maple state with the maple encoded node. | |
2048 | */ | |
2049 | static inline void mab_mas_cp(struct maple_big_node *b_node, | |
2050 | unsigned char mab_start, unsigned char mab_end, | |
2051 | struct ma_state *mas, bool new_max) | |
2052 | { | |
2053 | int i, j = 0; | |
2054 | enum maple_type mt = mte_node_type(mas->node); | |
2055 | struct maple_node *node = mte_to_node(mas->node); | |
2056 | void __rcu **slots = ma_slots(node, mt); | |
2057 | unsigned long *pivots = ma_pivots(node, mt); | |
2058 | unsigned long *gaps = NULL; | |
2059 | unsigned char end; | |
2060 | ||
2061 | if (mab_end - mab_start > mt_pivots[mt]) | |
2062 | mab_end--; | |
2063 | ||
2064 | if (!pivots[mt_pivots[mt] - 1]) | |
2065 | slots[mt_pivots[mt]] = NULL; | |
2066 | ||
2067 | i = mab_start; | |
2068 | do { | |
2069 | pivots[j++] = b_node->pivot[i++]; | |
2070 | } while (i <= mab_end && likely(b_node->pivot[i])); | |
2071 | ||
2072 | memcpy(slots, b_node->slot + mab_start, | |
2073 | sizeof(void *) * (i - mab_start)); | |
2074 | ||
2075 | if (new_max) | |
2076 | mas->max = b_node->pivot[i - 1]; | |
2077 | ||
2078 | end = j - 1; | |
2079 | if (likely(!ma_is_leaf(mt) && mt_is_alloc(mas->tree))) { | |
2080 | unsigned long max_gap = 0; | |
2081 | unsigned char offset = 15; | |
2082 | ||
2083 | gaps = ma_gaps(node, mt); | |
2084 | do { | |
2085 | gaps[--j] = b_node->gap[--i]; | |
2086 | if (gaps[j] > max_gap) { | |
2087 | offset = j; | |
2088 | max_gap = gaps[j]; | |
2089 | } | |
2090 | } while (j); | |
2091 | ||
2092 | ma_set_meta(node, mt, offset, end); | |
2093 | } else { | |
2094 | mas_leaf_set_meta(mas, node, pivots, mt, end); | |
2095 | } | |
2096 | } | |
2097 | ||
2098 | /* | |
2099 | * mas_descend_adopt() - Descend through a sub-tree and adopt children. | |
2100 | * @mas: the maple state with the maple encoded node of the sub-tree. | |
2101 | * | |
2102 | * Descend through a sub-tree and adopt children who do not have the correct | |
2103 | * parents set. Follow the parents which have the correct parents as they are | |
2104 | * the new entries which need to be followed to find other incorrectly set | |
2105 | * parents. | |
2106 | */ | |
2107 | static inline void mas_descend_adopt(struct ma_state *mas) | |
2108 | { | |
2109 | struct ma_state list[3], next[3]; | |
2110 | int i, n; | |
2111 | ||
2112 | /* | |
2113 | * At each level there may be up to 3 correct parent pointers which indicates | |
2114 | * the new nodes which need to be walked to find any new nodes at a lower level. | |
2115 | */ | |
2116 | ||
2117 | for (i = 0; i < 3; i++) { | |
2118 | list[i] = *mas; | |
2119 | list[i].offset = 0; | |
2120 | next[i].offset = 0; | |
2121 | } | |
2122 | next[0] = *mas; | |
2123 | ||
2124 | while (!mte_is_leaf(list[0].node)) { | |
2125 | n = 0; | |
2126 | for (i = 0; i < 3; i++) { | |
2127 | if (mas_is_none(&list[i])) | |
2128 | continue; | |
2129 | ||
2130 | if (i && list[i-1].node == list[i].node) | |
2131 | continue; | |
2132 | ||
2133 | while ((n < 3) && (mas_new_child(&list[i], &next[n]))) | |
2134 | n++; | |
2135 | ||
2136 | mas_adopt_children(&list[i], list[i].node); | |
2137 | } | |
2138 | ||
2139 | while (n < 3) | |
2140 | next[n++].node = MAS_NONE; | |
2141 | ||
2142 | /* descend by setting the list to the children */ | |
2143 | for (i = 0; i < 3; i++) | |
2144 | list[i] = next[i]; | |
2145 | } | |
2146 | } | |
2147 | ||
2148 | /* | |
2149 | * mas_bulk_rebalance() - Rebalance the end of a tree after a bulk insert. | |
2150 | * @mas: The maple state | |
2151 | * @end: The maple node end | |
2152 | * @mt: The maple node type | |
2153 | */ | |
2154 | static inline void mas_bulk_rebalance(struct ma_state *mas, unsigned char end, | |
2155 | enum maple_type mt) | |
2156 | { | |
2157 | if (!(mas->mas_flags & MA_STATE_BULK)) | |
2158 | return; | |
2159 | ||
2160 | if (mte_is_root(mas->node)) | |
2161 | return; | |
2162 | ||
2163 | if (end > mt_min_slots[mt]) { | |
2164 | mas->mas_flags &= ~MA_STATE_REBALANCE; | |
2165 | return; | |
2166 | } | |
2167 | } | |
2168 | ||
2169 | /* | |
2170 | * mas_store_b_node() - Store an @entry into the b_node while also copying the | |
2171 | * data from a maple encoded node. | |
2172 | * @wr_mas: the maple write state | |
2173 | * @b_node: the maple_big_node to fill with data | |
2174 | * @offset_end: the offset to end copying | |
2175 | * | |
2176 | * Return: The actual end of the data stored in @b_node | |
2177 | */ | |
44081c77 | 2178 | static noinline_for_kasan void mas_store_b_node(struct ma_wr_state *wr_mas, |
54a611b6 LH |
2179 | struct maple_big_node *b_node, unsigned char offset_end) |
2180 | { | |
2181 | unsigned char slot; | |
2182 | unsigned char b_end; | |
2183 | /* Possible underflow of piv will wrap back to 0 before use. */ | |
2184 | unsigned long piv; | |
2185 | struct ma_state *mas = wr_mas->mas; | |
2186 | ||
2187 | b_node->type = wr_mas->type; | |
2188 | b_end = 0; | |
2189 | slot = mas->offset; | |
2190 | if (slot) { | |
2191 | /* Copy start data up to insert. */ | |
2192 | mas_mab_cp(mas, 0, slot - 1, b_node, 0); | |
2193 | b_end = b_node->b_end; | |
2194 | piv = b_node->pivot[b_end - 1]; | |
2195 | } else | |
2196 | piv = mas->min - 1; | |
2197 | ||
2198 | if (piv + 1 < mas->index) { | |
2199 | /* Handle range starting after old range */ | |
2200 | b_node->slot[b_end] = wr_mas->content; | |
2201 | if (!wr_mas->content) | |
2202 | b_node->gap[b_end] = mas->index - 1 - piv; | |
2203 | b_node->pivot[b_end++] = mas->index - 1; | |
2204 | } | |
2205 | ||
2206 | /* Store the new entry. */ | |
2207 | mas->offset = b_end; | |
2208 | b_node->slot[b_end] = wr_mas->entry; | |
2209 | b_node->pivot[b_end] = mas->last; | |
2210 | ||
2211 | /* Appended. */ | |
2212 | if (mas->last >= mas->max) | |
2213 | goto b_end; | |
2214 | ||
2215 | /* Handle new range ending before old range ends */ | |
2216 | piv = mas_logical_pivot(mas, wr_mas->pivots, offset_end, wr_mas->type); | |
2217 | if (piv > mas->last) { | |
2218 | if (piv == ULONG_MAX) | |
2219 | mas_bulk_rebalance(mas, b_node->b_end, wr_mas->type); | |
2220 | ||
2221 | if (offset_end != slot) | |
2222 | wr_mas->content = mas_slot_locked(mas, wr_mas->slots, | |
2223 | offset_end); | |
2224 | ||
2225 | b_node->slot[++b_end] = wr_mas->content; | |
2226 | if (!wr_mas->content) | |
2227 | b_node->gap[b_end] = piv - mas->last + 1; | |
2228 | b_node->pivot[b_end] = piv; | |
2229 | } | |
2230 | ||
2231 | slot = offset_end + 1; | |
2232 | if (slot > wr_mas->node_end) | |
2233 | goto b_end; | |
2234 | ||
2235 | /* Copy end data to the end of the node. */ | |
2236 | mas_mab_cp(mas, slot, wr_mas->node_end + 1, b_node, ++b_end); | |
2237 | b_node->b_end--; | |
2238 | return; | |
2239 | ||
2240 | b_end: | |
2241 | b_node->b_end = b_end; | |
2242 | } | |
2243 | ||
2244 | /* | |
2245 | * mas_prev_sibling() - Find the previous node with the same parent. | |
2246 | * @mas: the maple state | |
2247 | * | |
2248 | * Return: True if there is a previous sibling, false otherwise. | |
2249 | */ | |
2250 | static inline bool mas_prev_sibling(struct ma_state *mas) | |
2251 | { | |
2252 | unsigned int p_slot = mte_parent_slot(mas->node); | |
2253 | ||
2254 | if (mte_is_root(mas->node)) | |
2255 | return false; | |
2256 | ||
2257 | if (!p_slot) | |
2258 | return false; | |
2259 | ||
2260 | mas_ascend(mas); | |
2261 | mas->offset = p_slot - 1; | |
2262 | mas_descend(mas); | |
2263 | return true; | |
2264 | } | |
2265 | ||
2266 | /* | |
2267 | * mas_next_sibling() - Find the next node with the same parent. | |
2268 | * @mas: the maple state | |
2269 | * | |
2270 | * Return: true if there is a next sibling, false otherwise. | |
2271 | */ | |
2272 | static inline bool mas_next_sibling(struct ma_state *mas) | |
2273 | { | |
2274 | MA_STATE(parent, mas->tree, mas->index, mas->last); | |
2275 | ||
2276 | if (mte_is_root(mas->node)) | |
2277 | return false; | |
2278 | ||
2279 | parent = *mas; | |
2280 | mas_ascend(&parent); | |
2281 | parent.offset = mte_parent_slot(mas->node) + 1; | |
2282 | if (parent.offset > mas_data_end(&parent)) | |
2283 | return false; | |
2284 | ||
2285 | *mas = parent; | |
2286 | mas_descend(mas); | |
2287 | return true; | |
2288 | } | |
2289 | ||
2290 | /* | |
2291 | * mte_node_or_node() - Return the encoded node or MAS_NONE. | |
2292 | * @enode: The encoded maple node. | |
2293 | * | |
2294 | * Shorthand to avoid setting %NULLs in the tree or maple_subtree_state. | |
2295 | * | |
2296 | * Return: @enode or MAS_NONE | |
2297 | */ | |
2298 | static inline struct maple_enode *mte_node_or_none(struct maple_enode *enode) | |
2299 | { | |
2300 | if (enode) | |
2301 | return enode; | |
2302 | ||
2303 | return ma_enode_ptr(MAS_NONE); | |
2304 | } | |
2305 | ||
2306 | /* | |
2307 | * mas_wr_node_walk() - Find the correct offset for the index in the @mas. | |
2308 | * @wr_mas: The maple write state | |
2309 | * | |
2310 | * Uses mas_slot_locked() and does not need to worry about dead nodes. | |
2311 | */ | |
2312 | static inline void mas_wr_node_walk(struct ma_wr_state *wr_mas) | |
2313 | { | |
2314 | struct ma_state *mas = wr_mas->mas; | |
97f7e094 | 2315 | unsigned char count, offset; |
54a611b6 LH |
2316 | |
2317 | if (unlikely(ma_is_dense(wr_mas->type))) { | |
2318 | wr_mas->r_max = wr_mas->r_min = mas->index; | |
2319 | mas->offset = mas->index = mas->min; | |
2320 | return; | |
2321 | } | |
2322 | ||
2323 | wr_mas->node = mas_mn(wr_mas->mas); | |
2324 | wr_mas->pivots = ma_pivots(wr_mas->node, wr_mas->type); | |
2325 | count = wr_mas->node_end = ma_data_end(wr_mas->node, wr_mas->type, | |
2326 | wr_mas->pivots, mas->max); | |
2327 | offset = mas->offset; | |
54a611b6 | 2328 | |
97f7e094 PZ |
2329 | while (offset < count && mas->index > wr_mas->pivots[offset]) |
2330 | offset++; | |
54a611b6 | 2331 | |
97f7e094 PZ |
2332 | wr_mas->r_max = offset < count ? wr_mas->pivots[offset] : mas->max; |
2333 | wr_mas->r_min = mas_safe_min(mas, wr_mas->pivots, offset); | |
54a611b6 LH |
2334 | wr_mas->offset_end = mas->offset = offset; |
2335 | } | |
2336 | ||
2337 | /* | |
2338 | * mas_topiary_range() - Add a range of slots to the topiary. | |
2339 | * @mas: The maple state | |
2340 | * @destroy: The topiary to add the slots (usually destroy) | |
2341 | * @start: The starting slot inclusively | |
2342 | * @end: The end slot inclusively | |
2343 | */ | |
2344 | static inline void mas_topiary_range(struct ma_state *mas, | |
2345 | struct ma_topiary *destroy, unsigned char start, unsigned char end) | |
2346 | { | |
2347 | void __rcu **slots; | |
2348 | unsigned char offset; | |
2349 | ||
2350 | MT_BUG_ON(mas->tree, mte_is_leaf(mas->node)); | |
2351 | slots = ma_slots(mas_mn(mas), mte_node_type(mas->node)); | |
2352 | for (offset = start; offset <= end; offset++) { | |
2353 | struct maple_enode *enode = mas_slot_locked(mas, slots, offset); | |
2354 | ||
2355 | if (mte_dead_node(enode)) | |
2356 | continue; | |
2357 | ||
2358 | mat_add(destroy, enode); | |
2359 | } | |
2360 | } | |
2361 | ||
2362 | /* | |
2363 | * mast_topiary() - Add the portions of the tree to the removal list; either to | |
2364 | * be freed or discarded (destroy walk). | |
2365 | * @mast: The maple_subtree_state. | |
2366 | */ | |
2367 | static inline void mast_topiary(struct maple_subtree_state *mast) | |
2368 | { | |
2369 | MA_WR_STATE(wr_mas, mast->orig_l, NULL); | |
2370 | unsigned char r_start, r_end; | |
2371 | unsigned char l_start, l_end; | |
2372 | void __rcu **l_slots, **r_slots; | |
2373 | ||
2374 | wr_mas.type = mte_node_type(mast->orig_l->node); | |
2375 | mast->orig_l->index = mast->orig_l->last; | |
2376 | mas_wr_node_walk(&wr_mas); | |
2377 | l_start = mast->orig_l->offset + 1; | |
2378 | l_end = mas_data_end(mast->orig_l); | |
2379 | r_start = 0; | |
2380 | r_end = mast->orig_r->offset; | |
2381 | ||
2382 | if (r_end) | |
2383 | r_end--; | |
2384 | ||
2385 | l_slots = ma_slots(mas_mn(mast->orig_l), | |
2386 | mte_node_type(mast->orig_l->node)); | |
2387 | ||
2388 | r_slots = ma_slots(mas_mn(mast->orig_r), | |
2389 | mte_node_type(mast->orig_r->node)); | |
2390 | ||
2391 | if ((l_start < l_end) && | |
2392 | mte_dead_node(mas_slot_locked(mast->orig_l, l_slots, l_start))) { | |
2393 | l_start++; | |
2394 | } | |
2395 | ||
2396 | if (mte_dead_node(mas_slot_locked(mast->orig_r, r_slots, r_end))) { | |
2397 | if (r_end) | |
2398 | r_end--; | |
2399 | } | |
2400 | ||
2401 | if ((l_start > r_end) && (mast->orig_l->node == mast->orig_r->node)) | |
2402 | return; | |
2403 | ||
2404 | /* At the node where left and right sides meet, add the parts between */ | |
2405 | if (mast->orig_l->node == mast->orig_r->node) { | |
2406 | return mas_topiary_range(mast->orig_l, mast->destroy, | |
2407 | l_start, r_end); | |
2408 | } | |
2409 | ||
2410 | /* mast->orig_r is different and consumed. */ | |
2411 | if (mte_is_leaf(mast->orig_r->node)) | |
2412 | return; | |
2413 | ||
2414 | if (mte_dead_node(mas_slot_locked(mast->orig_l, l_slots, l_end))) | |
2415 | l_end--; | |
2416 | ||
2417 | ||
2418 | if (l_start <= l_end) | |
2419 | mas_topiary_range(mast->orig_l, mast->destroy, l_start, l_end); | |
2420 | ||
2421 | if (mte_dead_node(mas_slot_locked(mast->orig_r, r_slots, r_start))) | |
2422 | r_start++; | |
2423 | ||
2424 | if (r_start <= r_end) | |
2425 | mas_topiary_range(mast->orig_r, mast->destroy, 0, r_end); | |
2426 | } | |
2427 | ||
2428 | /* | |
2429 | * mast_rebalance_next() - Rebalance against the next node | |
2430 | * @mast: The maple subtree state | |
2431 | * @old_r: The encoded maple node to the right (next node). | |
2432 | */ | |
2433 | static inline void mast_rebalance_next(struct maple_subtree_state *mast) | |
2434 | { | |
2435 | unsigned char b_end = mast->bn->b_end; | |
2436 | ||
2437 | mas_mab_cp(mast->orig_r, 0, mt_slot_count(mast->orig_r->node), | |
2438 | mast->bn, b_end); | |
2439 | mast->orig_r->last = mast->orig_r->max; | |
2440 | } | |
2441 | ||
2442 | /* | |
2443 | * mast_rebalance_prev() - Rebalance against the previous node | |
2444 | * @mast: The maple subtree state | |
2445 | * @old_l: The encoded maple node to the left (previous node) | |
2446 | */ | |
2447 | static inline void mast_rebalance_prev(struct maple_subtree_state *mast) | |
2448 | { | |
2449 | unsigned char end = mas_data_end(mast->orig_l) + 1; | |
2450 | unsigned char b_end = mast->bn->b_end; | |
2451 | ||
2452 | mab_shift_right(mast->bn, end); | |
2453 | mas_mab_cp(mast->orig_l, 0, end - 1, mast->bn, 0); | |
2454 | mast->l->min = mast->orig_l->min; | |
2455 | mast->orig_l->index = mast->orig_l->min; | |
2456 | mast->bn->b_end = end + b_end; | |
2457 | mast->l->offset += end; | |
2458 | } | |
2459 | ||
2460 | /* | |
2461 | * mast_spanning_rebalance() - Rebalance nodes with nearest neighbour favouring | |
2462 | * the node to the right. Checking the nodes to the right then the left at each | |
2463 | * level upwards until root is reached. Free and destroy as needed. | |
2464 | * Data is copied into the @mast->bn. | |
2465 | * @mast: The maple_subtree_state. | |
2466 | */ | |
2467 | static inline | |
2468 | bool mast_spanning_rebalance(struct maple_subtree_state *mast) | |
2469 | { | |
2470 | struct ma_state r_tmp = *mast->orig_r; | |
2471 | struct ma_state l_tmp = *mast->orig_l; | |
2472 | struct maple_enode *ancestor = NULL; | |
2473 | unsigned char start, end; | |
2474 | unsigned char depth = 0; | |
2475 | ||
2476 | r_tmp = *mast->orig_r; | |
2477 | l_tmp = *mast->orig_l; | |
2478 | do { | |
2479 | mas_ascend(mast->orig_r); | |
2480 | mas_ascend(mast->orig_l); | |
2481 | depth++; | |
2482 | if (!ancestor && | |
2483 | (mast->orig_r->node == mast->orig_l->node)) { | |
2484 | ancestor = mast->orig_r->node; | |
2485 | end = mast->orig_r->offset - 1; | |
2486 | start = mast->orig_l->offset + 1; | |
2487 | } | |
2488 | ||
2489 | if (mast->orig_r->offset < mas_data_end(mast->orig_r)) { | |
2490 | if (!ancestor) { | |
2491 | ancestor = mast->orig_r->node; | |
2492 | start = 0; | |
2493 | } | |
2494 | ||
2495 | mast->orig_r->offset++; | |
2496 | do { | |
2497 | mas_descend(mast->orig_r); | |
2498 | mast->orig_r->offset = 0; | |
2499 | depth--; | |
2500 | } while (depth); | |
2501 | ||
2502 | mast_rebalance_next(mast); | |
2503 | do { | |
2504 | unsigned char l_off = 0; | |
2505 | struct maple_enode *child = r_tmp.node; | |
2506 | ||
2507 | mas_ascend(&r_tmp); | |
2508 | if (ancestor == r_tmp.node) | |
2509 | l_off = start; | |
2510 | ||
2511 | if (r_tmp.offset) | |
2512 | r_tmp.offset--; | |
2513 | ||
2514 | if (l_off < r_tmp.offset) | |
2515 | mas_topiary_range(&r_tmp, mast->destroy, | |
2516 | l_off, r_tmp.offset); | |
2517 | ||
2518 | if (l_tmp.node != child) | |
2519 | mat_add(mast->free, child); | |
2520 | ||
2521 | } while (r_tmp.node != ancestor); | |
2522 | ||
2523 | *mast->orig_l = l_tmp; | |
2524 | return true; | |
2525 | ||
2526 | } else if (mast->orig_l->offset != 0) { | |
2527 | if (!ancestor) { | |
2528 | ancestor = mast->orig_l->node; | |
2529 | end = mas_data_end(mast->orig_l); | |
2530 | } | |
2531 | ||
2532 | mast->orig_l->offset--; | |
2533 | do { | |
2534 | mas_descend(mast->orig_l); | |
2535 | mast->orig_l->offset = | |
2536 | mas_data_end(mast->orig_l); | |
2537 | depth--; | |
2538 | } while (depth); | |
2539 | ||
2540 | mast_rebalance_prev(mast); | |
2541 | do { | |
2542 | unsigned char r_off; | |
2543 | struct maple_enode *child = l_tmp.node; | |
2544 | ||
2545 | mas_ascend(&l_tmp); | |
2546 | if (ancestor == l_tmp.node) | |
2547 | r_off = end; | |
2548 | else | |
2549 | r_off = mas_data_end(&l_tmp); | |
2550 | ||
2551 | if (l_tmp.offset < r_off) | |
2552 | l_tmp.offset++; | |
2553 | ||
2554 | if (l_tmp.offset < r_off) | |
2555 | mas_topiary_range(&l_tmp, mast->destroy, | |
2556 | l_tmp.offset, r_off); | |
2557 | ||
2558 | if (r_tmp.node != child) | |
2559 | mat_add(mast->free, child); | |
2560 | ||
2561 | } while (l_tmp.node != ancestor); | |
2562 | ||
2563 | *mast->orig_r = r_tmp; | |
2564 | return true; | |
2565 | } | |
2566 | } while (!mte_is_root(mast->orig_r->node)); | |
2567 | ||
2568 | *mast->orig_r = r_tmp; | |
2569 | *mast->orig_l = l_tmp; | |
2570 | return false; | |
2571 | } | |
2572 | ||
2573 | /* | |
2574 | * mast_ascend_free() - Add current original maple state nodes to the free list | |
2575 | * and ascend. | |
2576 | * @mast: the maple subtree state. | |
2577 | * | |
2578 | * Ascend the original left and right sides and add the previous nodes to the | |
2579 | * free list. Set the slots to point to the correct location in the new nodes. | |
2580 | */ | |
2581 | static inline void | |
2582 | mast_ascend_free(struct maple_subtree_state *mast) | |
2583 | { | |
2584 | MA_WR_STATE(wr_mas, mast->orig_r, NULL); | |
2585 | struct maple_enode *left = mast->orig_l->node; | |
2586 | struct maple_enode *right = mast->orig_r->node; | |
2587 | ||
2588 | mas_ascend(mast->orig_l); | |
2589 | mas_ascend(mast->orig_r); | |
2590 | mat_add(mast->free, left); | |
2591 | ||
2592 | if (left != right) | |
2593 | mat_add(mast->free, right); | |
2594 | ||
2595 | mast->orig_r->offset = 0; | |
2596 | mast->orig_r->index = mast->r->max; | |
2597 | /* last should be larger than or equal to index */ | |
2598 | if (mast->orig_r->last < mast->orig_r->index) | |
2599 | mast->orig_r->last = mast->orig_r->index; | |
2600 | /* | |
2601 | * The node may not contain the value so set slot to ensure all | |
2602 | * of the nodes contents are freed or destroyed. | |
2603 | */ | |
2604 | wr_mas.type = mte_node_type(mast->orig_r->node); | |
2605 | mas_wr_node_walk(&wr_mas); | |
2606 | /* Set up the left side of things */ | |
2607 | mast->orig_l->offset = 0; | |
2608 | mast->orig_l->index = mast->l->min; | |
2609 | wr_mas.mas = mast->orig_l; | |
2610 | wr_mas.type = mte_node_type(mast->orig_l->node); | |
2611 | mas_wr_node_walk(&wr_mas); | |
2612 | ||
2613 | mast->bn->type = wr_mas.type; | |
2614 | } | |
2615 | ||
2616 | /* | |
2617 | * mas_new_ma_node() - Create and return a new maple node. Helper function. | |
2618 | * @mas: the maple state with the allocations. | |
2619 | * @b_node: the maple_big_node with the type encoding. | |
2620 | * | |
2621 | * Use the node type from the maple_big_node to allocate a new node from the | |
2622 | * ma_state. This function exists mainly for code readability. | |
2623 | * | |
2624 | * Return: A new maple encoded node | |
2625 | */ | |
2626 | static inline struct maple_enode | |
2627 | *mas_new_ma_node(struct ma_state *mas, struct maple_big_node *b_node) | |
2628 | { | |
2629 | return mt_mk_node(ma_mnode_ptr(mas_pop_node(mas)), b_node->type); | |
2630 | } | |
2631 | ||
2632 | /* | |
2633 | * mas_mab_to_node() - Set up right and middle nodes | |
2634 | * | |
2635 | * @mas: the maple state that contains the allocations. | |
2636 | * @b_node: the node which contains the data. | |
2637 | * @left: The pointer which will have the left node | |
2638 | * @right: The pointer which may have the right node | |
2639 | * @middle: the pointer which may have the middle node (rare) | |
2640 | * @mid_split: the split location for the middle node | |
2641 | * | |
2642 | * Return: the split of left. | |
2643 | */ | |
2644 | static inline unsigned char mas_mab_to_node(struct ma_state *mas, | |
2645 | struct maple_big_node *b_node, struct maple_enode **left, | |
2646 | struct maple_enode **right, struct maple_enode **middle, | |
2647 | unsigned char *mid_split, unsigned long min) | |
2648 | { | |
2649 | unsigned char split = 0; | |
2650 | unsigned char slot_count = mt_slots[b_node->type]; | |
2651 | ||
2652 | *left = mas_new_ma_node(mas, b_node); | |
2653 | *right = NULL; | |
2654 | *middle = NULL; | |
2655 | *mid_split = 0; | |
2656 | ||
2657 | if (b_node->b_end < slot_count) { | |
2658 | split = b_node->b_end; | |
2659 | } else { | |
2660 | split = mab_calc_split(mas, b_node, mid_split, min); | |
2661 | *right = mas_new_ma_node(mas, b_node); | |
2662 | } | |
2663 | ||
2664 | if (*mid_split) | |
2665 | *middle = mas_new_ma_node(mas, b_node); | |
2666 | ||
2667 | return split; | |
2668 | ||
2669 | } | |
2670 | ||
2671 | /* | |
2672 | * mab_set_b_end() - Add entry to b_node at b_node->b_end and increment the end | |
2673 | * pointer. | |
2674 | * @b_node - the big node to add the entry | |
2675 | * @mas - the maple state to get the pivot (mas->max) | |
2676 | * @entry - the entry to add, if NULL nothing happens. | |
2677 | */ | |
2678 | static inline void mab_set_b_end(struct maple_big_node *b_node, | |
2679 | struct ma_state *mas, | |
2680 | void *entry) | |
2681 | { | |
2682 | if (!entry) | |
2683 | return; | |
2684 | ||
2685 | b_node->slot[b_node->b_end] = entry; | |
2686 | if (mt_is_alloc(mas->tree)) | |
2687 | b_node->gap[b_node->b_end] = mas_max_gap(mas); | |
2688 | b_node->pivot[b_node->b_end++] = mas->max; | |
2689 | } | |
2690 | ||
2691 | /* | |
2692 | * mas_set_split_parent() - combine_then_separate helper function. Sets the parent | |
2693 | * of @mas->node to either @left or @right, depending on @slot and @split | |
2694 | * | |
2695 | * @mas - the maple state with the node that needs a parent | |
2696 | * @left - possible parent 1 | |
2697 | * @right - possible parent 2 | |
2698 | * @slot - the slot the mas->node was placed | |
2699 | * @split - the split location between @left and @right | |
2700 | */ | |
2701 | static inline void mas_set_split_parent(struct ma_state *mas, | |
2702 | struct maple_enode *left, | |
2703 | struct maple_enode *right, | |
2704 | unsigned char *slot, unsigned char split) | |
2705 | { | |
2706 | if (mas_is_none(mas)) | |
2707 | return; | |
2708 | ||
2709 | if ((*slot) <= split) | |
2710 | mte_set_parent(mas->node, left, *slot); | |
2711 | else if (right) | |
2712 | mte_set_parent(mas->node, right, (*slot) - split - 1); | |
2713 | ||
2714 | (*slot)++; | |
2715 | } | |
2716 | ||
2717 | /* | |
2718 | * mte_mid_split_check() - Check if the next node passes the mid-split | |
2719 | * @**l: Pointer to left encoded maple node. | |
2720 | * @**m: Pointer to middle encoded maple node. | |
2721 | * @**r: Pointer to right encoded maple node. | |
2722 | * @slot: The offset | |
2723 | * @*split: The split location. | |
2724 | * @mid_split: The middle split. | |
2725 | */ | |
2726 | static inline void mte_mid_split_check(struct maple_enode **l, | |
2727 | struct maple_enode **r, | |
2728 | struct maple_enode *right, | |
2729 | unsigned char slot, | |
2730 | unsigned char *split, | |
2731 | unsigned char mid_split) | |
2732 | { | |
2733 | if (*r == right) | |
2734 | return; | |
2735 | ||
2736 | if (slot < mid_split) | |
2737 | return; | |
2738 | ||
2739 | *l = *r; | |
2740 | *r = right; | |
2741 | *split = mid_split; | |
2742 | } | |
2743 | ||
2744 | /* | |
2745 | * mast_set_split_parents() - Helper function to set three nodes parents. Slot | |
2746 | * is taken from @mast->l. | |
2747 | * @mast - the maple subtree state | |
2748 | * @left - the left node | |
2749 | * @right - the right node | |
2750 | * @split - the split location. | |
2751 | */ | |
2752 | static inline void mast_set_split_parents(struct maple_subtree_state *mast, | |
2753 | struct maple_enode *left, | |
2754 | struct maple_enode *middle, | |
2755 | struct maple_enode *right, | |
2756 | unsigned char split, | |
2757 | unsigned char mid_split) | |
2758 | { | |
2759 | unsigned char slot; | |
2760 | struct maple_enode *l = left; | |
2761 | struct maple_enode *r = right; | |
2762 | ||
2763 | if (mas_is_none(mast->l)) | |
2764 | return; | |
2765 | ||
2766 | if (middle) | |
2767 | r = middle; | |
2768 | ||
2769 | slot = mast->l->offset; | |
2770 | ||
2771 | mte_mid_split_check(&l, &r, right, slot, &split, mid_split); | |
2772 | mas_set_split_parent(mast->l, l, r, &slot, split); | |
2773 | ||
2774 | mte_mid_split_check(&l, &r, right, slot, &split, mid_split); | |
2775 | mas_set_split_parent(mast->m, l, r, &slot, split); | |
2776 | ||
2777 | mte_mid_split_check(&l, &r, right, slot, &split, mid_split); | |
2778 | mas_set_split_parent(mast->r, l, r, &slot, split); | |
2779 | } | |
2780 | ||
2781 | /* | |
2782 | * mas_wmb_replace() - Write memory barrier and replace | |
2783 | * @mas: The maple state | |
2784 | * @free: the maple topiary list of nodes to free | |
2785 | * @destroy: The maple topiary list of nodes to destroy (walk and free) | |
2786 | * | |
2787 | * Updates gap as necessary. | |
2788 | */ | |
2789 | static inline void mas_wmb_replace(struct ma_state *mas, | |
2790 | struct ma_topiary *free, | |
2791 | struct ma_topiary *destroy) | |
2792 | { | |
2793 | /* All nodes must see old data as dead prior to replacing that data */ | |
2794 | smp_wmb(); /* Needed for RCU */ | |
2795 | ||
2796 | /* Insert the new data in the tree */ | |
2797 | mas_replace(mas, true); | |
2798 | ||
2799 | if (!mte_is_leaf(mas->node)) | |
2800 | mas_descend_adopt(mas); | |
2801 | ||
2802 | mas_mat_free(mas, free); | |
2803 | ||
2804 | if (destroy) | |
2805 | mas_mat_destroy(mas, destroy); | |
2806 | ||
2807 | if (mte_is_leaf(mas->node)) | |
2808 | return; | |
2809 | ||
2810 | mas_update_gap(mas); | |
2811 | } | |
2812 | ||
2813 | /* | |
2814 | * mast_new_root() - Set a new tree root during subtree creation | |
2815 | * @mast: The maple subtree state | |
2816 | * @mas: The maple state | |
2817 | */ | |
2818 | static inline void mast_new_root(struct maple_subtree_state *mast, | |
2819 | struct ma_state *mas) | |
2820 | { | |
2821 | mas_mn(mast->l)->parent = | |
2822 | ma_parent_ptr(((unsigned long)mas->tree | MA_ROOT_PARENT)); | |
2823 | if (!mte_dead_node(mast->orig_l->node) && | |
2824 | !mte_is_root(mast->orig_l->node)) { | |
2825 | do { | |
2826 | mast_ascend_free(mast); | |
2827 | mast_topiary(mast); | |
2828 | } while (!mte_is_root(mast->orig_l->node)); | |
2829 | } | |
2830 | if ((mast->orig_l->node != mas->node) && | |
2831 | (mast->l->depth > mas_mt_height(mas))) { | |
2832 | mat_add(mast->free, mas->node); | |
2833 | } | |
2834 | } | |
2835 | ||
2836 | /* | |
2837 | * mast_cp_to_nodes() - Copy data out to nodes. | |
2838 | * @mast: The maple subtree state | |
2839 | * @left: The left encoded maple node | |
2840 | * @middle: The middle encoded maple node | |
2841 | * @right: The right encoded maple node | |
2842 | * @split: The location to split between left and (middle ? middle : right) | |
2843 | * @mid_split: The location to split between middle and right. | |
2844 | */ | |
2845 | static inline void mast_cp_to_nodes(struct maple_subtree_state *mast, | |
2846 | struct maple_enode *left, struct maple_enode *middle, | |
2847 | struct maple_enode *right, unsigned char split, unsigned char mid_split) | |
2848 | { | |
2849 | bool new_lmax = true; | |
2850 | ||
2851 | mast->l->node = mte_node_or_none(left); | |
2852 | mast->m->node = mte_node_or_none(middle); | |
2853 | mast->r->node = mte_node_or_none(right); | |
2854 | ||
2855 | mast->l->min = mast->orig_l->min; | |
2856 | if (split == mast->bn->b_end) { | |
2857 | mast->l->max = mast->orig_r->max; | |
2858 | new_lmax = false; | |
2859 | } | |
2860 | ||
2861 | mab_mas_cp(mast->bn, 0, split, mast->l, new_lmax); | |
2862 | ||
2863 | if (middle) { | |
2864 | mab_mas_cp(mast->bn, 1 + split, mid_split, mast->m, true); | |
2865 | mast->m->min = mast->bn->pivot[split] + 1; | |
2866 | split = mid_split; | |
2867 | } | |
2868 | ||
2869 | mast->r->max = mast->orig_r->max; | |
2870 | if (right) { | |
2871 | mab_mas_cp(mast->bn, 1 + split, mast->bn->b_end, mast->r, false); | |
2872 | mast->r->min = mast->bn->pivot[split] + 1; | |
2873 | } | |
2874 | } | |
2875 | ||
2876 | /* | |
2877 | * mast_combine_cp_left - Copy in the original left side of the tree into the | |
2878 | * combined data set in the maple subtree state big node. | |
2879 | * @mast: The maple subtree state | |
2880 | */ | |
2881 | static inline void mast_combine_cp_left(struct maple_subtree_state *mast) | |
2882 | { | |
2883 | unsigned char l_slot = mast->orig_l->offset; | |
2884 | ||
2885 | if (!l_slot) | |
2886 | return; | |
2887 | ||
2888 | mas_mab_cp(mast->orig_l, 0, l_slot - 1, mast->bn, 0); | |
2889 | } | |
2890 | ||
2891 | /* | |
2892 | * mast_combine_cp_right: Copy in the original right side of the tree into the | |
2893 | * combined data set in the maple subtree state big node. | |
2894 | * @mast: The maple subtree state | |
2895 | */ | |
2896 | static inline void mast_combine_cp_right(struct maple_subtree_state *mast) | |
2897 | { | |
2898 | if (mast->bn->pivot[mast->bn->b_end - 1] >= mast->orig_r->max) | |
2899 | return; | |
2900 | ||
2901 | mas_mab_cp(mast->orig_r, mast->orig_r->offset + 1, | |
2902 | mt_slot_count(mast->orig_r->node), mast->bn, | |
2903 | mast->bn->b_end); | |
2904 | mast->orig_r->last = mast->orig_r->max; | |
2905 | } | |
2906 | ||
2907 | /* | |
2908 | * mast_sufficient: Check if the maple subtree state has enough data in the big | |
2909 | * node to create at least one sufficient node | |
2910 | * @mast: the maple subtree state | |
2911 | */ | |
2912 | static inline bool mast_sufficient(struct maple_subtree_state *mast) | |
2913 | { | |
2914 | if (mast->bn->b_end > mt_min_slot_count(mast->orig_l->node)) | |
2915 | return true; | |
2916 | ||
2917 | return false; | |
2918 | } | |
2919 | ||
2920 | /* | |
2921 | * mast_overflow: Check if there is too much data in the subtree state for a | |
2922 | * single node. | |
2923 | * @mast: The maple subtree state | |
2924 | */ | |
2925 | static inline bool mast_overflow(struct maple_subtree_state *mast) | |
2926 | { | |
2927 | if (mast->bn->b_end >= mt_slot_count(mast->orig_l->node)) | |
2928 | return true; | |
2929 | ||
2930 | return false; | |
2931 | } | |
2932 | ||
2933 | static inline void *mtree_range_walk(struct ma_state *mas) | |
2934 | { | |
2935 | unsigned long *pivots; | |
2936 | unsigned char offset; | |
2937 | struct maple_node *node; | |
2938 | struct maple_enode *next, *last; | |
2939 | enum maple_type type; | |
2940 | void __rcu **slots; | |
2941 | unsigned char end; | |
2942 | unsigned long max, min; | |
2943 | unsigned long prev_max, prev_min; | |
2944 | ||
1b9c9183 LB |
2945 | next = mas->node; |
2946 | min = mas->min; | |
54a611b6 LH |
2947 | max = mas->max; |
2948 | do { | |
2949 | offset = 0; | |
2950 | last = next; | |
2951 | node = mte_to_node(next); | |
2952 | type = mte_node_type(next); | |
2953 | pivots = ma_pivots(node, type); | |
2954 | end = ma_data_end(node, type, pivots, max); | |
2955 | if (unlikely(ma_dead_node(node))) | |
2956 | goto dead_node; | |
2957 | ||
2958 | if (pivots[offset] >= mas->index) { | |
2959 | prev_max = max; | |
2960 | prev_min = min; | |
2961 | max = pivots[offset]; | |
2962 | goto next; | |
2963 | } | |
2964 | ||
2965 | do { | |
2966 | offset++; | |
2967 | } while ((offset < end) && (pivots[offset] < mas->index)); | |
2968 | ||
2969 | prev_min = min; | |
2970 | min = pivots[offset - 1] + 1; | |
2971 | prev_max = max; | |
2972 | if (likely(offset < end && pivots[offset])) | |
2973 | max = pivots[offset]; | |
2974 | ||
2975 | next: | |
2976 | slots = ma_slots(node, type); | |
2977 | next = mt_slot(mas->tree, slots, offset); | |
2978 | if (unlikely(ma_dead_node(node))) | |
2979 | goto dead_node; | |
2980 | } while (!ma_is_leaf(type)); | |
2981 | ||
2982 | mas->offset = offset; | |
2983 | mas->index = min; | |
2984 | mas->last = max; | |
2985 | mas->min = prev_min; | |
2986 | mas->max = prev_max; | |
2987 | mas->node = last; | |
831978e3 | 2988 | return (void *)next; |
54a611b6 LH |
2989 | |
2990 | dead_node: | |
2991 | mas_reset(mas); | |
2992 | return NULL; | |
2993 | } | |
2994 | ||
2995 | /* | |
2996 | * mas_spanning_rebalance() - Rebalance across two nodes which may not be peers. | |
2997 | * @mas: The starting maple state | |
2998 | * @mast: The maple_subtree_state, keeps track of 4 maple states. | |
2999 | * @count: The estimated count of iterations needed. | |
3000 | * | |
3001 | * Follow the tree upwards from @l_mas and @r_mas for @count, or until the root | |
3002 | * is hit. First @b_node is split into two entries which are inserted into the | |
3003 | * next iteration of the loop. @b_node is returned populated with the final | |
3004 | * iteration. @mas is used to obtain allocations. orig_l_mas keeps track of the | |
3005 | * nodes that will remain active by using orig_l_mas->index and orig_l_mas->last | |
3006 | * to account of what has been copied into the new sub-tree. The update of | |
3007 | * orig_l_mas->last is used in mas_consume to find the slots that will need to | |
3008 | * be either freed or destroyed. orig_l_mas->depth keeps track of the height of | |
3009 | * the new sub-tree in case the sub-tree becomes the full tree. | |
3010 | * | |
3011 | * Return: the number of elements in b_node during the last loop. | |
3012 | */ | |
3013 | static int mas_spanning_rebalance(struct ma_state *mas, | |
3014 | struct maple_subtree_state *mast, unsigned char count) | |
3015 | { | |
3016 | unsigned char split, mid_split; | |
3017 | unsigned char slot = 0; | |
3018 | struct maple_enode *left = NULL, *middle = NULL, *right = NULL; | |
3019 | ||
3020 | MA_STATE(l_mas, mas->tree, mas->index, mas->index); | |
3021 | MA_STATE(r_mas, mas->tree, mas->index, mas->last); | |
3022 | MA_STATE(m_mas, mas->tree, mas->index, mas->index); | |
3023 | MA_TOPIARY(free, mas->tree); | |
3024 | MA_TOPIARY(destroy, mas->tree); | |
3025 | ||
3026 | /* | |
3027 | * The tree needs to be rebalanced and leaves need to be kept at the same level. | |
3028 | * Rebalancing is done by use of the ``struct maple_topiary``. | |
3029 | */ | |
3030 | mast->l = &l_mas; | |
3031 | mast->m = &m_mas; | |
3032 | mast->r = &r_mas; | |
3033 | mast->free = &free; | |
3034 | mast->destroy = &destroy; | |
3035 | l_mas.node = r_mas.node = m_mas.node = MAS_NONE; | |
0abb964a LH |
3036 | |
3037 | /* Check if this is not root and has sufficient data. */ | |
3038 | if (((mast->orig_l->min != 0) || (mast->orig_r->max != ULONG_MAX)) && | |
54a611b6 LH |
3039 | unlikely(mast->bn->b_end <= mt_min_slots[mast->bn->type])) |
3040 | mast_spanning_rebalance(mast); | |
3041 | ||
3042 | mast->orig_l->depth = 0; | |
3043 | ||
3044 | /* | |
3045 | * Each level of the tree is examined and balanced, pushing data to the left or | |
3046 | * right, or rebalancing against left or right nodes is employed to avoid | |
3047 | * rippling up the tree to limit the amount of churn. Once a new sub-section of | |
3048 | * the tree is created, there may be a mix of new and old nodes. The old nodes | |
3049 | * will have the incorrect parent pointers and currently be in two trees: the | |
3050 | * original tree and the partially new tree. To remedy the parent pointers in | |
3051 | * the old tree, the new data is swapped into the active tree and a walk down | |
3052 | * the tree is performed and the parent pointers are updated. | |
3053 | * See mas_descend_adopt() for more information.. | |
3054 | */ | |
3055 | while (count--) { | |
3056 | mast->bn->b_end--; | |
3057 | mast->bn->type = mte_node_type(mast->orig_l->node); | |
3058 | split = mas_mab_to_node(mas, mast->bn, &left, &right, &middle, | |
3059 | &mid_split, mast->orig_l->min); | |
3060 | mast_set_split_parents(mast, left, middle, right, split, | |
3061 | mid_split); | |
3062 | mast_cp_to_nodes(mast, left, middle, right, split, mid_split); | |
3063 | ||
3064 | /* | |
3065 | * Copy data from next level in the tree to mast->bn from next | |
3066 | * iteration | |
3067 | */ | |
3068 | memset(mast->bn, 0, sizeof(struct maple_big_node)); | |
3069 | mast->bn->type = mte_node_type(left); | |
3070 | mast->orig_l->depth++; | |
3071 | ||
3072 | /* Root already stored in l->node. */ | |
3073 | if (mas_is_root_limits(mast->l)) | |
3074 | goto new_root; | |
3075 | ||
3076 | mast_ascend_free(mast); | |
3077 | mast_combine_cp_left(mast); | |
3078 | l_mas.offset = mast->bn->b_end; | |
3079 | mab_set_b_end(mast->bn, &l_mas, left); | |
3080 | mab_set_b_end(mast->bn, &m_mas, middle); | |
3081 | mab_set_b_end(mast->bn, &r_mas, right); | |
3082 | ||
3083 | /* Copy anything necessary out of the right node. */ | |
3084 | mast_combine_cp_right(mast); | |
3085 | mast_topiary(mast); | |
3086 | mast->orig_l->last = mast->orig_l->max; | |
3087 | ||
3088 | if (mast_sufficient(mast)) | |
3089 | continue; | |
3090 | ||
3091 | if (mast_overflow(mast)) | |
3092 | continue; | |
3093 | ||
3094 | /* May be a new root stored in mast->bn */ | |
3095 | if (mas_is_root_limits(mast->orig_l)) | |
3096 | break; | |
3097 | ||
3098 | mast_spanning_rebalance(mast); | |
3099 | ||
3100 | /* rebalancing from other nodes may require another loop. */ | |
3101 | if (!count) | |
3102 | count++; | |
3103 | } | |
3104 | ||
3105 | l_mas.node = mt_mk_node(ma_mnode_ptr(mas_pop_node(mas)), | |
3106 | mte_node_type(mast->orig_l->node)); | |
3107 | mast->orig_l->depth++; | |
3108 | mab_mas_cp(mast->bn, 0, mt_slots[mast->bn->type] - 1, &l_mas, true); | |
3109 | mte_set_parent(left, l_mas.node, slot); | |
3110 | if (middle) | |
3111 | mte_set_parent(middle, l_mas.node, ++slot); | |
3112 | ||
3113 | if (right) | |
3114 | mte_set_parent(right, l_mas.node, ++slot); | |
3115 | ||
3116 | if (mas_is_root_limits(mast->l)) { | |
3117 | new_root: | |
3118 | mast_new_root(mast, mas); | |
3119 | } else { | |
3120 | mas_mn(&l_mas)->parent = mas_mn(mast->orig_l)->parent; | |
3121 | } | |
3122 | ||
3123 | if (!mte_dead_node(mast->orig_l->node)) | |
3124 | mat_add(&free, mast->orig_l->node); | |
3125 | ||
3126 | mas->depth = mast->orig_l->depth; | |
3127 | *mast->orig_l = l_mas; | |
3128 | mte_set_node_dead(mas->node); | |
3129 | ||
3130 | /* Set up mas for insertion. */ | |
3131 | mast->orig_l->depth = mas->depth; | |
3132 | mast->orig_l->alloc = mas->alloc; | |
3133 | *mas = *mast->orig_l; | |
3134 | mas_wmb_replace(mas, &free, &destroy); | |
3135 | mtree_range_walk(mas); | |
3136 | return mast->bn->b_end; | |
3137 | } | |
3138 | ||
3139 | /* | |
3140 | * mas_rebalance() - Rebalance a given node. | |
3141 | * @mas: The maple state | |
3142 | * @b_node: The big maple node. | |
3143 | * | |
3144 | * Rebalance two nodes into a single node or two new nodes that are sufficient. | |
3145 | * Continue upwards until tree is sufficient. | |
3146 | * | |
3147 | * Return: the number of elements in b_node during the last loop. | |
3148 | */ | |
3149 | static inline int mas_rebalance(struct ma_state *mas, | |
3150 | struct maple_big_node *b_node) | |
3151 | { | |
3152 | char empty_count = mas_mt_height(mas); | |
3153 | struct maple_subtree_state mast; | |
3154 | unsigned char shift, b_end = ++b_node->b_end; | |
3155 | ||
3156 | MA_STATE(l_mas, mas->tree, mas->index, mas->last); | |
3157 | MA_STATE(r_mas, mas->tree, mas->index, mas->last); | |
3158 | ||
3159 | trace_ma_op(__func__, mas); | |
3160 | ||
3161 | /* | |
3162 | * Rebalancing occurs if a node is insufficient. Data is rebalanced | |
3163 | * against the node to the right if it exists, otherwise the node to the | |
3164 | * left of this node is rebalanced against this node. If rebalancing | |
3165 | * causes just one node to be produced instead of two, then the parent | |
3166 | * is also examined and rebalanced if it is insufficient. Every level | |
3167 | * tries to combine the data in the same way. If one node contains the | |
3168 | * entire range of the tree, then that node is used as a new root node. | |
3169 | */ | |
3170 | mas_node_count(mas, 1 + empty_count * 3); | |
3171 | if (mas_is_err(mas)) | |
3172 | return 0; | |
3173 | ||
3174 | mast.orig_l = &l_mas; | |
3175 | mast.orig_r = &r_mas; | |
3176 | mast.bn = b_node; | |
3177 | mast.bn->type = mte_node_type(mas->node); | |
3178 | ||
3179 | l_mas = r_mas = *mas; | |
3180 | ||
3181 | if (mas_next_sibling(&r_mas)) { | |
3182 | mas_mab_cp(&r_mas, 0, mt_slot_count(r_mas.node), b_node, b_end); | |
3183 | r_mas.last = r_mas.index = r_mas.max; | |
3184 | } else { | |
3185 | mas_prev_sibling(&l_mas); | |
3186 | shift = mas_data_end(&l_mas) + 1; | |
3187 | mab_shift_right(b_node, shift); | |
3188 | mas->offset += shift; | |
3189 | mas_mab_cp(&l_mas, 0, shift - 1, b_node, 0); | |
3190 | b_node->b_end = shift + b_end; | |
3191 | l_mas.index = l_mas.last = l_mas.min; | |
3192 | } | |
3193 | ||
3194 | return mas_spanning_rebalance(mas, &mast, empty_count); | |
3195 | } | |
3196 | ||
3197 | /* | |
3198 | * mas_destroy_rebalance() - Rebalance left-most node while destroying the maple | |
3199 | * state. | |
3200 | * @mas: The maple state | |
3201 | * @end: The end of the left-most node. | |
3202 | * | |
3203 | * During a mass-insert event (such as forking), it may be necessary to | |
3204 | * rebalance the left-most node when it is not sufficient. | |
3205 | */ | |
3206 | static inline void mas_destroy_rebalance(struct ma_state *mas, unsigned char end) | |
3207 | { | |
3208 | enum maple_type mt = mte_node_type(mas->node); | |
3209 | struct maple_node reuse, *newnode, *parent, *new_left, *left, *node; | |
3210 | struct maple_enode *eparent; | |
3211 | unsigned char offset, tmp, split = mt_slots[mt] / 2; | |
3212 | void __rcu **l_slots, **slots; | |
3213 | unsigned long *l_pivs, *pivs, gap; | |
3214 | bool in_rcu = mt_in_rcu(mas->tree); | |
3215 | ||
3216 | MA_STATE(l_mas, mas->tree, mas->index, mas->last); | |
3217 | ||
3218 | l_mas = *mas; | |
3219 | mas_prev_sibling(&l_mas); | |
3220 | ||
3221 | /* set up node. */ | |
3222 | if (in_rcu) { | |
3223 | /* Allocate for both left and right as well as parent. */ | |
3224 | mas_node_count(mas, 3); | |
3225 | if (mas_is_err(mas)) | |
3226 | return; | |
3227 | ||
3228 | newnode = mas_pop_node(mas); | |
3229 | } else { | |
3230 | newnode = &reuse; | |
3231 | } | |
3232 | ||
3233 | node = mas_mn(mas); | |
3234 | newnode->parent = node->parent; | |
3235 | slots = ma_slots(newnode, mt); | |
3236 | pivs = ma_pivots(newnode, mt); | |
3237 | left = mas_mn(&l_mas); | |
3238 | l_slots = ma_slots(left, mt); | |
3239 | l_pivs = ma_pivots(left, mt); | |
3240 | if (!l_slots[split]) | |
3241 | split++; | |
3242 | tmp = mas_data_end(&l_mas) - split; | |
3243 | ||
3244 | memcpy(slots, l_slots + split + 1, sizeof(void *) * tmp); | |
3245 | memcpy(pivs, l_pivs + split + 1, sizeof(unsigned long) * tmp); | |
3246 | pivs[tmp] = l_mas.max; | |
3247 | memcpy(slots + tmp, ma_slots(node, mt), sizeof(void *) * end); | |
3248 | memcpy(pivs + tmp, ma_pivots(node, mt), sizeof(unsigned long) * end); | |
3249 | ||
3250 | l_mas.max = l_pivs[split]; | |
3251 | mas->min = l_mas.max + 1; | |
3252 | eparent = mt_mk_node(mte_parent(l_mas.node), | |
3253 | mas_parent_enum(&l_mas, l_mas.node)); | |
3254 | tmp += end; | |
3255 | if (!in_rcu) { | |
3256 | unsigned char max_p = mt_pivots[mt]; | |
3257 | unsigned char max_s = mt_slots[mt]; | |
3258 | ||
3259 | if (tmp < max_p) | |
3260 | memset(pivs + tmp, 0, | |
fb20e99a | 3261 | sizeof(unsigned long) * (max_p - tmp)); |
54a611b6 LH |
3262 | |
3263 | if (tmp < mt_slots[mt]) | |
3264 | memset(slots + tmp, 0, sizeof(void *) * (max_s - tmp)); | |
3265 | ||
3266 | memcpy(node, newnode, sizeof(struct maple_node)); | |
3267 | ma_set_meta(node, mt, 0, tmp - 1); | |
3268 | mte_set_pivot(eparent, mte_parent_slot(l_mas.node), | |
3269 | l_pivs[split]); | |
3270 | ||
3271 | /* Remove data from l_pivs. */ | |
3272 | tmp = split + 1; | |
3273 | memset(l_pivs + tmp, 0, sizeof(unsigned long) * (max_p - tmp)); | |
3274 | memset(l_slots + tmp, 0, sizeof(void *) * (max_s - tmp)); | |
3275 | ma_set_meta(left, mt, 0, split); | |
3276 | ||
3277 | goto done; | |
3278 | } | |
3279 | ||
3280 | /* RCU requires replacing both l_mas, mas, and parent. */ | |
3281 | mas->node = mt_mk_node(newnode, mt); | |
3282 | ma_set_meta(newnode, mt, 0, tmp); | |
3283 | ||
3284 | new_left = mas_pop_node(mas); | |
3285 | new_left->parent = left->parent; | |
3286 | mt = mte_node_type(l_mas.node); | |
3287 | slots = ma_slots(new_left, mt); | |
3288 | pivs = ma_pivots(new_left, mt); | |
3289 | memcpy(slots, l_slots, sizeof(void *) * split); | |
3290 | memcpy(pivs, l_pivs, sizeof(unsigned long) * split); | |
3291 | ma_set_meta(new_left, mt, 0, split); | |
3292 | l_mas.node = mt_mk_node(new_left, mt); | |
3293 | ||
3294 | /* replace parent. */ | |
3295 | offset = mte_parent_slot(mas->node); | |
3296 | mt = mas_parent_enum(&l_mas, l_mas.node); | |
3297 | parent = mas_pop_node(mas); | |
3298 | slots = ma_slots(parent, mt); | |
3299 | pivs = ma_pivots(parent, mt); | |
3300 | memcpy(parent, mte_to_node(eparent), sizeof(struct maple_node)); | |
3301 | rcu_assign_pointer(slots[offset], mas->node); | |
3302 | rcu_assign_pointer(slots[offset - 1], l_mas.node); | |
3303 | pivs[offset - 1] = l_mas.max; | |
3304 | eparent = mt_mk_node(parent, mt); | |
3305 | done: | |
3306 | gap = mas_leaf_max_gap(mas); | |
3307 | mte_set_gap(eparent, mte_parent_slot(mas->node), gap); | |
3308 | gap = mas_leaf_max_gap(&l_mas); | |
3309 | mte_set_gap(eparent, mte_parent_slot(l_mas.node), gap); | |
3310 | mas_ascend(mas); | |
3311 | ||
3312 | if (in_rcu) | |
3313 | mas_replace(mas, false); | |
3314 | ||
3315 | mas_update_gap(mas); | |
3316 | } | |
3317 | ||
3318 | /* | |
3319 | * mas_split_final_node() - Split the final node in a subtree operation. | |
3320 | * @mast: the maple subtree state | |
3321 | * @mas: The maple state | |
3322 | * @height: The height of the tree in case it's a new root. | |
3323 | */ | |
3324 | static inline bool mas_split_final_node(struct maple_subtree_state *mast, | |
3325 | struct ma_state *mas, int height) | |
3326 | { | |
3327 | struct maple_enode *ancestor; | |
3328 | ||
3329 | if (mte_is_root(mas->node)) { | |
3330 | if (mt_is_alloc(mas->tree)) | |
3331 | mast->bn->type = maple_arange_64; | |
3332 | else | |
3333 | mast->bn->type = maple_range_64; | |
3334 | mas->depth = height; | |
3335 | } | |
3336 | /* | |
3337 | * Only a single node is used here, could be root. | |
3338 | * The Big_node data should just fit in a single node. | |
3339 | */ | |
3340 | ancestor = mas_new_ma_node(mas, mast->bn); | |
3341 | mte_set_parent(mast->l->node, ancestor, mast->l->offset); | |
3342 | mte_set_parent(mast->r->node, ancestor, mast->r->offset); | |
3343 | mte_to_node(ancestor)->parent = mas_mn(mas)->parent; | |
3344 | ||
3345 | mast->l->node = ancestor; | |
3346 | mab_mas_cp(mast->bn, 0, mt_slots[mast->bn->type] - 1, mast->l, true); | |
3347 | mas->offset = mast->bn->b_end - 1; | |
3348 | return true; | |
3349 | } | |
3350 | ||
3351 | /* | |
3352 | * mast_fill_bnode() - Copy data into the big node in the subtree state | |
3353 | * @mast: The maple subtree state | |
3354 | * @mas: the maple state | |
3355 | * @skip: The number of entries to skip for new nodes insertion. | |
3356 | */ | |
3357 | static inline void mast_fill_bnode(struct maple_subtree_state *mast, | |
3358 | struct ma_state *mas, | |
3359 | unsigned char skip) | |
3360 | { | |
3361 | bool cp = true; | |
3362 | struct maple_enode *old = mas->node; | |
3363 | unsigned char split; | |
3364 | ||
3365 | memset(mast->bn->gap, 0, sizeof(unsigned long) * ARRAY_SIZE(mast->bn->gap)); | |
3366 | memset(mast->bn->slot, 0, sizeof(unsigned long) * ARRAY_SIZE(mast->bn->slot)); | |
3367 | memset(mast->bn->pivot, 0, sizeof(unsigned long) * ARRAY_SIZE(mast->bn->pivot)); | |
3368 | mast->bn->b_end = 0; | |
3369 | ||
3370 | if (mte_is_root(mas->node)) { | |
3371 | cp = false; | |
3372 | } else { | |
3373 | mas_ascend(mas); | |
3374 | mat_add(mast->free, old); | |
3375 | mas->offset = mte_parent_slot(mas->node); | |
3376 | } | |
3377 | ||
3378 | if (cp && mast->l->offset) | |
3379 | mas_mab_cp(mas, 0, mast->l->offset - 1, mast->bn, 0); | |
3380 | ||
3381 | split = mast->bn->b_end; | |
3382 | mab_set_b_end(mast->bn, mast->l, mast->l->node); | |
3383 | mast->r->offset = mast->bn->b_end; | |
3384 | mab_set_b_end(mast->bn, mast->r, mast->r->node); | |
3385 | if (mast->bn->pivot[mast->bn->b_end - 1] == mas->max) | |
3386 | cp = false; | |
3387 | ||
3388 | if (cp) | |
3389 | mas_mab_cp(mas, split + skip, mt_slot_count(mas->node) - 1, | |
3390 | mast->bn, mast->bn->b_end); | |
3391 | ||
3392 | mast->bn->b_end--; | |
3393 | mast->bn->type = mte_node_type(mas->node); | |
3394 | } | |
3395 | ||
3396 | /* | |
3397 | * mast_split_data() - Split the data in the subtree state big node into regular | |
3398 | * nodes. | |
3399 | * @mast: The maple subtree state | |
3400 | * @mas: The maple state | |
3401 | * @split: The location to split the big node | |
3402 | */ | |
3403 | static inline void mast_split_data(struct maple_subtree_state *mast, | |
3404 | struct ma_state *mas, unsigned char split) | |
3405 | { | |
3406 | unsigned char p_slot; | |
3407 | ||
3408 | mab_mas_cp(mast->bn, 0, split, mast->l, true); | |
3409 | mte_set_pivot(mast->r->node, 0, mast->r->max); | |
3410 | mab_mas_cp(mast->bn, split + 1, mast->bn->b_end, mast->r, false); | |
3411 | mast->l->offset = mte_parent_slot(mas->node); | |
3412 | mast->l->max = mast->bn->pivot[split]; | |
3413 | mast->r->min = mast->l->max + 1; | |
3414 | if (mte_is_leaf(mas->node)) | |
3415 | return; | |
3416 | ||
3417 | p_slot = mast->orig_l->offset; | |
3418 | mas_set_split_parent(mast->orig_l, mast->l->node, mast->r->node, | |
3419 | &p_slot, split); | |
3420 | mas_set_split_parent(mast->orig_r, mast->l->node, mast->r->node, | |
3421 | &p_slot, split); | |
3422 | } | |
3423 | ||
3424 | /* | |
3425 | * mas_push_data() - Instead of splitting a node, it is beneficial to push the | |
3426 | * data to the right or left node if there is room. | |
3427 | * @mas: The maple state | |
3428 | * @height: The current height of the maple state | |
3429 | * @mast: The maple subtree state | |
3430 | * @left: Push left or not. | |
3431 | * | |
3432 | * Keeping the height of the tree low means faster lookups. | |
3433 | * | |
3434 | * Return: True if pushed, false otherwise. | |
3435 | */ | |
3436 | static inline bool mas_push_data(struct ma_state *mas, int height, | |
3437 | struct maple_subtree_state *mast, bool left) | |
3438 | { | |
3439 | unsigned char slot_total = mast->bn->b_end; | |
3440 | unsigned char end, space, split; | |
3441 | ||
3442 | MA_STATE(tmp_mas, mas->tree, mas->index, mas->last); | |
3443 | tmp_mas = *mas; | |
3444 | tmp_mas.depth = mast->l->depth; | |
3445 | ||
3446 | if (left && !mas_prev_sibling(&tmp_mas)) | |
3447 | return false; | |
3448 | else if (!left && !mas_next_sibling(&tmp_mas)) | |
3449 | return false; | |
3450 | ||
3451 | end = mas_data_end(&tmp_mas); | |
3452 | slot_total += end; | |
3453 | space = 2 * mt_slot_count(mas->node) - 2; | |
3454 | /* -2 instead of -1 to ensure there isn't a triple split */ | |
3455 | if (ma_is_leaf(mast->bn->type)) | |
3456 | space--; | |
3457 | ||
3458 | if (mas->max == ULONG_MAX) | |
3459 | space--; | |
3460 | ||
3461 | if (slot_total >= space) | |
3462 | return false; | |
3463 | ||
3464 | /* Get the data; Fill mast->bn */ | |
3465 | mast->bn->b_end++; | |
3466 | if (left) { | |
3467 | mab_shift_right(mast->bn, end + 1); | |
3468 | mas_mab_cp(&tmp_mas, 0, end, mast->bn, 0); | |
3469 | mast->bn->b_end = slot_total + 1; | |
3470 | } else { | |
3471 | mas_mab_cp(&tmp_mas, 0, end, mast->bn, mast->bn->b_end); | |
3472 | } | |
3473 | ||
3474 | /* Configure mast for splitting of mast->bn */ | |
3475 | split = mt_slots[mast->bn->type] - 2; | |
3476 | if (left) { | |
3477 | /* Switch mas to prev node */ | |
3478 | mat_add(mast->free, mas->node); | |
3479 | *mas = tmp_mas; | |
3480 | /* Start using mast->l for the left side. */ | |
3481 | tmp_mas.node = mast->l->node; | |
3482 | *mast->l = tmp_mas; | |
3483 | } else { | |
3484 | mat_add(mast->free, tmp_mas.node); | |
3485 | tmp_mas.node = mast->r->node; | |
3486 | *mast->r = tmp_mas; | |
3487 | split = slot_total - split; | |
3488 | } | |
3489 | split = mab_no_null_split(mast->bn, split, mt_slots[mast->bn->type]); | |
3490 | /* Update parent slot for split calculation. */ | |
3491 | if (left) | |
3492 | mast->orig_l->offset += end + 1; | |
3493 | ||
3494 | mast_split_data(mast, mas, split); | |
3495 | mast_fill_bnode(mast, mas, 2); | |
3496 | mas_split_final_node(mast, mas, height + 1); | |
3497 | return true; | |
3498 | } | |
3499 | ||
3500 | /* | |
3501 | * mas_split() - Split data that is too big for one node into two. | |
3502 | * @mas: The maple state | |
3503 | * @b_node: The maple big node | |
3504 | * Return: 1 on success, 0 on failure. | |
3505 | */ | |
3506 | static int mas_split(struct ma_state *mas, struct maple_big_node *b_node) | |
3507 | { | |
54a611b6 LH |
3508 | struct maple_subtree_state mast; |
3509 | int height = 0; | |
3510 | unsigned char mid_split, split = 0; | |
3511 | ||
3512 | /* | |
3513 | * Splitting is handled differently from any other B-tree; the Maple | |
3514 | * Tree splits upwards. Splitting up means that the split operation | |
3515 | * occurs when the walk of the tree hits the leaves and not on the way | |
3516 | * down. The reason for splitting up is that it is impossible to know | |
3517 | * how much space will be needed until the leaf is (or leaves are) | |
3518 | * reached. Since overwriting data is allowed and a range could | |
3519 | * overwrite more than one range or result in changing one entry into 3 | |
3520 | * entries, it is impossible to know if a split is required until the | |
3521 | * data is examined. | |
3522 | * | |
3523 | * Splitting is a balancing act between keeping allocations to a minimum | |
3524 | * and avoiding a 'jitter' event where a tree is expanded to make room | |
3525 | * for an entry followed by a contraction when the entry is removed. To | |
3526 | * accomplish the balance, there are empty slots remaining in both left | |
3527 | * and right nodes after a split. | |
3528 | */ | |
3529 | MA_STATE(l_mas, mas->tree, mas->index, mas->last); | |
3530 | MA_STATE(r_mas, mas->tree, mas->index, mas->last); | |
3531 | MA_STATE(prev_l_mas, mas->tree, mas->index, mas->last); | |
3532 | MA_STATE(prev_r_mas, mas->tree, mas->index, mas->last); | |
3533 | MA_TOPIARY(mat, mas->tree); | |
3534 | ||
3535 | trace_ma_op(__func__, mas); | |
3536 | mas->depth = mas_mt_height(mas); | |
3537 | /* Allocation failures will happen early. */ | |
3538 | mas_node_count(mas, 1 + mas->depth * 2); | |
3539 | if (mas_is_err(mas)) | |
3540 | return 0; | |
3541 | ||
3542 | mast.l = &l_mas; | |
3543 | mast.r = &r_mas; | |
3544 | mast.orig_l = &prev_l_mas; | |
3545 | mast.orig_r = &prev_r_mas; | |
3546 | mast.free = &mat; | |
3547 | mast.bn = b_node; | |
3548 | ||
3549 | while (height++ <= mas->depth) { | |
3550 | if (mt_slots[b_node->type] > b_node->b_end) { | |
3551 | mas_split_final_node(&mast, mas, height); | |
3552 | break; | |
3553 | } | |
3554 | ||
3555 | l_mas = r_mas = *mas; | |
3556 | l_mas.node = mas_new_ma_node(mas, b_node); | |
3557 | r_mas.node = mas_new_ma_node(mas, b_node); | |
3558 | /* | |
3559 | * Another way that 'jitter' is avoided is to terminate a split up early if the | |
3560 | * left or right node has space to spare. This is referred to as "pushing left" | |
3561 | * or "pushing right" and is similar to the B* tree, except the nodes left or | |
3562 | * right can rarely be reused due to RCU, but the ripple upwards is halted which | |
3563 | * is a significant savings. | |
3564 | */ | |
3565 | /* Try to push left. */ | |
3566 | if (mas_push_data(mas, height, &mast, true)) | |
3567 | break; | |
3568 | ||
3569 | /* Try to push right. */ | |
3570 | if (mas_push_data(mas, height, &mast, false)) | |
3571 | break; | |
3572 | ||
3573 | split = mab_calc_split(mas, b_node, &mid_split, prev_l_mas.min); | |
3574 | mast_split_data(&mast, mas, split); | |
3575 | /* | |
3576 | * Usually correct, mab_mas_cp in the above call overwrites | |
3577 | * r->max. | |
3578 | */ | |
3579 | mast.r->max = mas->max; | |
3580 | mast_fill_bnode(&mast, mas, 1); | |
3581 | prev_l_mas = *mast.l; | |
3582 | prev_r_mas = *mast.r; | |
3583 | } | |
3584 | ||
3585 | /* Set the original node as dead */ | |
3586 | mat_add(mast.free, mas->node); | |
3587 | mas->node = l_mas.node; | |
3588 | mas_wmb_replace(mas, mast.free, NULL); | |
3589 | mtree_range_walk(mas); | |
3590 | return 1; | |
3591 | } | |
3592 | ||
3593 | /* | |
3594 | * mas_reuse_node() - Reuse the node to store the data. | |
3595 | * @wr_mas: The maple write state | |
3596 | * @bn: The maple big node | |
3597 | * @end: The end of the data. | |
3598 | * | |
3599 | * Will always return false in RCU mode. | |
3600 | * | |
3601 | * Return: True if node was reused, false otherwise. | |
3602 | */ | |
3603 | static inline bool mas_reuse_node(struct ma_wr_state *wr_mas, | |
3604 | struct maple_big_node *bn, unsigned char end) | |
3605 | { | |
3606 | /* Need to be rcu safe. */ | |
3607 | if (mt_in_rcu(wr_mas->mas->tree)) | |
3608 | return false; | |
3609 | ||
3610 | if (end > bn->b_end) { | |
3611 | int clear = mt_slots[wr_mas->type] - bn->b_end; | |
3612 | ||
3613 | memset(wr_mas->slots + bn->b_end, 0, sizeof(void *) * clear--); | |
3614 | memset(wr_mas->pivots + bn->b_end, 0, sizeof(void *) * clear); | |
3615 | } | |
3616 | mab_mas_cp(bn, 0, bn->b_end, wr_mas->mas, false); | |
3617 | return true; | |
3618 | } | |
3619 | ||
3620 | /* | |
3621 | * mas_commit_b_node() - Commit the big node into the tree. | |
3622 | * @wr_mas: The maple write state | |
3623 | * @b_node: The maple big node | |
3624 | * @end: The end of the data. | |
3625 | */ | |
44081c77 | 3626 | static noinline_for_kasan int mas_commit_b_node(struct ma_wr_state *wr_mas, |
54a611b6 LH |
3627 | struct maple_big_node *b_node, unsigned char end) |
3628 | { | |
3629 | struct maple_node *node; | |
3630 | unsigned char b_end = b_node->b_end; | |
3631 | enum maple_type b_type = b_node->type; | |
3632 | ||
3633 | if ((b_end < mt_min_slots[b_type]) && | |
3634 | (!mte_is_root(wr_mas->mas->node)) && | |
3635 | (mas_mt_height(wr_mas->mas) > 1)) | |
3636 | return mas_rebalance(wr_mas->mas, b_node); | |
3637 | ||
3638 | if (b_end >= mt_slots[b_type]) | |
3639 | return mas_split(wr_mas->mas, b_node); | |
3640 | ||
3641 | if (mas_reuse_node(wr_mas, b_node, end)) | |
3642 | goto reuse_node; | |
3643 | ||
3644 | mas_node_count(wr_mas->mas, 1); | |
3645 | if (mas_is_err(wr_mas->mas)) | |
3646 | return 0; | |
3647 | ||
3648 | node = mas_pop_node(wr_mas->mas); | |
3649 | node->parent = mas_mn(wr_mas->mas)->parent; | |
3650 | wr_mas->mas->node = mt_mk_node(node, b_type); | |
7dc5ba62 | 3651 | mab_mas_cp(b_node, 0, b_end, wr_mas->mas, false); |
54a611b6 LH |
3652 | mas_replace(wr_mas->mas, false); |
3653 | reuse_node: | |
3654 | mas_update_gap(wr_mas->mas); | |
3655 | return 1; | |
3656 | } | |
3657 | ||
3658 | /* | |
3659 | * mas_root_expand() - Expand a root to a node | |
3660 | * @mas: The maple state | |
3661 | * @entry: The entry to store into the tree | |
3662 | */ | |
3663 | static inline int mas_root_expand(struct ma_state *mas, void *entry) | |
3664 | { | |
3665 | void *contents = mas_root_locked(mas); | |
3666 | enum maple_type type = maple_leaf_64; | |
3667 | struct maple_node *node; | |
3668 | void __rcu **slots; | |
3669 | unsigned long *pivots; | |
3670 | int slot = 0; | |
3671 | ||
3672 | mas_node_count(mas, 1); | |
3673 | if (unlikely(mas_is_err(mas))) | |
3674 | return 0; | |
3675 | ||
3676 | node = mas_pop_node(mas); | |
3677 | pivots = ma_pivots(node, type); | |
3678 | slots = ma_slots(node, type); | |
3679 | node->parent = ma_parent_ptr( | |
3680 | ((unsigned long)mas->tree | MA_ROOT_PARENT)); | |
3681 | mas->node = mt_mk_node(node, type); | |
3682 | ||
3683 | if (mas->index) { | |
3684 | if (contents) { | |
3685 | rcu_assign_pointer(slots[slot], contents); | |
3686 | if (likely(mas->index > 1)) | |
3687 | slot++; | |
3688 | } | |
3689 | pivots[slot++] = mas->index - 1; | |
3690 | } | |
3691 | ||
3692 | rcu_assign_pointer(slots[slot], entry); | |
3693 | mas->offset = slot; | |
3694 | pivots[slot] = mas->last; | |
3695 | if (mas->last != ULONG_MAX) | |
3696 | slot++; | |
3697 | mas->depth = 1; | |
3698 | mas_set_height(mas); | |
c45ea315 | 3699 | ma_set_meta(node, maple_leaf_64, 0, slot); |
54a611b6 LH |
3700 | /* swap the new root into the tree */ |
3701 | rcu_assign_pointer(mas->tree->ma_root, mte_mk_root(mas->node)); | |
54a611b6 LH |
3702 | return slot; |
3703 | } | |
3704 | ||
3705 | static inline void mas_store_root(struct ma_state *mas, void *entry) | |
3706 | { | |
3707 | if (likely((mas->last != 0) || (mas->index != 0))) | |
3708 | mas_root_expand(mas, entry); | |
3709 | else if (((unsigned long) (entry) & 3) == 2) | |
3710 | mas_root_expand(mas, entry); | |
3711 | else { | |
3712 | rcu_assign_pointer(mas->tree->ma_root, entry); | |
3713 | mas->node = MAS_START; | |
3714 | } | |
3715 | } | |
3716 | ||
3717 | /* | |
3718 | * mas_is_span_wr() - Check if the write needs to be treated as a write that | |
3719 | * spans the node. | |
3720 | * @mas: The maple state | |
3721 | * @piv: The pivot value being written | |
3722 | * @type: The maple node type | |
3723 | * @entry: The data to write | |
3724 | * | |
3725 | * Spanning writes are writes that start in one node and end in another OR if | |
3726 | * the write of a %NULL will cause the node to end with a %NULL. | |
3727 | * | |
3728 | * Return: True if this is a spanning write, false otherwise. | |
3729 | */ | |
3730 | static bool mas_is_span_wr(struct ma_wr_state *wr_mas) | |
3731 | { | |
3732 | unsigned long max; | |
3733 | unsigned long last = wr_mas->mas->last; | |
3734 | unsigned long piv = wr_mas->r_max; | |
3735 | enum maple_type type = wr_mas->type; | |
3736 | void *entry = wr_mas->entry; | |
3737 | ||
3738 | /* Contained in this pivot */ | |
3739 | if (piv > last) | |
3740 | return false; | |
3741 | ||
3742 | max = wr_mas->mas->max; | |
3743 | if (unlikely(ma_is_leaf(type))) { | |
3744 | /* Fits in the node, but may span slots. */ | |
3745 | if (last < max) | |
3746 | return false; | |
3747 | ||
3748 | /* Writes to the end of the node but not null. */ | |
3749 | if ((last == max) && entry) | |
3750 | return false; | |
3751 | ||
3752 | /* | |
3753 | * Writing ULONG_MAX is not a spanning write regardless of the | |
3754 | * value being written as long as the range fits in the node. | |
3755 | */ | |
3756 | if ((last == ULONG_MAX) && (last == max)) | |
3757 | return false; | |
3758 | } else if (piv == last) { | |
3759 | if (entry) | |
3760 | return false; | |
3761 | ||
3762 | /* Detect spanning store wr walk */ | |
3763 | if (last == ULONG_MAX) | |
3764 | return false; | |
3765 | } | |
3766 | ||
3767 | trace_ma_write(__func__, wr_mas->mas, piv, entry); | |
3768 | ||
3769 | return true; | |
3770 | } | |
3771 | ||
3772 | static inline void mas_wr_walk_descend(struct ma_wr_state *wr_mas) | |
3773 | { | |
54a611b6 LH |
3774 | wr_mas->type = mte_node_type(wr_mas->mas->node); |
3775 | mas_wr_node_walk(wr_mas); | |
3776 | wr_mas->slots = ma_slots(wr_mas->node, wr_mas->type); | |
3777 | } | |
3778 | ||
3779 | static inline void mas_wr_walk_traverse(struct ma_wr_state *wr_mas) | |
3780 | { | |
3781 | wr_mas->mas->max = wr_mas->r_max; | |
3782 | wr_mas->mas->min = wr_mas->r_min; | |
3783 | wr_mas->mas->node = wr_mas->content; | |
3784 | wr_mas->mas->offset = 0; | |
9bbba563 | 3785 | wr_mas->mas->depth++; |
54a611b6 LH |
3786 | } |
3787 | /* | |
3788 | * mas_wr_walk() - Walk the tree for a write. | |
3789 | * @wr_mas: The maple write state | |
3790 | * | |
3791 | * Uses mas_slot_locked() and does not need to worry about dead nodes. | |
3792 | * | |
3793 | * Return: True if it's contained in a node, false on spanning write. | |
3794 | */ | |
3795 | static bool mas_wr_walk(struct ma_wr_state *wr_mas) | |
3796 | { | |
3797 | struct ma_state *mas = wr_mas->mas; | |
3798 | ||
3799 | while (true) { | |
3800 | mas_wr_walk_descend(wr_mas); | |
3801 | if (unlikely(mas_is_span_wr(wr_mas))) | |
3802 | return false; | |
3803 | ||
3804 | wr_mas->content = mas_slot_locked(mas, wr_mas->slots, | |
3805 | mas->offset); | |
3806 | if (ma_is_leaf(wr_mas->type)) | |
3807 | return true; | |
3808 | ||
3809 | mas_wr_walk_traverse(wr_mas); | |
3810 | } | |
3811 | ||
3812 | return true; | |
3813 | } | |
3814 | ||
3815 | static bool mas_wr_walk_index(struct ma_wr_state *wr_mas) | |
3816 | { | |
3817 | struct ma_state *mas = wr_mas->mas; | |
3818 | ||
3819 | while (true) { | |
3820 | mas_wr_walk_descend(wr_mas); | |
3821 | wr_mas->content = mas_slot_locked(mas, wr_mas->slots, | |
3822 | mas->offset); | |
3823 | if (ma_is_leaf(wr_mas->type)) | |
3824 | return true; | |
3825 | mas_wr_walk_traverse(wr_mas); | |
3826 | ||
3827 | } | |
3828 | return true; | |
3829 | } | |
3830 | /* | |
3831 | * mas_extend_spanning_null() - Extend a store of a %NULL to include surrounding %NULLs. | |
3832 | * @l_wr_mas: The left maple write state | |
3833 | * @r_wr_mas: The right maple write state | |
3834 | */ | |
3835 | static inline void mas_extend_spanning_null(struct ma_wr_state *l_wr_mas, | |
3836 | struct ma_wr_state *r_wr_mas) | |
3837 | { | |
3838 | struct ma_state *r_mas = r_wr_mas->mas; | |
3839 | struct ma_state *l_mas = l_wr_mas->mas; | |
3840 | unsigned char l_slot; | |
3841 | ||
3842 | l_slot = l_mas->offset; | |
3843 | if (!l_wr_mas->content) | |
3844 | l_mas->index = l_wr_mas->r_min; | |
3845 | ||
3846 | if ((l_mas->index == l_wr_mas->r_min) && | |
3847 | (l_slot && | |
3848 | !mas_slot_locked(l_mas, l_wr_mas->slots, l_slot - 1))) { | |
3849 | if (l_slot > 1) | |
3850 | l_mas->index = l_wr_mas->pivots[l_slot - 2] + 1; | |
3851 | else | |
3852 | l_mas->index = l_mas->min; | |
3853 | ||
3854 | l_mas->offset = l_slot - 1; | |
3855 | } | |
3856 | ||
3857 | if (!r_wr_mas->content) { | |
3858 | if (r_mas->last < r_wr_mas->r_max) | |
3859 | r_mas->last = r_wr_mas->r_max; | |
3860 | r_mas->offset++; | |
3861 | } else if ((r_mas->last == r_wr_mas->r_max) && | |
3862 | (r_mas->last < r_mas->max) && | |
3863 | !mas_slot_locked(r_mas, r_wr_mas->slots, r_mas->offset + 1)) { | |
3864 | r_mas->last = mas_safe_pivot(r_mas, r_wr_mas->pivots, | |
3865 | r_wr_mas->type, r_mas->offset + 1); | |
3866 | r_mas->offset++; | |
3867 | } | |
3868 | } | |
3869 | ||
3870 | static inline void *mas_state_walk(struct ma_state *mas) | |
3871 | { | |
3872 | void *entry; | |
3873 | ||
3874 | entry = mas_start(mas); | |
3875 | if (mas_is_none(mas)) | |
3876 | return NULL; | |
3877 | ||
3878 | if (mas_is_ptr(mas)) | |
3879 | return entry; | |
3880 | ||
3881 | return mtree_range_walk(mas); | |
3882 | } | |
3883 | ||
3884 | /* | |
3885 | * mtree_lookup_walk() - Internal quick lookup that does not keep maple state up | |
3886 | * to date. | |
3887 | * | |
3888 | * @mas: The maple state. | |
3889 | * | |
3890 | * Note: Leaves mas in undesirable state. | |
3891 | * Return: The entry for @mas->index or %NULL on dead node. | |
3892 | */ | |
3893 | static inline void *mtree_lookup_walk(struct ma_state *mas) | |
3894 | { | |
3895 | unsigned long *pivots; | |
3896 | unsigned char offset; | |
3897 | struct maple_node *node; | |
3898 | struct maple_enode *next; | |
3899 | enum maple_type type; | |
3900 | void __rcu **slots; | |
3901 | unsigned char end; | |
3902 | unsigned long max; | |
3903 | ||
3904 | next = mas->node; | |
3905 | max = ULONG_MAX; | |
3906 | do { | |
3907 | offset = 0; | |
3908 | node = mte_to_node(next); | |
3909 | type = mte_node_type(next); | |
3910 | pivots = ma_pivots(node, type); | |
3911 | end = ma_data_end(node, type, pivots, max); | |
3912 | if (unlikely(ma_dead_node(node))) | |
3913 | goto dead_node; | |
54a611b6 | 3914 | do { |
ec07967d PZ |
3915 | if (pivots[offset] >= mas->index) { |
3916 | max = pivots[offset]; | |
3917 | break; | |
3918 | } | |
3919 | } while (++offset < end); | |
54a611b6 | 3920 | |
54a611b6 LH |
3921 | slots = ma_slots(node, type); |
3922 | next = mt_slot(mas->tree, slots, offset); | |
3923 | if (unlikely(ma_dead_node(node))) | |
3924 | goto dead_node; | |
3925 | } while (!ma_is_leaf(type)); | |
3926 | ||
831978e3 | 3927 | return (void *)next; |
54a611b6 LH |
3928 | |
3929 | dead_node: | |
3930 | mas_reset(mas); | |
3931 | return NULL; | |
3932 | } | |
3933 | ||
3934 | /* | |
3935 | * mas_new_root() - Create a new root node that only contains the entry passed | |
3936 | * in. | |
3937 | * @mas: The maple state | |
3938 | * @entry: The entry to store. | |
3939 | * | |
3940 | * Only valid when the index == 0 and the last == ULONG_MAX | |
3941 | * | |
3942 | * Return 0 on error, 1 on success. | |
3943 | */ | |
3944 | static inline int mas_new_root(struct ma_state *mas, void *entry) | |
3945 | { | |
3946 | struct maple_enode *root = mas_root_locked(mas); | |
3947 | enum maple_type type = maple_leaf_64; | |
3948 | struct maple_node *node; | |
3949 | void __rcu **slots; | |
3950 | unsigned long *pivots; | |
3951 | ||
3952 | if (!entry && !mas->index && mas->last == ULONG_MAX) { | |
3953 | mas->depth = 0; | |
3954 | mas_set_height(mas); | |
3955 | rcu_assign_pointer(mas->tree->ma_root, entry); | |
3956 | mas->node = MAS_START; | |
3957 | goto done; | |
3958 | } | |
3959 | ||
3960 | mas_node_count(mas, 1); | |
3961 | if (mas_is_err(mas)) | |
3962 | return 0; | |
3963 | ||
3964 | node = mas_pop_node(mas); | |
3965 | pivots = ma_pivots(node, type); | |
3966 | slots = ma_slots(node, type); | |
3967 | node->parent = ma_parent_ptr( | |
3968 | ((unsigned long)mas->tree | MA_ROOT_PARENT)); | |
3969 | mas->node = mt_mk_node(node, type); | |
3970 | rcu_assign_pointer(slots[0], entry); | |
3971 | pivots[0] = mas->last; | |
3972 | mas->depth = 1; | |
3973 | mas_set_height(mas); | |
3974 | rcu_assign_pointer(mas->tree->ma_root, mte_mk_root(mas->node)); | |
3975 | ||
3976 | done: | |
3977 | if (xa_is_node(root)) | |
3978 | mte_destroy_walk(root, mas->tree); | |
3979 | ||
3980 | return 1; | |
3981 | } | |
3982 | /* | |
3983 | * mas_wr_spanning_store() - Create a subtree with the store operation completed | |
3984 | * and new nodes where necessary, then place the sub-tree in the actual tree. | |
3985 | * Note that mas is expected to point to the node which caused the store to | |
3986 | * span. | |
3987 | * @wr_mas: The maple write state | |
3988 | * | |
3989 | * Return: 0 on error, positive on success. | |
3990 | */ | |
3991 | static inline int mas_wr_spanning_store(struct ma_wr_state *wr_mas) | |
3992 | { | |
3993 | struct maple_subtree_state mast; | |
3994 | struct maple_big_node b_node; | |
3995 | struct ma_state *mas; | |
3996 | unsigned char height; | |
3997 | ||
3998 | /* Left and Right side of spanning store */ | |
3999 | MA_STATE(l_mas, NULL, 0, 0); | |
4000 | MA_STATE(r_mas, NULL, 0, 0); | |
4001 | ||
4002 | MA_WR_STATE(r_wr_mas, &r_mas, wr_mas->entry); | |
4003 | MA_WR_STATE(l_wr_mas, &l_mas, wr_mas->entry); | |
4004 | ||
4005 | /* | |
4006 | * A store operation that spans multiple nodes is called a spanning | |
4007 | * store and is handled early in the store call stack by the function | |
4008 | * mas_is_span_wr(). When a spanning store is identified, the maple | |
4009 | * state is duplicated. The first maple state walks the left tree path | |
4010 | * to ``index``, the duplicate walks the right tree path to ``last``. | |
4011 | * The data in the two nodes are combined into a single node, two nodes, | |
4012 | * or possibly three nodes (see the 3-way split above). A ``NULL`` | |
4013 | * written to the last entry of a node is considered a spanning store as | |
4014 | * a rebalance is required for the operation to complete and an overflow | |
4015 | * of data may happen. | |
4016 | */ | |
4017 | mas = wr_mas->mas; | |
4018 | trace_ma_op(__func__, mas); | |
4019 | ||
4020 | if (unlikely(!mas->index && mas->last == ULONG_MAX)) | |
4021 | return mas_new_root(mas, wr_mas->entry); | |
4022 | /* | |
4023 | * Node rebalancing may occur due to this store, so there may be three new | |
4024 | * entries per level plus a new root. | |
4025 | */ | |
4026 | height = mas_mt_height(mas); | |
4027 | mas_node_count(mas, 1 + height * 3); | |
4028 | if (mas_is_err(mas)) | |
4029 | return 0; | |
4030 | ||
4031 | /* | |
4032 | * Set up right side. Need to get to the next offset after the spanning | |
4033 | * store to ensure it's not NULL and to combine both the next node and | |
4034 | * the node with the start together. | |
4035 | */ | |
4036 | r_mas = *mas; | |
4037 | /* Avoid overflow, walk to next slot in the tree. */ | |
4038 | if (r_mas.last + 1) | |
4039 | r_mas.last++; | |
4040 | ||
4041 | r_mas.index = r_mas.last; | |
4042 | mas_wr_walk_index(&r_wr_mas); | |
4043 | r_mas.last = r_mas.index = mas->last; | |
4044 | ||
4045 | /* Set up left side. */ | |
4046 | l_mas = *mas; | |
4047 | mas_wr_walk_index(&l_wr_mas); | |
4048 | ||
4049 | if (!wr_mas->entry) { | |
4050 | mas_extend_spanning_null(&l_wr_mas, &r_wr_mas); | |
4051 | mas->offset = l_mas.offset; | |
4052 | mas->index = l_mas.index; | |
4053 | mas->last = l_mas.last = r_mas.last; | |
4054 | } | |
4055 | ||
4056 | /* expanding NULLs may make this cover the entire range */ | |
4057 | if (!l_mas.index && r_mas.last == ULONG_MAX) { | |
4058 | mas_set_range(mas, 0, ULONG_MAX); | |
4059 | return mas_new_root(mas, wr_mas->entry); | |
4060 | } | |
4061 | ||
4062 | memset(&b_node, 0, sizeof(struct maple_big_node)); | |
4063 | /* Copy l_mas and store the value in b_node. */ | |
4064 | mas_store_b_node(&l_wr_mas, &b_node, l_wr_mas.node_end); | |
4065 | /* Copy r_mas into b_node. */ | |
4066 | if (r_mas.offset <= r_wr_mas.node_end) | |
4067 | mas_mab_cp(&r_mas, r_mas.offset, r_wr_mas.node_end, | |
4068 | &b_node, b_node.b_end + 1); | |
4069 | else | |
4070 | b_node.b_end++; | |
4071 | ||
4072 | /* Stop spanning searches by searching for just index. */ | |
4073 | l_mas.index = l_mas.last = mas->index; | |
4074 | ||
4075 | mast.bn = &b_node; | |
4076 | mast.orig_l = &l_mas; | |
4077 | mast.orig_r = &r_mas; | |
4078 | /* Combine l_mas and r_mas and split them up evenly again. */ | |
4079 | return mas_spanning_rebalance(mas, &mast, height + 1); | |
4080 | } | |
4081 | ||
4082 | /* | |
4083 | * mas_wr_node_store() - Attempt to store the value in a node | |
4084 | * @wr_mas: The maple write state | |
4085 | * | |
4086 | * Attempts to reuse the node, but may allocate. | |
4087 | * | |
4088 | * Return: True if stored, false otherwise | |
4089 | */ | |
4090 | static inline bool mas_wr_node_store(struct ma_wr_state *wr_mas) | |
4091 | { | |
4092 | struct ma_state *mas = wr_mas->mas; | |
4093 | void __rcu **dst_slots; | |
4094 | unsigned long *dst_pivots; | |
4095 | unsigned char dst_offset; | |
4096 | unsigned char new_end = wr_mas->node_end; | |
4097 | unsigned char offset; | |
4098 | unsigned char node_slots = mt_slots[wr_mas->type]; | |
4099 | struct maple_node reuse, *newnode; | |
4100 | unsigned char copy_size, max_piv = mt_pivots[wr_mas->type]; | |
4101 | bool in_rcu = mt_in_rcu(mas->tree); | |
4102 | ||
4103 | offset = mas->offset; | |
4104 | if (mas->last == wr_mas->r_max) { | |
4105 | /* runs right to the end of the node */ | |
4106 | if (mas->last == mas->max) | |
4107 | new_end = offset; | |
4108 | /* don't copy this offset */ | |
4109 | wr_mas->offset_end++; | |
4110 | } else if (mas->last < wr_mas->r_max) { | |
4111 | /* new range ends in this range */ | |
4112 | if (unlikely(wr_mas->r_max == ULONG_MAX)) | |
4113 | mas_bulk_rebalance(mas, wr_mas->node_end, wr_mas->type); | |
4114 | ||
4115 | new_end++; | |
4116 | } else { | |
4117 | if (wr_mas->end_piv == mas->last) | |
4118 | wr_mas->offset_end++; | |
4119 | ||
4120 | new_end -= wr_mas->offset_end - offset - 1; | |
4121 | } | |
4122 | ||
4123 | /* new range starts within a range */ | |
4124 | if (wr_mas->r_min < mas->index) | |
4125 | new_end++; | |
4126 | ||
4127 | /* Not enough room */ | |
4128 | if (new_end >= node_slots) | |
4129 | return false; | |
4130 | ||
4131 | /* Not enough data. */ | |
4132 | if (!mte_is_root(mas->node) && (new_end <= mt_min_slots[wr_mas->type]) && | |
4133 | !(mas->mas_flags & MA_STATE_BULK)) | |
4134 | return false; | |
4135 | ||
4136 | /* set up node. */ | |
4137 | if (in_rcu) { | |
4138 | mas_node_count(mas, 1); | |
4139 | if (mas_is_err(mas)) | |
4140 | return false; | |
4141 | ||
4142 | newnode = mas_pop_node(mas); | |
4143 | } else { | |
4144 | memset(&reuse, 0, sizeof(struct maple_node)); | |
4145 | newnode = &reuse; | |
4146 | } | |
4147 | ||
4148 | newnode->parent = mas_mn(mas)->parent; | |
4149 | dst_pivots = ma_pivots(newnode, wr_mas->type); | |
4150 | dst_slots = ma_slots(newnode, wr_mas->type); | |
4151 | /* Copy from start to insert point */ | |
4152 | memcpy(dst_pivots, wr_mas->pivots, sizeof(unsigned long) * (offset + 1)); | |
4153 | memcpy(dst_slots, wr_mas->slots, sizeof(void *) * (offset + 1)); | |
4154 | dst_offset = offset; | |
4155 | ||
4156 | /* Handle insert of new range starting after old range */ | |
4157 | if (wr_mas->r_min < mas->index) { | |
4158 | mas->offset++; | |
4159 | rcu_assign_pointer(dst_slots[dst_offset], wr_mas->content); | |
4160 | dst_pivots[dst_offset++] = mas->index - 1; | |
4161 | } | |
4162 | ||
4163 | /* Store the new entry and range end. */ | |
4164 | if (dst_offset < max_piv) | |
4165 | dst_pivots[dst_offset] = mas->last; | |
4166 | mas->offset = dst_offset; | |
4167 | rcu_assign_pointer(dst_slots[dst_offset], wr_mas->entry); | |
4168 | ||
4169 | /* | |
4170 | * this range wrote to the end of the node or it overwrote the rest of | |
4171 | * the data | |
4172 | */ | |
4173 | if (wr_mas->offset_end > wr_mas->node_end || mas->last >= mas->max) { | |
4174 | new_end = dst_offset; | |
4175 | goto done; | |
4176 | } | |
4177 | ||
4178 | dst_offset++; | |
4179 | /* Copy to the end of node if necessary. */ | |
4180 | copy_size = wr_mas->node_end - wr_mas->offset_end + 1; | |
4181 | memcpy(dst_slots + dst_offset, wr_mas->slots + wr_mas->offset_end, | |
4182 | sizeof(void *) * copy_size); | |
4183 | if (dst_offset < max_piv) { | |
4184 | if (copy_size > max_piv - dst_offset) | |
4185 | copy_size = max_piv - dst_offset; | |
4186 | ||
4187 | memcpy(dst_pivots + dst_offset, | |
4188 | wr_mas->pivots + wr_mas->offset_end, | |
4189 | sizeof(unsigned long) * copy_size); | |
4190 | } | |
4191 | ||
4192 | if ((wr_mas->node_end == node_slots - 1) && (new_end < node_slots - 1)) | |
4193 | dst_pivots[new_end] = mas->max; | |
4194 | ||
4195 | done: | |
4196 | mas_leaf_set_meta(mas, newnode, dst_pivots, maple_leaf_64, new_end); | |
4197 | if (in_rcu) { | |
c13af03d | 4198 | mte_set_node_dead(mas->node); |
54a611b6 LH |
4199 | mas->node = mt_mk_node(newnode, wr_mas->type); |
4200 | mas_replace(mas, false); | |
4201 | } else { | |
4202 | memcpy(wr_mas->node, newnode, sizeof(struct maple_node)); | |
4203 | } | |
4204 | trace_ma_write(__func__, mas, 0, wr_mas->entry); | |
4205 | mas_update_gap(mas); | |
4206 | return true; | |
4207 | } | |
4208 | ||
4209 | /* | |
4210 | * mas_wr_slot_store: Attempt to store a value in a slot. | |
4211 | * @wr_mas: the maple write state | |
4212 | * | |
4213 | * Return: True if stored, false otherwise | |
4214 | */ | |
4215 | static inline bool mas_wr_slot_store(struct ma_wr_state *wr_mas) | |
4216 | { | |
4217 | struct ma_state *mas = wr_mas->mas; | |
4218 | unsigned long lmax; /* Logical max. */ | |
4219 | unsigned char offset = mas->offset; | |
4220 | ||
4221 | if ((wr_mas->r_max > mas->last) && ((wr_mas->r_min != mas->index) || | |
4222 | (offset != wr_mas->node_end))) | |
4223 | return false; | |
4224 | ||
4225 | if (offset == wr_mas->node_end - 1) | |
4226 | lmax = mas->max; | |
4227 | else | |
4228 | lmax = wr_mas->pivots[offset + 1]; | |
4229 | ||
4230 | /* going to overwrite too many slots. */ | |
4231 | if (lmax < mas->last) | |
4232 | return false; | |
4233 | ||
4234 | if (wr_mas->r_min == mas->index) { | |
4235 | /* overwriting two or more ranges with one. */ | |
4236 | if (lmax == mas->last) | |
4237 | return false; | |
4238 | ||
4239 | /* Overwriting all of offset and a portion of offset + 1. */ | |
4240 | rcu_assign_pointer(wr_mas->slots[offset], wr_mas->entry); | |
4241 | wr_mas->pivots[offset] = mas->last; | |
4242 | goto done; | |
4243 | } | |
4244 | ||
4245 | /* Doesn't end on the next range end. */ | |
4246 | if (lmax != mas->last) | |
4247 | return false; | |
4248 | ||
4249 | /* Overwriting a portion of offset and all of offset + 1 */ | |
4250 | if ((offset + 1 < mt_pivots[wr_mas->type]) && | |
4251 | (wr_mas->entry || wr_mas->pivots[offset + 1])) | |
4252 | wr_mas->pivots[offset + 1] = mas->last; | |
4253 | ||
4254 | rcu_assign_pointer(wr_mas->slots[offset + 1], wr_mas->entry); | |
4255 | wr_mas->pivots[offset] = mas->index - 1; | |
4256 | mas->offset++; /* Keep mas accurate. */ | |
4257 | ||
4258 | done: | |
4259 | trace_ma_write(__func__, mas, 0, wr_mas->entry); | |
4260 | mas_update_gap(mas); | |
4261 | return true; | |
4262 | } | |
4263 | ||
4264 | static inline void mas_wr_end_piv(struct ma_wr_state *wr_mas) | |
4265 | { | |
4266 | while ((wr_mas->mas->last > wr_mas->end_piv) && | |
4267 | (wr_mas->offset_end < wr_mas->node_end)) | |
4268 | wr_mas->end_piv = wr_mas->pivots[++wr_mas->offset_end]; | |
4269 | ||
4270 | if (wr_mas->mas->last > wr_mas->end_piv) | |
4271 | wr_mas->end_piv = wr_mas->mas->max; | |
4272 | } | |
4273 | ||
4274 | static inline void mas_wr_extend_null(struct ma_wr_state *wr_mas) | |
4275 | { | |
4276 | struct ma_state *mas = wr_mas->mas; | |
4277 | ||
4278 | if (mas->last < wr_mas->end_piv && !wr_mas->slots[wr_mas->offset_end]) | |
4279 | mas->last = wr_mas->end_piv; | |
4280 | ||
4281 | /* Check next slot(s) if we are overwriting the end */ | |
4282 | if ((mas->last == wr_mas->end_piv) && | |
4283 | (wr_mas->node_end != wr_mas->offset_end) && | |
4284 | !wr_mas->slots[wr_mas->offset_end + 1]) { | |
4285 | wr_mas->offset_end++; | |
4286 | if (wr_mas->offset_end == wr_mas->node_end) | |
4287 | mas->last = mas->max; | |
4288 | else | |
4289 | mas->last = wr_mas->pivots[wr_mas->offset_end]; | |
4290 | wr_mas->end_piv = mas->last; | |
4291 | } | |
4292 | ||
4293 | if (!wr_mas->content) { | |
4294 | /* If this one is null, the next and prev are not */ | |
4295 | mas->index = wr_mas->r_min; | |
4296 | } else { | |
4297 | /* Check prev slot if we are overwriting the start */ | |
4298 | if (mas->index == wr_mas->r_min && mas->offset && | |
4299 | !wr_mas->slots[mas->offset - 1]) { | |
4300 | mas->offset--; | |
4301 | wr_mas->r_min = mas->index = | |
4302 | mas_safe_min(mas, wr_mas->pivots, mas->offset); | |
4303 | wr_mas->r_max = wr_mas->pivots[mas->offset]; | |
4304 | } | |
4305 | } | |
4306 | } | |
4307 | ||
4308 | static inline bool mas_wr_append(struct ma_wr_state *wr_mas) | |
4309 | { | |
4310 | unsigned char end = wr_mas->node_end; | |
4311 | unsigned char new_end = end + 1; | |
4312 | struct ma_state *mas = wr_mas->mas; | |
4313 | unsigned char node_pivots = mt_pivots[wr_mas->type]; | |
4314 | ||
4315 | if ((mas->index != wr_mas->r_min) && (mas->last == wr_mas->r_max)) { | |
4316 | if (new_end < node_pivots) | |
4317 | wr_mas->pivots[new_end] = wr_mas->pivots[end]; | |
4318 | ||
4319 | if (new_end < node_pivots) | |
4320 | ma_set_meta(wr_mas->node, maple_leaf_64, 0, new_end); | |
4321 | ||
4322 | rcu_assign_pointer(wr_mas->slots[new_end], wr_mas->entry); | |
4323 | mas->offset = new_end; | |
4324 | wr_mas->pivots[end] = mas->index - 1; | |
4325 | ||
4326 | return true; | |
4327 | } | |
4328 | ||
4329 | if ((mas->index == wr_mas->r_min) && (mas->last < wr_mas->r_max)) { | |
4330 | if (new_end < node_pivots) | |
4331 | wr_mas->pivots[new_end] = wr_mas->pivots[end]; | |
4332 | ||
4333 | rcu_assign_pointer(wr_mas->slots[new_end], wr_mas->content); | |
4334 | if (new_end < node_pivots) | |
4335 | ma_set_meta(wr_mas->node, maple_leaf_64, 0, new_end); | |
4336 | ||
4337 | wr_mas->pivots[end] = mas->last; | |
4338 | rcu_assign_pointer(wr_mas->slots[end], wr_mas->entry); | |
4339 | return true; | |
4340 | } | |
4341 | ||
4342 | return false; | |
4343 | } | |
4344 | ||
4345 | /* | |
4346 | * mas_wr_bnode() - Slow path for a modification. | |
4347 | * @wr_mas: The write maple state | |
4348 | * | |
4349 | * This is where split, rebalance end up. | |
4350 | */ | |
4351 | static void mas_wr_bnode(struct ma_wr_state *wr_mas) | |
4352 | { | |
4353 | struct maple_big_node b_node; | |
4354 | ||
4355 | trace_ma_write(__func__, wr_mas->mas, 0, wr_mas->entry); | |
4356 | memset(&b_node, 0, sizeof(struct maple_big_node)); | |
4357 | mas_store_b_node(wr_mas, &b_node, wr_mas->offset_end); | |
4358 | mas_commit_b_node(wr_mas, &b_node, wr_mas->node_end); | |
4359 | } | |
4360 | ||
4361 | static inline void mas_wr_modify(struct ma_wr_state *wr_mas) | |
4362 | { | |
4363 | unsigned char node_slots; | |
4364 | unsigned char node_size; | |
4365 | struct ma_state *mas = wr_mas->mas; | |
4366 | ||
4367 | /* Direct replacement */ | |
4368 | if (wr_mas->r_min == mas->index && wr_mas->r_max == mas->last) { | |
4369 | rcu_assign_pointer(wr_mas->slots[mas->offset], wr_mas->entry); | |
4370 | if (!!wr_mas->entry ^ !!wr_mas->content) | |
4371 | mas_update_gap(mas); | |
4372 | return; | |
4373 | } | |
4374 | ||
4375 | /* Attempt to append */ | |
4376 | node_slots = mt_slots[wr_mas->type]; | |
4377 | node_size = wr_mas->node_end - wr_mas->offset_end + mas->offset + 2; | |
4378 | if (mas->max == ULONG_MAX) | |
4379 | node_size++; | |
4380 | ||
4381 | /* slot and node store will not fit, go to the slow path */ | |
4382 | if (unlikely(node_size >= node_slots)) | |
4383 | goto slow_path; | |
4384 | ||
4385 | if (wr_mas->entry && (wr_mas->node_end < node_slots - 1) && | |
4386 | (mas->offset == wr_mas->node_end) && mas_wr_append(wr_mas)) { | |
4387 | if (!wr_mas->content || !wr_mas->entry) | |
4388 | mas_update_gap(mas); | |
4389 | return; | |
4390 | } | |
4391 | ||
4392 | if ((wr_mas->offset_end - mas->offset <= 1) && mas_wr_slot_store(wr_mas)) | |
4393 | return; | |
4394 | else if (mas_wr_node_store(wr_mas)) | |
4395 | return; | |
4396 | ||
4397 | if (mas_is_err(mas)) | |
4398 | return; | |
4399 | ||
4400 | slow_path: | |
4401 | mas_wr_bnode(wr_mas); | |
4402 | } | |
4403 | ||
4404 | /* | |
4405 | * mas_wr_store_entry() - Internal call to store a value | |
4406 | * @mas: The maple state | |
4407 | * @entry: The entry to store. | |
4408 | * | |
4409 | * Return: The contents that was stored at the index. | |
4410 | */ | |
4411 | static inline void *mas_wr_store_entry(struct ma_wr_state *wr_mas) | |
4412 | { | |
4413 | struct ma_state *mas = wr_mas->mas; | |
4414 | ||
4415 | wr_mas->content = mas_start(mas); | |
4416 | if (mas_is_none(mas) || mas_is_ptr(mas)) { | |
4417 | mas_store_root(mas, wr_mas->entry); | |
4418 | return wr_mas->content; | |
4419 | } | |
4420 | ||
4421 | if (unlikely(!mas_wr_walk(wr_mas))) { | |
4422 | mas_wr_spanning_store(wr_mas); | |
4423 | return wr_mas->content; | |
4424 | } | |
4425 | ||
4426 | /* At this point, we are at the leaf node that needs to be altered. */ | |
4427 | wr_mas->end_piv = wr_mas->r_max; | |
4428 | mas_wr_end_piv(wr_mas); | |
4429 | ||
4430 | if (!wr_mas->entry) | |
4431 | mas_wr_extend_null(wr_mas); | |
4432 | ||
4433 | /* New root for a single pointer */ | |
4434 | if (unlikely(!mas->index && mas->last == ULONG_MAX)) { | |
4435 | mas_new_root(mas, wr_mas->entry); | |
4436 | return wr_mas->content; | |
4437 | } | |
4438 | ||
4439 | mas_wr_modify(wr_mas); | |
4440 | return wr_mas->content; | |
4441 | } | |
4442 | ||
4443 | /** | |
4444 | * mas_insert() - Internal call to insert a value | |
4445 | * @mas: The maple state | |
4446 | * @entry: The entry to store | |
4447 | * | |
4448 | * Return: %NULL or the contents that already exists at the requested index | |
4449 | * otherwise. The maple state needs to be checked for error conditions. | |
4450 | */ | |
4451 | static inline void *mas_insert(struct ma_state *mas, void *entry) | |
4452 | { | |
4453 | MA_WR_STATE(wr_mas, mas, entry); | |
4454 | ||
4455 | /* | |
4456 | * Inserting a new range inserts either 0, 1, or 2 pivots within the | |
4457 | * tree. If the insert fits exactly into an existing gap with a value | |
4458 | * of NULL, then the slot only needs to be written with the new value. | |
4459 | * If the range being inserted is adjacent to another range, then only a | |
4460 | * single pivot needs to be inserted (as well as writing the entry). If | |
4461 | * the new range is within a gap but does not touch any other ranges, | |
4462 | * then two pivots need to be inserted: the start - 1, and the end. As | |
4463 | * usual, the entry must be written. Most operations require a new node | |
4464 | * to be allocated and replace an existing node to ensure RCU safety, | |
4465 | * when in RCU mode. The exception to requiring a newly allocated node | |
4466 | * is when inserting at the end of a node (appending). When done | |
4467 | * carefully, appending can reuse the node in place. | |
4468 | */ | |
4469 | wr_mas.content = mas_start(mas); | |
4470 | if (wr_mas.content) | |
4471 | goto exists; | |
4472 | ||
4473 | if (mas_is_none(mas) || mas_is_ptr(mas)) { | |
4474 | mas_store_root(mas, entry); | |
4475 | return NULL; | |
4476 | } | |
4477 | ||
4478 | /* spanning writes always overwrite something */ | |
4479 | if (!mas_wr_walk(&wr_mas)) | |
4480 | goto exists; | |
4481 | ||
4482 | /* At this point, we are at the leaf node that needs to be altered. */ | |
4483 | wr_mas.offset_end = mas->offset; | |
4484 | wr_mas.end_piv = wr_mas.r_max; | |
4485 | ||
4486 | if (wr_mas.content || (mas->last > wr_mas.r_max)) | |
4487 | goto exists; | |
4488 | ||
4489 | if (!entry) | |
4490 | return NULL; | |
4491 | ||
4492 | mas_wr_modify(&wr_mas); | |
4493 | return wr_mas.content; | |
4494 | ||
4495 | exists: | |
4496 | mas_set_err(mas, -EEXIST); | |
4497 | return wr_mas.content; | |
4498 | ||
4499 | } | |
4500 | ||
4501 | /* | |
4502 | * mas_prev_node() - Find the prev non-null entry at the same level in the | |
4503 | * tree. The prev value will be mas->node[mas->offset] or MAS_NONE. | |
4504 | * @mas: The maple state | |
4505 | * @min: The lower limit to search | |
4506 | * | |
4507 | * The prev node value will be mas->node[mas->offset] or MAS_NONE. | |
4508 | * Return: 1 if the node is dead, 0 otherwise. | |
4509 | */ | |
4510 | static inline int mas_prev_node(struct ma_state *mas, unsigned long min) | |
4511 | { | |
4512 | enum maple_type mt; | |
4513 | int offset, level; | |
4514 | void __rcu **slots; | |
4515 | struct maple_node *node; | |
4516 | struct maple_enode *enode; | |
4517 | unsigned long *pivots; | |
4518 | ||
4519 | if (mas_is_none(mas)) | |
4520 | return 0; | |
4521 | ||
4522 | level = 0; | |
4523 | do { | |
4524 | node = mas_mn(mas); | |
4525 | if (ma_is_root(node)) | |
4526 | goto no_entry; | |
4527 | ||
4528 | /* Walk up. */ | |
4529 | if (unlikely(mas_ascend(mas))) | |
4530 | return 1; | |
4531 | offset = mas->offset; | |
4532 | level++; | |
4533 | } while (!offset); | |
4534 | ||
4535 | offset--; | |
4536 | mt = mte_node_type(mas->node); | |
4537 | node = mas_mn(mas); | |
4538 | slots = ma_slots(node, mt); | |
4539 | pivots = ma_pivots(node, mt); | |
39d0bd86 LH |
4540 | if (unlikely(ma_dead_node(node))) |
4541 | return 1; | |
4542 | ||
54a611b6 LH |
4543 | mas->max = pivots[offset]; |
4544 | if (offset) | |
4545 | mas->min = pivots[offset - 1] + 1; | |
4546 | if (unlikely(ma_dead_node(node))) | |
4547 | return 1; | |
4548 | ||
4549 | if (mas->max < min) | |
4550 | goto no_entry_min; | |
4551 | ||
4552 | while (level > 1) { | |
4553 | level--; | |
4554 | enode = mas_slot(mas, slots, offset); | |
4555 | if (unlikely(ma_dead_node(node))) | |
4556 | return 1; | |
4557 | ||
4558 | mas->node = enode; | |
4559 | mt = mte_node_type(mas->node); | |
4560 | node = mas_mn(mas); | |
4561 | slots = ma_slots(node, mt); | |
4562 | pivots = ma_pivots(node, mt); | |
4563 | offset = ma_data_end(node, mt, pivots, mas->max); | |
39d0bd86 LH |
4564 | if (unlikely(ma_dead_node(node))) |
4565 | return 1; | |
4566 | ||
54a611b6 LH |
4567 | if (offset) |
4568 | mas->min = pivots[offset - 1] + 1; | |
4569 | ||
4570 | if (offset < mt_pivots[mt]) | |
4571 | mas->max = pivots[offset]; | |
4572 | ||
4573 | if (mas->max < min) | |
4574 | goto no_entry; | |
4575 | } | |
4576 | ||
4577 | mas->node = mas_slot(mas, slots, offset); | |
4578 | if (unlikely(ma_dead_node(node))) | |
4579 | return 1; | |
4580 | ||
4581 | mas->offset = mas_data_end(mas); | |
4582 | if (unlikely(mte_dead_node(mas->node))) | |
4583 | return 1; | |
4584 | ||
4585 | return 0; | |
4586 | ||
4587 | no_entry_min: | |
4588 | mas->offset = offset; | |
4589 | if (offset) | |
4590 | mas->min = pivots[offset - 1] + 1; | |
4591 | no_entry: | |
4592 | if (unlikely(ma_dead_node(node))) | |
4593 | return 1; | |
4594 | ||
4595 | mas->node = MAS_NONE; | |
4596 | return 0; | |
4597 | } | |
4598 | ||
4599 | /* | |
4600 | * mas_next_node() - Get the next node at the same level in the tree. | |
4601 | * @mas: The maple state | |
4602 | * @max: The maximum pivot value to check. | |
4603 | * | |
4604 | * The next value will be mas->node[mas->offset] or MAS_NONE. | |
4605 | * Return: 1 on dead node, 0 otherwise. | |
4606 | */ | |
4607 | static inline int mas_next_node(struct ma_state *mas, struct maple_node *node, | |
4608 | unsigned long max) | |
4609 | { | |
4610 | unsigned long min, pivot; | |
4611 | unsigned long *pivots; | |
4612 | struct maple_enode *enode; | |
4613 | int level = 0; | |
4614 | unsigned char offset; | |
39d0bd86 | 4615 | unsigned char node_end; |
54a611b6 LH |
4616 | enum maple_type mt; |
4617 | void __rcu **slots; | |
4618 | ||
4619 | if (mas->max >= max) | |
4620 | goto no_entry; | |
4621 | ||
4622 | level = 0; | |
4623 | do { | |
4624 | if (ma_is_root(node)) | |
4625 | goto no_entry; | |
4626 | ||
4627 | min = mas->max + 1; | |
4628 | if (min > max) | |
4629 | goto no_entry; | |
4630 | ||
4631 | if (unlikely(mas_ascend(mas))) | |
4632 | return 1; | |
4633 | ||
4634 | offset = mas->offset; | |
4635 | level++; | |
4636 | node = mas_mn(mas); | |
4637 | mt = mte_node_type(mas->node); | |
4638 | pivots = ma_pivots(node, mt); | |
39d0bd86 LH |
4639 | node_end = ma_data_end(node, mt, pivots, mas->max); |
4640 | if (unlikely(ma_dead_node(node))) | |
4641 | return 1; | |
4642 | ||
4643 | } while (unlikely(offset == node_end)); | |
54a611b6 LH |
4644 | |
4645 | slots = ma_slots(node, mt); | |
4646 | pivot = mas_safe_pivot(mas, pivots, ++offset, mt); | |
4647 | while (unlikely(level > 1)) { | |
4648 | /* Descend, if necessary */ | |
4649 | enode = mas_slot(mas, slots, offset); | |
4650 | if (unlikely(ma_dead_node(node))) | |
4651 | return 1; | |
4652 | ||
4653 | mas->node = enode; | |
4654 | level--; | |
4655 | node = mas_mn(mas); | |
4656 | mt = mte_node_type(mas->node); | |
4657 | slots = ma_slots(node, mt); | |
4658 | pivots = ma_pivots(node, mt); | |
39d0bd86 LH |
4659 | if (unlikely(ma_dead_node(node))) |
4660 | return 1; | |
4661 | ||
54a611b6 LH |
4662 | offset = 0; |
4663 | pivot = pivots[0]; | |
4664 | } | |
4665 | ||
4666 | enode = mas_slot(mas, slots, offset); | |
4667 | if (unlikely(ma_dead_node(node))) | |
4668 | return 1; | |
4669 | ||
4670 | mas->node = enode; | |
4671 | mas->min = min; | |
4672 | mas->max = pivot; | |
4673 | return 0; | |
4674 | ||
4675 | no_entry: | |
4676 | if (unlikely(ma_dead_node(node))) | |
4677 | return 1; | |
4678 | ||
4679 | mas->node = MAS_NONE; | |
4680 | return 0; | |
4681 | } | |
4682 | ||
4683 | /* | |
4684 | * mas_next_nentry() - Get the next node entry | |
4685 | * @mas: The maple state | |
4686 | * @max: The maximum value to check | |
4687 | * @*range_start: Pointer to store the start of the range. | |
4688 | * | |
4689 | * Sets @mas->offset to the offset of the next node entry, @mas->last to the | |
4690 | * pivot of the entry. | |
4691 | * | |
4692 | * Return: The next entry, %NULL otherwise | |
4693 | */ | |
4694 | static inline void *mas_next_nentry(struct ma_state *mas, | |
4695 | struct maple_node *node, unsigned long max, enum maple_type type) | |
4696 | { | |
4697 | unsigned char count; | |
4698 | unsigned long pivot; | |
4699 | unsigned long *pivots; | |
4700 | void __rcu **slots; | |
4701 | void *entry; | |
4702 | ||
4703 | if (mas->last == mas->max) { | |
4704 | mas->index = mas->max; | |
4705 | return NULL; | |
4706 | } | |
4707 | ||
54a611b6 | 4708 | slots = ma_slots(node, type); |
39d0bd86 | 4709 | pivots = ma_pivots(node, type); |
65be6f05 | 4710 | count = ma_data_end(node, type, pivots, mas->max); |
39d0bd86 LH |
4711 | if (unlikely(ma_dead_node(node))) |
4712 | return NULL; | |
4713 | ||
4714 | mas->index = mas_safe_min(mas, pivots, mas->offset); | |
4715 | if (unlikely(ma_dead_node(node))) | |
54a611b6 LH |
4716 | return NULL; |
4717 | ||
4718 | if (mas->index > max) | |
4719 | return NULL; | |
4720 | ||
54a611b6 LH |
4721 | if (mas->offset > count) |
4722 | return NULL; | |
4723 | ||
4724 | while (mas->offset < count) { | |
4725 | pivot = pivots[mas->offset]; | |
4726 | entry = mas_slot(mas, slots, mas->offset); | |
4727 | if (ma_dead_node(node)) | |
4728 | return NULL; | |
4729 | ||
4730 | if (entry) | |
4731 | goto found; | |
4732 | ||
4733 | if (pivot >= max) | |
4734 | return NULL; | |
4735 | ||
4736 | mas->index = pivot + 1; | |
4737 | mas->offset++; | |
4738 | } | |
4739 | ||
4740 | if (mas->index > mas->max) { | |
4741 | mas->index = mas->last; | |
4742 | return NULL; | |
4743 | } | |
4744 | ||
4745 | pivot = mas_safe_pivot(mas, pivots, mas->offset, type); | |
4746 | entry = mas_slot(mas, slots, mas->offset); | |
4747 | if (ma_dead_node(node)) | |
4748 | return NULL; | |
4749 | ||
4750 | if (!pivot) | |
4751 | return NULL; | |
4752 | ||
4753 | if (!entry) | |
4754 | return NULL; | |
4755 | ||
4756 | found: | |
4757 | mas->last = pivot; | |
4758 | return entry; | |
4759 | } | |
4760 | ||
4761 | static inline void mas_rewalk(struct ma_state *mas, unsigned long index) | |
4762 | { | |
54a611b6 LH |
4763 | retry: |
4764 | mas_set(mas, index); | |
4765 | mas_state_walk(mas); | |
4766 | if (mas_is_start(mas)) | |
4767 | goto retry; | |
54a611b6 LH |
4768 | } |
4769 | ||
4770 | /* | |
4771 | * mas_next_entry() - Internal function to get the next entry. | |
4772 | * @mas: The maple state | |
4773 | * @limit: The maximum range start. | |
4774 | * | |
4775 | * Set the @mas->node to the next entry and the range_start to | |
4776 | * the beginning value for the entry. Does not check beyond @limit. | |
4777 | * Sets @mas->index and @mas->last to the limit if it is hit. | |
4778 | * Restarts on dead nodes. | |
4779 | * | |
4780 | * Return: the next entry or %NULL. | |
4781 | */ | |
4782 | static inline void *mas_next_entry(struct ma_state *mas, unsigned long limit) | |
4783 | { | |
4784 | void *entry = NULL; | |
4785 | struct maple_enode *prev_node; | |
4786 | struct maple_node *node; | |
4787 | unsigned char offset; | |
4788 | unsigned long last; | |
4789 | enum maple_type mt; | |
4790 | ||
50e81c82 LH |
4791 | if (mas->index > limit) { |
4792 | mas->index = mas->last = limit; | |
4793 | mas_pause(mas); | |
4794 | return NULL; | |
4795 | } | |
54a611b6 LH |
4796 | last = mas->last; |
4797 | retry: | |
4798 | offset = mas->offset; | |
4799 | prev_node = mas->node; | |
4800 | node = mas_mn(mas); | |
4801 | mt = mte_node_type(mas->node); | |
4802 | mas->offset++; | |
4803 | if (unlikely(mas->offset >= mt_slots[mt])) { | |
4804 | mas->offset = mt_slots[mt] - 1; | |
4805 | goto next_node; | |
4806 | } | |
4807 | ||
4808 | while (!mas_is_none(mas)) { | |
4809 | entry = mas_next_nentry(mas, node, limit, mt); | |
4810 | if (unlikely(ma_dead_node(node))) { | |
4811 | mas_rewalk(mas, last); | |
4812 | goto retry; | |
4813 | } | |
4814 | ||
4815 | if (likely(entry)) | |
4816 | return entry; | |
4817 | ||
4818 | if (unlikely((mas->index > limit))) | |
4819 | break; | |
4820 | ||
4821 | next_node: | |
4822 | prev_node = mas->node; | |
4823 | offset = mas->offset; | |
4824 | if (unlikely(mas_next_node(mas, node, limit))) { | |
4825 | mas_rewalk(mas, last); | |
4826 | goto retry; | |
4827 | } | |
4828 | mas->offset = 0; | |
4829 | node = mas_mn(mas); | |
4830 | mt = mte_node_type(mas->node); | |
4831 | } | |
4832 | ||
4833 | mas->index = mas->last = limit; | |
4834 | mas->offset = offset; | |
4835 | mas->node = prev_node; | |
4836 | return NULL; | |
4837 | } | |
4838 | ||
4839 | /* | |
4840 | * mas_prev_nentry() - Get the previous node entry. | |
4841 | * @mas: The maple state. | |
4842 | * @limit: The lower limit to check for a value. | |
4843 | * | |
4844 | * Return: the entry, %NULL otherwise. | |
4845 | */ | |
4846 | static inline void *mas_prev_nentry(struct ma_state *mas, unsigned long limit, | |
4847 | unsigned long index) | |
4848 | { | |
4849 | unsigned long pivot, min; | |
4850 | unsigned char offset; | |
4851 | struct maple_node *mn; | |
4852 | enum maple_type mt; | |
4853 | unsigned long *pivots; | |
4854 | void __rcu **slots; | |
4855 | void *entry; | |
4856 | ||
4857 | retry: | |
4858 | if (!mas->offset) | |
4859 | return NULL; | |
4860 | ||
4861 | mn = mas_mn(mas); | |
4862 | mt = mte_node_type(mas->node); | |
4863 | offset = mas->offset - 1; | |
4864 | if (offset >= mt_slots[mt]) | |
4865 | offset = mt_slots[mt] - 1; | |
4866 | ||
4867 | slots = ma_slots(mn, mt); | |
4868 | pivots = ma_pivots(mn, mt); | |
39d0bd86 LH |
4869 | if (unlikely(ma_dead_node(mn))) { |
4870 | mas_rewalk(mas, index); | |
4871 | goto retry; | |
4872 | } | |
4873 | ||
54a611b6 LH |
4874 | if (offset == mt_pivots[mt]) |
4875 | pivot = mas->max; | |
4876 | else | |
4877 | pivot = pivots[offset]; | |
4878 | ||
4879 | if (unlikely(ma_dead_node(mn))) { | |
4880 | mas_rewalk(mas, index); | |
4881 | goto retry; | |
4882 | } | |
4883 | ||
4884 | while (offset && ((!mas_slot(mas, slots, offset) && pivot >= limit) || | |
4885 | !pivot)) | |
4886 | pivot = pivots[--offset]; | |
4887 | ||
4888 | min = mas_safe_min(mas, pivots, offset); | |
4889 | entry = mas_slot(mas, slots, offset); | |
4890 | if (unlikely(ma_dead_node(mn))) { | |
4891 | mas_rewalk(mas, index); | |
4892 | goto retry; | |
4893 | } | |
4894 | ||
4895 | if (likely(entry)) { | |
4896 | mas->offset = offset; | |
4897 | mas->last = pivot; | |
4898 | mas->index = min; | |
4899 | } | |
4900 | return entry; | |
4901 | } | |
4902 | ||
4903 | static inline void *mas_prev_entry(struct ma_state *mas, unsigned long min) | |
4904 | { | |
4905 | void *entry; | |
4906 | ||
50e81c82 LH |
4907 | if (mas->index < min) { |
4908 | mas->index = mas->last = min; | |
17dc622c | 4909 | mas->node = MAS_NONE; |
50e81c82 LH |
4910 | return NULL; |
4911 | } | |
54a611b6 LH |
4912 | retry: |
4913 | while (likely(!mas_is_none(mas))) { | |
4914 | entry = mas_prev_nentry(mas, min, mas->index); | |
4915 | if (unlikely(mas->last < min)) | |
4916 | goto not_found; | |
4917 | ||
4918 | if (likely(entry)) | |
4919 | return entry; | |
4920 | ||
4921 | if (unlikely(mas_prev_node(mas, min))) { | |
4922 | mas_rewalk(mas, mas->index); | |
4923 | goto retry; | |
4924 | } | |
4925 | ||
4926 | mas->offset++; | |
4927 | } | |
4928 | ||
4929 | mas->offset--; | |
4930 | not_found: | |
4931 | mas->index = mas->last = min; | |
4932 | return NULL; | |
4933 | } | |
4934 | ||
4935 | /* | |
4936 | * mas_rev_awalk() - Internal function. Reverse allocation walk. Find the | |
4937 | * highest gap address of a given size in a given node and descend. | |
4938 | * @mas: The maple state | |
4939 | * @size: The needed size. | |
4940 | * | |
4941 | * Return: True if found in a leaf, false otherwise. | |
4942 | * | |
4943 | */ | |
fad8e429 LH |
4944 | static bool mas_rev_awalk(struct ma_state *mas, unsigned long size, |
4945 | unsigned long *gap_min, unsigned long *gap_max) | |
54a611b6 LH |
4946 | { |
4947 | enum maple_type type = mte_node_type(mas->node); | |
4948 | struct maple_node *node = mas_mn(mas); | |
4949 | unsigned long *pivots, *gaps; | |
4950 | void __rcu **slots; | |
4951 | unsigned long gap = 0; | |
7327e811 | 4952 | unsigned long max, min; |
54a611b6 LH |
4953 | unsigned char offset; |
4954 | ||
4955 | if (unlikely(mas_is_err(mas))) | |
4956 | return true; | |
4957 | ||
4958 | if (ma_is_dense(type)) { | |
4959 | /* dense nodes. */ | |
4960 | mas->offset = (unsigned char)(mas->index - mas->min); | |
4961 | return true; | |
4962 | } | |
4963 | ||
4964 | pivots = ma_pivots(node, type); | |
4965 | slots = ma_slots(node, type); | |
4966 | gaps = ma_gaps(node, type); | |
4967 | offset = mas->offset; | |
4968 | min = mas_safe_min(mas, pivots, offset); | |
4969 | /* Skip out of bounds. */ | |
4970 | while (mas->last < min) | |
4971 | min = mas_safe_min(mas, pivots, --offset); | |
4972 | ||
4973 | max = mas_safe_pivot(mas, pivots, offset, type); | |
7327e811 | 4974 | while (mas->index <= max) { |
54a611b6 LH |
4975 | gap = 0; |
4976 | if (gaps) | |
4977 | gap = gaps[offset]; | |
4978 | else if (!mas_slot(mas, slots, offset)) | |
4979 | gap = max - min + 1; | |
4980 | ||
4981 | if (gap) { | |
4982 | if ((size <= gap) && (size <= mas->last - min + 1)) | |
4983 | break; | |
4984 | ||
4985 | if (!gaps) { | |
4986 | /* Skip the next slot, it cannot be a gap. */ | |
4987 | if (offset < 2) | |
4988 | goto ascend; | |
4989 | ||
4990 | offset -= 2; | |
4991 | max = pivots[offset]; | |
4992 | min = mas_safe_min(mas, pivots, offset); | |
4993 | continue; | |
4994 | } | |
4995 | } | |
4996 | ||
4997 | if (!offset) | |
4998 | goto ascend; | |
4999 | ||
5000 | offset--; | |
5001 | max = min - 1; | |
5002 | min = mas_safe_min(mas, pivots, offset); | |
5003 | } | |
5004 | ||
7327e811 LH |
5005 | if (unlikely((mas->index > max) || (size - 1 > max - mas->index))) |
5006 | goto no_space; | |
54a611b6 LH |
5007 | |
5008 | if (unlikely(ma_is_leaf(type))) { | |
5009 | mas->offset = offset; | |
fad8e429 LH |
5010 | *gap_min = min; |
5011 | *gap_max = min + gap - 1; | |
54a611b6 LH |
5012 | return true; |
5013 | } | |
5014 | ||
5015 | /* descend, only happens under lock. */ | |
5016 | mas->node = mas_slot(mas, slots, offset); | |
5017 | mas->min = min; | |
5018 | mas->max = max; | |
5019 | mas->offset = mas_data_end(mas); | |
5020 | return false; | |
5021 | ||
5022 | ascend: | |
7327e811 LH |
5023 | if (!mte_is_root(mas->node)) |
5024 | return false; | |
54a611b6 | 5025 | |
7327e811 LH |
5026 | no_space: |
5027 | mas_set_err(mas, -EBUSY); | |
54a611b6 LH |
5028 | return false; |
5029 | } | |
5030 | ||
5031 | static inline bool mas_anode_descend(struct ma_state *mas, unsigned long size) | |
5032 | { | |
5033 | enum maple_type type = mte_node_type(mas->node); | |
5034 | unsigned long pivot, min, gap = 0; | |
06e8fd99 LH |
5035 | unsigned char offset, data_end; |
5036 | unsigned long *gaps, *pivots; | |
5037 | void __rcu **slots; | |
5038 | struct maple_node *node; | |
54a611b6 LH |
5039 | bool found = false; |
5040 | ||
5041 | if (ma_is_dense(type)) { | |
5042 | mas->offset = (unsigned char)(mas->index - mas->min); | |
5043 | return true; | |
5044 | } | |
5045 | ||
06e8fd99 LH |
5046 | node = mas_mn(mas); |
5047 | pivots = ma_pivots(node, type); | |
5048 | slots = ma_slots(node, type); | |
5049 | gaps = ma_gaps(node, type); | |
54a611b6 | 5050 | offset = mas->offset; |
54a611b6 | 5051 | min = mas_safe_min(mas, pivots, offset); |
06e8fd99 LH |
5052 | data_end = ma_data_end(node, type, pivots, mas->max); |
5053 | for (; offset <= data_end; offset++) { | |
5054 | pivot = mas_logical_pivot(mas, pivots, offset, type); | |
54a611b6 LH |
5055 | |
5056 | /* Not within lower bounds */ | |
5057 | if (mas->index > pivot) | |
5058 | goto next_slot; | |
5059 | ||
5060 | if (gaps) | |
5061 | gap = gaps[offset]; | |
5062 | else if (!mas_slot(mas, slots, offset)) | |
5063 | gap = min(pivot, mas->last) - max(mas->index, min) + 1; | |
5064 | else | |
5065 | goto next_slot; | |
5066 | ||
5067 | if (gap >= size) { | |
5068 | if (ma_is_leaf(type)) { | |
5069 | found = true; | |
5070 | goto done; | |
5071 | } | |
5072 | if (mas->index <= pivot) { | |
5073 | mas->node = mas_slot(mas, slots, offset); | |
5074 | mas->min = min; | |
5075 | mas->max = pivot; | |
5076 | offset = 0; | |
54a611b6 LH |
5077 | break; |
5078 | } | |
5079 | } | |
5080 | next_slot: | |
5081 | min = pivot + 1; | |
5082 | if (mas->last <= pivot) { | |
5083 | mas_set_err(mas, -EBUSY); | |
5084 | return true; | |
5085 | } | |
5086 | } | |
5087 | ||
5088 | if (mte_is_root(mas->node)) | |
5089 | found = true; | |
5090 | done: | |
5091 | mas->offset = offset; | |
5092 | return found; | |
5093 | } | |
5094 | ||
5095 | /** | |
5096 | * mas_walk() - Search for @mas->index in the tree. | |
5097 | * @mas: The maple state. | |
5098 | * | |
5099 | * mas->index and mas->last will be set to the range if there is a value. If | |
5100 | * mas->node is MAS_NONE, reset to MAS_START. | |
5101 | * | |
5102 | * Return: the entry at the location or %NULL. | |
5103 | */ | |
5104 | void *mas_walk(struct ma_state *mas) | |
5105 | { | |
5106 | void *entry; | |
5107 | ||
5108 | retry: | |
5109 | entry = mas_state_walk(mas); | |
5110 | if (mas_is_start(mas)) | |
5111 | goto retry; | |
5112 | ||
5113 | if (mas_is_ptr(mas)) { | |
5114 | if (!mas->index) { | |
5115 | mas->last = 0; | |
5116 | } else { | |
5117 | mas->index = 1; | |
5118 | mas->last = ULONG_MAX; | |
5119 | } | |
5120 | return entry; | |
5121 | } | |
5122 | ||
5123 | if (mas_is_none(mas)) { | |
5124 | mas->index = 0; | |
5125 | mas->last = ULONG_MAX; | |
5126 | } | |
5127 | ||
5128 | return entry; | |
5129 | } | |
120b1162 | 5130 | EXPORT_SYMBOL_GPL(mas_walk); |
54a611b6 LH |
5131 | |
5132 | static inline bool mas_rewind_node(struct ma_state *mas) | |
5133 | { | |
5134 | unsigned char slot; | |
5135 | ||
5136 | do { | |
5137 | if (mte_is_root(mas->node)) { | |
5138 | slot = mas->offset; | |
5139 | if (!slot) | |
5140 | return false; | |
5141 | } else { | |
5142 | mas_ascend(mas); | |
5143 | slot = mas->offset; | |
5144 | } | |
5145 | } while (!slot); | |
5146 | ||
5147 | mas->offset = --slot; | |
5148 | return true; | |
5149 | } | |
5150 | ||
5151 | /* | |
5152 | * mas_skip_node() - Internal function. Skip over a node. | |
5153 | * @mas: The maple state. | |
5154 | * | |
5155 | * Return: true if there is another node, false otherwise. | |
5156 | */ | |
5157 | static inline bool mas_skip_node(struct ma_state *mas) | |
5158 | { | |
0fa99fdf LH |
5159 | if (mas_is_err(mas)) |
5160 | return false; | |
54a611b6 | 5161 | |
54a611b6 LH |
5162 | do { |
5163 | if (mte_is_root(mas->node)) { | |
0fa99fdf | 5164 | if (mas->offset >= mas_data_end(mas)) { |
54a611b6 LH |
5165 | mas_set_err(mas, -EBUSY); |
5166 | return false; | |
5167 | } | |
5168 | } else { | |
5169 | mas_ascend(mas); | |
54a611b6 | 5170 | } |
0fa99fdf | 5171 | } while (mas->offset >= mas_data_end(mas)); |
54a611b6 | 5172 | |
0fa99fdf | 5173 | mas->offset++; |
54a611b6 LH |
5174 | return true; |
5175 | } | |
5176 | ||
5177 | /* | |
5178 | * mas_awalk() - Allocation walk. Search from low address to high, for a gap of | |
5179 | * @size | |
5180 | * @mas: The maple state | |
5181 | * @size: The size of the gap required | |
5182 | * | |
5183 | * Search between @mas->index and @mas->last for a gap of @size. | |
5184 | */ | |
5185 | static inline void mas_awalk(struct ma_state *mas, unsigned long size) | |
5186 | { | |
5187 | struct maple_enode *last = NULL; | |
5188 | ||
5189 | /* | |
5190 | * There are 4 options: | |
5191 | * go to child (descend) | |
5192 | * go back to parent (ascend) | |
5193 | * no gap found. (return, slot == MAPLE_NODE_SLOTS) | |
5194 | * found the gap. (return, slot != MAPLE_NODE_SLOTS) | |
5195 | */ | |
5196 | while (!mas_is_err(mas) && !mas_anode_descend(mas, size)) { | |
5197 | if (last == mas->node) | |
5198 | mas_skip_node(mas); | |
5199 | else | |
5200 | last = mas->node; | |
5201 | } | |
5202 | } | |
5203 | ||
5204 | /* | |
5205 | * mas_fill_gap() - Fill a located gap with @entry. | |
5206 | * @mas: The maple state | |
5207 | * @entry: The value to store | |
5208 | * @slot: The offset into the node to store the @entry | |
5209 | * @size: The size of the entry | |
5210 | * @index: The start location | |
5211 | */ | |
5212 | static inline void mas_fill_gap(struct ma_state *mas, void *entry, | |
5213 | unsigned char slot, unsigned long size, unsigned long *index) | |
5214 | { | |
5215 | MA_WR_STATE(wr_mas, mas, entry); | |
5216 | unsigned char pslot = mte_parent_slot(mas->node); | |
5217 | struct maple_enode *mn = mas->node; | |
5218 | unsigned long *pivots; | |
5219 | enum maple_type ptype; | |
5220 | /* | |
5221 | * mas->index is the start address for the search | |
5222 | * which may no longer be needed. | |
5223 | * mas->last is the end address for the search | |
5224 | */ | |
5225 | ||
5226 | *index = mas->index; | |
5227 | mas->last = mas->index + size - 1; | |
5228 | ||
5229 | /* | |
5230 | * It is possible that using mas->max and mas->min to correctly | |
5231 | * calculate the index and last will cause an issue in the gap | |
5232 | * calculation, so fix the ma_state here | |
5233 | */ | |
5234 | mas_ascend(mas); | |
5235 | ptype = mte_node_type(mas->node); | |
5236 | pivots = ma_pivots(mas_mn(mas), ptype); | |
5237 | mas->max = mas_safe_pivot(mas, pivots, pslot, ptype); | |
5238 | mas->min = mas_safe_min(mas, pivots, pslot); | |
5239 | mas->node = mn; | |
5240 | mas->offset = slot; | |
5241 | mas_wr_store_entry(&wr_mas); | |
5242 | } | |
5243 | ||
5244 | /* | |
5245 | * mas_sparse_area() - Internal function. Return upper or lower limit when | |
5246 | * searching for a gap in an empty tree. | |
5247 | * @mas: The maple state | |
5248 | * @min: the minimum range | |
5249 | * @max: The maximum range | |
5250 | * @size: The size of the gap | |
5251 | * @fwd: Searching forward or back | |
5252 | */ | |
29ad6bb3 | 5253 | static inline int mas_sparse_area(struct ma_state *mas, unsigned long min, |
54a611b6 LH |
5254 | unsigned long max, unsigned long size, bool fwd) |
5255 | { | |
29ad6bb3 PZ |
5256 | if (!unlikely(mas_is_none(mas)) && min == 0) { |
5257 | min++; | |
5258 | /* | |
5259 | * At this time, min is increased, we need to recheck whether | |
5260 | * the size is satisfied. | |
5261 | */ | |
5262 | if (min > max || max - min + 1 < size) | |
5263 | return -EBUSY; | |
5264 | } | |
54a611b6 LH |
5265 | /* mas_is_ptr */ |
5266 | ||
54a611b6 | 5267 | if (fwd) { |
29ad6bb3 PZ |
5268 | mas->index = min; |
5269 | mas->last = min + size - 1; | |
5270 | } else { | |
5271 | mas->last = max; | |
5272 | mas->index = max - size + 1; | |
54a611b6 | 5273 | } |
29ad6bb3 | 5274 | return 0; |
54a611b6 LH |
5275 | } |
5276 | ||
5277 | /* | |
5278 | * mas_empty_area() - Get the lowest address within the range that is | |
5279 | * sufficient for the size requested. | |
5280 | * @mas: The maple state | |
5281 | * @min: The lowest value of the range | |
5282 | * @max: The highest value of the range | |
5283 | * @size: The size needed | |
5284 | */ | |
5285 | int mas_empty_area(struct ma_state *mas, unsigned long min, | |
5286 | unsigned long max, unsigned long size) | |
5287 | { | |
5288 | unsigned char offset; | |
5289 | unsigned long *pivots; | |
5290 | enum maple_type mt; | |
5291 | ||
fad8e429 LH |
5292 | if (min >= max) |
5293 | return -EINVAL; | |
5294 | ||
54a611b6 LH |
5295 | if (mas_is_start(mas)) |
5296 | mas_start(mas); | |
5297 | else if (mas->offset >= 2) | |
5298 | mas->offset -= 2; | |
5299 | else if (!mas_skip_node(mas)) | |
5300 | return -EBUSY; | |
5301 | ||
5302 | /* Empty set */ | |
29ad6bb3 PZ |
5303 | if (mas_is_none(mas) || mas_is_ptr(mas)) |
5304 | return mas_sparse_area(mas, min, max, size, true); | |
54a611b6 LH |
5305 | |
5306 | /* The start of the window can only be within these values */ | |
5307 | mas->index = min; | |
5308 | mas->last = max; | |
5309 | mas_awalk(mas, size); | |
5310 | ||
5311 | if (unlikely(mas_is_err(mas))) | |
5312 | return xa_err(mas->node); | |
5313 | ||
5314 | offset = mas->offset; | |
5315 | if (unlikely(offset == MAPLE_NODE_SLOTS)) | |
5316 | return -EBUSY; | |
5317 | ||
5318 | mt = mte_node_type(mas->node); | |
5319 | pivots = ma_pivots(mas_mn(mas), mt); | |
5320 | if (offset) | |
5321 | mas->min = pivots[offset - 1] + 1; | |
5322 | ||
5323 | if (offset < mt_pivots[mt]) | |
5324 | mas->max = pivots[offset]; | |
5325 | ||
5326 | if (mas->index < mas->min) | |
5327 | mas->index = mas->min; | |
5328 | ||
5329 | mas->last = mas->index + size - 1; | |
5330 | return 0; | |
5331 | } | |
120b1162 | 5332 | EXPORT_SYMBOL_GPL(mas_empty_area); |
54a611b6 LH |
5333 | |
5334 | /* | |
5335 | * mas_empty_area_rev() - Get the highest address within the range that is | |
5336 | * sufficient for the size requested. | |
5337 | * @mas: The maple state | |
5338 | * @min: The lowest value of the range | |
5339 | * @max: The highest value of the range | |
5340 | * @size: The size needed | |
5341 | */ | |
5342 | int mas_empty_area_rev(struct ma_state *mas, unsigned long min, | |
5343 | unsigned long max, unsigned long size) | |
5344 | { | |
5345 | struct maple_enode *last = mas->node; | |
5346 | ||
fad8e429 LH |
5347 | if (min >= max) |
5348 | return -EINVAL; | |
5349 | ||
54a611b6 LH |
5350 | if (mas_is_start(mas)) { |
5351 | mas_start(mas); | |
5352 | mas->offset = mas_data_end(mas); | |
5353 | } else if (mas->offset >= 2) { | |
5354 | mas->offset -= 2; | |
5355 | } else if (!mas_rewind_node(mas)) { | |
5356 | return -EBUSY; | |
5357 | } | |
5358 | ||
5359 | /* Empty set. */ | |
29ad6bb3 PZ |
5360 | if (mas_is_none(mas) || mas_is_ptr(mas)) |
5361 | return mas_sparse_area(mas, min, max, size, false); | |
54a611b6 LH |
5362 | |
5363 | /* The start of the window can only be within these values. */ | |
5364 | mas->index = min; | |
5365 | mas->last = max; | |
5366 | ||
fad8e429 | 5367 | while (!mas_rev_awalk(mas, size, &min, &max)) { |
54a611b6 LH |
5368 | if (last == mas->node) { |
5369 | if (!mas_rewind_node(mas)) | |
5370 | return -EBUSY; | |
5371 | } else { | |
5372 | last = mas->node; | |
5373 | } | |
5374 | } | |
5375 | ||
5376 | if (mas_is_err(mas)) | |
5377 | return xa_err(mas->node); | |
5378 | ||
5379 | if (unlikely(mas->offset == MAPLE_NODE_SLOTS)) | |
5380 | return -EBUSY; | |
5381 | ||
54a611b6 | 5382 | /* Trim the upper limit to the max. */ |
fad8e429 LH |
5383 | if (max <= mas->last) |
5384 | mas->last = max; | |
54a611b6 LH |
5385 | |
5386 | mas->index = mas->last - size + 1; | |
5387 | return 0; | |
5388 | } | |
120b1162 | 5389 | EXPORT_SYMBOL_GPL(mas_empty_area_rev); |
54a611b6 LH |
5390 | |
5391 | static inline int mas_alloc(struct ma_state *mas, void *entry, | |
5392 | unsigned long size, unsigned long *index) | |
5393 | { | |
5394 | unsigned long min; | |
5395 | ||
5396 | mas_start(mas); | |
5397 | if (mas_is_none(mas) || mas_is_ptr(mas)) { | |
5398 | mas_root_expand(mas, entry); | |
5399 | if (mas_is_err(mas)) | |
5400 | return xa_err(mas->node); | |
5401 | ||
5402 | if (!mas->index) | |
5403 | return mte_pivot(mas->node, 0); | |
5404 | return mte_pivot(mas->node, 1); | |
5405 | } | |
5406 | ||
5407 | /* Must be walking a tree. */ | |
5408 | mas_awalk(mas, size); | |
5409 | if (mas_is_err(mas)) | |
5410 | return xa_err(mas->node); | |
5411 | ||
5412 | if (mas->offset == MAPLE_NODE_SLOTS) | |
5413 | goto no_gap; | |
5414 | ||
5415 | /* | |
5416 | * At this point, mas->node points to the right node and we have an | |
5417 | * offset that has a sufficient gap. | |
5418 | */ | |
5419 | min = mas->min; | |
5420 | if (mas->offset) | |
5421 | min = mte_pivot(mas->node, mas->offset - 1) + 1; | |
5422 | ||
5423 | if (mas->index < min) | |
5424 | mas->index = min; | |
5425 | ||
5426 | mas_fill_gap(mas, entry, mas->offset, size, index); | |
5427 | return 0; | |
5428 | ||
5429 | no_gap: | |
5430 | return -EBUSY; | |
5431 | } | |
5432 | ||
5433 | static inline int mas_rev_alloc(struct ma_state *mas, unsigned long min, | |
5434 | unsigned long max, void *entry, | |
5435 | unsigned long size, unsigned long *index) | |
5436 | { | |
5437 | int ret = 0; | |
5438 | ||
5439 | ret = mas_empty_area_rev(mas, min, max, size); | |
5440 | if (ret) | |
5441 | return ret; | |
5442 | ||
5443 | if (mas_is_err(mas)) | |
5444 | return xa_err(mas->node); | |
5445 | ||
5446 | if (mas->offset == MAPLE_NODE_SLOTS) | |
5447 | goto no_gap; | |
5448 | ||
5449 | mas_fill_gap(mas, entry, mas->offset, size, index); | |
5450 | return 0; | |
5451 | ||
5452 | no_gap: | |
5453 | return -EBUSY; | |
5454 | } | |
5455 | ||
5456 | /* | |
790e1fa8 | 5457 | * mte_dead_leaves() - Mark all leaves of a node as dead. |
54a611b6 LH |
5458 | * @mas: The maple state |
5459 | * @slots: Pointer to the slot array | |
2e5b4921 | 5460 | * @type: The maple node type |
54a611b6 LH |
5461 | * |
5462 | * Must hold the write lock. | |
5463 | * | |
5464 | * Return: The number of leaves marked as dead. | |
5465 | */ | |
5466 | static inline | |
790e1fa8 LH |
5467 | unsigned char mte_dead_leaves(struct maple_enode *enode, struct maple_tree *mt, |
5468 | void __rcu **slots) | |
54a611b6 LH |
5469 | { |
5470 | struct maple_node *node; | |
5471 | enum maple_type type; | |
5472 | void *entry; | |
5473 | int offset; | |
5474 | ||
790e1fa8 LH |
5475 | for (offset = 0; offset < mt_slot_count(enode); offset++) { |
5476 | entry = mt_slot(mt, slots, offset); | |
54a611b6 LH |
5477 | type = mte_node_type(entry); |
5478 | node = mte_to_node(entry); | |
5479 | /* Use both node and type to catch LE & BE metadata */ | |
5480 | if (!node || !type) | |
5481 | break; | |
5482 | ||
5483 | mte_set_node_dead(entry); | |
54a611b6 LH |
5484 | node->type = type; |
5485 | rcu_assign_pointer(slots[offset], node); | |
5486 | } | |
5487 | ||
5488 | return offset; | |
5489 | } | |
5490 | ||
790e1fa8 LH |
5491 | /** |
5492 | * mte_dead_walk() - Walk down a dead tree to just before the leaves | |
5493 | * @enode: The maple encoded node | |
5494 | * @offset: The starting offset | |
5495 | * | |
5496 | * Note: This can only be used from the RCU callback context. | |
5497 | */ | |
5498 | static void __rcu **mte_dead_walk(struct maple_enode **enode, unsigned char offset) | |
54a611b6 LH |
5499 | { |
5500 | struct maple_node *node, *next; | |
5501 | void __rcu **slots = NULL; | |
5502 | ||
790e1fa8 | 5503 | next = mte_to_node(*enode); |
54a611b6 | 5504 | do { |
790e1fa8 LH |
5505 | *enode = ma_enode_ptr(next); |
5506 | node = mte_to_node(*enode); | |
54a611b6 | 5507 | slots = ma_slots(node, node->type); |
790e1fa8 LH |
5508 | next = rcu_dereference_protected(slots[offset], |
5509 | lock_is_held(&rcu_callback_map)); | |
54a611b6 LH |
5510 | offset = 0; |
5511 | } while (!ma_is_leaf(next->type)); | |
5512 | ||
5513 | return slots; | |
5514 | } | |
5515 | ||
790e1fa8 LH |
5516 | /** |
5517 | * mt_free_walk() - Walk & free a tree in the RCU callback context | |
5518 | * @head: The RCU head that's within the node. | |
5519 | * | |
5520 | * Note: This can only be used from the RCU callback context. | |
5521 | */ | |
54a611b6 LH |
5522 | static void mt_free_walk(struct rcu_head *head) |
5523 | { | |
5524 | void __rcu **slots; | |
5525 | struct maple_node *node, *start; | |
790e1fa8 | 5526 | struct maple_enode *enode; |
54a611b6 LH |
5527 | unsigned char offset; |
5528 | enum maple_type type; | |
54a611b6 LH |
5529 | |
5530 | node = container_of(head, struct maple_node, rcu); | |
5531 | ||
5532 | if (ma_is_leaf(node->type)) | |
5533 | goto free_leaf; | |
5534 | ||
54a611b6 | 5535 | start = node; |
790e1fa8 LH |
5536 | enode = mt_mk_node(node, node->type); |
5537 | slots = mte_dead_walk(&enode, 0); | |
5538 | node = mte_to_node(enode); | |
54a611b6 LH |
5539 | do { |
5540 | mt_free_bulk(node->slot_len, slots); | |
5541 | offset = node->parent_slot + 1; | |
790e1fa8 LH |
5542 | enode = node->piv_parent; |
5543 | if (mte_to_node(enode) == node) | |
5544 | goto free_leaf; | |
5545 | ||
5546 | type = mte_node_type(enode); | |
5547 | slots = ma_slots(mte_to_node(enode), type); | |
5548 | if ((offset < mt_slots[type]) && | |
5549 | rcu_dereference_protected(slots[offset], | |
5550 | lock_is_held(&rcu_callback_map))) | |
5551 | slots = mte_dead_walk(&enode, offset); | |
5552 | node = mte_to_node(enode); | |
54a611b6 LH |
5553 | } while ((node != start) || (node->slot_len < offset)); |
5554 | ||
5555 | slots = ma_slots(node, node->type); | |
5556 | mt_free_bulk(node->slot_len, slots); | |
5557 | ||
54a611b6 LH |
5558 | free_leaf: |
5559 | mt_free_rcu(&node->rcu); | |
5560 | } | |
5561 | ||
790e1fa8 LH |
5562 | static inline void __rcu **mte_destroy_descend(struct maple_enode **enode, |
5563 | struct maple_tree *mt, struct maple_enode *prev, unsigned char offset) | |
54a611b6 LH |
5564 | { |
5565 | struct maple_node *node; | |
790e1fa8 | 5566 | struct maple_enode *next = *enode; |
54a611b6 | 5567 | void __rcu **slots = NULL; |
790e1fa8 LH |
5568 | enum maple_type type; |
5569 | unsigned char next_offset = 0; | |
54a611b6 LH |
5570 | |
5571 | do { | |
790e1fa8 LH |
5572 | *enode = next; |
5573 | node = mte_to_node(*enode); | |
5574 | type = mte_node_type(*enode); | |
5575 | slots = ma_slots(node, type); | |
5576 | next = mt_slot_locked(mt, slots, next_offset); | |
54a611b6 | 5577 | if ((mte_dead_node(next))) |
790e1fa8 | 5578 | next = mt_slot_locked(mt, slots, ++next_offset); |
54a611b6 | 5579 | |
790e1fa8 LH |
5580 | mte_set_node_dead(*enode); |
5581 | node->type = type; | |
54a611b6 LH |
5582 | node->piv_parent = prev; |
5583 | node->parent_slot = offset; | |
790e1fa8 LH |
5584 | offset = next_offset; |
5585 | next_offset = 0; | |
5586 | prev = *enode; | |
54a611b6 LH |
5587 | } while (!mte_is_leaf(next)); |
5588 | ||
5589 | return slots; | |
5590 | } | |
5591 | ||
790e1fa8 | 5592 | static void mt_destroy_walk(struct maple_enode *enode, struct maple_tree *mt, |
54a611b6 LH |
5593 | bool free) |
5594 | { | |
5595 | void __rcu **slots; | |
5596 | struct maple_node *node = mte_to_node(enode); | |
5597 | struct maple_enode *start; | |
54a611b6 | 5598 | |
2e5b4921 LH |
5599 | if (mte_is_leaf(enode)) { |
5600 | node->type = mte_node_type(enode); | |
54a611b6 | 5601 | goto free_leaf; |
2e5b4921 | 5602 | } |
54a611b6 | 5603 | |
2e5b4921 | 5604 | start = enode; |
790e1fa8 LH |
5605 | slots = mte_destroy_descend(&enode, mt, start, 0); |
5606 | node = mte_to_node(enode); // Updated in the above call. | |
54a611b6 LH |
5607 | do { |
5608 | enum maple_type type; | |
5609 | unsigned char offset; | |
5610 | struct maple_enode *parent, *tmp; | |
5611 | ||
790e1fa8 | 5612 | node->slot_len = mte_dead_leaves(enode, mt, slots); |
54a611b6 LH |
5613 | if (free) |
5614 | mt_free_bulk(node->slot_len, slots); | |
5615 | offset = node->parent_slot + 1; | |
790e1fa8 LH |
5616 | enode = node->piv_parent; |
5617 | if (mte_to_node(enode) == node) | |
5618 | goto free_leaf; | |
54a611b6 | 5619 | |
790e1fa8 LH |
5620 | type = mte_node_type(enode); |
5621 | slots = ma_slots(mte_to_node(enode), type); | |
54a611b6 LH |
5622 | if (offset >= mt_slots[type]) |
5623 | goto next; | |
5624 | ||
790e1fa8 | 5625 | tmp = mt_slot_locked(mt, slots, offset); |
54a611b6 | 5626 | if (mte_node_type(tmp) && mte_to_node(tmp)) { |
790e1fa8 LH |
5627 | parent = enode; |
5628 | enode = tmp; | |
5629 | slots = mte_destroy_descend(&enode, mt, parent, offset); | |
54a611b6 LH |
5630 | } |
5631 | next: | |
790e1fa8 LH |
5632 | node = mte_to_node(enode); |
5633 | } while (start != enode); | |
54a611b6 | 5634 | |
790e1fa8 LH |
5635 | node = mte_to_node(enode); |
5636 | node->slot_len = mte_dead_leaves(enode, mt, slots); | |
54a611b6 LH |
5637 | if (free) |
5638 | mt_free_bulk(node->slot_len, slots); | |
5639 | ||
54a611b6 LH |
5640 | free_leaf: |
5641 | if (free) | |
5642 | mt_free_rcu(&node->rcu); | |
2e5b4921 | 5643 | else |
790e1fa8 | 5644 | mt_clear_meta(mt, node, node->type); |
54a611b6 LH |
5645 | } |
5646 | ||
5647 | /* | |
5648 | * mte_destroy_walk() - Free a tree or sub-tree. | |
f942b0f0 VY |
5649 | * @enode: the encoded maple node (maple_enode) to start |
5650 | * @mt: the tree to free - needed for node types. | |
54a611b6 LH |
5651 | * |
5652 | * Must hold the write lock. | |
5653 | */ | |
5654 | static inline void mte_destroy_walk(struct maple_enode *enode, | |
5655 | struct maple_tree *mt) | |
5656 | { | |
5657 | struct maple_node *node = mte_to_node(enode); | |
5658 | ||
5659 | if (mt_in_rcu(mt)) { | |
790e1fa8 | 5660 | mt_destroy_walk(enode, mt, false); |
54a611b6 LH |
5661 | call_rcu(&node->rcu, mt_free_walk); |
5662 | } else { | |
790e1fa8 | 5663 | mt_destroy_walk(enode, mt, true); |
54a611b6 LH |
5664 | } |
5665 | } | |
5666 | ||
5667 | static void mas_wr_store_setup(struct ma_wr_state *wr_mas) | |
5668 | { | |
1202700c LH |
5669 | if (unlikely(mas_is_paused(wr_mas->mas))) |
5670 | mas_reset(wr_mas->mas); | |
5671 | ||
54a611b6 LH |
5672 | if (!mas_is_start(wr_mas->mas)) { |
5673 | if (mas_is_none(wr_mas->mas)) { | |
5674 | mas_reset(wr_mas->mas); | |
5675 | } else { | |
5676 | wr_mas->r_max = wr_mas->mas->max; | |
5677 | wr_mas->type = mte_node_type(wr_mas->mas->node); | |
5678 | if (mas_is_span_wr(wr_mas)) | |
5679 | mas_reset(wr_mas->mas); | |
5680 | } | |
5681 | } | |
54a611b6 LH |
5682 | } |
5683 | ||
5684 | /* Interface */ | |
5685 | ||
5686 | /** | |
5687 | * mas_store() - Store an @entry. | |
5688 | * @mas: The maple state. | |
5689 | * @entry: The entry to store. | |
5690 | * | |
5691 | * The @mas->index and @mas->last is used to set the range for the @entry. | |
5692 | * Note: The @mas should have pre-allocated entries to ensure there is memory to | |
5693 | * store the entry. Please see mas_expected_entries()/mas_destroy() for more details. | |
5694 | * | |
5695 | * Return: the first entry between mas->index and mas->last or %NULL. | |
5696 | */ | |
5697 | void *mas_store(struct ma_state *mas, void *entry) | |
5698 | { | |
5699 | MA_WR_STATE(wr_mas, mas, entry); | |
5700 | ||
5701 | trace_ma_write(__func__, mas, 0, entry); | |
5702 | #ifdef CONFIG_DEBUG_MAPLE_TREE | |
5703 | if (mas->index > mas->last) | |
5704 | pr_err("Error %lu > %lu %p\n", mas->index, mas->last, entry); | |
5705 | MT_BUG_ON(mas->tree, mas->index > mas->last); | |
5706 | if (mas->index > mas->last) { | |
5707 | mas_set_err(mas, -EINVAL); | |
5708 | return NULL; | |
5709 | } | |
5710 | ||
5711 | #endif | |
5712 | ||
5713 | /* | |
5714 | * Storing is the same operation as insert with the added caveat that it | |
5715 | * can overwrite entries. Although this seems simple enough, one may | |
5716 | * want to examine what happens if a single store operation was to | |
5717 | * overwrite multiple entries within a self-balancing B-Tree. | |
5718 | */ | |
5719 | mas_wr_store_setup(&wr_mas); | |
5720 | mas_wr_store_entry(&wr_mas); | |
5721 | return wr_mas.content; | |
5722 | } | |
120b1162 | 5723 | EXPORT_SYMBOL_GPL(mas_store); |
54a611b6 LH |
5724 | |
5725 | /** | |
5726 | * mas_store_gfp() - Store a value into the tree. | |
5727 | * @mas: The maple state | |
5728 | * @entry: The entry to store | |
5729 | * @gfp: The GFP_FLAGS to use for allocations if necessary. | |
5730 | * | |
5731 | * Return: 0 on success, -EINVAL on invalid request, -ENOMEM if memory could not | |
5732 | * be allocated. | |
5733 | */ | |
5734 | int mas_store_gfp(struct ma_state *mas, void *entry, gfp_t gfp) | |
5735 | { | |
5736 | MA_WR_STATE(wr_mas, mas, entry); | |
5737 | ||
5738 | mas_wr_store_setup(&wr_mas); | |
5739 | trace_ma_write(__func__, mas, 0, entry); | |
5740 | retry: | |
5741 | mas_wr_store_entry(&wr_mas); | |
5742 | if (unlikely(mas_nomem(mas, gfp))) | |
5743 | goto retry; | |
5744 | ||
5745 | if (unlikely(mas_is_err(mas))) | |
5746 | return xa_err(mas->node); | |
5747 | ||
5748 | return 0; | |
5749 | } | |
120b1162 | 5750 | EXPORT_SYMBOL_GPL(mas_store_gfp); |
54a611b6 LH |
5751 | |
5752 | /** | |
5753 | * mas_store_prealloc() - Store a value into the tree using memory | |
5754 | * preallocated in the maple state. | |
5755 | * @mas: The maple state | |
5756 | * @entry: The entry to store. | |
5757 | */ | |
5758 | void mas_store_prealloc(struct ma_state *mas, void *entry) | |
5759 | { | |
5760 | MA_WR_STATE(wr_mas, mas, entry); | |
5761 | ||
5762 | mas_wr_store_setup(&wr_mas); | |
5763 | trace_ma_write(__func__, mas, 0, entry); | |
5764 | mas_wr_store_entry(&wr_mas); | |
5765 | BUG_ON(mas_is_err(mas)); | |
5766 | mas_destroy(mas); | |
5767 | } | |
120b1162 | 5768 | EXPORT_SYMBOL_GPL(mas_store_prealloc); |
54a611b6 LH |
5769 | |
5770 | /** | |
5771 | * mas_preallocate() - Preallocate enough nodes for a store operation | |
5772 | * @mas: The maple state | |
54a611b6 LH |
5773 | * @gfp: The GFP_FLAGS to use for allocations. |
5774 | * | |
5775 | * Return: 0 on success, -ENOMEM if memory could not be allocated. | |
5776 | */ | |
c5d5546e | 5777 | int mas_preallocate(struct ma_state *mas, gfp_t gfp) |
54a611b6 LH |
5778 | { |
5779 | int ret; | |
5780 | ||
5781 | mas_node_count_gfp(mas, 1 + mas_mt_height(mas) * 3, gfp); | |
5782 | mas->mas_flags |= MA_STATE_PREALLOC; | |
5783 | if (likely(!mas_is_err(mas))) | |
5784 | return 0; | |
5785 | ||
5786 | mas_set_alloc_req(mas, 0); | |
5787 | ret = xa_err(mas->node); | |
5788 | mas_reset(mas); | |
5789 | mas_destroy(mas); | |
5790 | mas_reset(mas); | |
5791 | return ret; | |
5792 | } | |
5c63a7c3 | 5793 | EXPORT_SYMBOL_GPL(mas_preallocate); |
54a611b6 LH |
5794 | |
5795 | /* | |
5796 | * mas_destroy() - destroy a maple state. | |
5797 | * @mas: The maple state | |
5798 | * | |
5799 | * Upon completion, check the left-most node and rebalance against the node to | |
5800 | * the right if necessary. Frees any allocated nodes associated with this maple | |
5801 | * state. | |
5802 | */ | |
5803 | void mas_destroy(struct ma_state *mas) | |
5804 | { | |
5805 | struct maple_alloc *node; | |
541e06b7 | 5806 | unsigned long total; |
54a611b6 LH |
5807 | |
5808 | /* | |
5809 | * When using mas_for_each() to insert an expected number of elements, | |
5810 | * it is possible that the number inserted is less than the expected | |
5811 | * number. To fix an invalid final node, a check is performed here to | |
5812 | * rebalance the previous node with the final node. | |
5813 | */ | |
5814 | if (mas->mas_flags & MA_STATE_REBALANCE) { | |
5815 | unsigned char end; | |
5816 | ||
5817 | if (mas_is_start(mas)) | |
5818 | mas_start(mas); | |
5819 | ||
5820 | mtree_range_walk(mas); | |
5821 | end = mas_data_end(mas) + 1; | |
5822 | if (end < mt_min_slot_count(mas->node) - 1) | |
5823 | mas_destroy_rebalance(mas, end); | |
5824 | ||
5825 | mas->mas_flags &= ~MA_STATE_REBALANCE; | |
5826 | } | |
5827 | mas->mas_flags &= ~(MA_STATE_BULK|MA_STATE_PREALLOC); | |
5828 | ||
541e06b7 LH |
5829 | total = mas_allocated(mas); |
5830 | while (total) { | |
54a611b6 LH |
5831 | node = mas->alloc; |
5832 | mas->alloc = node->slot[0]; | |
541e06b7 LH |
5833 | if (node->node_count > 1) { |
5834 | size_t count = node->node_count - 1; | |
5835 | ||
5836 | mt_free_bulk(count, (void __rcu **)&node->slot[1]); | |
5837 | total -= count; | |
5838 | } | |
54a611b6 | 5839 | kmem_cache_free(maple_node_cache, node); |
541e06b7 | 5840 | total--; |
54a611b6 | 5841 | } |
541e06b7 | 5842 | |
54a611b6 LH |
5843 | mas->alloc = NULL; |
5844 | } | |
120b1162 | 5845 | EXPORT_SYMBOL_GPL(mas_destroy); |
54a611b6 LH |
5846 | |
5847 | /* | |
5848 | * mas_expected_entries() - Set the expected number of entries that will be inserted. | |
5849 | * @mas: The maple state | |
5850 | * @nr_entries: The number of expected entries. | |
5851 | * | |
5852 | * This will attempt to pre-allocate enough nodes to store the expected number | |
5853 | * of entries. The allocations will occur using the bulk allocator interface | |
5854 | * for speed. Please call mas_destroy() on the @mas after inserting the entries | |
5855 | * to ensure any unused nodes are freed. | |
5856 | * | |
5857 | * Return: 0 on success, -ENOMEM if memory could not be allocated. | |
5858 | */ | |
5859 | int mas_expected_entries(struct ma_state *mas, unsigned long nr_entries) | |
5860 | { | |
5861 | int nonleaf_cap = MAPLE_ARANGE64_SLOTS - 2; | |
5862 | struct maple_enode *enode = mas->node; | |
5863 | int nr_nodes; | |
5864 | int ret; | |
5865 | ||
5866 | /* | |
5867 | * Sometimes it is necessary to duplicate a tree to a new tree, such as | |
5868 | * forking a process and duplicating the VMAs from one tree to a new | |
5869 | * tree. When such a situation arises, it is known that the new tree is | |
5870 | * not going to be used until the entire tree is populated. For | |
5871 | * performance reasons, it is best to use a bulk load with RCU disabled. | |
5872 | * This allows for optimistic splitting that favours the left and reuse | |
5873 | * of nodes during the operation. | |
5874 | */ | |
5875 | ||
5876 | /* Optimize splitting for bulk insert in-order */ | |
5877 | mas->mas_flags |= MA_STATE_BULK; | |
5878 | ||
5879 | /* | |
5880 | * Avoid overflow, assume a gap between each entry and a trailing null. | |
5881 | * If this is wrong, it just means allocation can happen during | |
5882 | * insertion of entries. | |
5883 | */ | |
5884 | nr_nodes = max(nr_entries, nr_entries * 2 + 1); | |
5885 | if (!mt_is_alloc(mas->tree)) | |
5886 | nonleaf_cap = MAPLE_RANGE64_SLOTS - 2; | |
5887 | ||
5888 | /* Leaves; reduce slots to keep space for expansion */ | |
5889 | nr_nodes = DIV_ROUND_UP(nr_nodes, MAPLE_RANGE64_SLOTS - 2); | |
5890 | /* Internal nodes */ | |
5891 | nr_nodes += DIV_ROUND_UP(nr_nodes, nonleaf_cap); | |
5892 | /* Add working room for split (2 nodes) + new parents */ | |
5893 | mas_node_count(mas, nr_nodes + 3); | |
5894 | ||
5895 | /* Detect if allocations run out */ | |
5896 | mas->mas_flags |= MA_STATE_PREALLOC; | |
5897 | ||
5898 | if (!mas_is_err(mas)) | |
5899 | return 0; | |
5900 | ||
5901 | ret = xa_err(mas->node); | |
5902 | mas->node = enode; | |
5903 | mas_destroy(mas); | |
5904 | return ret; | |
5905 | ||
5906 | } | |
120b1162 | 5907 | EXPORT_SYMBOL_GPL(mas_expected_entries); |
54a611b6 LH |
5908 | |
5909 | /** | |
5910 | * mas_next() - Get the next entry. | |
5911 | * @mas: The maple state | |
5912 | * @max: The maximum index to check. | |
5913 | * | |
5914 | * Returns the next entry after @mas->index. | |
5915 | * Must hold rcu_read_lock or the write lock. | |
5916 | * Can return the zero entry. | |
5917 | * | |
5918 | * Return: The next entry or %NULL | |
5919 | */ | |
5920 | void *mas_next(struct ma_state *mas, unsigned long max) | |
5921 | { | |
5922 | if (mas_is_none(mas) || mas_is_paused(mas)) | |
5923 | mas->node = MAS_START; | |
5924 | ||
5925 | if (mas_is_start(mas)) | |
5926 | mas_walk(mas); /* Retries on dead nodes handled by mas_walk */ | |
5927 | ||
5928 | if (mas_is_ptr(mas)) { | |
5929 | if (!mas->index) { | |
5930 | mas->index = 1; | |
5931 | mas->last = ULONG_MAX; | |
5932 | } | |
5933 | return NULL; | |
5934 | } | |
5935 | ||
5936 | if (mas->last == ULONG_MAX) | |
5937 | return NULL; | |
5938 | ||
5939 | /* Retries on dead nodes handled by mas_next_entry */ | |
5940 | return mas_next_entry(mas, max); | |
5941 | } | |
5942 | EXPORT_SYMBOL_GPL(mas_next); | |
5943 | ||
5944 | /** | |
5945 | * mt_next() - get the next value in the maple tree | |
5946 | * @mt: The maple tree | |
5947 | * @index: The start index | |
5948 | * @max: The maximum index to check | |
5949 | * | |
5950 | * Return: The entry at @index or higher, or %NULL if nothing is found. | |
5951 | */ | |
5952 | void *mt_next(struct maple_tree *mt, unsigned long index, unsigned long max) | |
5953 | { | |
5954 | void *entry = NULL; | |
5955 | MA_STATE(mas, mt, index, index); | |
5956 | ||
5957 | rcu_read_lock(); | |
5958 | entry = mas_next(&mas, max); | |
5959 | rcu_read_unlock(); | |
5960 | return entry; | |
5961 | } | |
5962 | EXPORT_SYMBOL_GPL(mt_next); | |
5963 | ||
5964 | /** | |
5965 | * mas_prev() - Get the previous entry | |
5966 | * @mas: The maple state | |
5967 | * @min: The minimum value to check. | |
5968 | * | |
5969 | * Must hold rcu_read_lock or the write lock. | |
5970 | * Will reset mas to MAS_START if the node is MAS_NONE. Will stop on not | |
5971 | * searchable nodes. | |
5972 | * | |
5973 | * Return: the previous value or %NULL. | |
5974 | */ | |
5975 | void *mas_prev(struct ma_state *mas, unsigned long min) | |
5976 | { | |
5977 | if (!mas->index) { | |
5978 | /* Nothing comes before 0 */ | |
5979 | mas->last = 0; | |
17dc622c | 5980 | mas->node = MAS_NONE; |
54a611b6 LH |
5981 | return NULL; |
5982 | } | |
5983 | ||
5984 | if (unlikely(mas_is_ptr(mas))) | |
5985 | return NULL; | |
5986 | ||
5987 | if (mas_is_none(mas) || mas_is_paused(mas)) | |
5988 | mas->node = MAS_START; | |
5989 | ||
5990 | if (mas_is_start(mas)) { | |
5991 | mas_walk(mas); | |
5992 | if (!mas->index) | |
5993 | return NULL; | |
5994 | } | |
5995 | ||
5996 | if (mas_is_ptr(mas)) { | |
5997 | if (!mas->index) { | |
5998 | mas->last = 0; | |
5999 | return NULL; | |
6000 | } | |
6001 | ||
6002 | mas->index = mas->last = 0; | |
6003 | return mas_root_locked(mas); | |
6004 | } | |
6005 | return mas_prev_entry(mas, min); | |
6006 | } | |
6007 | EXPORT_SYMBOL_GPL(mas_prev); | |
6008 | ||
6009 | /** | |
6010 | * mt_prev() - get the previous value in the maple tree | |
6011 | * @mt: The maple tree | |
6012 | * @index: The start index | |
6013 | * @min: The minimum index to check | |
6014 | * | |
6015 | * Return: The entry at @index or lower, or %NULL if nothing is found. | |
6016 | */ | |
6017 | void *mt_prev(struct maple_tree *mt, unsigned long index, unsigned long min) | |
6018 | { | |
6019 | void *entry = NULL; | |
6020 | MA_STATE(mas, mt, index, index); | |
6021 | ||
6022 | rcu_read_lock(); | |
6023 | entry = mas_prev(&mas, min); | |
6024 | rcu_read_unlock(); | |
6025 | return entry; | |
6026 | } | |
6027 | EXPORT_SYMBOL_GPL(mt_prev); | |
6028 | ||
6029 | /** | |
6030 | * mas_pause() - Pause a mas_find/mas_for_each to drop the lock. | |
6031 | * @mas: The maple state to pause | |
6032 | * | |
6033 | * Some users need to pause a walk and drop the lock they're holding in | |
6034 | * order to yield to a higher priority thread or carry out an operation | |
6035 | * on an entry. Those users should call this function before they drop | |
6036 | * the lock. It resets the @mas to be suitable for the next iteration | |
6037 | * of the loop after the user has reacquired the lock. If most entries | |
6038 | * found during a walk require you to call mas_pause(), the mt_for_each() | |
6039 | * iterator may be more appropriate. | |
6040 | * | |
6041 | */ | |
6042 | void mas_pause(struct ma_state *mas) | |
6043 | { | |
6044 | mas->node = MAS_PAUSE; | |
6045 | } | |
6046 | EXPORT_SYMBOL_GPL(mas_pause); | |
6047 | ||
6048 | /** | |
6049 | * mas_find() - On the first call, find the entry at or after mas->index up to | |
6050 | * %max. Otherwise, find the entry after mas->index. | |
6051 | * @mas: The maple state | |
6052 | * @max: The maximum value to check. | |
6053 | * | |
6054 | * Must hold rcu_read_lock or the write lock. | |
6055 | * If an entry exists, last and index are updated accordingly. | |
6056 | * May set @mas->node to MAS_NONE. | |
6057 | * | |
6058 | * Return: The entry or %NULL. | |
6059 | */ | |
6060 | void *mas_find(struct ma_state *mas, unsigned long max) | |
6061 | { | |
6062 | if (unlikely(mas_is_paused(mas))) { | |
6063 | if (unlikely(mas->last == ULONG_MAX)) { | |
6064 | mas->node = MAS_NONE; | |
6065 | return NULL; | |
6066 | } | |
6067 | mas->node = MAS_START; | |
6068 | mas->index = ++mas->last; | |
6069 | } | |
6070 | ||
17dc622c LH |
6071 | if (unlikely(mas_is_none(mas))) |
6072 | mas->node = MAS_START; | |
6073 | ||
54a611b6 LH |
6074 | if (unlikely(mas_is_start(mas))) { |
6075 | /* First run or continue */ | |
6076 | void *entry; | |
6077 | ||
6078 | if (mas->index > max) | |
6079 | return NULL; | |
6080 | ||
6081 | entry = mas_walk(mas); | |
6082 | if (entry) | |
6083 | return entry; | |
6084 | } | |
6085 | ||
6086 | if (unlikely(!mas_searchable(mas))) | |
6087 | return NULL; | |
6088 | ||
6089 | /* Retries on dead nodes handled by mas_next_entry */ | |
6090 | return mas_next_entry(mas, max); | |
6091 | } | |
120b1162 | 6092 | EXPORT_SYMBOL_GPL(mas_find); |
54a611b6 LH |
6093 | |
6094 | /** | |
6095 | * mas_find_rev: On the first call, find the first non-null entry at or below | |
6096 | * mas->index down to %min. Otherwise find the first non-null entry below | |
6097 | * mas->index down to %min. | |
6098 | * @mas: The maple state | |
6099 | * @min: The minimum value to check. | |
6100 | * | |
6101 | * Must hold rcu_read_lock or the write lock. | |
6102 | * If an entry exists, last and index are updated accordingly. | |
6103 | * May set @mas->node to MAS_NONE. | |
6104 | * | |
6105 | * Return: The entry or %NULL. | |
6106 | */ | |
6107 | void *mas_find_rev(struct ma_state *mas, unsigned long min) | |
6108 | { | |
6109 | if (unlikely(mas_is_paused(mas))) { | |
6110 | if (unlikely(mas->last == ULONG_MAX)) { | |
6111 | mas->node = MAS_NONE; | |
6112 | return NULL; | |
6113 | } | |
6114 | mas->node = MAS_START; | |
6115 | mas->last = --mas->index; | |
6116 | } | |
6117 | ||
6118 | if (unlikely(mas_is_start(mas))) { | |
6119 | /* First run or continue */ | |
6120 | void *entry; | |
6121 | ||
6122 | if (mas->index < min) | |
6123 | return NULL; | |
6124 | ||
6125 | entry = mas_walk(mas); | |
6126 | if (entry) | |
6127 | return entry; | |
6128 | } | |
6129 | ||
6130 | if (unlikely(!mas_searchable(mas))) | |
6131 | return NULL; | |
6132 | ||
6133 | if (mas->index < min) | |
6134 | return NULL; | |
6135 | ||
d98c86b9 | 6136 | /* Retries on dead nodes handled by mas_prev_entry */ |
54a611b6 LH |
6137 | return mas_prev_entry(mas, min); |
6138 | } | |
120b1162 | 6139 | EXPORT_SYMBOL_GPL(mas_find_rev); |
54a611b6 LH |
6140 | |
6141 | /** | |
6142 | * mas_erase() - Find the range in which index resides and erase the entire | |
6143 | * range. | |
6144 | * @mas: The maple state | |
6145 | * | |
6146 | * Must hold the write lock. | |
6147 | * Searches for @mas->index, sets @mas->index and @mas->last to the range and | |
6148 | * erases that range. | |
6149 | * | |
6150 | * Return: the entry that was erased or %NULL, @mas->index and @mas->last are updated. | |
6151 | */ | |
6152 | void *mas_erase(struct ma_state *mas) | |
6153 | { | |
6154 | void *entry; | |
6155 | MA_WR_STATE(wr_mas, mas, NULL); | |
6156 | ||
6157 | if (mas_is_none(mas) || mas_is_paused(mas)) | |
6158 | mas->node = MAS_START; | |
6159 | ||
6160 | /* Retry unnecessary when holding the write lock. */ | |
6161 | entry = mas_state_walk(mas); | |
6162 | if (!entry) | |
6163 | return NULL; | |
6164 | ||
6165 | write_retry: | |
6166 | /* Must reset to ensure spanning writes of last slot are detected */ | |
6167 | mas_reset(mas); | |
6168 | mas_wr_store_setup(&wr_mas); | |
6169 | mas_wr_store_entry(&wr_mas); | |
6170 | if (mas_nomem(mas, GFP_KERNEL)) | |
6171 | goto write_retry; | |
6172 | ||
6173 | return entry; | |
6174 | } | |
6175 | EXPORT_SYMBOL_GPL(mas_erase); | |
6176 | ||
6177 | /** | |
6178 | * mas_nomem() - Check if there was an error allocating and do the allocation | |
6179 | * if necessary If there are allocations, then free them. | |
6180 | * @mas: The maple state | |
6181 | * @gfp: The GFP_FLAGS to use for allocations | |
6182 | * Return: true on allocation, false otherwise. | |
6183 | */ | |
6184 | bool mas_nomem(struct ma_state *mas, gfp_t gfp) | |
6185 | __must_hold(mas->tree->lock) | |
6186 | { | |
6187 | if (likely(mas->node != MA_ERROR(-ENOMEM))) { | |
6188 | mas_destroy(mas); | |
6189 | return false; | |
6190 | } | |
6191 | ||
6192 | if (gfpflags_allow_blocking(gfp) && !mt_external_lock(mas->tree)) { | |
6193 | mtree_unlock(mas->tree); | |
6194 | mas_alloc_nodes(mas, gfp); | |
6195 | mtree_lock(mas->tree); | |
6196 | } else { | |
6197 | mas_alloc_nodes(mas, gfp); | |
6198 | } | |
6199 | ||
6200 | if (!mas_allocated(mas)) | |
6201 | return false; | |
6202 | ||
6203 | mas->node = MAS_START; | |
6204 | return true; | |
6205 | } | |
6206 | ||
6207 | void __init maple_tree_init(void) | |
6208 | { | |
6209 | maple_node_cache = kmem_cache_create("maple_node", | |
6210 | sizeof(struct maple_node), sizeof(struct maple_node), | |
6211 | SLAB_PANIC, NULL); | |
6212 | } | |
6213 | ||
6214 | /** | |
6215 | * mtree_load() - Load a value stored in a maple tree | |
6216 | * @mt: The maple tree | |
6217 | * @index: The index to load | |
6218 | * | |
6219 | * Return: the entry or %NULL | |
6220 | */ | |
6221 | void *mtree_load(struct maple_tree *mt, unsigned long index) | |
6222 | { | |
6223 | MA_STATE(mas, mt, index, index); | |
6224 | void *entry; | |
6225 | ||
6226 | trace_ma_read(__func__, &mas); | |
6227 | rcu_read_lock(); | |
6228 | retry: | |
6229 | entry = mas_start(&mas); | |
6230 | if (unlikely(mas_is_none(&mas))) | |
6231 | goto unlock; | |
6232 | ||
6233 | if (unlikely(mas_is_ptr(&mas))) { | |
6234 | if (index) | |
6235 | entry = NULL; | |
6236 | ||
6237 | goto unlock; | |
6238 | } | |
6239 | ||
6240 | entry = mtree_lookup_walk(&mas); | |
6241 | if (!entry && unlikely(mas_is_start(&mas))) | |
6242 | goto retry; | |
6243 | unlock: | |
6244 | rcu_read_unlock(); | |
6245 | if (xa_is_zero(entry)) | |
6246 | return NULL; | |
6247 | ||
6248 | return entry; | |
6249 | } | |
6250 | EXPORT_SYMBOL(mtree_load); | |
6251 | ||
6252 | /** | |
6253 | * mtree_store_range() - Store an entry at a given range. | |
6254 | * @mt: The maple tree | |
6255 | * @index: The start of the range | |
6256 | * @last: The end of the range | |
6257 | * @entry: The entry to store | |
6258 | * @gfp: The GFP_FLAGS to use for allocations | |
6259 | * | |
6260 | * Return: 0 on success, -EINVAL on invalid request, -ENOMEM if memory could not | |
6261 | * be allocated. | |
6262 | */ | |
6263 | int mtree_store_range(struct maple_tree *mt, unsigned long index, | |
6264 | unsigned long last, void *entry, gfp_t gfp) | |
6265 | { | |
6266 | MA_STATE(mas, mt, index, last); | |
6267 | MA_WR_STATE(wr_mas, &mas, entry); | |
6268 | ||
6269 | trace_ma_write(__func__, &mas, 0, entry); | |
6270 | if (WARN_ON_ONCE(xa_is_advanced(entry))) | |
6271 | return -EINVAL; | |
6272 | ||
6273 | if (index > last) | |
6274 | return -EINVAL; | |
6275 | ||
6276 | mtree_lock(mt); | |
6277 | retry: | |
6278 | mas_wr_store_entry(&wr_mas); | |
6279 | if (mas_nomem(&mas, gfp)) | |
6280 | goto retry; | |
6281 | ||
6282 | mtree_unlock(mt); | |
6283 | if (mas_is_err(&mas)) | |
6284 | return xa_err(mas.node); | |
6285 | ||
6286 | return 0; | |
6287 | } | |
6288 | EXPORT_SYMBOL(mtree_store_range); | |
6289 | ||
6290 | /** | |
6291 | * mtree_store() - Store an entry at a given index. | |
6292 | * @mt: The maple tree | |
6293 | * @index: The index to store the value | |
6294 | * @entry: The entry to store | |
6295 | * @gfp: The GFP_FLAGS to use for allocations | |
6296 | * | |
6297 | * Return: 0 on success, -EINVAL on invalid request, -ENOMEM if memory could not | |
6298 | * be allocated. | |
6299 | */ | |
6300 | int mtree_store(struct maple_tree *mt, unsigned long index, void *entry, | |
6301 | gfp_t gfp) | |
6302 | { | |
6303 | return mtree_store_range(mt, index, index, entry, gfp); | |
6304 | } | |
6305 | EXPORT_SYMBOL(mtree_store); | |
6306 | ||
6307 | /** | |
6308 | * mtree_insert_range() - Insert an entry at a give range if there is no value. | |
6309 | * @mt: The maple tree | |
6310 | * @first: The start of the range | |
6311 | * @last: The end of the range | |
6312 | * @entry: The entry to store | |
6313 | * @gfp: The GFP_FLAGS to use for allocations. | |
6314 | * | |
6315 | * Return: 0 on success, -EEXISTS if the range is occupied, -EINVAL on invalid | |
6316 | * request, -ENOMEM if memory could not be allocated. | |
6317 | */ | |
6318 | int mtree_insert_range(struct maple_tree *mt, unsigned long first, | |
6319 | unsigned long last, void *entry, gfp_t gfp) | |
6320 | { | |
6321 | MA_STATE(ms, mt, first, last); | |
6322 | ||
6323 | if (WARN_ON_ONCE(xa_is_advanced(entry))) | |
6324 | return -EINVAL; | |
6325 | ||
6326 | if (first > last) | |
6327 | return -EINVAL; | |
6328 | ||
6329 | mtree_lock(mt); | |
6330 | retry: | |
6331 | mas_insert(&ms, entry); | |
6332 | if (mas_nomem(&ms, gfp)) | |
6333 | goto retry; | |
6334 | ||
6335 | mtree_unlock(mt); | |
6336 | if (mas_is_err(&ms)) | |
6337 | return xa_err(ms.node); | |
6338 | ||
6339 | return 0; | |
6340 | } | |
6341 | EXPORT_SYMBOL(mtree_insert_range); | |
6342 | ||
6343 | /** | |
6344 | * mtree_insert() - Insert an entry at a give index if there is no value. | |
6345 | * @mt: The maple tree | |
6346 | * @index : The index to store the value | |
6347 | * @entry: The entry to store | |
6348 | * @gfp: The FGP_FLAGS to use for allocations. | |
6349 | * | |
6350 | * Return: 0 on success, -EEXISTS if the range is occupied, -EINVAL on invalid | |
6351 | * request, -ENOMEM if memory could not be allocated. | |
6352 | */ | |
6353 | int mtree_insert(struct maple_tree *mt, unsigned long index, void *entry, | |
6354 | gfp_t gfp) | |
6355 | { | |
6356 | return mtree_insert_range(mt, index, index, entry, gfp); | |
6357 | } | |
6358 | EXPORT_SYMBOL(mtree_insert); | |
6359 | ||
6360 | int mtree_alloc_range(struct maple_tree *mt, unsigned long *startp, | |
6361 | void *entry, unsigned long size, unsigned long min, | |
6362 | unsigned long max, gfp_t gfp) | |
6363 | { | |
6364 | int ret = 0; | |
6365 | ||
6366 | MA_STATE(mas, mt, min, max - size); | |
6367 | if (!mt_is_alloc(mt)) | |
6368 | return -EINVAL; | |
6369 | ||
6370 | if (WARN_ON_ONCE(mt_is_reserved(entry))) | |
6371 | return -EINVAL; | |
6372 | ||
6373 | if (min > max) | |
6374 | return -EINVAL; | |
6375 | ||
6376 | if (max < size) | |
6377 | return -EINVAL; | |
6378 | ||
6379 | if (!size) | |
6380 | return -EINVAL; | |
6381 | ||
6382 | mtree_lock(mt); | |
6383 | retry: | |
6384 | mas.offset = 0; | |
6385 | mas.index = min; | |
6386 | mas.last = max - size; | |
6387 | ret = mas_alloc(&mas, entry, size, startp); | |
6388 | if (mas_nomem(&mas, gfp)) | |
6389 | goto retry; | |
6390 | ||
6391 | mtree_unlock(mt); | |
6392 | return ret; | |
6393 | } | |
6394 | EXPORT_SYMBOL(mtree_alloc_range); | |
6395 | ||
6396 | int mtree_alloc_rrange(struct maple_tree *mt, unsigned long *startp, | |
6397 | void *entry, unsigned long size, unsigned long min, | |
6398 | unsigned long max, gfp_t gfp) | |
6399 | { | |
6400 | int ret = 0; | |
6401 | ||
6402 | MA_STATE(mas, mt, min, max - size); | |
6403 | if (!mt_is_alloc(mt)) | |
6404 | return -EINVAL; | |
6405 | ||
6406 | if (WARN_ON_ONCE(mt_is_reserved(entry))) | |
6407 | return -EINVAL; | |
6408 | ||
6409 | if (min >= max) | |
6410 | return -EINVAL; | |
6411 | ||
6412 | if (max < size - 1) | |
6413 | return -EINVAL; | |
6414 | ||
6415 | if (!size) | |
6416 | return -EINVAL; | |
6417 | ||
6418 | mtree_lock(mt); | |
6419 | retry: | |
6420 | ret = mas_rev_alloc(&mas, min, max, entry, size, startp); | |
6421 | if (mas_nomem(&mas, gfp)) | |
6422 | goto retry; | |
6423 | ||
6424 | mtree_unlock(mt); | |
6425 | return ret; | |
6426 | } | |
6427 | EXPORT_SYMBOL(mtree_alloc_rrange); | |
6428 | ||
6429 | /** | |
6430 | * mtree_erase() - Find an index and erase the entire range. | |
6431 | * @mt: The maple tree | |
6432 | * @index: The index to erase | |
6433 | * | |
6434 | * Erasing is the same as a walk to an entry then a store of a NULL to that | |
6435 | * ENTIRE range. In fact, it is implemented as such using the advanced API. | |
6436 | * | |
6437 | * Return: The entry stored at the @index or %NULL | |
6438 | */ | |
6439 | void *mtree_erase(struct maple_tree *mt, unsigned long index) | |
6440 | { | |
6441 | void *entry = NULL; | |
6442 | ||
6443 | MA_STATE(mas, mt, index, index); | |
6444 | trace_ma_op(__func__, &mas); | |
6445 | ||
6446 | mtree_lock(mt); | |
6447 | entry = mas_erase(&mas); | |
6448 | mtree_unlock(mt); | |
6449 | ||
6450 | return entry; | |
6451 | } | |
6452 | EXPORT_SYMBOL(mtree_erase); | |
6453 | ||
6454 | /** | |
6455 | * __mt_destroy() - Walk and free all nodes of a locked maple tree. | |
6456 | * @mt: The maple tree | |
6457 | * | |
6458 | * Note: Does not handle locking. | |
6459 | */ | |
6460 | void __mt_destroy(struct maple_tree *mt) | |
6461 | { | |
6462 | void *root = mt_root_locked(mt); | |
6463 | ||
6464 | rcu_assign_pointer(mt->ma_root, NULL); | |
6465 | if (xa_is_node(root)) | |
6466 | mte_destroy_walk(root, mt); | |
6467 | ||
6468 | mt->ma_flags = 0; | |
6469 | } | |
6470 | EXPORT_SYMBOL_GPL(__mt_destroy); | |
6471 | ||
6472 | /** | |
6473 | * mtree_destroy() - Destroy a maple tree | |
6474 | * @mt: The maple tree | |
6475 | * | |
6476 | * Frees all resources used by the tree. Handles locking. | |
6477 | */ | |
6478 | void mtree_destroy(struct maple_tree *mt) | |
6479 | { | |
6480 | mtree_lock(mt); | |
6481 | __mt_destroy(mt); | |
6482 | mtree_unlock(mt); | |
6483 | } | |
6484 | EXPORT_SYMBOL(mtree_destroy); | |
6485 | ||
6486 | /** | |
6487 | * mt_find() - Search from the start up until an entry is found. | |
6488 | * @mt: The maple tree | |
6489 | * @index: Pointer which contains the start location of the search | |
6490 | * @max: The maximum value to check | |
6491 | * | |
6492 | * Handles locking. @index will be incremented to one beyond the range. | |
6493 | * | |
6494 | * Return: The entry at or after the @index or %NULL | |
6495 | */ | |
6496 | void *mt_find(struct maple_tree *mt, unsigned long *index, unsigned long max) | |
6497 | { | |
6498 | MA_STATE(mas, mt, *index, *index); | |
6499 | void *entry; | |
6500 | #ifdef CONFIG_DEBUG_MAPLE_TREE | |
6501 | unsigned long copy = *index; | |
6502 | #endif | |
6503 | ||
6504 | trace_ma_read(__func__, &mas); | |
6505 | ||
6506 | if ((*index) > max) | |
6507 | return NULL; | |
6508 | ||
6509 | rcu_read_lock(); | |
6510 | retry: | |
6511 | entry = mas_state_walk(&mas); | |
6512 | if (mas_is_start(&mas)) | |
6513 | goto retry; | |
6514 | ||
6515 | if (unlikely(xa_is_zero(entry))) | |
6516 | entry = NULL; | |
6517 | ||
6518 | if (entry) | |
6519 | goto unlock; | |
6520 | ||
6521 | while (mas_searchable(&mas) && (mas.index < max)) { | |
6522 | entry = mas_next_entry(&mas, max); | |
6523 | if (likely(entry && !xa_is_zero(entry))) | |
6524 | break; | |
6525 | } | |
6526 | ||
6527 | if (unlikely(xa_is_zero(entry))) | |
6528 | entry = NULL; | |
6529 | unlock: | |
6530 | rcu_read_unlock(); | |
6531 | if (likely(entry)) { | |
6532 | *index = mas.last + 1; | |
6533 | #ifdef CONFIG_DEBUG_MAPLE_TREE | |
6534 | if ((*index) && (*index) <= copy) | |
6535 | pr_err("index not increased! %lx <= %lx\n", | |
6536 | *index, copy); | |
6537 | MT_BUG_ON(mt, (*index) && ((*index) <= copy)); | |
6538 | #endif | |
6539 | } | |
6540 | ||
6541 | return entry; | |
6542 | } | |
6543 | EXPORT_SYMBOL(mt_find); | |
6544 | ||
6545 | /** | |
6546 | * mt_find_after() - Search from the start up until an entry is found. | |
6547 | * @mt: The maple tree | |
6548 | * @index: Pointer which contains the start location of the search | |
6549 | * @max: The maximum value to check | |
6550 | * | |
6551 | * Handles locking, detects wrapping on index == 0 | |
6552 | * | |
6553 | * Return: The entry at or after the @index or %NULL | |
6554 | */ | |
6555 | void *mt_find_after(struct maple_tree *mt, unsigned long *index, | |
6556 | unsigned long max) | |
6557 | { | |
6558 | if (!(*index)) | |
6559 | return NULL; | |
6560 | ||
6561 | return mt_find(mt, index, max); | |
6562 | } | |
6563 | EXPORT_SYMBOL(mt_find_after); | |
6564 | ||
6565 | #ifdef CONFIG_DEBUG_MAPLE_TREE | |
6566 | atomic_t maple_tree_tests_run; | |
6567 | EXPORT_SYMBOL_GPL(maple_tree_tests_run); | |
6568 | atomic_t maple_tree_tests_passed; | |
6569 | EXPORT_SYMBOL_GPL(maple_tree_tests_passed); | |
6570 | ||
6571 | #ifndef __KERNEL__ | |
6572 | extern void kmem_cache_set_non_kernel(struct kmem_cache *, unsigned int); | |
6573 | void mt_set_non_kernel(unsigned int val) | |
6574 | { | |
6575 | kmem_cache_set_non_kernel(maple_node_cache, val); | |
6576 | } | |
6577 | ||
6578 | extern unsigned long kmem_cache_get_alloc(struct kmem_cache *); | |
6579 | unsigned long mt_get_alloc_size(void) | |
6580 | { | |
6581 | return kmem_cache_get_alloc(maple_node_cache); | |
6582 | } | |
6583 | ||
6584 | extern void kmem_cache_zero_nr_tallocated(struct kmem_cache *); | |
6585 | void mt_zero_nr_tallocated(void) | |
6586 | { | |
6587 | kmem_cache_zero_nr_tallocated(maple_node_cache); | |
6588 | } | |
6589 | ||
6590 | extern unsigned int kmem_cache_nr_tallocated(struct kmem_cache *); | |
6591 | unsigned int mt_nr_tallocated(void) | |
6592 | { | |
6593 | return kmem_cache_nr_tallocated(maple_node_cache); | |
6594 | } | |
6595 | ||
6596 | extern unsigned int kmem_cache_nr_allocated(struct kmem_cache *); | |
6597 | unsigned int mt_nr_allocated(void) | |
6598 | { | |
6599 | return kmem_cache_nr_allocated(maple_node_cache); | |
6600 | } | |
6601 | ||
6602 | /* | |
6603 | * mas_dead_node() - Check if the maple state is pointing to a dead node. | |
6604 | * @mas: The maple state | |
6605 | * @index: The index to restore in @mas. | |
6606 | * | |
6607 | * Used in test code. | |
6608 | * Return: 1 if @mas has been reset to MAS_START, 0 otherwise. | |
6609 | */ | |
6610 | static inline int mas_dead_node(struct ma_state *mas, unsigned long index) | |
6611 | { | |
6612 | if (unlikely(!mas_searchable(mas) || mas_is_start(mas))) | |
6613 | return 0; | |
6614 | ||
6615 | if (likely(!mte_dead_node(mas->node))) | |
6616 | return 0; | |
6617 | ||
6618 | mas_rewalk(mas, index); | |
6619 | return 1; | |
6620 | } | |
54a611b6 | 6621 | |
120b1162 LH |
6622 | void mt_cache_shrink(void) |
6623 | { | |
6624 | } | |
6625 | #else | |
6626 | /* | |
6627 | * mt_cache_shrink() - For testing, don't use this. | |
6628 | * | |
6629 | * Certain testcases can trigger an OOM when combined with other memory | |
6630 | * debugging configuration options. This function is used to reduce the | |
6631 | * possibility of an out of memory even due to kmem_cache objects remaining | |
6632 | * around for longer than usual. | |
6633 | */ | |
6634 | void mt_cache_shrink(void) | |
6635 | { | |
6636 | kmem_cache_shrink(maple_node_cache); | |
6637 | ||
6638 | } | |
6639 | EXPORT_SYMBOL_GPL(mt_cache_shrink); | |
6640 | ||
6641 | #endif /* not defined __KERNEL__ */ | |
54a611b6 LH |
6642 | /* |
6643 | * mas_get_slot() - Get the entry in the maple state node stored at @offset. | |
6644 | * @mas: The maple state | |
6645 | * @offset: The offset into the slot array to fetch. | |
6646 | * | |
6647 | * Return: The entry stored at @offset. | |
6648 | */ | |
6649 | static inline struct maple_enode *mas_get_slot(struct ma_state *mas, | |
6650 | unsigned char offset) | |
6651 | { | |
6652 | return mas_slot(mas, ma_slots(mas_mn(mas), mte_node_type(mas->node)), | |
6653 | offset); | |
6654 | } | |
6655 | ||
6656 | ||
6657 | /* | |
6658 | * mas_first_entry() - Go the first leaf and find the first entry. | |
6659 | * @mas: the maple state. | |
6660 | * @limit: the maximum index to check. | |
6661 | * @*r_start: Pointer to set to the range start. | |
6662 | * | |
6663 | * Sets mas->offset to the offset of the entry, r_start to the range minimum. | |
6664 | * | |
6665 | * Return: The first entry or MAS_NONE. | |
6666 | */ | |
6667 | static inline void *mas_first_entry(struct ma_state *mas, struct maple_node *mn, | |
6668 | unsigned long limit, enum maple_type mt) | |
6669 | ||
6670 | { | |
6671 | unsigned long max; | |
6672 | unsigned long *pivots; | |
6673 | void __rcu **slots; | |
6674 | void *entry = NULL; | |
6675 | ||
6676 | mas->index = mas->min; | |
6677 | if (mas->index > limit) | |
6678 | goto none; | |
6679 | ||
6680 | max = mas->max; | |
6681 | mas->offset = 0; | |
6682 | while (likely(!ma_is_leaf(mt))) { | |
6683 | MT_BUG_ON(mas->tree, mte_dead_node(mas->node)); | |
6684 | slots = ma_slots(mn, mt); | |
54a611b6 | 6685 | entry = mas_slot(mas, slots, 0); |
39d0bd86 | 6686 | pivots = ma_pivots(mn, mt); |
54a611b6 LH |
6687 | if (unlikely(ma_dead_node(mn))) |
6688 | return NULL; | |
39d0bd86 | 6689 | max = pivots[0]; |
54a611b6 LH |
6690 | mas->node = entry; |
6691 | mn = mas_mn(mas); | |
6692 | mt = mte_node_type(mas->node); | |
6693 | } | |
6694 | MT_BUG_ON(mas->tree, mte_dead_node(mas->node)); | |
6695 | ||
6696 | mas->max = max; | |
6697 | slots = ma_slots(mn, mt); | |
6698 | entry = mas_slot(mas, slots, 0); | |
6699 | if (unlikely(ma_dead_node(mn))) | |
6700 | return NULL; | |
6701 | ||
6702 | /* Slot 0 or 1 must be set */ | |
6703 | if (mas->index > limit) | |
6704 | goto none; | |
6705 | ||
6706 | if (likely(entry)) | |
6707 | return entry; | |
6708 | ||
54a611b6 LH |
6709 | mas->offset = 1; |
6710 | entry = mas_slot(mas, slots, 1); | |
39d0bd86 | 6711 | pivots = ma_pivots(mn, mt); |
54a611b6 LH |
6712 | if (unlikely(ma_dead_node(mn))) |
6713 | return NULL; | |
6714 | ||
39d0bd86 | 6715 | mas->index = pivots[0] + 1; |
54a611b6 LH |
6716 | if (mas->index > limit) |
6717 | goto none; | |
6718 | ||
6719 | if (likely(entry)) | |
6720 | return entry; | |
6721 | ||
6722 | none: | |
6723 | if (likely(!ma_dead_node(mn))) | |
6724 | mas->node = MAS_NONE; | |
6725 | return NULL; | |
6726 | } | |
6727 | ||
6728 | /* Depth first search, post-order */ | |
6729 | static void mas_dfs_postorder(struct ma_state *mas, unsigned long max) | |
6730 | { | |
6731 | ||
6732 | struct maple_enode *p = MAS_NONE, *mn = mas->node; | |
6733 | unsigned long p_min, p_max; | |
6734 | ||
6735 | mas_next_node(mas, mas_mn(mas), max); | |
6736 | if (!mas_is_none(mas)) | |
6737 | return; | |
6738 | ||
6739 | if (mte_is_root(mn)) | |
6740 | return; | |
6741 | ||
6742 | mas->node = mn; | |
6743 | mas_ascend(mas); | |
6744 | while (mas->node != MAS_NONE) { | |
6745 | p = mas->node; | |
6746 | p_min = mas->min; | |
6747 | p_max = mas->max; | |
6748 | mas_prev_node(mas, 0); | |
6749 | } | |
6750 | ||
6751 | if (p == MAS_NONE) | |
6752 | return; | |
6753 | ||
6754 | mas->node = p; | |
6755 | mas->max = p_max; | |
6756 | mas->min = p_min; | |
6757 | } | |
6758 | ||
6759 | /* Tree validations */ | |
6760 | static void mt_dump_node(const struct maple_tree *mt, void *entry, | |
6761 | unsigned long min, unsigned long max, unsigned int depth); | |
6762 | static void mt_dump_range(unsigned long min, unsigned long max, | |
6763 | unsigned int depth) | |
6764 | { | |
6765 | static const char spaces[] = " "; | |
6766 | ||
6767 | if (min == max) | |
6768 | pr_info("%.*s%lu: ", depth * 2, spaces, min); | |
6769 | else | |
6770 | pr_info("%.*s%lu-%lu: ", depth * 2, spaces, min, max); | |
6771 | } | |
6772 | ||
6773 | static void mt_dump_entry(void *entry, unsigned long min, unsigned long max, | |
6774 | unsigned int depth) | |
6775 | { | |
6776 | mt_dump_range(min, max, depth); | |
6777 | ||
6778 | if (xa_is_value(entry)) | |
6779 | pr_cont("value %ld (0x%lx) [%p]\n", xa_to_value(entry), | |
6780 | xa_to_value(entry), entry); | |
6781 | else if (xa_is_zero(entry)) | |
6782 | pr_cont("zero (%ld)\n", xa_to_internal(entry)); | |
6783 | else if (mt_is_reserved(entry)) | |
6784 | pr_cont("UNKNOWN ENTRY (%p)\n", entry); | |
6785 | else | |
6786 | pr_cont("%p\n", entry); | |
6787 | } | |
6788 | ||
6789 | static void mt_dump_range64(const struct maple_tree *mt, void *entry, | |
6790 | unsigned long min, unsigned long max, unsigned int depth) | |
6791 | { | |
6792 | struct maple_range_64 *node = &mte_to_node(entry)->mr64; | |
6793 | bool leaf = mte_is_leaf(entry); | |
6794 | unsigned long first = min; | |
6795 | int i; | |
6796 | ||
6797 | pr_cont(" contents: "); | |
6798 | for (i = 0; i < MAPLE_RANGE64_SLOTS - 1; i++) | |
6799 | pr_cont("%p %lu ", node->slot[i], node->pivot[i]); | |
6800 | pr_cont("%p\n", node->slot[i]); | |
6801 | for (i = 0; i < MAPLE_RANGE64_SLOTS; i++) { | |
6802 | unsigned long last = max; | |
6803 | ||
6804 | if (i < (MAPLE_RANGE64_SLOTS - 1)) | |
6805 | last = node->pivot[i]; | |
bd592703 | 6806 | else if (!node->slot[i] && max != mt_node_max(entry)) |
54a611b6 LH |
6807 | break; |
6808 | if (last == 0 && i > 0) | |
6809 | break; | |
6810 | if (leaf) | |
6811 | mt_dump_entry(mt_slot(mt, node->slot, i), | |
6812 | first, last, depth + 1); | |
6813 | else if (node->slot[i]) | |
6814 | mt_dump_node(mt, mt_slot(mt, node->slot, i), | |
6815 | first, last, depth + 1); | |
6816 | ||
6817 | if (last == max) | |
6818 | break; | |
6819 | if (last > max) { | |
6820 | pr_err("node %p last (%lu) > max (%lu) at pivot %d!\n", | |
6821 | node, last, max, i); | |
6822 | break; | |
6823 | } | |
6824 | first = last + 1; | |
6825 | } | |
6826 | } | |
6827 | ||
6828 | static void mt_dump_arange64(const struct maple_tree *mt, void *entry, | |
6829 | unsigned long min, unsigned long max, unsigned int depth) | |
6830 | { | |
6831 | struct maple_arange_64 *node = &mte_to_node(entry)->ma64; | |
6832 | bool leaf = mte_is_leaf(entry); | |
6833 | unsigned long first = min; | |
6834 | int i; | |
6835 | ||
6836 | pr_cont(" contents: "); | |
6837 | for (i = 0; i < MAPLE_ARANGE64_SLOTS; i++) | |
6838 | pr_cont("%lu ", node->gap[i]); | |
6839 | pr_cont("| %02X %02X| ", node->meta.end, node->meta.gap); | |
6840 | for (i = 0; i < MAPLE_ARANGE64_SLOTS - 1; i++) | |
6841 | pr_cont("%p %lu ", node->slot[i], node->pivot[i]); | |
6842 | pr_cont("%p\n", node->slot[i]); | |
6843 | for (i = 0; i < MAPLE_ARANGE64_SLOTS; i++) { | |
6844 | unsigned long last = max; | |
6845 | ||
6846 | if (i < (MAPLE_ARANGE64_SLOTS - 1)) | |
6847 | last = node->pivot[i]; | |
6848 | else if (!node->slot[i]) | |
6849 | break; | |
6850 | if (last == 0 && i > 0) | |
6851 | break; | |
6852 | if (leaf) | |
6853 | mt_dump_entry(mt_slot(mt, node->slot, i), | |
6854 | first, last, depth + 1); | |
6855 | else if (node->slot[i]) | |
6856 | mt_dump_node(mt, mt_slot(mt, node->slot, i), | |
6857 | first, last, depth + 1); | |
6858 | ||
6859 | if (last == max) | |
6860 | break; | |
6861 | if (last > max) { | |
6862 | pr_err("node %p last (%lu) > max (%lu) at pivot %d!\n", | |
6863 | node, last, max, i); | |
6864 | break; | |
6865 | } | |
6866 | first = last + 1; | |
6867 | } | |
6868 | } | |
6869 | ||
6870 | static void mt_dump_node(const struct maple_tree *mt, void *entry, | |
6871 | unsigned long min, unsigned long max, unsigned int depth) | |
6872 | { | |
6873 | struct maple_node *node = mte_to_node(entry); | |
6874 | unsigned int type = mte_node_type(entry); | |
6875 | unsigned int i; | |
6876 | ||
6877 | mt_dump_range(min, max, depth); | |
6878 | ||
6879 | pr_cont("node %p depth %d type %d parent %p", node, depth, type, | |
6880 | node ? node->parent : NULL); | |
6881 | switch (type) { | |
6882 | case maple_dense: | |
6883 | pr_cont("\n"); | |
6884 | for (i = 0; i < MAPLE_NODE_SLOTS; i++) { | |
6885 | if (min + i > max) | |
6886 | pr_cont("OUT OF RANGE: "); | |
6887 | mt_dump_entry(mt_slot(mt, node->slot, i), | |
6888 | min + i, min + i, depth); | |
6889 | } | |
6890 | break; | |
6891 | case maple_leaf_64: | |
6892 | case maple_range_64: | |
6893 | mt_dump_range64(mt, entry, min, max, depth); | |
6894 | break; | |
6895 | case maple_arange_64: | |
6896 | mt_dump_arange64(mt, entry, min, max, depth); | |
6897 | break; | |
6898 | ||
6899 | default: | |
6900 | pr_cont(" UNKNOWN TYPE\n"); | |
6901 | } | |
6902 | } | |
6903 | ||
6904 | void mt_dump(const struct maple_tree *mt) | |
6905 | { | |
6906 | void *entry = rcu_dereference_check(mt->ma_root, mt_locked(mt)); | |
6907 | ||
6908 | pr_info("maple_tree(%p) flags %X, height %u root %p\n", | |
6909 | mt, mt->ma_flags, mt_height(mt), entry); | |
6910 | if (!xa_is_node(entry)) | |
6911 | mt_dump_entry(entry, 0, 0, 0); | |
6912 | else if (entry) | |
bd592703 | 6913 | mt_dump_node(mt, entry, 0, mt_node_max(entry), 0); |
54a611b6 | 6914 | } |
120b1162 | 6915 | EXPORT_SYMBOL_GPL(mt_dump); |
54a611b6 LH |
6916 | |
6917 | /* | |
6918 | * Calculate the maximum gap in a node and check if that's what is reported in | |
6919 | * the parent (unless root). | |
6920 | */ | |
6921 | static void mas_validate_gaps(struct ma_state *mas) | |
6922 | { | |
6923 | struct maple_enode *mte = mas->node; | |
6924 | struct maple_node *p_mn; | |
6925 | unsigned long gap = 0, max_gap = 0; | |
6926 | unsigned long p_end, p_start = mas->min; | |
6927 | unsigned char p_slot; | |
6928 | unsigned long *gaps = NULL; | |
6929 | unsigned long *pivots = ma_pivots(mte_to_node(mte), mte_node_type(mte)); | |
6930 | int i; | |
6931 | ||
6932 | if (ma_is_dense(mte_node_type(mte))) { | |
6933 | for (i = 0; i < mt_slot_count(mte); i++) { | |
6934 | if (mas_get_slot(mas, i)) { | |
6935 | if (gap > max_gap) | |
6936 | max_gap = gap; | |
6937 | gap = 0; | |
6938 | continue; | |
6939 | } | |
6940 | gap++; | |
6941 | } | |
6942 | goto counted; | |
6943 | } | |
6944 | ||
6945 | gaps = ma_gaps(mte_to_node(mte), mte_node_type(mte)); | |
6946 | for (i = 0; i < mt_slot_count(mte); i++) { | |
6947 | p_end = mas_logical_pivot(mas, pivots, i, mte_node_type(mte)); | |
6948 | ||
6949 | if (!gaps) { | |
6950 | if (mas_get_slot(mas, i)) { | |
6951 | gap = 0; | |
6952 | goto not_empty; | |
6953 | } | |
6954 | ||
6955 | gap += p_end - p_start + 1; | |
6956 | } else { | |
6957 | void *entry = mas_get_slot(mas, i); | |
6958 | ||
6959 | gap = gaps[i]; | |
6960 | if (!entry) { | |
6961 | if (gap != p_end - p_start + 1) { | |
6962 | pr_err("%p[%u] -> %p %lu != %lu - %lu + 1\n", | |
6963 | mas_mn(mas), i, | |
6964 | mas_get_slot(mas, i), gap, | |
6965 | p_end, p_start); | |
6966 | mt_dump(mas->tree); | |
6967 | ||
6968 | MT_BUG_ON(mas->tree, | |
6969 | gap != p_end - p_start + 1); | |
6970 | } | |
6971 | } else { | |
6972 | if (gap > p_end - p_start + 1) { | |
6973 | pr_err("%p[%u] %lu >= %lu - %lu + 1 (%lu)\n", | |
6974 | mas_mn(mas), i, gap, p_end, p_start, | |
6975 | p_end - p_start + 1); | |
6976 | MT_BUG_ON(mas->tree, | |
6977 | gap > p_end - p_start + 1); | |
6978 | } | |
6979 | } | |
6980 | } | |
6981 | ||
6982 | if (gap > max_gap) | |
6983 | max_gap = gap; | |
6984 | not_empty: | |
6985 | p_start = p_end + 1; | |
6986 | if (p_end >= mas->max) | |
6987 | break; | |
6988 | } | |
6989 | ||
6990 | counted: | |
6991 | if (mte_is_root(mte)) | |
6992 | return; | |
6993 | ||
6994 | p_slot = mte_parent_slot(mas->node); | |
6995 | p_mn = mte_parent(mte); | |
6996 | MT_BUG_ON(mas->tree, max_gap > mas->max); | |
6997 | if (ma_gaps(p_mn, mas_parent_enum(mas, mte))[p_slot] != max_gap) { | |
6998 | pr_err("gap %p[%u] != %lu\n", p_mn, p_slot, max_gap); | |
6999 | mt_dump(mas->tree); | |
7000 | } | |
7001 | ||
7002 | MT_BUG_ON(mas->tree, | |
7003 | ma_gaps(p_mn, mas_parent_enum(mas, mte))[p_slot] != max_gap); | |
7004 | } | |
7005 | ||
7006 | static void mas_validate_parent_slot(struct ma_state *mas) | |
7007 | { | |
7008 | struct maple_node *parent; | |
7009 | struct maple_enode *node; | |
7010 | enum maple_type p_type = mas_parent_enum(mas, mas->node); | |
7011 | unsigned char p_slot = mte_parent_slot(mas->node); | |
7012 | void __rcu **slots; | |
7013 | int i; | |
7014 | ||
7015 | if (mte_is_root(mas->node)) | |
7016 | return; | |
7017 | ||
7018 | parent = mte_parent(mas->node); | |
7019 | slots = ma_slots(parent, p_type); | |
7020 | MT_BUG_ON(mas->tree, mas_mn(mas) == parent); | |
7021 | ||
7022 | /* Check prev/next parent slot for duplicate node entry */ | |
7023 | ||
7024 | for (i = 0; i < mt_slots[p_type]; i++) { | |
7025 | node = mas_slot(mas, slots, i); | |
7026 | if (i == p_slot) { | |
7027 | if (node != mas->node) | |
7028 | pr_err("parent %p[%u] does not have %p\n", | |
7029 | parent, i, mas_mn(mas)); | |
7030 | MT_BUG_ON(mas->tree, node != mas->node); | |
7031 | } else if (node == mas->node) { | |
7032 | pr_err("Invalid child %p at parent %p[%u] p_slot %u\n", | |
7033 | mas_mn(mas), parent, i, p_slot); | |
7034 | MT_BUG_ON(mas->tree, node == mas->node); | |
7035 | } | |
7036 | } | |
7037 | } | |
7038 | ||
7039 | static void mas_validate_child_slot(struct ma_state *mas) | |
7040 | { | |
7041 | enum maple_type type = mte_node_type(mas->node); | |
7042 | void __rcu **slots = ma_slots(mte_to_node(mas->node), type); | |
7043 | unsigned long *pivots = ma_pivots(mte_to_node(mas->node), type); | |
7044 | struct maple_enode *child; | |
7045 | unsigned char i; | |
7046 | ||
7047 | if (mte_is_leaf(mas->node)) | |
7048 | return; | |
7049 | ||
7050 | for (i = 0; i < mt_slots[type]; i++) { | |
7051 | child = mas_slot(mas, slots, i); | |
7052 | if (!pivots[i] || pivots[i] == mas->max) | |
7053 | break; | |
7054 | ||
7055 | if (!child) | |
7056 | break; | |
7057 | ||
7058 | if (mte_parent_slot(child) != i) { | |
7059 | pr_err("Slot error at %p[%u]: child %p has pslot %u\n", | |
7060 | mas_mn(mas), i, mte_to_node(child), | |
7061 | mte_parent_slot(child)); | |
7062 | MT_BUG_ON(mas->tree, 1); | |
7063 | } | |
7064 | ||
7065 | if (mte_parent(child) != mte_to_node(mas->node)) { | |
7066 | pr_err("child %p has parent %p not %p\n", | |
7067 | mte_to_node(child), mte_parent(child), | |
7068 | mte_to_node(mas->node)); | |
7069 | MT_BUG_ON(mas->tree, 1); | |
7070 | } | |
7071 | } | |
7072 | } | |
7073 | ||
7074 | /* | |
7075 | * Validate all pivots are within mas->min and mas->max. | |
7076 | */ | |
7077 | static void mas_validate_limits(struct ma_state *mas) | |
7078 | { | |
7079 | int i; | |
7080 | unsigned long prev_piv = 0; | |
7081 | enum maple_type type = mte_node_type(mas->node); | |
7082 | void __rcu **slots = ma_slots(mte_to_node(mas->node), type); | |
7083 | unsigned long *pivots = ma_pivots(mas_mn(mas), type); | |
7084 | ||
7085 | /* all limits are fine here. */ | |
7086 | if (mte_is_root(mas->node)) | |
7087 | return; | |
7088 | ||
7089 | for (i = 0; i < mt_slots[type]; i++) { | |
7090 | unsigned long piv; | |
7091 | ||
7092 | piv = mas_safe_pivot(mas, pivots, i, type); | |
7093 | ||
7094 | if (!piv && (i != 0)) | |
7095 | break; | |
7096 | ||
7097 | if (!mte_is_leaf(mas->node)) { | |
7098 | void *entry = mas_slot(mas, slots, i); | |
7099 | ||
7100 | if (!entry) | |
7101 | pr_err("%p[%u] cannot be null\n", | |
7102 | mas_mn(mas), i); | |
7103 | ||
7104 | MT_BUG_ON(mas->tree, !entry); | |
7105 | } | |
7106 | ||
7107 | if (prev_piv > piv) { | |
7108 | pr_err("%p[%u] piv %lu < prev_piv %lu\n", | |
7109 | mas_mn(mas), i, piv, prev_piv); | |
7110 | MT_BUG_ON(mas->tree, piv < prev_piv); | |
7111 | } | |
7112 | ||
7113 | if (piv < mas->min) { | |
7114 | pr_err("%p[%u] %lu < %lu\n", mas_mn(mas), i, | |
7115 | piv, mas->min); | |
7116 | MT_BUG_ON(mas->tree, piv < mas->min); | |
7117 | } | |
7118 | if (piv > mas->max) { | |
7119 | pr_err("%p[%u] %lu > %lu\n", mas_mn(mas), i, | |
7120 | piv, mas->max); | |
7121 | MT_BUG_ON(mas->tree, piv > mas->max); | |
7122 | } | |
7123 | prev_piv = piv; | |
7124 | if (piv == mas->max) | |
7125 | break; | |
7126 | } | |
7127 | for (i += 1; i < mt_slots[type]; i++) { | |
7128 | void *entry = mas_slot(mas, slots, i); | |
7129 | ||
7130 | if (entry && (i != mt_slots[type] - 1)) { | |
7131 | pr_err("%p[%u] should not have entry %p\n", mas_mn(mas), | |
7132 | i, entry); | |
7133 | MT_BUG_ON(mas->tree, entry != NULL); | |
7134 | } | |
7135 | ||
7136 | if (i < mt_pivots[type]) { | |
7137 | unsigned long piv = pivots[i]; | |
7138 | ||
7139 | if (!piv) | |
7140 | continue; | |
7141 | ||
7142 | pr_err("%p[%u] should not have piv %lu\n", | |
7143 | mas_mn(mas), i, piv); | |
7144 | MT_BUG_ON(mas->tree, i < mt_pivots[type] - 1); | |
7145 | } | |
7146 | } | |
7147 | } | |
7148 | ||
7149 | static void mt_validate_nulls(struct maple_tree *mt) | |
7150 | { | |
7151 | void *entry, *last = (void *)1; | |
7152 | unsigned char offset = 0; | |
7153 | void __rcu **slots; | |
7154 | MA_STATE(mas, mt, 0, 0); | |
7155 | ||
7156 | mas_start(&mas); | |
7157 | if (mas_is_none(&mas) || (mas.node == MAS_ROOT)) | |
7158 | return; | |
7159 | ||
7160 | while (!mte_is_leaf(mas.node)) | |
7161 | mas_descend(&mas); | |
7162 | ||
7163 | slots = ma_slots(mte_to_node(mas.node), mte_node_type(mas.node)); | |
7164 | do { | |
7165 | entry = mas_slot(&mas, slots, offset); | |
7166 | if (!last && !entry) { | |
7167 | pr_err("Sequential nulls end at %p[%u]\n", | |
7168 | mas_mn(&mas), offset); | |
7169 | } | |
7170 | MT_BUG_ON(mt, !last && !entry); | |
7171 | last = entry; | |
7172 | if (offset == mas_data_end(&mas)) { | |
7173 | mas_next_node(&mas, mas_mn(&mas), ULONG_MAX); | |
7174 | if (mas_is_none(&mas)) | |
7175 | return; | |
7176 | offset = 0; | |
7177 | slots = ma_slots(mte_to_node(mas.node), | |
7178 | mte_node_type(mas.node)); | |
7179 | } else { | |
7180 | offset++; | |
7181 | } | |
7182 | ||
7183 | } while (!mas_is_none(&mas)); | |
7184 | } | |
7185 | ||
7186 | /* | |
7187 | * validate a maple tree by checking: | |
7188 | * 1. The limits (pivots are within mas->min to mas->max) | |
7189 | * 2. The gap is correctly set in the parents | |
7190 | */ | |
7191 | void mt_validate(struct maple_tree *mt) | |
7192 | { | |
7193 | unsigned char end; | |
7194 | ||
7195 | MA_STATE(mas, mt, 0, 0); | |
7196 | rcu_read_lock(); | |
7197 | mas_start(&mas); | |
7198 | if (!mas_searchable(&mas)) | |
7199 | goto done; | |
7200 | ||
7201 | mas_first_entry(&mas, mas_mn(&mas), ULONG_MAX, mte_node_type(mas.node)); | |
7202 | while (!mas_is_none(&mas)) { | |
7203 | MT_BUG_ON(mas.tree, mte_dead_node(mas.node)); | |
7204 | if (!mte_is_root(mas.node)) { | |
7205 | end = mas_data_end(&mas); | |
7206 | if ((end < mt_min_slot_count(mas.node)) && | |
7207 | (mas.max != ULONG_MAX)) { | |
7208 | pr_err("Invalid size %u of %p\n", end, | |
7209 | mas_mn(&mas)); | |
7210 | MT_BUG_ON(mas.tree, 1); | |
7211 | } | |
7212 | ||
7213 | } | |
7214 | mas_validate_parent_slot(&mas); | |
7215 | mas_validate_child_slot(&mas); | |
7216 | mas_validate_limits(&mas); | |
7217 | if (mt_is_alloc(mt)) | |
7218 | mas_validate_gaps(&mas); | |
7219 | mas_dfs_postorder(&mas, ULONG_MAX); | |
7220 | } | |
7221 | mt_validate_nulls(mt); | |
7222 | done: | |
7223 | rcu_read_unlock(); | |
7224 | ||
7225 | } | |
120b1162 | 7226 | EXPORT_SYMBOL_GPL(mt_validate); |
54a611b6 LH |
7227 | |
7228 | #endif /* CONFIG_DEBUG_MAPLE_TREE */ |