]>
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; | |
5950ada9 | 197 | MAS_BUG_ON(mas, mas->depth > MAPLE_HEIGHT_MAX); |
54a611b6 LH |
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 | ||
f0a1f866 | 243 | static inline bool mas_is_ptr(const struct ma_state *mas) |
54a611b6 LH |
244 | { |
245 | return mas->node == MAS_ROOT; | |
246 | } | |
247 | ||
f0a1f866 | 248 | static inline bool mas_is_start(const struct ma_state *mas) |
54a611b6 LH |
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 | /* | |
afc754c6 | 428 | * mas_parent_type() - Return the maple_type of the parent from the stored |
54a611b6 LH |
429 | * parent type. |
430 | * @mas: The maple state | |
afc754c6 | 431 | * @enode: The maple_enode to extract the parent's enum |
54a611b6 LH |
432 | * Return: The node->parent maple_type |
433 | */ | |
434 | static inline | |
afc754c6 | 435 | enum maple_type mas_parent_type(struct ma_state *mas, struct maple_enode *enode) |
54a611b6 LH |
436 | { |
437 | unsigned long p_type; | |
438 | ||
afc754c6 LH |
439 | p_type = (unsigned long)mte_to_node(enode)->parent; |
440 | if (WARN_ON(p_type & MAPLE_PARENT_ROOT)) | |
441 | return 0; | |
54a611b6 LH |
442 | |
443 | p_type &= MAPLE_NODE_MASK; | |
afc754c6 | 444 | p_type &= ~mte_parent_slot_mask(p_type); |
54a611b6 LH |
445 | switch (p_type) { |
446 | case MAPLE_PARENT_RANGE64: /* or MAPLE_PARENT_ARANGE64 */ | |
afc754c6 | 447 | if (mt_is_alloc(mas->tree)) |
54a611b6 LH |
448 | return maple_arange_64; |
449 | return maple_range_64; | |
450 | } | |
451 | ||
452 | return 0; | |
453 | } | |
454 | ||
54a611b6 | 455 | /* |
bf96715e | 456 | * mas_set_parent() - Set the parent node and encode the slot |
54a611b6 LH |
457 | * @enode: The encoded maple node. |
458 | * @parent: The encoded maple node that is the parent of @enode. | |
459 | * @slot: The slot that @enode resides in @parent. | |
460 | * | |
461 | * Slot number is encoded in the enode->parent bit 3-6 or 2-6, depending on the | |
462 | * parent type. | |
463 | */ | |
464 | static inline | |
bf96715e LH |
465 | void mas_set_parent(struct ma_state *mas, struct maple_enode *enode, |
466 | const struct maple_enode *parent, unsigned char slot) | |
54a611b6 | 467 | { |
831978e3 | 468 | unsigned long val = (unsigned long)parent; |
54a611b6 LH |
469 | unsigned long shift; |
470 | unsigned long type; | |
471 | enum maple_type p_type = mte_node_type(parent); | |
472 | ||
bf96715e LH |
473 | MAS_BUG_ON(mas, p_type == maple_dense); |
474 | MAS_BUG_ON(mas, p_type == maple_leaf_64); | |
54a611b6 LH |
475 | |
476 | switch (p_type) { | |
477 | case maple_range_64: | |
478 | case maple_arange_64: | |
479 | shift = MAPLE_PARENT_SLOT_SHIFT; | |
480 | type = MAPLE_PARENT_RANGE64; | |
481 | break; | |
482 | default: | |
483 | case maple_dense: | |
484 | case maple_leaf_64: | |
485 | shift = type = 0; | |
486 | break; | |
487 | } | |
488 | ||
489 | val &= ~MAPLE_NODE_MASK; /* Clear all node metadata in parent */ | |
490 | val |= (slot << shift) | type; | |
491 | mte_to_node(enode)->parent = ma_parent_ptr(val); | |
492 | } | |
493 | ||
494 | /* | |
495 | * mte_parent_slot() - get the parent slot of @enode. | |
496 | * @enode: The encoded maple node. | |
497 | * | |
498 | * Return: The slot in the parent node where @enode resides. | |
499 | */ | |
500 | static inline unsigned int mte_parent_slot(const struct maple_enode *enode) | |
501 | { | |
831978e3 | 502 | unsigned long val = (unsigned long)mte_to_node(enode)->parent; |
54a611b6 | 503 | |
84fd3e1e | 504 | if (val & MA_ROOT_PARENT) |
54a611b6 LH |
505 | return 0; |
506 | ||
507 | /* | |
508 | * Okay to use MAPLE_PARENT_16B_SLOT_MASK as the last bit will be lost | |
509 | * by shift if the parent shift is MAPLE_PARENT_SLOT_SHIFT | |
510 | */ | |
511 | return (val & MAPLE_PARENT_16B_SLOT_MASK) >> mte_parent_shift(val); | |
512 | } | |
513 | ||
514 | /* | |
515 | * mte_parent() - Get the parent of @node. | |
516 | * @node: The encoded maple node. | |
517 | * | |
518 | * Return: The parent maple node. | |
519 | */ | |
520 | static inline struct maple_node *mte_parent(const struct maple_enode *enode) | |
521 | { | |
522 | return (void *)((unsigned long) | |
523 | (mte_to_node(enode)->parent) & ~MAPLE_NODE_MASK); | |
524 | } | |
525 | ||
526 | /* | |
527 | * ma_dead_node() - check if the @enode is dead. | |
528 | * @enode: The encoded maple node | |
529 | * | |
530 | * Return: true if dead, false otherwise. | |
531 | */ | |
532 | static inline bool ma_dead_node(const struct maple_node *node) | |
533 | { | |
0a2b18d9 | 534 | struct maple_node *parent; |
54a611b6 | 535 | |
0a2b18d9 LH |
536 | /* Do not reorder reads from the node prior to the parent check */ |
537 | smp_rmb(); | |
538 | parent = (void *)((unsigned long) node->parent & ~MAPLE_NODE_MASK); | |
54a611b6 LH |
539 | return (parent == node); |
540 | } | |
39d0bd86 | 541 | |
54a611b6 LH |
542 | /* |
543 | * mte_dead_node() - check if the @enode is dead. | |
544 | * @enode: The encoded maple node | |
545 | * | |
546 | * Return: true if dead, false otherwise. | |
547 | */ | |
548 | static inline bool mte_dead_node(const struct maple_enode *enode) | |
549 | { | |
550 | struct maple_node *parent, *node; | |
551 | ||
552 | node = mte_to_node(enode); | |
0a2b18d9 LH |
553 | /* Do not reorder reads from the node prior to the parent check */ |
554 | smp_rmb(); | |
54a611b6 LH |
555 | parent = mte_parent(enode); |
556 | return (parent == node); | |
557 | } | |
558 | ||
559 | /* | |
560 | * mas_allocated() - Get the number of nodes allocated in a maple state. | |
561 | * @mas: The maple state | |
562 | * | |
563 | * The ma_state alloc member is overloaded to hold a pointer to the first | |
564 | * allocated node or to the number of requested nodes to allocate. If bit 0 is | |
565 | * set, then the alloc contains the number of requested nodes. If there is an | |
566 | * allocated node, then the total allocated nodes is in that node. | |
567 | * | |
568 | * Return: The total number of nodes allocated | |
569 | */ | |
570 | static inline unsigned long mas_allocated(const struct ma_state *mas) | |
571 | { | |
572 | if (!mas->alloc || ((unsigned long)mas->alloc & 0x1)) | |
573 | return 0; | |
574 | ||
575 | return mas->alloc->total; | |
576 | } | |
577 | ||
578 | /* | |
579 | * mas_set_alloc_req() - Set the requested number of allocations. | |
580 | * @mas: the maple state | |
581 | * @count: the number of allocations. | |
582 | * | |
583 | * The requested number of allocations is either in the first allocated node, | |
584 | * located in @mas->alloc->request_count, or directly in @mas->alloc if there is | |
585 | * no allocated node. Set the request either in the node or do the necessary | |
586 | * encoding to store in @mas->alloc directly. | |
587 | */ | |
588 | static inline void mas_set_alloc_req(struct ma_state *mas, unsigned long count) | |
589 | { | |
590 | if (!mas->alloc || ((unsigned long)mas->alloc & 0x1)) { | |
591 | if (!count) | |
592 | mas->alloc = NULL; | |
593 | else | |
594 | mas->alloc = (struct maple_alloc *)(((count) << 1U) | 1U); | |
595 | return; | |
596 | } | |
597 | ||
598 | mas->alloc->request_count = count; | |
599 | } | |
600 | ||
601 | /* | |
602 | * mas_alloc_req() - get the requested number of allocations. | |
603 | * @mas: The maple state | |
604 | * | |
605 | * The alloc count is either stored directly in @mas, or in | |
606 | * @mas->alloc->request_count if there is at least one node allocated. Decode | |
607 | * the request count if it's stored directly in @mas->alloc. | |
608 | * | |
609 | * Return: The allocation request count. | |
610 | */ | |
611 | static inline unsigned int mas_alloc_req(const struct ma_state *mas) | |
612 | { | |
613 | if ((unsigned long)mas->alloc & 0x1) | |
614 | return (unsigned long)(mas->alloc) >> 1; | |
615 | else if (mas->alloc) | |
616 | return mas->alloc->request_count; | |
617 | return 0; | |
618 | } | |
619 | ||
620 | /* | |
621 | * ma_pivots() - Get a pointer to the maple node pivots. | |
622 | * @node - the maple node | |
623 | * @type - the node type | |
624 | * | |
39d0bd86 LH |
625 | * In the event of a dead node, this array may be %NULL |
626 | * | |
54a611b6 LH |
627 | * Return: A pointer to the maple node pivots |
628 | */ | |
629 | static inline unsigned long *ma_pivots(struct maple_node *node, | |
630 | enum maple_type type) | |
631 | { | |
632 | switch (type) { | |
633 | case maple_arange_64: | |
634 | return node->ma64.pivot; | |
635 | case maple_range_64: | |
636 | case maple_leaf_64: | |
637 | return node->mr64.pivot; | |
638 | case maple_dense: | |
639 | return NULL; | |
640 | } | |
641 | return NULL; | |
642 | } | |
643 | ||
644 | /* | |
645 | * ma_gaps() - Get a pointer to the maple node gaps. | |
646 | * @node - the maple node | |
647 | * @type - the node type | |
648 | * | |
649 | * Return: A pointer to the maple node gaps | |
650 | */ | |
651 | static inline unsigned long *ma_gaps(struct maple_node *node, | |
652 | enum maple_type type) | |
653 | { | |
654 | switch (type) { | |
655 | case maple_arange_64: | |
656 | return node->ma64.gap; | |
657 | case maple_range_64: | |
658 | case maple_leaf_64: | |
659 | case maple_dense: | |
660 | return NULL; | |
661 | } | |
662 | return NULL; | |
663 | } | |
664 | ||
665 | /* | |
acd4de60 LH |
666 | * mas_pivot() - Get the pivot at @piv of the maple encoded node. |
667 | * @mas: The maple state. | |
54a611b6 LH |
668 | * @piv: The pivot. |
669 | * | |
670 | * Return: the pivot at @piv of @mn. | |
671 | */ | |
acd4de60 | 672 | static inline unsigned long mas_pivot(struct ma_state *mas, unsigned char piv) |
54a611b6 | 673 | { |
acd4de60 LH |
674 | struct maple_node *node = mas_mn(mas); |
675 | enum maple_type type = mte_node_type(mas->node); | |
54a611b6 | 676 | |
acd4de60 LH |
677 | if (MAS_WARN_ON(mas, piv >= mt_pivots[type])) { |
678 | mas_set_err(mas, -EIO); | |
54a611b6 LH |
679 | return 0; |
680 | } | |
acd4de60 | 681 | |
ab6ef70a | 682 | switch (type) { |
54a611b6 LH |
683 | case maple_arange_64: |
684 | return node->ma64.pivot[piv]; | |
685 | case maple_range_64: | |
686 | case maple_leaf_64: | |
687 | return node->mr64.pivot[piv]; | |
688 | case maple_dense: | |
689 | return 0; | |
690 | } | |
691 | return 0; | |
692 | } | |
693 | ||
694 | /* | |
695 | * mas_safe_pivot() - get the pivot at @piv or mas->max. | |
696 | * @mas: The maple state | |
697 | * @pivots: The pointer to the maple node pivots | |
698 | * @piv: The pivot to fetch | |
699 | * @type: The maple node type | |
700 | * | |
701 | * Return: The pivot at @piv within the limit of the @pivots array, @mas->max | |
702 | * otherwise. | |
703 | */ | |
704 | static inline unsigned long | |
705 | mas_safe_pivot(const struct ma_state *mas, unsigned long *pivots, | |
706 | unsigned char piv, enum maple_type type) | |
707 | { | |
708 | if (piv >= mt_pivots[type]) | |
709 | return mas->max; | |
710 | ||
711 | return pivots[piv]; | |
712 | } | |
713 | ||
714 | /* | |
715 | * mas_safe_min() - Return the minimum for a given offset. | |
716 | * @mas: The maple state | |
717 | * @pivots: The pointer to the maple node pivots | |
718 | * @offset: The offset into the pivot array | |
719 | * | |
720 | * Return: The minimum range value that is contained in @offset. | |
721 | */ | |
722 | static inline unsigned long | |
723 | mas_safe_min(struct ma_state *mas, unsigned long *pivots, unsigned char offset) | |
724 | { | |
725 | if (likely(offset)) | |
726 | return pivots[offset - 1] + 1; | |
727 | ||
728 | return mas->min; | |
729 | } | |
730 | ||
54a611b6 LH |
731 | /* |
732 | * mte_set_pivot() - Set a pivot to a value in an encoded maple node. | |
733 | * @mn: The encoded maple node | |
734 | * @piv: The pivot offset | |
735 | * @val: The value of the pivot | |
736 | */ | |
737 | static inline void mte_set_pivot(struct maple_enode *mn, unsigned char piv, | |
738 | unsigned long val) | |
739 | { | |
740 | struct maple_node *node = mte_to_node(mn); | |
741 | enum maple_type type = mte_node_type(mn); | |
742 | ||
743 | BUG_ON(piv >= mt_pivots[type]); | |
744 | switch (type) { | |
745 | default: | |
746 | case maple_range_64: | |
747 | case maple_leaf_64: | |
748 | node->mr64.pivot[piv] = val; | |
749 | break; | |
750 | case maple_arange_64: | |
751 | node->ma64.pivot[piv] = val; | |
752 | break; | |
753 | case maple_dense: | |
754 | break; | |
755 | } | |
756 | ||
757 | } | |
758 | ||
759 | /* | |
760 | * ma_slots() - Get a pointer to the maple node slots. | |
761 | * @mn: The maple node | |
762 | * @mt: The maple node type | |
763 | * | |
764 | * Return: A pointer to the maple node slots | |
765 | */ | |
766 | static inline void __rcu **ma_slots(struct maple_node *mn, enum maple_type mt) | |
767 | { | |
768 | switch (mt) { | |
769 | default: | |
770 | case maple_arange_64: | |
771 | return mn->ma64.slot; | |
772 | case maple_range_64: | |
773 | case maple_leaf_64: | |
774 | return mn->mr64.slot; | |
775 | case maple_dense: | |
776 | return mn->slot; | |
777 | } | |
778 | } | |
779 | ||
19a462f0 LH |
780 | static inline bool mt_write_locked(const struct maple_tree *mt) |
781 | { | |
782 | return mt_external_lock(mt) ? mt_write_lock_is_held(mt) : | |
783 | lockdep_is_held(&mt->ma_lock); | |
784 | } | |
785 | ||
54a611b6 LH |
786 | static inline bool mt_locked(const struct maple_tree *mt) |
787 | { | |
788 | return mt_external_lock(mt) ? mt_lock_is_held(mt) : | |
789 | lockdep_is_held(&mt->ma_lock); | |
790 | } | |
791 | ||
792 | static inline void *mt_slot(const struct maple_tree *mt, | |
793 | void __rcu **slots, unsigned char offset) | |
794 | { | |
795 | return rcu_dereference_check(slots[offset], mt_locked(mt)); | |
796 | } | |
797 | ||
790e1fa8 LH |
798 | static inline void *mt_slot_locked(struct maple_tree *mt, void __rcu **slots, |
799 | unsigned char offset) | |
800 | { | |
19a462f0 | 801 | return rcu_dereference_protected(slots[offset], mt_write_locked(mt)); |
790e1fa8 | 802 | } |
54a611b6 LH |
803 | /* |
804 | * mas_slot_locked() - Get the slot value when holding the maple tree lock. | |
805 | * @mas: The maple state | |
806 | * @slots: The pointer to the slots | |
807 | * @offset: The offset into the slots array to fetch | |
808 | * | |
809 | * Return: The entry stored in @slots at the @offset. | |
810 | */ | |
811 | static inline void *mas_slot_locked(struct ma_state *mas, void __rcu **slots, | |
812 | unsigned char offset) | |
813 | { | |
790e1fa8 | 814 | return mt_slot_locked(mas->tree, slots, offset); |
54a611b6 LH |
815 | } |
816 | ||
817 | /* | |
818 | * mas_slot() - Get the slot value when not holding the maple tree lock. | |
819 | * @mas: The maple state | |
820 | * @slots: The pointer to the slots | |
821 | * @offset: The offset into the slots array to fetch | |
822 | * | |
823 | * Return: The entry stored in @slots at the @offset | |
824 | */ | |
825 | static inline void *mas_slot(struct ma_state *mas, void __rcu **slots, | |
826 | unsigned char offset) | |
827 | { | |
828 | return mt_slot(mas->tree, slots, offset); | |
829 | } | |
830 | ||
831 | /* | |
832 | * mas_root() - Get the maple tree root. | |
833 | * @mas: The maple state. | |
834 | * | |
835 | * Return: The pointer to the root of the tree | |
836 | */ | |
837 | static inline void *mas_root(struct ma_state *mas) | |
838 | { | |
839 | return rcu_dereference_check(mas->tree->ma_root, mt_locked(mas->tree)); | |
840 | } | |
841 | ||
842 | static inline void *mt_root_locked(struct maple_tree *mt) | |
843 | { | |
19a462f0 | 844 | return rcu_dereference_protected(mt->ma_root, mt_write_locked(mt)); |
54a611b6 LH |
845 | } |
846 | ||
847 | /* | |
848 | * mas_root_locked() - Get the maple tree root when holding the maple tree lock. | |
849 | * @mas: The maple state. | |
850 | * | |
851 | * Return: The pointer to the root of the tree | |
852 | */ | |
853 | static inline void *mas_root_locked(struct ma_state *mas) | |
854 | { | |
855 | return mt_root_locked(mas->tree); | |
856 | } | |
857 | ||
858 | static inline struct maple_metadata *ma_meta(struct maple_node *mn, | |
859 | enum maple_type mt) | |
860 | { | |
861 | switch (mt) { | |
862 | case maple_arange_64: | |
863 | return &mn->ma64.meta; | |
864 | default: | |
865 | return &mn->mr64.meta; | |
866 | } | |
867 | } | |
868 | ||
869 | /* | |
870 | * ma_set_meta() - Set the metadata information of a node. | |
871 | * @mn: The maple node | |
872 | * @mt: The maple node type | |
873 | * @offset: The offset of the highest sub-gap in this node. | |
874 | * @end: The end of the data in this node. | |
875 | */ | |
876 | static inline void ma_set_meta(struct maple_node *mn, enum maple_type mt, | |
877 | unsigned char offset, unsigned char end) | |
878 | { | |
879 | struct maple_metadata *meta = ma_meta(mn, mt); | |
880 | ||
881 | meta->gap = offset; | |
882 | meta->end = end; | |
883 | } | |
884 | ||
2e5b4921 | 885 | /* |
790e1fa8 LH |
886 | * mt_clear_meta() - clear the metadata information of a node, if it exists |
887 | * @mt: The maple tree | |
2e5b4921 | 888 | * @mn: The maple node |
790e1fa8 | 889 | * @type: The maple node type |
2e5b4921 LH |
890 | * @offset: The offset of the highest sub-gap in this node. |
891 | * @end: The end of the data in this node. | |
892 | */ | |
790e1fa8 LH |
893 | static inline void mt_clear_meta(struct maple_tree *mt, struct maple_node *mn, |
894 | enum maple_type type) | |
2e5b4921 LH |
895 | { |
896 | struct maple_metadata *meta; | |
897 | unsigned long *pivots; | |
898 | void __rcu **slots; | |
899 | void *next; | |
900 | ||
790e1fa8 | 901 | switch (type) { |
2e5b4921 LH |
902 | case maple_range_64: |
903 | pivots = mn->mr64.pivot; | |
904 | if (unlikely(pivots[MAPLE_RANGE64_SLOTS - 2])) { | |
905 | slots = mn->mr64.slot; | |
790e1fa8 LH |
906 | next = mt_slot_locked(mt, slots, |
907 | MAPLE_RANGE64_SLOTS - 1); | |
908 | if (unlikely((mte_to_node(next) && | |
909 | mte_node_type(next)))) | |
910 | return; /* no metadata, could be node */ | |
2e5b4921 LH |
911 | } |
912 | fallthrough; | |
913 | case maple_arange_64: | |
790e1fa8 | 914 | meta = ma_meta(mn, type); |
2e5b4921 LH |
915 | break; |
916 | default: | |
917 | return; | |
918 | } | |
919 | ||
920 | meta->gap = 0; | |
921 | meta->end = 0; | |
922 | } | |
923 | ||
54a611b6 LH |
924 | /* |
925 | * ma_meta_end() - Get the data end of a node from the metadata | |
926 | * @mn: The maple node | |
927 | * @mt: The maple node type | |
928 | */ | |
929 | static inline unsigned char ma_meta_end(struct maple_node *mn, | |
930 | enum maple_type mt) | |
931 | { | |
932 | struct maple_metadata *meta = ma_meta(mn, mt); | |
933 | ||
934 | return meta->end; | |
935 | } | |
936 | ||
937 | /* | |
938 | * ma_meta_gap() - Get the largest gap location of a node from the metadata | |
939 | * @mn: The maple node | |
940 | * @mt: The maple node type | |
941 | */ | |
942 | static inline unsigned char ma_meta_gap(struct maple_node *mn, | |
943 | enum maple_type mt) | |
944 | { | |
54a611b6 LH |
945 | return mn->ma64.meta.gap; |
946 | } | |
947 | ||
948 | /* | |
949 | * ma_set_meta_gap() - Set the largest gap location in a nodes metadata | |
950 | * @mn: The maple node | |
951 | * @mn: The maple node type | |
952 | * @offset: The location of the largest gap. | |
953 | */ | |
954 | static inline void ma_set_meta_gap(struct maple_node *mn, enum maple_type mt, | |
955 | unsigned char offset) | |
956 | { | |
957 | ||
958 | struct maple_metadata *meta = ma_meta(mn, mt); | |
959 | ||
960 | meta->gap = offset; | |
961 | } | |
962 | ||
963 | /* | |
964 | * mat_add() - Add a @dead_enode to the ma_topiary of a list of dead nodes. | |
965 | * @mat - the ma_topiary, a linked list of dead nodes. | |
966 | * @dead_enode - the node to be marked as dead and added to the tail of the list | |
967 | * | |
968 | * Add the @dead_enode to the linked list in @mat. | |
969 | */ | |
970 | static inline void mat_add(struct ma_topiary *mat, | |
971 | struct maple_enode *dead_enode) | |
972 | { | |
973 | mte_set_node_dead(dead_enode); | |
974 | mte_to_mat(dead_enode)->next = NULL; | |
975 | if (!mat->tail) { | |
976 | mat->tail = mat->head = dead_enode; | |
977 | return; | |
978 | } | |
979 | ||
980 | mte_to_mat(mat->tail)->next = dead_enode; | |
981 | mat->tail = dead_enode; | |
982 | } | |
983 | ||
984 | static void mte_destroy_walk(struct maple_enode *, struct maple_tree *); | |
985 | static inline void mas_free(struct ma_state *mas, struct maple_enode *used); | |
986 | ||
987 | /* | |
988 | * mas_mat_free() - Free all nodes in a dead list. | |
989 | * @mas - the maple state | |
990 | * @mat - the ma_topiary linked list of dead nodes to free. | |
991 | * | |
992 | * Free walk a dead list. | |
993 | */ | |
994 | static void mas_mat_free(struct ma_state *mas, struct ma_topiary *mat) | |
995 | { | |
996 | struct maple_enode *next; | |
997 | ||
998 | while (mat->head) { | |
999 | next = mte_to_mat(mat->head)->next; | |
1000 | mas_free(mas, mat->head); | |
1001 | mat->head = next; | |
1002 | } | |
1003 | } | |
1004 | ||
1005 | /* | |
1006 | * mas_mat_destroy() - Free all nodes and subtrees in a dead list. | |
1007 | * @mas - the maple state | |
1008 | * @mat - the ma_topiary linked list of dead nodes to free. | |
1009 | * | |
1010 | * Destroy walk a dead list. | |
1011 | */ | |
1012 | static void mas_mat_destroy(struct ma_state *mas, struct ma_topiary *mat) | |
1013 | { | |
1014 | struct maple_enode *next; | |
1015 | ||
1016 | while (mat->head) { | |
1017 | next = mte_to_mat(mat->head)->next; | |
1018 | mte_destroy_walk(mat->head, mat->mtree); | |
1019 | mat->head = next; | |
1020 | } | |
1021 | } | |
1022 | /* | |
1023 | * mas_descend() - Descend into the slot stored in the ma_state. | |
1024 | * @mas - the maple state. | |
1025 | * | |
1026 | * Note: Not RCU safe, only use in write side or debug code. | |
1027 | */ | |
1028 | static inline void mas_descend(struct ma_state *mas) | |
1029 | { | |
1030 | enum maple_type type; | |
1031 | unsigned long *pivots; | |
1032 | struct maple_node *node; | |
1033 | void __rcu **slots; | |
1034 | ||
1035 | node = mas_mn(mas); | |
1036 | type = mte_node_type(mas->node); | |
1037 | pivots = ma_pivots(node, type); | |
1038 | slots = ma_slots(node, type); | |
1039 | ||
1040 | if (mas->offset) | |
1041 | mas->min = pivots[mas->offset - 1] + 1; | |
1042 | mas->max = mas_safe_pivot(mas, pivots, mas->offset, type); | |
1043 | mas->node = mas_slot(mas, slots, mas->offset); | |
1044 | } | |
1045 | ||
1046 | /* | |
1047 | * mte_set_gap() - Set a maple node gap. | |
1048 | * @mn: The encoded maple node | |
1049 | * @gap: The offset of the gap to set | |
1050 | * @val: The gap value | |
1051 | */ | |
1052 | static inline void mte_set_gap(const struct maple_enode *mn, | |
1053 | unsigned char gap, unsigned long val) | |
1054 | { | |
1055 | switch (mte_node_type(mn)) { | |
1056 | default: | |
1057 | break; | |
1058 | case maple_arange_64: | |
1059 | mte_to_node(mn)->ma64.gap[gap] = val; | |
1060 | break; | |
1061 | } | |
1062 | } | |
1063 | ||
1064 | /* | |
1065 | * mas_ascend() - Walk up a level of the tree. | |
1066 | * @mas: The maple state | |
1067 | * | |
1068 | * Sets the @mas->max and @mas->min to the correct values when walking up. This | |
1069 | * may cause several levels of walking up to find the correct min and max. | |
1070 | * May find a dead node which will cause a premature return. | |
1071 | * Return: 1 on dead node, 0 otherwise | |
1072 | */ | |
1073 | static int mas_ascend(struct ma_state *mas) | |
1074 | { | |
1075 | struct maple_enode *p_enode; /* parent enode. */ | |
1076 | struct maple_enode *a_enode; /* ancestor enode. */ | |
1077 | struct maple_node *a_node; /* ancestor node. */ | |
1078 | struct maple_node *p_node; /* parent node. */ | |
1079 | unsigned char a_slot; | |
1080 | enum maple_type a_type; | |
1081 | unsigned long min, max; | |
1082 | unsigned long *pivots; | |
54a611b6 LH |
1083 | bool set_max = false, set_min = false; |
1084 | ||
1085 | a_node = mas_mn(mas); | |
1086 | if (ma_is_root(a_node)) { | |
1087 | mas->offset = 0; | |
1088 | return 0; | |
1089 | } | |
1090 | ||
1091 | p_node = mte_parent(mas->node); | |
1092 | if (unlikely(a_node == p_node)) | |
1093 | return 1; | |
633769c9 | 1094 | |
afc754c6 | 1095 | a_type = mas_parent_type(mas, mas->node); |
633769c9 | 1096 | mas->offset = mte_parent_slot(mas->node); |
54a611b6 LH |
1097 | a_enode = mt_mk_node(p_node, a_type); |
1098 | ||
1099 | /* Check to make sure all parent information is still accurate */ | |
1100 | if (p_node != mte_parent(mas->node)) | |
1101 | return 1; | |
1102 | ||
1103 | mas->node = a_enode; | |
54a611b6 LH |
1104 | |
1105 | if (mte_is_root(a_enode)) { | |
1106 | mas->max = ULONG_MAX; | |
1107 | mas->min = 0; | |
1108 | return 0; | |
1109 | } | |
1110 | ||
633769c9 LH |
1111 | if (!mas->min) |
1112 | set_min = true; | |
1113 | ||
1114 | if (mas->max == ULONG_MAX) | |
1115 | set_max = true; | |
1116 | ||
54a611b6 LH |
1117 | min = 0; |
1118 | max = ULONG_MAX; | |
1119 | do { | |
1120 | p_enode = a_enode; | |
afc754c6 | 1121 | a_type = mas_parent_type(mas, p_enode); |
54a611b6 LH |
1122 | a_node = mte_parent(p_enode); |
1123 | a_slot = mte_parent_slot(p_enode); | |
54a611b6 | 1124 | a_enode = mt_mk_node(a_node, a_type); |
39d0bd86 LH |
1125 | pivots = ma_pivots(a_node, a_type); |
1126 | ||
1127 | if (unlikely(ma_dead_node(a_node))) | |
1128 | return 1; | |
54a611b6 LH |
1129 | |
1130 | if (!set_min && a_slot) { | |
1131 | set_min = true; | |
1132 | min = pivots[a_slot - 1] + 1; | |
1133 | } | |
1134 | ||
1135 | if (!set_max && a_slot < mt_pivots[a_type]) { | |
1136 | set_max = true; | |
1137 | max = pivots[a_slot]; | |
1138 | } | |
1139 | ||
1140 | if (unlikely(ma_dead_node(a_node))) | |
1141 | return 1; | |
1142 | ||
1143 | if (unlikely(ma_is_root(a_node))) | |
1144 | break; | |
1145 | ||
1146 | } while (!set_min || !set_max); | |
1147 | ||
1148 | mas->max = max; | |
1149 | mas->min = min; | |
1150 | return 0; | |
1151 | } | |
1152 | ||
1153 | /* | |
1154 | * mas_pop_node() - Get a previously allocated maple node from the maple state. | |
1155 | * @mas: The maple state | |
1156 | * | |
1157 | * Return: A pointer to a maple node. | |
1158 | */ | |
1159 | static inline struct maple_node *mas_pop_node(struct ma_state *mas) | |
1160 | { | |
1161 | struct maple_alloc *ret, *node = mas->alloc; | |
1162 | unsigned long total = mas_allocated(mas); | |
541e06b7 | 1163 | unsigned int req = mas_alloc_req(mas); |
54a611b6 LH |
1164 | |
1165 | /* nothing or a request pending. */ | |
541e06b7 | 1166 | if (WARN_ON(!total)) |
54a611b6 LH |
1167 | return NULL; |
1168 | ||
1169 | if (total == 1) { | |
1170 | /* single allocation in this ma_state */ | |
1171 | mas->alloc = NULL; | |
1172 | ret = node; | |
1173 | goto single_node; | |
1174 | } | |
1175 | ||
541e06b7 | 1176 | if (node->node_count == 1) { |
54a611b6 LH |
1177 | /* Single allocation in this node. */ |
1178 | mas->alloc = node->slot[0]; | |
54a611b6 LH |
1179 | mas->alloc->total = node->total - 1; |
1180 | ret = node; | |
1181 | goto new_head; | |
1182 | } | |
54a611b6 | 1183 | node->total--; |
541e06b7 LH |
1184 | ret = node->slot[--node->node_count]; |
1185 | node->slot[node->node_count] = NULL; | |
54a611b6 LH |
1186 | |
1187 | single_node: | |
1188 | new_head: | |
541e06b7 LH |
1189 | if (req) { |
1190 | req++; | |
1191 | mas_set_alloc_req(mas, req); | |
54a611b6 | 1192 | } |
541e06b7 LH |
1193 | |
1194 | memset(ret, 0, sizeof(*ret)); | |
54a611b6 LH |
1195 | return (struct maple_node *)ret; |
1196 | } | |
1197 | ||
1198 | /* | |
1199 | * mas_push_node() - Push a node back on the maple state allocation. | |
1200 | * @mas: The maple state | |
1201 | * @used: The used maple node | |
1202 | * | |
1203 | * Stores the maple node back into @mas->alloc for reuse. Updates allocated and | |
1204 | * requested node count as necessary. | |
1205 | */ | |
1206 | static inline void mas_push_node(struct ma_state *mas, struct maple_node *used) | |
1207 | { | |
1208 | struct maple_alloc *reuse = (struct maple_alloc *)used; | |
1209 | struct maple_alloc *head = mas->alloc; | |
1210 | unsigned long count; | |
1211 | unsigned int requested = mas_alloc_req(mas); | |
1212 | ||
54a611b6 LH |
1213 | count = mas_allocated(mas); |
1214 | ||
541e06b7 LH |
1215 | reuse->request_count = 0; |
1216 | reuse->node_count = 0; | |
1217 | if (count && (head->node_count < MAPLE_ALLOC_SLOTS)) { | |
1218 | head->slot[head->node_count++] = reuse; | |
54a611b6 LH |
1219 | head->total++; |
1220 | goto done; | |
1221 | } | |
1222 | ||
1223 | reuse->total = 1; | |
1224 | if ((head) && !((unsigned long)head & 0x1)) { | |
54a611b6 | 1225 | reuse->slot[0] = head; |
541e06b7 | 1226 | reuse->node_count = 1; |
54a611b6 LH |
1227 | reuse->total += head->total; |
1228 | } | |
1229 | ||
1230 | mas->alloc = reuse; | |
1231 | done: | |
1232 | if (requested > 1) | |
1233 | mas_set_alloc_req(mas, requested - 1); | |
1234 | } | |
1235 | ||
1236 | /* | |
1237 | * mas_alloc_nodes() - Allocate nodes into a maple state | |
1238 | * @mas: The maple state | |
1239 | * @gfp: The GFP Flags | |
1240 | */ | |
1241 | static inline void mas_alloc_nodes(struct ma_state *mas, gfp_t gfp) | |
1242 | { | |
1243 | struct maple_alloc *node; | |
54a611b6 | 1244 | unsigned long allocated = mas_allocated(mas); |
54a611b6 LH |
1245 | unsigned int requested = mas_alloc_req(mas); |
1246 | unsigned int count; | |
1247 | void **slots = NULL; | |
1248 | unsigned int max_req = 0; | |
1249 | ||
1250 | if (!requested) | |
1251 | return; | |
1252 | ||
1253 | mas_set_alloc_req(mas, 0); | |
1254 | if (mas->mas_flags & MA_STATE_PREALLOC) { | |
1255 | if (allocated) | |
1256 | return; | |
1257 | WARN_ON(!allocated); | |
1258 | } | |
1259 | ||
541e06b7 | 1260 | if (!allocated || mas->alloc->node_count == MAPLE_ALLOC_SLOTS) { |
54a611b6 LH |
1261 | node = (struct maple_alloc *)mt_alloc_one(gfp); |
1262 | if (!node) | |
1263 | goto nomem_one; | |
1264 | ||
541e06b7 | 1265 | if (allocated) { |
54a611b6 | 1266 | node->slot[0] = mas->alloc; |
541e06b7 LH |
1267 | node->node_count = 1; |
1268 | } else { | |
1269 | node->node_count = 0; | |
1270 | } | |
54a611b6 | 1271 | |
54a611b6 | 1272 | mas->alloc = node; |
541e06b7 | 1273 | node->total = ++allocated; |
54a611b6 LH |
1274 | requested--; |
1275 | } | |
1276 | ||
1277 | node = mas->alloc; | |
541e06b7 | 1278 | node->request_count = 0; |
54a611b6 | 1279 | while (requested) { |
1f5f12ec PZ |
1280 | max_req = MAPLE_ALLOC_SLOTS - node->node_count; |
1281 | slots = (void **)&node->slot[node->node_count]; | |
54a611b6 LH |
1282 | max_req = min(requested, max_req); |
1283 | count = mt_alloc_bulk(gfp, max_req, slots); | |
1284 | if (!count) | |
1285 | goto nomem_bulk; | |
1286 | ||
1f5f12ec PZ |
1287 | if (node->node_count == 0) { |
1288 | node->slot[0]->node_count = 0; | |
1289 | node->slot[0]->request_count = 0; | |
1290 | } | |
1291 | ||
54a611b6 | 1292 | node->node_count += count; |
541e06b7 | 1293 | allocated += count; |
c61b3a2b | 1294 | node = node->slot[0]; |
54a611b6 LH |
1295 | requested -= count; |
1296 | } | |
541e06b7 | 1297 | mas->alloc->total = allocated; |
54a611b6 LH |
1298 | return; |
1299 | ||
1300 | nomem_bulk: | |
1301 | /* Clean up potential freed allocations on bulk failure */ | |
1302 | memset(slots, 0, max_req * sizeof(unsigned long)); | |
1303 | nomem_one: | |
1304 | mas_set_alloc_req(mas, requested); | |
1305 | if (mas->alloc && !(((unsigned long)mas->alloc & 0x1))) | |
541e06b7 | 1306 | mas->alloc->total = allocated; |
54a611b6 | 1307 | mas_set_err(mas, -ENOMEM); |
54a611b6 LH |
1308 | } |
1309 | ||
1310 | /* | |
1311 | * mas_free() - Free an encoded maple node | |
1312 | * @mas: The maple state | |
1313 | * @used: The encoded maple node to free. | |
1314 | * | |
1315 | * Uses rcu free if necessary, pushes @used back on the maple state allocations | |
1316 | * otherwise. | |
1317 | */ | |
1318 | static inline void mas_free(struct ma_state *mas, struct maple_enode *used) | |
1319 | { | |
1320 | struct maple_node *tmp = mte_to_node(used); | |
1321 | ||
1322 | if (mt_in_rcu(mas->tree)) | |
1323 | ma_free_rcu(tmp); | |
1324 | else | |
1325 | mas_push_node(mas, tmp); | |
1326 | } | |
1327 | ||
1328 | /* | |
1329 | * mas_node_count() - Check if enough nodes are allocated and request more if | |
1330 | * there is not enough nodes. | |
1331 | * @mas: The maple state | |
1332 | * @count: The number of nodes needed | |
1333 | * @gfp: the gfp flags | |
1334 | */ | |
1335 | static void mas_node_count_gfp(struct ma_state *mas, int count, gfp_t gfp) | |
1336 | { | |
1337 | unsigned long allocated = mas_allocated(mas); | |
1338 | ||
1339 | if (allocated < count) { | |
1340 | mas_set_alloc_req(mas, count - allocated); | |
1341 | mas_alloc_nodes(mas, gfp); | |
1342 | } | |
1343 | } | |
1344 | ||
1345 | /* | |
1346 | * mas_node_count() - Check if enough nodes are allocated and request more if | |
1347 | * there is not enough nodes. | |
1348 | * @mas: The maple state | |
1349 | * @count: The number of nodes needed | |
1350 | * | |
1351 | * Note: Uses GFP_NOWAIT | __GFP_NOWARN for gfp flags. | |
1352 | */ | |
1353 | static void mas_node_count(struct ma_state *mas, int count) | |
1354 | { | |
1355 | return mas_node_count_gfp(mas, count, GFP_NOWAIT | __GFP_NOWARN); | |
1356 | } | |
1357 | ||
1358 | /* | |
1359 | * mas_start() - Sets up maple state for operations. | |
1360 | * @mas: The maple state. | |
1361 | * | |
46b34584 | 1362 | * If mas->node == MAS_START, then set the min, max and depth to |
54a611b6 LH |
1363 | * defaults. |
1364 | * | |
1365 | * Return: | |
1366 | * - If mas->node is an error or not MAS_START, return NULL. | |
1367 | * - If it's an empty tree: NULL & mas->node == MAS_NONE | |
1368 | * - If it's a single entry: The entry & mas->node == MAS_ROOT | |
1369 | * - If it's a tree: NULL & mas->node == safe root node. | |
1370 | */ | |
1371 | static inline struct maple_enode *mas_start(struct ma_state *mas) | |
1372 | { | |
1373 | if (likely(mas_is_start(mas))) { | |
1374 | struct maple_enode *root; | |
1375 | ||
54a611b6 LH |
1376 | mas->min = 0; |
1377 | mas->max = ULONG_MAX; | |
54a611b6 | 1378 | |
a7b92d59 | 1379 | retry: |
d0411860 | 1380 | mas->depth = 0; |
54a611b6 LH |
1381 | root = mas_root(mas); |
1382 | /* Tree with nodes */ | |
1383 | if (likely(xa_is_node(root))) { | |
9bbba563 | 1384 | mas->depth = 1; |
54a611b6 | 1385 | mas->node = mte_safe_root(root); |
46b34584 | 1386 | mas->offset = 0; |
a7b92d59 LH |
1387 | if (mte_dead_node(mas->node)) |
1388 | goto retry; | |
1389 | ||
54a611b6 LH |
1390 | return NULL; |
1391 | } | |
1392 | ||
1393 | /* empty tree */ | |
1394 | if (unlikely(!root)) { | |
46b34584 | 1395 | mas->node = MAS_NONE; |
54a611b6 LH |
1396 | mas->offset = MAPLE_NODE_SLOTS; |
1397 | return NULL; | |
1398 | } | |
1399 | ||
1400 | /* Single entry tree */ | |
1401 | mas->node = MAS_ROOT; | |
1402 | mas->offset = MAPLE_NODE_SLOTS; | |
1403 | ||
1404 | /* Single entry tree. */ | |
1405 | if (mas->index > 0) | |
1406 | return NULL; | |
1407 | ||
1408 | return root; | |
1409 | } | |
1410 | ||
1411 | return NULL; | |
1412 | } | |
1413 | ||
1414 | /* | |
1415 | * ma_data_end() - Find the end of the data in a node. | |
1416 | * @node: The maple node | |
1417 | * @type: The maple node type | |
1418 | * @pivots: The array of pivots in the node | |
1419 | * @max: The maximum value in the node | |
1420 | * | |
1421 | * Uses metadata to find the end of the data when possible. | |
1422 | * Return: The zero indexed last slot with data (may be null). | |
1423 | */ | |
1424 | static inline unsigned char ma_data_end(struct maple_node *node, | |
1425 | enum maple_type type, | |
1426 | unsigned long *pivots, | |
1427 | unsigned long max) | |
1428 | { | |
1429 | unsigned char offset; | |
1430 | ||
39d0bd86 LH |
1431 | if (!pivots) |
1432 | return 0; | |
1433 | ||
54a611b6 LH |
1434 | if (type == maple_arange_64) |
1435 | return ma_meta_end(node, type); | |
1436 | ||
1437 | offset = mt_pivots[type] - 1; | |
1438 | if (likely(!pivots[offset])) | |
1439 | return ma_meta_end(node, type); | |
1440 | ||
1441 | if (likely(pivots[offset] == max)) | |
1442 | return offset; | |
1443 | ||
1444 | return mt_pivots[type]; | |
1445 | } | |
1446 | ||
1447 | /* | |
1448 | * mas_data_end() - Find the end of the data (slot). | |
1449 | * @mas: the maple state | |
1450 | * | |
1451 | * This method is optimized to check the metadata of a node if the node type | |
1452 | * supports data end metadata. | |
1453 | * | |
1454 | * Return: The zero indexed last slot with data (may be null). | |
1455 | */ | |
1456 | static inline unsigned char mas_data_end(struct ma_state *mas) | |
1457 | { | |
1458 | enum maple_type type; | |
1459 | struct maple_node *node; | |
1460 | unsigned char offset; | |
1461 | unsigned long *pivots; | |
1462 | ||
1463 | type = mte_node_type(mas->node); | |
1464 | node = mas_mn(mas); | |
1465 | if (type == maple_arange_64) | |
1466 | return ma_meta_end(node, type); | |
1467 | ||
1468 | pivots = ma_pivots(node, type); | |
39d0bd86 LH |
1469 | if (unlikely(ma_dead_node(node))) |
1470 | return 0; | |
1471 | ||
54a611b6 LH |
1472 | offset = mt_pivots[type] - 1; |
1473 | if (likely(!pivots[offset])) | |
1474 | return ma_meta_end(node, type); | |
1475 | ||
1476 | if (likely(pivots[offset] == mas->max)) | |
1477 | return offset; | |
1478 | ||
1479 | return mt_pivots[type]; | |
1480 | } | |
1481 | ||
1482 | /* | |
1483 | * mas_leaf_max_gap() - Returns the largest gap in a leaf node | |
1484 | * @mas - the maple state | |
1485 | * | |
1486 | * Return: The maximum gap in the leaf. | |
1487 | */ | |
1488 | static unsigned long mas_leaf_max_gap(struct ma_state *mas) | |
1489 | { | |
1490 | enum maple_type mt; | |
1491 | unsigned long pstart, gap, max_gap; | |
1492 | struct maple_node *mn; | |
1493 | unsigned long *pivots; | |
1494 | void __rcu **slots; | |
1495 | unsigned char i; | |
1496 | unsigned char max_piv; | |
1497 | ||
1498 | mt = mte_node_type(mas->node); | |
1499 | mn = mas_mn(mas); | |
1500 | slots = ma_slots(mn, mt); | |
1501 | max_gap = 0; | |
1502 | if (unlikely(ma_is_dense(mt))) { | |
1503 | gap = 0; | |
1504 | for (i = 0; i < mt_slots[mt]; i++) { | |
1505 | if (slots[i]) { | |
1506 | if (gap > max_gap) | |
1507 | max_gap = gap; | |
1508 | gap = 0; | |
1509 | } else { | |
1510 | gap++; | |
1511 | } | |
1512 | } | |
1513 | if (gap > max_gap) | |
1514 | max_gap = gap; | |
1515 | return max_gap; | |
1516 | } | |
1517 | ||
1518 | /* | |
1519 | * Check the first implied pivot optimizes the loop below and slot 1 may | |
1520 | * be skipped if there is a gap in slot 0. | |
1521 | */ | |
1522 | pivots = ma_pivots(mn, mt); | |
1523 | if (likely(!slots[0])) { | |
1524 | max_gap = pivots[0] - mas->min + 1; | |
1525 | i = 2; | |
1526 | } else { | |
1527 | i = 1; | |
1528 | } | |
1529 | ||
1530 | /* reduce max_piv as the special case is checked before the loop */ | |
1531 | max_piv = ma_data_end(mn, mt, pivots, mas->max) - 1; | |
1532 | /* | |
1533 | * Check end implied pivot which can only be a gap on the right most | |
1534 | * node. | |
1535 | */ | |
1536 | if (unlikely(mas->max == ULONG_MAX) && !slots[max_piv + 1]) { | |
1537 | gap = ULONG_MAX - pivots[max_piv]; | |
1538 | if (gap > max_gap) | |
1539 | max_gap = gap; | |
1540 | } | |
1541 | ||
1542 | for (; i <= max_piv; i++) { | |
1543 | /* data == no gap. */ | |
1544 | if (likely(slots[i])) | |
1545 | continue; | |
1546 | ||
1547 | pstart = pivots[i - 1]; | |
1548 | gap = pivots[i] - pstart; | |
1549 | if (gap > max_gap) | |
1550 | max_gap = gap; | |
1551 | ||
1552 | /* There cannot be two gaps in a row. */ | |
1553 | i++; | |
1554 | } | |
1555 | return max_gap; | |
1556 | } | |
1557 | ||
1558 | /* | |
1559 | * ma_max_gap() - Get the maximum gap in a maple node (non-leaf) | |
1560 | * @node: The maple node | |
1561 | * @gaps: The pointer to the gaps | |
1562 | * @mt: The maple node type | |
1563 | * @*off: Pointer to store the offset location of the gap. | |
1564 | * | |
1565 | * Uses the metadata data end to scan backwards across set gaps. | |
1566 | * | |
1567 | * Return: The maximum gap value | |
1568 | */ | |
1569 | static inline unsigned long | |
1570 | ma_max_gap(struct maple_node *node, unsigned long *gaps, enum maple_type mt, | |
1571 | unsigned char *off) | |
1572 | { | |
1573 | unsigned char offset, i; | |
1574 | unsigned long max_gap = 0; | |
1575 | ||
1576 | i = offset = ma_meta_end(node, mt); | |
1577 | do { | |
1578 | if (gaps[i] > max_gap) { | |
1579 | max_gap = gaps[i]; | |
1580 | offset = i; | |
1581 | } | |
1582 | } while (i--); | |
1583 | ||
1584 | *off = offset; | |
1585 | return max_gap; | |
1586 | } | |
1587 | ||
1588 | /* | |
1589 | * mas_max_gap() - find the largest gap in a non-leaf node and set the slot. | |
1590 | * @mas: The maple state. | |
1591 | * | |
54a611b6 LH |
1592 | * Return: The gap value. |
1593 | */ | |
1594 | static inline unsigned long mas_max_gap(struct ma_state *mas) | |
1595 | { | |
1596 | unsigned long *gaps; | |
1597 | unsigned char offset; | |
1598 | enum maple_type mt; | |
1599 | struct maple_node *node; | |
1600 | ||
1601 | mt = mte_node_type(mas->node); | |
1602 | if (ma_is_leaf(mt)) | |
1603 | return mas_leaf_max_gap(mas); | |
1604 | ||
1605 | node = mas_mn(mas); | |
bec1b51e | 1606 | MAS_BUG_ON(mas, mt != maple_arange_64); |
54a611b6 | 1607 | offset = ma_meta_gap(node, mt); |
54a611b6 LH |
1608 | gaps = ma_gaps(node, mt); |
1609 | return gaps[offset]; | |
1610 | } | |
1611 | ||
1612 | /* | |
1613 | * mas_parent_gap() - Set the parent gap and any gaps above, as needed | |
1614 | * @mas: The maple state | |
1615 | * @offset: The gap offset in the parent to set | |
1616 | * @new: The new gap value. | |
1617 | * | |
1618 | * Set the parent gap then continue to set the gap upwards, using the metadata | |
1619 | * of the parent to see if it is necessary to check the node above. | |
1620 | */ | |
1621 | static inline void mas_parent_gap(struct ma_state *mas, unsigned char offset, | |
1622 | unsigned long new) | |
1623 | { | |
1624 | unsigned long meta_gap = 0; | |
1625 | struct maple_node *pnode; | |
1626 | struct maple_enode *penode; | |
1627 | unsigned long *pgaps; | |
1628 | unsigned char meta_offset; | |
1629 | enum maple_type pmt; | |
1630 | ||
1631 | pnode = mte_parent(mas->node); | |
afc754c6 | 1632 | pmt = mas_parent_type(mas, mas->node); |
54a611b6 LH |
1633 | penode = mt_mk_node(pnode, pmt); |
1634 | pgaps = ma_gaps(pnode, pmt); | |
1635 | ||
1636 | ascend: | |
bec1b51e | 1637 | MAS_BUG_ON(mas, pmt != maple_arange_64); |
54a611b6 | 1638 | meta_offset = ma_meta_gap(pnode, pmt); |
d695c30a | 1639 | meta_gap = pgaps[meta_offset]; |
54a611b6 LH |
1640 | |
1641 | pgaps[offset] = new; | |
1642 | ||
1643 | if (meta_gap == new) | |
1644 | return; | |
1645 | ||
1646 | if (offset != meta_offset) { | |
1647 | if (meta_gap > new) | |
1648 | return; | |
1649 | ||
1650 | ma_set_meta_gap(pnode, pmt, offset); | |
1651 | } else if (new < meta_gap) { | |
54a611b6 LH |
1652 | new = ma_max_gap(pnode, pgaps, pmt, &meta_offset); |
1653 | ma_set_meta_gap(pnode, pmt, meta_offset); | |
1654 | } | |
1655 | ||
1656 | if (ma_is_root(pnode)) | |
1657 | return; | |
1658 | ||
1659 | /* Go to the parent node. */ | |
1660 | pnode = mte_parent(penode); | |
afc754c6 | 1661 | pmt = mas_parent_type(mas, penode); |
54a611b6 LH |
1662 | pgaps = ma_gaps(pnode, pmt); |
1663 | offset = mte_parent_slot(penode); | |
1664 | penode = mt_mk_node(pnode, pmt); | |
1665 | goto ascend; | |
1666 | } | |
1667 | ||
1668 | /* | |
1669 | * mas_update_gap() - Update a nodes gaps and propagate up if necessary. | |
1670 | * @mas - the maple state. | |
1671 | */ | |
1672 | static inline void mas_update_gap(struct ma_state *mas) | |
1673 | { | |
1674 | unsigned char pslot; | |
1675 | unsigned long p_gap; | |
1676 | unsigned long max_gap; | |
1677 | ||
1678 | if (!mt_is_alloc(mas->tree)) | |
1679 | return; | |
1680 | ||
1681 | if (mte_is_root(mas->node)) | |
1682 | return; | |
1683 | ||
1684 | max_gap = mas_max_gap(mas); | |
1685 | ||
1686 | pslot = mte_parent_slot(mas->node); | |
1687 | p_gap = ma_gaps(mte_parent(mas->node), | |
afc754c6 | 1688 | mas_parent_type(mas, mas->node))[pslot]; |
54a611b6 LH |
1689 | |
1690 | if (p_gap != max_gap) | |
1691 | mas_parent_gap(mas, pslot, max_gap); | |
1692 | } | |
1693 | ||
1694 | /* | |
1695 | * mas_adopt_children() - Set the parent pointer of all nodes in @parent to | |
1696 | * @parent with the slot encoded. | |
1697 | * @mas - the maple state (for the tree) | |
1698 | * @parent - the maple encoded node containing the children. | |
1699 | */ | |
1700 | static inline void mas_adopt_children(struct ma_state *mas, | |
1701 | struct maple_enode *parent) | |
1702 | { | |
1703 | enum maple_type type = mte_node_type(parent); | |
1704 | struct maple_node *node = mas_mn(mas); | |
1705 | void __rcu **slots = ma_slots(node, type); | |
1706 | unsigned long *pivots = ma_pivots(node, type); | |
1707 | struct maple_enode *child; | |
1708 | unsigned char offset; | |
1709 | ||
1710 | offset = ma_data_end(node, type, pivots, mas->max); | |
1711 | do { | |
1712 | child = mas_slot_locked(mas, slots, offset); | |
bf96715e | 1713 | mas_set_parent(mas, child, parent, offset); |
54a611b6 LH |
1714 | } while (offset--); |
1715 | } | |
1716 | ||
1717 | /* | |
1718 | * mas_replace() - Replace a maple node in the tree with mas->node. Uses the | |
1719 | * parent encoding to locate the maple node in the tree. | |
1720 | * @mas - the ma_state to use for operations. | |
1721 | * @advanced - boolean to adopt the child nodes and free the old node (false) or | |
1722 | * leave the node (true) and handle the adoption and free elsewhere. | |
1723 | */ | |
1724 | static inline void mas_replace(struct ma_state *mas, bool advanced) | |
14c4b5ab | 1725 | __must_hold(mas->tree->ma_lock) |
54a611b6 LH |
1726 | { |
1727 | struct maple_node *mn = mas_mn(mas); | |
1728 | struct maple_enode *old_enode; | |
1729 | unsigned char offset = 0; | |
1730 | void __rcu **slots = NULL; | |
1731 | ||
1732 | if (ma_is_root(mn)) { | |
1733 | old_enode = mas_root_locked(mas); | |
1734 | } else { | |
1735 | offset = mte_parent_slot(mas->node); | |
1736 | slots = ma_slots(mte_parent(mas->node), | |
afc754c6 | 1737 | mas_parent_type(mas, mas->node)); |
54a611b6 LH |
1738 | old_enode = mas_slot_locked(mas, slots, offset); |
1739 | } | |
1740 | ||
1741 | if (!advanced && !mte_is_leaf(mas->node)) | |
1742 | mas_adopt_children(mas, mas->node); | |
1743 | ||
1744 | if (mte_is_root(mas->node)) { | |
1745 | mn->parent = ma_parent_ptr( | |
1746 | ((unsigned long)mas->tree | MA_ROOT_PARENT)); | |
1747 | rcu_assign_pointer(mas->tree->ma_root, mte_mk_root(mas->node)); | |
1748 | mas_set_height(mas); | |
1749 | } else { | |
1750 | rcu_assign_pointer(slots[offset], mas->node); | |
1751 | } | |
1752 | ||
c13af03d LH |
1753 | if (!advanced) { |
1754 | mte_set_node_dead(old_enode); | |
54a611b6 | 1755 | mas_free(mas, old_enode); |
c13af03d | 1756 | } |
54a611b6 LH |
1757 | } |
1758 | ||
1759 | /* | |
1760 | * mas_new_child() - Find the new child of a node. | |
1761 | * @mas: the maple state | |
1762 | * @child: the maple state to store the child. | |
1763 | */ | |
1764 | static inline bool mas_new_child(struct ma_state *mas, struct ma_state *child) | |
14c4b5ab | 1765 | __must_hold(mas->tree->ma_lock) |
54a611b6 LH |
1766 | { |
1767 | enum maple_type mt; | |
1768 | unsigned char offset; | |
1769 | unsigned char end; | |
1770 | unsigned long *pivots; | |
1771 | struct maple_enode *entry; | |
1772 | struct maple_node *node; | |
1773 | void __rcu **slots; | |
1774 | ||
1775 | mt = mte_node_type(mas->node); | |
1776 | node = mas_mn(mas); | |
1777 | slots = ma_slots(node, mt); | |
1778 | pivots = ma_pivots(node, mt); | |
1779 | end = ma_data_end(node, mt, pivots, mas->max); | |
1780 | for (offset = mas->offset; offset <= end; offset++) { | |
1781 | entry = mas_slot_locked(mas, slots, offset); | |
1782 | if (mte_parent(entry) == node) { | |
1783 | *child = *mas; | |
1784 | mas->offset = offset + 1; | |
1785 | child->offset = offset; | |
1786 | mas_descend(child); | |
1787 | child->offset = 0; | |
1788 | return true; | |
1789 | } | |
1790 | } | |
1791 | return false; | |
1792 | } | |
1793 | ||
1794 | /* | |
1795 | * mab_shift_right() - Shift the data in mab right. Note, does not clean out the | |
1796 | * old data or set b_node->b_end. | |
1797 | * @b_node: the maple_big_node | |
1798 | * @shift: the shift count | |
1799 | */ | |
1800 | static inline void mab_shift_right(struct maple_big_node *b_node, | |
1801 | unsigned char shift) | |
1802 | { | |
1803 | unsigned long size = b_node->b_end * sizeof(unsigned long); | |
1804 | ||
1805 | memmove(b_node->pivot + shift, b_node->pivot, size); | |
1806 | memmove(b_node->slot + shift, b_node->slot, size); | |
1807 | if (b_node->type == maple_arange_64) | |
1808 | memmove(b_node->gap + shift, b_node->gap, size); | |
1809 | } | |
1810 | ||
1811 | /* | |
1812 | * mab_middle_node() - Check if a middle node is needed (unlikely) | |
1813 | * @b_node: the maple_big_node that contains the data. | |
1814 | * @size: the amount of data in the b_node | |
1815 | * @split: the potential split location | |
1816 | * @slot_count: the size that can be stored in a single node being considered. | |
1817 | * | |
1818 | * Return: true if a middle node is required. | |
1819 | */ | |
1820 | static inline bool mab_middle_node(struct maple_big_node *b_node, int split, | |
1821 | unsigned char slot_count) | |
1822 | { | |
1823 | unsigned char size = b_node->b_end; | |
1824 | ||
1825 | if (size >= 2 * slot_count) | |
1826 | return true; | |
1827 | ||
1828 | if (!b_node->slot[split] && (size >= 2 * slot_count - 1)) | |
1829 | return true; | |
1830 | ||
1831 | return false; | |
1832 | } | |
1833 | ||
1834 | /* | |
1835 | * mab_no_null_split() - ensure the split doesn't fall on a NULL | |
1836 | * @b_node: the maple_big_node with the data | |
1837 | * @split: the suggested split location | |
1838 | * @slot_count: the number of slots in the node being considered. | |
1839 | * | |
1840 | * Return: the split location. | |
1841 | */ | |
1842 | static inline int mab_no_null_split(struct maple_big_node *b_node, | |
1843 | unsigned char split, unsigned char slot_count) | |
1844 | { | |
1845 | if (!b_node->slot[split]) { | |
1846 | /* | |
1847 | * If the split is less than the max slot && the right side will | |
1848 | * still be sufficient, then increment the split on NULL. | |
1849 | */ | |
1850 | if ((split < slot_count - 1) && | |
1851 | (b_node->b_end - split) > (mt_min_slots[b_node->type])) | |
1852 | split++; | |
1853 | else | |
1854 | split--; | |
1855 | } | |
1856 | return split; | |
1857 | } | |
1858 | ||
1859 | /* | |
1860 | * mab_calc_split() - Calculate the split location and if there needs to be two | |
1861 | * splits. | |
1862 | * @bn: The maple_big_node with the data | |
1863 | * @mid_split: The second split, if required. 0 otherwise. | |
1864 | * | |
1865 | * Return: The first split location. The middle split is set in @mid_split. | |
1866 | */ | |
1867 | static inline int mab_calc_split(struct ma_state *mas, | |
1868 | struct maple_big_node *bn, unsigned char *mid_split, unsigned long min) | |
1869 | { | |
1870 | unsigned char b_end = bn->b_end; | |
1871 | int split = b_end / 2; /* Assume equal split. */ | |
1872 | unsigned char slot_min, slot_count = mt_slots[bn->type]; | |
1873 | ||
1874 | /* | |
1875 | * To support gap tracking, all NULL entries are kept together and a node cannot | |
1876 | * end on a NULL entry, with the exception of the left-most leaf. The | |
1877 | * limitation means that the split of a node must be checked for this condition | |
1878 | * and be able to put more data in one direction or the other. | |
1879 | */ | |
1880 | if (unlikely((mas->mas_flags & MA_STATE_BULK))) { | |
1881 | *mid_split = 0; | |
1882 | split = b_end - mt_min_slots[bn->type]; | |
1883 | ||
1884 | if (!ma_is_leaf(bn->type)) | |
1885 | return split; | |
1886 | ||
1887 | mas->mas_flags |= MA_STATE_REBALANCE; | |
1888 | if (!bn->slot[split]) | |
1889 | split--; | |
1890 | return split; | |
1891 | } | |
1892 | ||
1893 | /* | |
1894 | * Although extremely rare, it is possible to enter what is known as the 3-way | |
1895 | * split scenario. The 3-way split comes about by means of a store of a range | |
1896 | * that overwrites the end and beginning of two full nodes. The result is a set | |
1897 | * of entries that cannot be stored in 2 nodes. Sometimes, these two nodes can | |
1898 | * also be located in different parent nodes which are also full. This can | |
1899 | * carry upwards all the way to the root in the worst case. | |
1900 | */ | |
1901 | if (unlikely(mab_middle_node(bn, split, slot_count))) { | |
1902 | split = b_end / 3; | |
1903 | *mid_split = split * 2; | |
1904 | } else { | |
1905 | slot_min = mt_min_slots[bn->type]; | |
1906 | ||
1907 | *mid_split = 0; | |
1908 | /* | |
1909 | * Avoid having a range less than the slot count unless it | |
1910 | * causes one node to be deficient. | |
1911 | * NOTE: mt_min_slots is 1 based, b_end and split are zero. | |
1912 | */ | |
5729e06c LH |
1913 | while ((split < slot_count - 1) && |
1914 | ((bn->pivot[split] - min) < slot_count - 1) && | |
1915 | (b_end - split > slot_min)) | |
54a611b6 LH |
1916 | split++; |
1917 | } | |
1918 | ||
1919 | /* Avoid ending a node on a NULL entry */ | |
1920 | split = mab_no_null_split(bn, split, slot_count); | |
54a611b6 | 1921 | |
e11cb683 VY |
1922 | if (unlikely(*mid_split)) |
1923 | *mid_split = mab_no_null_split(bn, *mid_split, slot_count); | |
54a611b6 LH |
1924 | |
1925 | return split; | |
1926 | } | |
1927 | ||
1928 | /* | |
1929 | * mas_mab_cp() - Copy data from a maple state inclusively to a maple_big_node | |
1930 | * and set @b_node->b_end to the next free slot. | |
1931 | * @mas: The maple state | |
1932 | * @mas_start: The starting slot to copy | |
1933 | * @mas_end: The end slot to copy (inclusively) | |
1934 | * @b_node: The maple_big_node to place the data | |
1935 | * @mab_start: The starting location in maple_big_node to store the data. | |
1936 | */ | |
1937 | static inline void mas_mab_cp(struct ma_state *mas, unsigned char mas_start, | |
1938 | unsigned char mas_end, struct maple_big_node *b_node, | |
1939 | unsigned char mab_start) | |
1940 | { | |
1941 | enum maple_type mt; | |
1942 | struct maple_node *node; | |
1943 | void __rcu **slots; | |
1944 | unsigned long *pivots, *gaps; | |
1945 | int i = mas_start, j = mab_start; | |
1946 | unsigned char piv_end; | |
1947 | ||
1948 | node = mas_mn(mas); | |
1949 | mt = mte_node_type(mas->node); | |
1950 | pivots = ma_pivots(node, mt); | |
1951 | if (!i) { | |
1952 | b_node->pivot[j] = pivots[i++]; | |
1953 | if (unlikely(i > mas_end)) | |
1954 | goto complete; | |
1955 | j++; | |
1956 | } | |
1957 | ||
1958 | piv_end = min(mas_end, mt_pivots[mt]); | |
1959 | for (; i < piv_end; i++, j++) { | |
1960 | b_node->pivot[j] = pivots[i]; | |
1961 | if (unlikely(!b_node->pivot[j])) | |
1962 | break; | |
1963 | ||
1964 | if (unlikely(mas->max == b_node->pivot[j])) | |
1965 | goto complete; | |
1966 | } | |
1967 | ||
1968 | if (likely(i <= mas_end)) | |
1969 | b_node->pivot[j] = mas_safe_pivot(mas, pivots, i, mt); | |
1970 | ||
1971 | complete: | |
1972 | b_node->b_end = ++j; | |
1973 | j -= mab_start; | |
1974 | slots = ma_slots(node, mt); | |
1975 | memcpy(b_node->slot + mab_start, slots + mas_start, sizeof(void *) * j); | |
1976 | if (!ma_is_leaf(mt) && mt_is_alloc(mas->tree)) { | |
1977 | gaps = ma_gaps(node, mt); | |
1978 | memcpy(b_node->gap + mab_start, gaps + mas_start, | |
1979 | sizeof(unsigned long) * j); | |
1980 | } | |
1981 | } | |
1982 | ||
1983 | /* | |
1984 | * mas_leaf_set_meta() - Set the metadata of a leaf if possible. | |
1985 | * @mas: The maple state | |
1986 | * @node: The maple node | |
1987 | * @pivots: pointer to the maple node pivots | |
1988 | * @mt: The maple type | |
1989 | * @end: The assumed end | |
1990 | * | |
1991 | * Note, end may be incremented within this function but not modified at the | |
1992 | * source. This is fine since the metadata is the last thing to be stored in a | |
1993 | * node during a write. | |
1994 | */ | |
1995 | static inline void mas_leaf_set_meta(struct ma_state *mas, | |
1996 | struct maple_node *node, unsigned long *pivots, | |
1997 | enum maple_type mt, unsigned char end) | |
1998 | { | |
1999 | /* There is no room for metadata already */ | |
2000 | if (mt_pivots[mt] <= end) | |
2001 | return; | |
2002 | ||
2003 | if (pivots[end] && pivots[end] < mas->max) | |
2004 | end++; | |
2005 | ||
2006 | if (end < mt_slots[mt] - 1) | |
2007 | ma_set_meta(node, mt, 0, end); | |
2008 | } | |
2009 | ||
2010 | /* | |
2011 | * mab_mas_cp() - Copy data from maple_big_node to a maple encoded node. | |
2012 | * @b_node: the maple_big_node that has the data | |
2013 | * @mab_start: the start location in @b_node. | |
2014 | * @mab_end: The end location in @b_node (inclusively) | |
2015 | * @mas: The maple state with the maple encoded node. | |
2016 | */ | |
2017 | static inline void mab_mas_cp(struct maple_big_node *b_node, | |
2018 | unsigned char mab_start, unsigned char mab_end, | |
2019 | struct ma_state *mas, bool new_max) | |
2020 | { | |
2021 | int i, j = 0; | |
2022 | enum maple_type mt = mte_node_type(mas->node); | |
2023 | struct maple_node *node = mte_to_node(mas->node); | |
2024 | void __rcu **slots = ma_slots(node, mt); | |
2025 | unsigned long *pivots = ma_pivots(node, mt); | |
2026 | unsigned long *gaps = NULL; | |
2027 | unsigned char end; | |
2028 | ||
2029 | if (mab_end - mab_start > mt_pivots[mt]) | |
2030 | mab_end--; | |
2031 | ||
2032 | if (!pivots[mt_pivots[mt] - 1]) | |
2033 | slots[mt_pivots[mt]] = NULL; | |
2034 | ||
2035 | i = mab_start; | |
2036 | do { | |
2037 | pivots[j++] = b_node->pivot[i++]; | |
2038 | } while (i <= mab_end && likely(b_node->pivot[i])); | |
2039 | ||
2040 | memcpy(slots, b_node->slot + mab_start, | |
2041 | sizeof(void *) * (i - mab_start)); | |
2042 | ||
2043 | if (new_max) | |
2044 | mas->max = b_node->pivot[i - 1]; | |
2045 | ||
2046 | end = j - 1; | |
2047 | if (likely(!ma_is_leaf(mt) && mt_is_alloc(mas->tree))) { | |
2048 | unsigned long max_gap = 0; | |
d695c30a | 2049 | unsigned char offset = 0; |
54a611b6 LH |
2050 | |
2051 | gaps = ma_gaps(node, mt); | |
2052 | do { | |
2053 | gaps[--j] = b_node->gap[--i]; | |
2054 | if (gaps[j] > max_gap) { | |
2055 | offset = j; | |
2056 | max_gap = gaps[j]; | |
2057 | } | |
2058 | } while (j); | |
2059 | ||
2060 | ma_set_meta(node, mt, offset, end); | |
2061 | } else { | |
2062 | mas_leaf_set_meta(mas, node, pivots, mt, end); | |
2063 | } | |
2064 | } | |
2065 | ||
2066 | /* | |
2067 | * mas_descend_adopt() - Descend through a sub-tree and adopt children. | |
2068 | * @mas: the maple state with the maple encoded node of the sub-tree. | |
2069 | * | |
2070 | * Descend through a sub-tree and adopt children who do not have the correct | |
2071 | * parents set. Follow the parents which have the correct parents as they are | |
2072 | * the new entries which need to be followed to find other incorrectly set | |
2073 | * parents. | |
2074 | */ | |
2075 | static inline void mas_descend_adopt(struct ma_state *mas) | |
2076 | { | |
2077 | struct ma_state list[3], next[3]; | |
2078 | int i, n; | |
2079 | ||
2080 | /* | |
2081 | * At each level there may be up to 3 correct parent pointers which indicates | |
2082 | * the new nodes which need to be walked to find any new nodes at a lower level. | |
2083 | */ | |
2084 | ||
2085 | for (i = 0; i < 3; i++) { | |
2086 | list[i] = *mas; | |
2087 | list[i].offset = 0; | |
2088 | next[i].offset = 0; | |
2089 | } | |
2090 | next[0] = *mas; | |
2091 | ||
2092 | while (!mte_is_leaf(list[0].node)) { | |
2093 | n = 0; | |
2094 | for (i = 0; i < 3; i++) { | |
2095 | if (mas_is_none(&list[i])) | |
2096 | continue; | |
2097 | ||
2098 | if (i && list[i-1].node == list[i].node) | |
2099 | continue; | |
2100 | ||
2101 | while ((n < 3) && (mas_new_child(&list[i], &next[n]))) | |
2102 | n++; | |
2103 | ||
2104 | mas_adopt_children(&list[i], list[i].node); | |
2105 | } | |
2106 | ||
2107 | while (n < 3) | |
2108 | next[n++].node = MAS_NONE; | |
2109 | ||
2110 | /* descend by setting the list to the children */ | |
2111 | for (i = 0; i < 3; i++) | |
2112 | list[i] = next[i]; | |
2113 | } | |
2114 | } | |
2115 | ||
2116 | /* | |
2117 | * mas_bulk_rebalance() - Rebalance the end of a tree after a bulk insert. | |
2118 | * @mas: The maple state | |
2119 | * @end: The maple node end | |
2120 | * @mt: The maple node type | |
2121 | */ | |
2122 | static inline void mas_bulk_rebalance(struct ma_state *mas, unsigned char end, | |
2123 | enum maple_type mt) | |
2124 | { | |
2125 | if (!(mas->mas_flags & MA_STATE_BULK)) | |
2126 | return; | |
2127 | ||
2128 | if (mte_is_root(mas->node)) | |
2129 | return; | |
2130 | ||
2131 | if (end > mt_min_slots[mt]) { | |
2132 | mas->mas_flags &= ~MA_STATE_REBALANCE; | |
2133 | return; | |
2134 | } | |
2135 | } | |
2136 | ||
2137 | /* | |
2138 | * mas_store_b_node() - Store an @entry into the b_node while also copying the | |
2139 | * data from a maple encoded node. | |
2140 | * @wr_mas: the maple write state | |
2141 | * @b_node: the maple_big_node to fill with data | |
2142 | * @offset_end: the offset to end copying | |
2143 | * | |
2144 | * Return: The actual end of the data stored in @b_node | |
2145 | */ | |
44081c77 | 2146 | static noinline_for_kasan void mas_store_b_node(struct ma_wr_state *wr_mas, |
54a611b6 LH |
2147 | struct maple_big_node *b_node, unsigned char offset_end) |
2148 | { | |
2149 | unsigned char slot; | |
2150 | unsigned char b_end; | |
2151 | /* Possible underflow of piv will wrap back to 0 before use. */ | |
2152 | unsigned long piv; | |
2153 | struct ma_state *mas = wr_mas->mas; | |
2154 | ||
2155 | b_node->type = wr_mas->type; | |
2156 | b_end = 0; | |
2157 | slot = mas->offset; | |
2158 | if (slot) { | |
2159 | /* Copy start data up to insert. */ | |
2160 | mas_mab_cp(mas, 0, slot - 1, b_node, 0); | |
2161 | b_end = b_node->b_end; | |
2162 | piv = b_node->pivot[b_end - 1]; | |
2163 | } else | |
2164 | piv = mas->min - 1; | |
2165 | ||
2166 | if (piv + 1 < mas->index) { | |
2167 | /* Handle range starting after old range */ | |
2168 | b_node->slot[b_end] = wr_mas->content; | |
2169 | if (!wr_mas->content) | |
2170 | b_node->gap[b_end] = mas->index - 1 - piv; | |
2171 | b_node->pivot[b_end++] = mas->index - 1; | |
2172 | } | |
2173 | ||
2174 | /* Store the new entry. */ | |
2175 | mas->offset = b_end; | |
2176 | b_node->slot[b_end] = wr_mas->entry; | |
2177 | b_node->pivot[b_end] = mas->last; | |
2178 | ||
2179 | /* Appended. */ | |
2180 | if (mas->last >= mas->max) | |
2181 | goto b_end; | |
2182 | ||
2183 | /* Handle new range ending before old range ends */ | |
29b2681f | 2184 | piv = mas_safe_pivot(mas, wr_mas->pivots, offset_end, wr_mas->type); |
54a611b6 LH |
2185 | if (piv > mas->last) { |
2186 | if (piv == ULONG_MAX) | |
2187 | mas_bulk_rebalance(mas, b_node->b_end, wr_mas->type); | |
2188 | ||
2189 | if (offset_end != slot) | |
2190 | wr_mas->content = mas_slot_locked(mas, wr_mas->slots, | |
2191 | offset_end); | |
2192 | ||
2193 | b_node->slot[++b_end] = wr_mas->content; | |
2194 | if (!wr_mas->content) | |
2195 | b_node->gap[b_end] = piv - mas->last + 1; | |
2196 | b_node->pivot[b_end] = piv; | |
2197 | } | |
2198 | ||
2199 | slot = offset_end + 1; | |
2200 | if (slot > wr_mas->node_end) | |
2201 | goto b_end; | |
2202 | ||
2203 | /* Copy end data to the end of the node. */ | |
2204 | mas_mab_cp(mas, slot, wr_mas->node_end + 1, b_node, ++b_end); | |
2205 | b_node->b_end--; | |
2206 | return; | |
2207 | ||
2208 | b_end: | |
2209 | b_node->b_end = b_end; | |
2210 | } | |
2211 | ||
2212 | /* | |
2213 | * mas_prev_sibling() - Find the previous node with the same parent. | |
2214 | * @mas: the maple state | |
2215 | * | |
2216 | * Return: True if there is a previous sibling, false otherwise. | |
2217 | */ | |
2218 | static inline bool mas_prev_sibling(struct ma_state *mas) | |
2219 | { | |
2220 | unsigned int p_slot = mte_parent_slot(mas->node); | |
2221 | ||
2222 | if (mte_is_root(mas->node)) | |
2223 | return false; | |
2224 | ||
2225 | if (!p_slot) | |
2226 | return false; | |
2227 | ||
2228 | mas_ascend(mas); | |
2229 | mas->offset = p_slot - 1; | |
2230 | mas_descend(mas); | |
2231 | return true; | |
2232 | } | |
2233 | ||
2234 | /* | |
2235 | * mas_next_sibling() - Find the next node with the same parent. | |
2236 | * @mas: the maple state | |
2237 | * | |
2238 | * Return: true if there is a next sibling, false otherwise. | |
2239 | */ | |
2240 | static inline bool mas_next_sibling(struct ma_state *mas) | |
2241 | { | |
2242 | MA_STATE(parent, mas->tree, mas->index, mas->last); | |
2243 | ||
2244 | if (mte_is_root(mas->node)) | |
2245 | return false; | |
2246 | ||
2247 | parent = *mas; | |
2248 | mas_ascend(&parent); | |
2249 | parent.offset = mte_parent_slot(mas->node) + 1; | |
2250 | if (parent.offset > mas_data_end(&parent)) | |
2251 | return false; | |
2252 | ||
2253 | *mas = parent; | |
2254 | mas_descend(mas); | |
2255 | return true; | |
2256 | } | |
2257 | ||
2258 | /* | |
2259 | * mte_node_or_node() - Return the encoded node or MAS_NONE. | |
2260 | * @enode: The encoded maple node. | |
2261 | * | |
2262 | * Shorthand to avoid setting %NULLs in the tree or maple_subtree_state. | |
2263 | * | |
2264 | * Return: @enode or MAS_NONE | |
2265 | */ | |
2266 | static inline struct maple_enode *mte_node_or_none(struct maple_enode *enode) | |
2267 | { | |
2268 | if (enode) | |
2269 | return enode; | |
2270 | ||
2271 | return ma_enode_ptr(MAS_NONE); | |
2272 | } | |
2273 | ||
2274 | /* | |
2275 | * mas_wr_node_walk() - Find the correct offset for the index in the @mas. | |
2276 | * @wr_mas: The maple write state | |
2277 | * | |
2278 | * Uses mas_slot_locked() and does not need to worry about dead nodes. | |
2279 | */ | |
2280 | static inline void mas_wr_node_walk(struct ma_wr_state *wr_mas) | |
2281 | { | |
2282 | struct ma_state *mas = wr_mas->mas; | |
97f7e094 | 2283 | unsigned char count, offset; |
54a611b6 LH |
2284 | |
2285 | if (unlikely(ma_is_dense(wr_mas->type))) { | |
2286 | wr_mas->r_max = wr_mas->r_min = mas->index; | |
2287 | mas->offset = mas->index = mas->min; | |
2288 | return; | |
2289 | } | |
2290 | ||
2291 | wr_mas->node = mas_mn(wr_mas->mas); | |
2292 | wr_mas->pivots = ma_pivots(wr_mas->node, wr_mas->type); | |
2293 | count = wr_mas->node_end = ma_data_end(wr_mas->node, wr_mas->type, | |
2294 | wr_mas->pivots, mas->max); | |
2295 | offset = mas->offset; | |
54a611b6 | 2296 | |
97f7e094 PZ |
2297 | while (offset < count && mas->index > wr_mas->pivots[offset]) |
2298 | offset++; | |
54a611b6 | 2299 | |
97f7e094 PZ |
2300 | wr_mas->r_max = offset < count ? wr_mas->pivots[offset] : mas->max; |
2301 | wr_mas->r_min = mas_safe_min(mas, wr_mas->pivots, offset); | |
54a611b6 LH |
2302 | wr_mas->offset_end = mas->offset = offset; |
2303 | } | |
2304 | ||
2305 | /* | |
2306 | * mas_topiary_range() - Add a range of slots to the topiary. | |
2307 | * @mas: The maple state | |
2308 | * @destroy: The topiary to add the slots (usually destroy) | |
2309 | * @start: The starting slot inclusively | |
2310 | * @end: The end slot inclusively | |
2311 | */ | |
2312 | static inline void mas_topiary_range(struct ma_state *mas, | |
2313 | struct ma_topiary *destroy, unsigned char start, unsigned char end) | |
2314 | { | |
2315 | void __rcu **slots; | |
2316 | unsigned char offset; | |
2317 | ||
4bbd1748 LH |
2318 | MAS_BUG_ON(mas, mte_is_leaf(mas->node)); |
2319 | ||
54a611b6 LH |
2320 | slots = ma_slots(mas_mn(mas), mte_node_type(mas->node)); |
2321 | for (offset = start; offset <= end; offset++) { | |
2322 | struct maple_enode *enode = mas_slot_locked(mas, slots, offset); | |
2323 | ||
2324 | if (mte_dead_node(enode)) | |
2325 | continue; | |
2326 | ||
2327 | mat_add(destroy, enode); | |
2328 | } | |
2329 | } | |
2330 | ||
2331 | /* | |
2332 | * mast_topiary() - Add the portions of the tree to the removal list; either to | |
2333 | * be freed or discarded (destroy walk). | |
2334 | * @mast: The maple_subtree_state. | |
2335 | */ | |
2336 | static inline void mast_topiary(struct maple_subtree_state *mast) | |
2337 | { | |
2338 | MA_WR_STATE(wr_mas, mast->orig_l, NULL); | |
2339 | unsigned char r_start, r_end; | |
2340 | unsigned char l_start, l_end; | |
2341 | void __rcu **l_slots, **r_slots; | |
2342 | ||
2343 | wr_mas.type = mte_node_type(mast->orig_l->node); | |
2344 | mast->orig_l->index = mast->orig_l->last; | |
2345 | mas_wr_node_walk(&wr_mas); | |
2346 | l_start = mast->orig_l->offset + 1; | |
2347 | l_end = mas_data_end(mast->orig_l); | |
2348 | r_start = 0; | |
2349 | r_end = mast->orig_r->offset; | |
2350 | ||
2351 | if (r_end) | |
2352 | r_end--; | |
2353 | ||
2354 | l_slots = ma_slots(mas_mn(mast->orig_l), | |
2355 | mte_node_type(mast->orig_l->node)); | |
2356 | ||
2357 | r_slots = ma_slots(mas_mn(mast->orig_r), | |
2358 | mte_node_type(mast->orig_r->node)); | |
2359 | ||
2360 | if ((l_start < l_end) && | |
2361 | mte_dead_node(mas_slot_locked(mast->orig_l, l_slots, l_start))) { | |
2362 | l_start++; | |
2363 | } | |
2364 | ||
2365 | if (mte_dead_node(mas_slot_locked(mast->orig_r, r_slots, r_end))) { | |
2366 | if (r_end) | |
2367 | r_end--; | |
2368 | } | |
2369 | ||
2370 | if ((l_start > r_end) && (mast->orig_l->node == mast->orig_r->node)) | |
2371 | return; | |
2372 | ||
2373 | /* At the node where left and right sides meet, add the parts between */ | |
2374 | if (mast->orig_l->node == mast->orig_r->node) { | |
2375 | return mas_topiary_range(mast->orig_l, mast->destroy, | |
2376 | l_start, r_end); | |
2377 | } | |
2378 | ||
2379 | /* mast->orig_r is different and consumed. */ | |
2380 | if (mte_is_leaf(mast->orig_r->node)) | |
2381 | return; | |
2382 | ||
2383 | if (mte_dead_node(mas_slot_locked(mast->orig_l, l_slots, l_end))) | |
2384 | l_end--; | |
2385 | ||
2386 | ||
2387 | if (l_start <= l_end) | |
2388 | mas_topiary_range(mast->orig_l, mast->destroy, l_start, l_end); | |
2389 | ||
2390 | if (mte_dead_node(mas_slot_locked(mast->orig_r, r_slots, r_start))) | |
2391 | r_start++; | |
2392 | ||
2393 | if (r_start <= r_end) | |
2394 | mas_topiary_range(mast->orig_r, mast->destroy, 0, r_end); | |
2395 | } | |
2396 | ||
2397 | /* | |
2398 | * mast_rebalance_next() - Rebalance against the next node | |
2399 | * @mast: The maple subtree state | |
2400 | * @old_r: The encoded maple node to the right (next node). | |
2401 | */ | |
2402 | static inline void mast_rebalance_next(struct maple_subtree_state *mast) | |
2403 | { | |
2404 | unsigned char b_end = mast->bn->b_end; | |
2405 | ||
2406 | mas_mab_cp(mast->orig_r, 0, mt_slot_count(mast->orig_r->node), | |
2407 | mast->bn, b_end); | |
2408 | mast->orig_r->last = mast->orig_r->max; | |
2409 | } | |
2410 | ||
2411 | /* | |
2412 | * mast_rebalance_prev() - Rebalance against the previous node | |
2413 | * @mast: The maple subtree state | |
2414 | * @old_l: The encoded maple node to the left (previous node) | |
2415 | */ | |
2416 | static inline void mast_rebalance_prev(struct maple_subtree_state *mast) | |
2417 | { | |
2418 | unsigned char end = mas_data_end(mast->orig_l) + 1; | |
2419 | unsigned char b_end = mast->bn->b_end; | |
2420 | ||
2421 | mab_shift_right(mast->bn, end); | |
2422 | mas_mab_cp(mast->orig_l, 0, end - 1, mast->bn, 0); | |
2423 | mast->l->min = mast->orig_l->min; | |
2424 | mast->orig_l->index = mast->orig_l->min; | |
2425 | mast->bn->b_end = end + b_end; | |
2426 | mast->l->offset += end; | |
2427 | } | |
2428 | ||
2429 | /* | |
2430 | * mast_spanning_rebalance() - Rebalance nodes with nearest neighbour favouring | |
2431 | * the node to the right. Checking the nodes to the right then the left at each | |
2432 | * level upwards until root is reached. Free and destroy as needed. | |
2433 | * Data is copied into the @mast->bn. | |
2434 | * @mast: The maple_subtree_state. | |
2435 | */ | |
2436 | static inline | |
2437 | bool mast_spanning_rebalance(struct maple_subtree_state *mast) | |
2438 | { | |
2439 | struct ma_state r_tmp = *mast->orig_r; | |
2440 | struct ma_state l_tmp = *mast->orig_l; | |
2441 | struct maple_enode *ancestor = NULL; | |
2442 | unsigned char start, end; | |
2443 | unsigned char depth = 0; | |
2444 | ||
2445 | r_tmp = *mast->orig_r; | |
2446 | l_tmp = *mast->orig_l; | |
2447 | do { | |
2448 | mas_ascend(mast->orig_r); | |
2449 | mas_ascend(mast->orig_l); | |
2450 | depth++; | |
2451 | if (!ancestor && | |
2452 | (mast->orig_r->node == mast->orig_l->node)) { | |
2453 | ancestor = mast->orig_r->node; | |
2454 | end = mast->orig_r->offset - 1; | |
2455 | start = mast->orig_l->offset + 1; | |
2456 | } | |
2457 | ||
2458 | if (mast->orig_r->offset < mas_data_end(mast->orig_r)) { | |
2459 | if (!ancestor) { | |
2460 | ancestor = mast->orig_r->node; | |
2461 | start = 0; | |
2462 | } | |
2463 | ||
2464 | mast->orig_r->offset++; | |
2465 | do { | |
2466 | mas_descend(mast->orig_r); | |
2467 | mast->orig_r->offset = 0; | |
2468 | depth--; | |
2469 | } while (depth); | |
2470 | ||
2471 | mast_rebalance_next(mast); | |
2472 | do { | |
2473 | unsigned char l_off = 0; | |
2474 | struct maple_enode *child = r_tmp.node; | |
2475 | ||
2476 | mas_ascend(&r_tmp); | |
2477 | if (ancestor == r_tmp.node) | |
2478 | l_off = start; | |
2479 | ||
2480 | if (r_tmp.offset) | |
2481 | r_tmp.offset--; | |
2482 | ||
2483 | if (l_off < r_tmp.offset) | |
2484 | mas_topiary_range(&r_tmp, mast->destroy, | |
2485 | l_off, r_tmp.offset); | |
2486 | ||
2487 | if (l_tmp.node != child) | |
2488 | mat_add(mast->free, child); | |
2489 | ||
2490 | } while (r_tmp.node != ancestor); | |
2491 | ||
2492 | *mast->orig_l = l_tmp; | |
2493 | return true; | |
2494 | ||
2495 | } else if (mast->orig_l->offset != 0) { | |
2496 | if (!ancestor) { | |
2497 | ancestor = mast->orig_l->node; | |
2498 | end = mas_data_end(mast->orig_l); | |
2499 | } | |
2500 | ||
2501 | mast->orig_l->offset--; | |
2502 | do { | |
2503 | mas_descend(mast->orig_l); | |
2504 | mast->orig_l->offset = | |
2505 | mas_data_end(mast->orig_l); | |
2506 | depth--; | |
2507 | } while (depth); | |
2508 | ||
2509 | mast_rebalance_prev(mast); | |
2510 | do { | |
2511 | unsigned char r_off; | |
2512 | struct maple_enode *child = l_tmp.node; | |
2513 | ||
2514 | mas_ascend(&l_tmp); | |
2515 | if (ancestor == l_tmp.node) | |
2516 | r_off = end; | |
2517 | else | |
2518 | r_off = mas_data_end(&l_tmp); | |
2519 | ||
2520 | if (l_tmp.offset < r_off) | |
2521 | l_tmp.offset++; | |
2522 | ||
2523 | if (l_tmp.offset < r_off) | |
2524 | mas_topiary_range(&l_tmp, mast->destroy, | |
2525 | l_tmp.offset, r_off); | |
2526 | ||
2527 | if (r_tmp.node != child) | |
2528 | mat_add(mast->free, child); | |
2529 | ||
2530 | } while (l_tmp.node != ancestor); | |
2531 | ||
2532 | *mast->orig_r = r_tmp; | |
2533 | return true; | |
2534 | } | |
2535 | } while (!mte_is_root(mast->orig_r->node)); | |
2536 | ||
2537 | *mast->orig_r = r_tmp; | |
2538 | *mast->orig_l = l_tmp; | |
2539 | return false; | |
2540 | } | |
2541 | ||
2542 | /* | |
2543 | * mast_ascend_free() - Add current original maple state nodes to the free list | |
2544 | * and ascend. | |
2545 | * @mast: the maple subtree state. | |
2546 | * | |
2547 | * Ascend the original left and right sides and add the previous nodes to the | |
2548 | * free list. Set the slots to point to the correct location in the new nodes. | |
2549 | */ | |
2550 | static inline void | |
2551 | mast_ascend_free(struct maple_subtree_state *mast) | |
2552 | { | |
2553 | MA_WR_STATE(wr_mas, mast->orig_r, NULL); | |
2554 | struct maple_enode *left = mast->orig_l->node; | |
2555 | struct maple_enode *right = mast->orig_r->node; | |
2556 | ||
2557 | mas_ascend(mast->orig_l); | |
2558 | mas_ascend(mast->orig_r); | |
2559 | mat_add(mast->free, left); | |
2560 | ||
2561 | if (left != right) | |
2562 | mat_add(mast->free, right); | |
2563 | ||
2564 | mast->orig_r->offset = 0; | |
2565 | mast->orig_r->index = mast->r->max; | |
2566 | /* last should be larger than or equal to index */ | |
2567 | if (mast->orig_r->last < mast->orig_r->index) | |
2568 | mast->orig_r->last = mast->orig_r->index; | |
2569 | /* | |
2570 | * The node may not contain the value so set slot to ensure all | |
2571 | * of the nodes contents are freed or destroyed. | |
2572 | */ | |
2573 | wr_mas.type = mte_node_type(mast->orig_r->node); | |
2574 | mas_wr_node_walk(&wr_mas); | |
2575 | /* Set up the left side of things */ | |
2576 | mast->orig_l->offset = 0; | |
2577 | mast->orig_l->index = mast->l->min; | |
2578 | wr_mas.mas = mast->orig_l; | |
2579 | wr_mas.type = mte_node_type(mast->orig_l->node); | |
2580 | mas_wr_node_walk(&wr_mas); | |
2581 | ||
2582 | mast->bn->type = wr_mas.type; | |
2583 | } | |
2584 | ||
2585 | /* | |
2586 | * mas_new_ma_node() - Create and return a new maple node. Helper function. | |
2587 | * @mas: the maple state with the allocations. | |
2588 | * @b_node: the maple_big_node with the type encoding. | |
2589 | * | |
2590 | * Use the node type from the maple_big_node to allocate a new node from the | |
2591 | * ma_state. This function exists mainly for code readability. | |
2592 | * | |
2593 | * Return: A new maple encoded node | |
2594 | */ | |
2595 | static inline struct maple_enode | |
2596 | *mas_new_ma_node(struct ma_state *mas, struct maple_big_node *b_node) | |
2597 | { | |
2598 | return mt_mk_node(ma_mnode_ptr(mas_pop_node(mas)), b_node->type); | |
2599 | } | |
2600 | ||
2601 | /* | |
2602 | * mas_mab_to_node() - Set up right and middle nodes | |
2603 | * | |
2604 | * @mas: the maple state that contains the allocations. | |
2605 | * @b_node: the node which contains the data. | |
2606 | * @left: The pointer which will have the left node | |
2607 | * @right: The pointer which may have the right node | |
2608 | * @middle: the pointer which may have the middle node (rare) | |
2609 | * @mid_split: the split location for the middle node | |
2610 | * | |
2611 | * Return: the split of left. | |
2612 | */ | |
2613 | static inline unsigned char mas_mab_to_node(struct ma_state *mas, | |
2614 | struct maple_big_node *b_node, struct maple_enode **left, | |
2615 | struct maple_enode **right, struct maple_enode **middle, | |
2616 | unsigned char *mid_split, unsigned long min) | |
2617 | { | |
2618 | unsigned char split = 0; | |
2619 | unsigned char slot_count = mt_slots[b_node->type]; | |
2620 | ||
2621 | *left = mas_new_ma_node(mas, b_node); | |
2622 | *right = NULL; | |
2623 | *middle = NULL; | |
2624 | *mid_split = 0; | |
2625 | ||
2626 | if (b_node->b_end < slot_count) { | |
2627 | split = b_node->b_end; | |
2628 | } else { | |
2629 | split = mab_calc_split(mas, b_node, mid_split, min); | |
2630 | *right = mas_new_ma_node(mas, b_node); | |
2631 | } | |
2632 | ||
2633 | if (*mid_split) | |
2634 | *middle = mas_new_ma_node(mas, b_node); | |
2635 | ||
2636 | return split; | |
2637 | ||
2638 | } | |
2639 | ||
2640 | /* | |
2641 | * mab_set_b_end() - Add entry to b_node at b_node->b_end and increment the end | |
2642 | * pointer. | |
2643 | * @b_node - the big node to add the entry | |
2644 | * @mas - the maple state to get the pivot (mas->max) | |
2645 | * @entry - the entry to add, if NULL nothing happens. | |
2646 | */ | |
2647 | static inline void mab_set_b_end(struct maple_big_node *b_node, | |
2648 | struct ma_state *mas, | |
2649 | void *entry) | |
2650 | { | |
2651 | if (!entry) | |
2652 | return; | |
2653 | ||
2654 | b_node->slot[b_node->b_end] = entry; | |
2655 | if (mt_is_alloc(mas->tree)) | |
2656 | b_node->gap[b_node->b_end] = mas_max_gap(mas); | |
2657 | b_node->pivot[b_node->b_end++] = mas->max; | |
2658 | } | |
2659 | ||
2660 | /* | |
2661 | * mas_set_split_parent() - combine_then_separate helper function. Sets the parent | |
2662 | * of @mas->node to either @left or @right, depending on @slot and @split | |
2663 | * | |
2664 | * @mas - the maple state with the node that needs a parent | |
2665 | * @left - possible parent 1 | |
2666 | * @right - possible parent 2 | |
2667 | * @slot - the slot the mas->node was placed | |
2668 | * @split - the split location between @left and @right | |
2669 | */ | |
2670 | static inline void mas_set_split_parent(struct ma_state *mas, | |
2671 | struct maple_enode *left, | |
2672 | struct maple_enode *right, | |
2673 | unsigned char *slot, unsigned char split) | |
2674 | { | |
2675 | if (mas_is_none(mas)) | |
2676 | return; | |
2677 | ||
2678 | if ((*slot) <= split) | |
bf96715e | 2679 | mas_set_parent(mas, mas->node, left, *slot); |
54a611b6 | 2680 | else if (right) |
bf96715e | 2681 | mas_set_parent(mas, mas->node, right, (*slot) - split - 1); |
54a611b6 LH |
2682 | |
2683 | (*slot)++; | |
2684 | } | |
2685 | ||
2686 | /* | |
2687 | * mte_mid_split_check() - Check if the next node passes the mid-split | |
2688 | * @**l: Pointer to left encoded maple node. | |
2689 | * @**m: Pointer to middle encoded maple node. | |
2690 | * @**r: Pointer to right encoded maple node. | |
2691 | * @slot: The offset | |
2692 | * @*split: The split location. | |
2693 | * @mid_split: The middle split. | |
2694 | */ | |
2695 | static inline void mte_mid_split_check(struct maple_enode **l, | |
2696 | struct maple_enode **r, | |
2697 | struct maple_enode *right, | |
2698 | unsigned char slot, | |
2699 | unsigned char *split, | |
2700 | unsigned char mid_split) | |
2701 | { | |
2702 | if (*r == right) | |
2703 | return; | |
2704 | ||
2705 | if (slot < mid_split) | |
2706 | return; | |
2707 | ||
2708 | *l = *r; | |
2709 | *r = right; | |
2710 | *split = mid_split; | |
2711 | } | |
2712 | ||
2713 | /* | |
2714 | * mast_set_split_parents() - Helper function to set three nodes parents. Slot | |
2715 | * is taken from @mast->l. | |
2716 | * @mast - the maple subtree state | |
2717 | * @left - the left node | |
2718 | * @right - the right node | |
2719 | * @split - the split location. | |
2720 | */ | |
2721 | static inline void mast_set_split_parents(struct maple_subtree_state *mast, | |
2722 | struct maple_enode *left, | |
2723 | struct maple_enode *middle, | |
2724 | struct maple_enode *right, | |
2725 | unsigned char split, | |
2726 | unsigned char mid_split) | |
2727 | { | |
2728 | unsigned char slot; | |
2729 | struct maple_enode *l = left; | |
2730 | struct maple_enode *r = right; | |
2731 | ||
2732 | if (mas_is_none(mast->l)) | |
2733 | return; | |
2734 | ||
2735 | if (middle) | |
2736 | r = middle; | |
2737 | ||
2738 | slot = mast->l->offset; | |
2739 | ||
2740 | mte_mid_split_check(&l, &r, right, slot, &split, mid_split); | |
2741 | mas_set_split_parent(mast->l, l, r, &slot, split); | |
2742 | ||
2743 | mte_mid_split_check(&l, &r, right, slot, &split, mid_split); | |
2744 | mas_set_split_parent(mast->m, l, r, &slot, split); | |
2745 | ||
2746 | mte_mid_split_check(&l, &r, right, slot, &split, mid_split); | |
2747 | mas_set_split_parent(mast->r, l, r, &slot, split); | |
2748 | } | |
2749 | ||
2750 | /* | |
2751 | * mas_wmb_replace() - Write memory barrier and replace | |
2752 | * @mas: The maple state | |
2753 | * @free: the maple topiary list of nodes to free | |
2754 | * @destroy: The maple topiary list of nodes to destroy (walk and free) | |
2755 | * | |
2756 | * Updates gap as necessary. | |
2757 | */ | |
2758 | static inline void mas_wmb_replace(struct ma_state *mas, | |
2759 | struct ma_topiary *free, | |
2760 | struct ma_topiary *destroy) | |
2761 | { | |
2762 | /* All nodes must see old data as dead prior to replacing that data */ | |
2763 | smp_wmb(); /* Needed for RCU */ | |
2764 | ||
2765 | /* Insert the new data in the tree */ | |
2766 | mas_replace(mas, true); | |
2767 | ||
2768 | if (!mte_is_leaf(mas->node)) | |
2769 | mas_descend_adopt(mas); | |
2770 | ||
2771 | mas_mat_free(mas, free); | |
2772 | ||
2773 | if (destroy) | |
2774 | mas_mat_destroy(mas, destroy); | |
2775 | ||
2776 | if (mte_is_leaf(mas->node)) | |
2777 | return; | |
2778 | ||
2779 | mas_update_gap(mas); | |
2780 | } | |
2781 | ||
2782 | /* | |
2783 | * mast_new_root() - Set a new tree root during subtree creation | |
2784 | * @mast: The maple subtree state | |
2785 | * @mas: The maple state | |
2786 | */ | |
2787 | static inline void mast_new_root(struct maple_subtree_state *mast, | |
2788 | struct ma_state *mas) | |
2789 | { | |
2790 | mas_mn(mast->l)->parent = | |
2791 | ma_parent_ptr(((unsigned long)mas->tree | MA_ROOT_PARENT)); | |
2792 | if (!mte_dead_node(mast->orig_l->node) && | |
2793 | !mte_is_root(mast->orig_l->node)) { | |
2794 | do { | |
2795 | mast_ascend_free(mast); | |
2796 | mast_topiary(mast); | |
2797 | } while (!mte_is_root(mast->orig_l->node)); | |
2798 | } | |
2799 | if ((mast->orig_l->node != mas->node) && | |
2800 | (mast->l->depth > mas_mt_height(mas))) { | |
2801 | mat_add(mast->free, mas->node); | |
2802 | } | |
2803 | } | |
2804 | ||
2805 | /* | |
2806 | * mast_cp_to_nodes() - Copy data out to nodes. | |
2807 | * @mast: The maple subtree state | |
2808 | * @left: The left encoded maple node | |
2809 | * @middle: The middle encoded maple node | |
2810 | * @right: The right encoded maple node | |
2811 | * @split: The location to split between left and (middle ? middle : right) | |
2812 | * @mid_split: The location to split between middle and right. | |
2813 | */ | |
2814 | static inline void mast_cp_to_nodes(struct maple_subtree_state *mast, | |
2815 | struct maple_enode *left, struct maple_enode *middle, | |
2816 | struct maple_enode *right, unsigned char split, unsigned char mid_split) | |
2817 | { | |
2818 | bool new_lmax = true; | |
2819 | ||
2820 | mast->l->node = mte_node_or_none(left); | |
2821 | mast->m->node = mte_node_or_none(middle); | |
2822 | mast->r->node = mte_node_or_none(right); | |
2823 | ||
2824 | mast->l->min = mast->orig_l->min; | |
2825 | if (split == mast->bn->b_end) { | |
2826 | mast->l->max = mast->orig_r->max; | |
2827 | new_lmax = false; | |
2828 | } | |
2829 | ||
2830 | mab_mas_cp(mast->bn, 0, split, mast->l, new_lmax); | |
2831 | ||
2832 | if (middle) { | |
2833 | mab_mas_cp(mast->bn, 1 + split, mid_split, mast->m, true); | |
2834 | mast->m->min = mast->bn->pivot[split] + 1; | |
2835 | split = mid_split; | |
2836 | } | |
2837 | ||
2838 | mast->r->max = mast->orig_r->max; | |
2839 | if (right) { | |
2840 | mab_mas_cp(mast->bn, 1 + split, mast->bn->b_end, mast->r, false); | |
2841 | mast->r->min = mast->bn->pivot[split] + 1; | |
2842 | } | |
2843 | } | |
2844 | ||
2845 | /* | |
2846 | * mast_combine_cp_left - Copy in the original left side of the tree into the | |
2847 | * combined data set in the maple subtree state big node. | |
2848 | * @mast: The maple subtree state | |
2849 | */ | |
2850 | static inline void mast_combine_cp_left(struct maple_subtree_state *mast) | |
2851 | { | |
2852 | unsigned char l_slot = mast->orig_l->offset; | |
2853 | ||
2854 | if (!l_slot) | |
2855 | return; | |
2856 | ||
2857 | mas_mab_cp(mast->orig_l, 0, l_slot - 1, mast->bn, 0); | |
2858 | } | |
2859 | ||
2860 | /* | |
2861 | * mast_combine_cp_right: Copy in the original right side of the tree into the | |
2862 | * combined data set in the maple subtree state big node. | |
2863 | * @mast: The maple subtree state | |
2864 | */ | |
2865 | static inline void mast_combine_cp_right(struct maple_subtree_state *mast) | |
2866 | { | |
2867 | if (mast->bn->pivot[mast->bn->b_end - 1] >= mast->orig_r->max) | |
2868 | return; | |
2869 | ||
2870 | mas_mab_cp(mast->orig_r, mast->orig_r->offset + 1, | |
2871 | mt_slot_count(mast->orig_r->node), mast->bn, | |
2872 | mast->bn->b_end); | |
2873 | mast->orig_r->last = mast->orig_r->max; | |
2874 | } | |
2875 | ||
2876 | /* | |
2877 | * mast_sufficient: Check if the maple subtree state has enough data in the big | |
2878 | * node to create at least one sufficient node | |
2879 | * @mast: the maple subtree state | |
2880 | */ | |
2881 | static inline bool mast_sufficient(struct maple_subtree_state *mast) | |
2882 | { | |
2883 | if (mast->bn->b_end > mt_min_slot_count(mast->orig_l->node)) | |
2884 | return true; | |
2885 | ||
2886 | return false; | |
2887 | } | |
2888 | ||
2889 | /* | |
2890 | * mast_overflow: Check if there is too much data in the subtree state for a | |
2891 | * single node. | |
2892 | * @mast: The maple subtree state | |
2893 | */ | |
2894 | static inline bool mast_overflow(struct maple_subtree_state *mast) | |
2895 | { | |
2896 | if (mast->bn->b_end >= mt_slot_count(mast->orig_l->node)) | |
2897 | return true; | |
2898 | ||
2899 | return false; | |
2900 | } | |
2901 | ||
2902 | static inline void *mtree_range_walk(struct ma_state *mas) | |
2903 | { | |
2904 | unsigned long *pivots; | |
2905 | unsigned char offset; | |
2906 | struct maple_node *node; | |
2907 | struct maple_enode *next, *last; | |
2908 | enum maple_type type; | |
2909 | void __rcu **slots; | |
2910 | unsigned char end; | |
2911 | unsigned long max, min; | |
2912 | unsigned long prev_max, prev_min; | |
2913 | ||
1b9c9183 LB |
2914 | next = mas->node; |
2915 | min = mas->min; | |
54a611b6 LH |
2916 | max = mas->max; |
2917 | do { | |
2918 | offset = 0; | |
2919 | last = next; | |
2920 | node = mte_to_node(next); | |
2921 | type = mte_node_type(next); | |
2922 | pivots = ma_pivots(node, type); | |
2923 | end = ma_data_end(node, type, pivots, max); | |
2924 | if (unlikely(ma_dead_node(node))) | |
2925 | goto dead_node; | |
2926 | ||
2927 | if (pivots[offset] >= mas->index) { | |
2928 | prev_max = max; | |
2929 | prev_min = min; | |
2930 | max = pivots[offset]; | |
2931 | goto next; | |
2932 | } | |
2933 | ||
2934 | do { | |
2935 | offset++; | |
2936 | } while ((offset < end) && (pivots[offset] < mas->index)); | |
2937 | ||
2938 | prev_min = min; | |
2939 | min = pivots[offset - 1] + 1; | |
2940 | prev_max = max; | |
2941 | if (likely(offset < end && pivots[offset])) | |
2942 | max = pivots[offset]; | |
2943 | ||
2944 | next: | |
2945 | slots = ma_slots(node, type); | |
2946 | next = mt_slot(mas->tree, slots, offset); | |
2947 | if (unlikely(ma_dead_node(node))) | |
2948 | goto dead_node; | |
2949 | } while (!ma_is_leaf(type)); | |
2950 | ||
2951 | mas->offset = offset; | |
2952 | mas->index = min; | |
2953 | mas->last = max; | |
2954 | mas->min = prev_min; | |
2955 | mas->max = prev_max; | |
2956 | mas->node = last; | |
831978e3 | 2957 | return (void *)next; |
54a611b6 LH |
2958 | |
2959 | dead_node: | |
2960 | mas_reset(mas); | |
2961 | return NULL; | |
2962 | } | |
2963 | ||
2964 | /* | |
2965 | * mas_spanning_rebalance() - Rebalance across two nodes which may not be peers. | |
2966 | * @mas: The starting maple state | |
2967 | * @mast: The maple_subtree_state, keeps track of 4 maple states. | |
2968 | * @count: The estimated count of iterations needed. | |
2969 | * | |
2970 | * Follow the tree upwards from @l_mas and @r_mas for @count, or until the root | |
2971 | * is hit. First @b_node is split into two entries which are inserted into the | |
2972 | * next iteration of the loop. @b_node is returned populated with the final | |
2973 | * iteration. @mas is used to obtain allocations. orig_l_mas keeps track of the | |
2974 | * nodes that will remain active by using orig_l_mas->index and orig_l_mas->last | |
2975 | * to account of what has been copied into the new sub-tree. The update of | |
2976 | * orig_l_mas->last is used in mas_consume to find the slots that will need to | |
2977 | * be either freed or destroyed. orig_l_mas->depth keeps track of the height of | |
2978 | * the new sub-tree in case the sub-tree becomes the full tree. | |
2979 | * | |
2980 | * Return: the number of elements in b_node during the last loop. | |
2981 | */ | |
2982 | static int mas_spanning_rebalance(struct ma_state *mas, | |
2983 | struct maple_subtree_state *mast, unsigned char count) | |
2984 | { | |
2985 | unsigned char split, mid_split; | |
2986 | unsigned char slot = 0; | |
2987 | struct maple_enode *left = NULL, *middle = NULL, *right = NULL; | |
2988 | ||
2989 | MA_STATE(l_mas, mas->tree, mas->index, mas->index); | |
2990 | MA_STATE(r_mas, mas->tree, mas->index, mas->last); | |
2991 | MA_STATE(m_mas, mas->tree, mas->index, mas->index); | |
2992 | MA_TOPIARY(free, mas->tree); | |
2993 | MA_TOPIARY(destroy, mas->tree); | |
2994 | ||
2995 | /* | |
2996 | * The tree needs to be rebalanced and leaves need to be kept at the same level. | |
2997 | * Rebalancing is done by use of the ``struct maple_topiary``. | |
2998 | */ | |
2999 | mast->l = &l_mas; | |
3000 | mast->m = &m_mas; | |
3001 | mast->r = &r_mas; | |
3002 | mast->free = &free; | |
3003 | mast->destroy = &destroy; | |
3004 | l_mas.node = r_mas.node = m_mas.node = MAS_NONE; | |
0abb964a LH |
3005 | |
3006 | /* Check if this is not root and has sufficient data. */ | |
3007 | if (((mast->orig_l->min != 0) || (mast->orig_r->max != ULONG_MAX)) && | |
54a611b6 LH |
3008 | unlikely(mast->bn->b_end <= mt_min_slots[mast->bn->type])) |
3009 | mast_spanning_rebalance(mast); | |
3010 | ||
3011 | mast->orig_l->depth = 0; | |
3012 | ||
3013 | /* | |
3014 | * Each level of the tree is examined and balanced, pushing data to the left or | |
3015 | * right, or rebalancing against left or right nodes is employed to avoid | |
3016 | * rippling up the tree to limit the amount of churn. Once a new sub-section of | |
3017 | * the tree is created, there may be a mix of new and old nodes. The old nodes | |
3018 | * will have the incorrect parent pointers and currently be in two trees: the | |
3019 | * original tree and the partially new tree. To remedy the parent pointers in | |
3020 | * the old tree, the new data is swapped into the active tree and a walk down | |
3021 | * the tree is performed and the parent pointers are updated. | |
3022 | * See mas_descend_adopt() for more information.. | |
3023 | */ | |
3024 | while (count--) { | |
3025 | mast->bn->b_end--; | |
3026 | mast->bn->type = mte_node_type(mast->orig_l->node); | |
3027 | split = mas_mab_to_node(mas, mast->bn, &left, &right, &middle, | |
3028 | &mid_split, mast->orig_l->min); | |
3029 | mast_set_split_parents(mast, left, middle, right, split, | |
3030 | mid_split); | |
3031 | mast_cp_to_nodes(mast, left, middle, right, split, mid_split); | |
3032 | ||
3033 | /* | |
3034 | * Copy data from next level in the tree to mast->bn from next | |
3035 | * iteration | |
3036 | */ | |
3037 | memset(mast->bn, 0, sizeof(struct maple_big_node)); | |
3038 | mast->bn->type = mte_node_type(left); | |
3039 | mast->orig_l->depth++; | |
3040 | ||
3041 | /* Root already stored in l->node. */ | |
3042 | if (mas_is_root_limits(mast->l)) | |
3043 | goto new_root; | |
3044 | ||
3045 | mast_ascend_free(mast); | |
3046 | mast_combine_cp_left(mast); | |
3047 | l_mas.offset = mast->bn->b_end; | |
3048 | mab_set_b_end(mast->bn, &l_mas, left); | |
3049 | mab_set_b_end(mast->bn, &m_mas, middle); | |
3050 | mab_set_b_end(mast->bn, &r_mas, right); | |
3051 | ||
3052 | /* Copy anything necessary out of the right node. */ | |
3053 | mast_combine_cp_right(mast); | |
3054 | mast_topiary(mast); | |
3055 | mast->orig_l->last = mast->orig_l->max; | |
3056 | ||
3057 | if (mast_sufficient(mast)) | |
3058 | continue; | |
3059 | ||
3060 | if (mast_overflow(mast)) | |
3061 | continue; | |
3062 | ||
3063 | /* May be a new root stored in mast->bn */ | |
3064 | if (mas_is_root_limits(mast->orig_l)) | |
3065 | break; | |
3066 | ||
3067 | mast_spanning_rebalance(mast); | |
3068 | ||
3069 | /* rebalancing from other nodes may require another loop. */ | |
3070 | if (!count) | |
3071 | count++; | |
3072 | } | |
3073 | ||
3074 | l_mas.node = mt_mk_node(ma_mnode_ptr(mas_pop_node(mas)), | |
3075 | mte_node_type(mast->orig_l->node)); | |
3076 | mast->orig_l->depth++; | |
3077 | mab_mas_cp(mast->bn, 0, mt_slots[mast->bn->type] - 1, &l_mas, true); | |
bf96715e | 3078 | mas_set_parent(mas, left, l_mas.node, slot); |
54a611b6 | 3079 | if (middle) |
bf96715e | 3080 | mas_set_parent(mas, middle, l_mas.node, ++slot); |
54a611b6 LH |
3081 | |
3082 | if (right) | |
bf96715e | 3083 | mas_set_parent(mas, right, l_mas.node, ++slot); |
54a611b6 LH |
3084 | |
3085 | if (mas_is_root_limits(mast->l)) { | |
3086 | new_root: | |
3087 | mast_new_root(mast, mas); | |
3088 | } else { | |
3089 | mas_mn(&l_mas)->parent = mas_mn(mast->orig_l)->parent; | |
3090 | } | |
3091 | ||
3092 | if (!mte_dead_node(mast->orig_l->node)) | |
3093 | mat_add(&free, mast->orig_l->node); | |
3094 | ||
3095 | mas->depth = mast->orig_l->depth; | |
3096 | *mast->orig_l = l_mas; | |
3097 | mte_set_node_dead(mas->node); | |
3098 | ||
3099 | /* Set up mas for insertion. */ | |
3100 | mast->orig_l->depth = mas->depth; | |
3101 | mast->orig_l->alloc = mas->alloc; | |
3102 | *mas = *mast->orig_l; | |
3103 | mas_wmb_replace(mas, &free, &destroy); | |
3104 | mtree_range_walk(mas); | |
3105 | return mast->bn->b_end; | |
3106 | } | |
3107 | ||
3108 | /* | |
3109 | * mas_rebalance() - Rebalance a given node. | |
3110 | * @mas: The maple state | |
3111 | * @b_node: The big maple node. | |
3112 | * | |
3113 | * Rebalance two nodes into a single node or two new nodes that are sufficient. | |
3114 | * Continue upwards until tree is sufficient. | |
3115 | * | |
3116 | * Return: the number of elements in b_node during the last loop. | |
3117 | */ | |
3118 | static inline int mas_rebalance(struct ma_state *mas, | |
3119 | struct maple_big_node *b_node) | |
3120 | { | |
3121 | char empty_count = mas_mt_height(mas); | |
3122 | struct maple_subtree_state mast; | |
3123 | unsigned char shift, b_end = ++b_node->b_end; | |
3124 | ||
3125 | MA_STATE(l_mas, mas->tree, mas->index, mas->last); | |
3126 | MA_STATE(r_mas, mas->tree, mas->index, mas->last); | |
3127 | ||
3128 | trace_ma_op(__func__, mas); | |
3129 | ||
3130 | /* | |
3131 | * Rebalancing occurs if a node is insufficient. Data is rebalanced | |
3132 | * against the node to the right if it exists, otherwise the node to the | |
3133 | * left of this node is rebalanced against this node. If rebalancing | |
3134 | * causes just one node to be produced instead of two, then the parent | |
3135 | * is also examined and rebalanced if it is insufficient. Every level | |
3136 | * tries to combine the data in the same way. If one node contains the | |
3137 | * entire range of the tree, then that node is used as a new root node. | |
3138 | */ | |
c108df76 | 3139 | mas_node_count(mas, empty_count * 2 - 1); |
54a611b6 LH |
3140 | if (mas_is_err(mas)) |
3141 | return 0; | |
3142 | ||
3143 | mast.orig_l = &l_mas; | |
3144 | mast.orig_r = &r_mas; | |
3145 | mast.bn = b_node; | |
3146 | mast.bn->type = mte_node_type(mas->node); | |
3147 | ||
3148 | l_mas = r_mas = *mas; | |
3149 | ||
3150 | if (mas_next_sibling(&r_mas)) { | |
3151 | mas_mab_cp(&r_mas, 0, mt_slot_count(r_mas.node), b_node, b_end); | |
3152 | r_mas.last = r_mas.index = r_mas.max; | |
3153 | } else { | |
3154 | mas_prev_sibling(&l_mas); | |
3155 | shift = mas_data_end(&l_mas) + 1; | |
3156 | mab_shift_right(b_node, shift); | |
3157 | mas->offset += shift; | |
3158 | mas_mab_cp(&l_mas, 0, shift - 1, b_node, 0); | |
3159 | b_node->b_end = shift + b_end; | |
3160 | l_mas.index = l_mas.last = l_mas.min; | |
3161 | } | |
3162 | ||
3163 | return mas_spanning_rebalance(mas, &mast, empty_count); | |
3164 | } | |
3165 | ||
3166 | /* | |
3167 | * mas_destroy_rebalance() - Rebalance left-most node while destroying the maple | |
3168 | * state. | |
3169 | * @mas: The maple state | |
3170 | * @end: The end of the left-most node. | |
3171 | * | |
3172 | * During a mass-insert event (such as forking), it may be necessary to | |
3173 | * rebalance the left-most node when it is not sufficient. | |
3174 | */ | |
3175 | static inline void mas_destroy_rebalance(struct ma_state *mas, unsigned char end) | |
3176 | { | |
3177 | enum maple_type mt = mte_node_type(mas->node); | |
3178 | struct maple_node reuse, *newnode, *parent, *new_left, *left, *node; | |
3179 | struct maple_enode *eparent; | |
3180 | unsigned char offset, tmp, split = mt_slots[mt] / 2; | |
3181 | void __rcu **l_slots, **slots; | |
3182 | unsigned long *l_pivs, *pivs, gap; | |
3183 | bool in_rcu = mt_in_rcu(mas->tree); | |
3184 | ||
3185 | MA_STATE(l_mas, mas->tree, mas->index, mas->last); | |
3186 | ||
3187 | l_mas = *mas; | |
3188 | mas_prev_sibling(&l_mas); | |
3189 | ||
3190 | /* set up node. */ | |
3191 | if (in_rcu) { | |
3192 | /* Allocate for both left and right as well as parent. */ | |
3193 | mas_node_count(mas, 3); | |
3194 | if (mas_is_err(mas)) | |
3195 | return; | |
3196 | ||
3197 | newnode = mas_pop_node(mas); | |
3198 | } else { | |
3199 | newnode = &reuse; | |
3200 | } | |
3201 | ||
3202 | node = mas_mn(mas); | |
3203 | newnode->parent = node->parent; | |
3204 | slots = ma_slots(newnode, mt); | |
3205 | pivs = ma_pivots(newnode, mt); | |
3206 | left = mas_mn(&l_mas); | |
3207 | l_slots = ma_slots(left, mt); | |
3208 | l_pivs = ma_pivots(left, mt); | |
3209 | if (!l_slots[split]) | |
3210 | split++; | |
3211 | tmp = mas_data_end(&l_mas) - split; | |
3212 | ||
3213 | memcpy(slots, l_slots + split + 1, sizeof(void *) * tmp); | |
3214 | memcpy(pivs, l_pivs + split + 1, sizeof(unsigned long) * tmp); | |
3215 | pivs[tmp] = l_mas.max; | |
3216 | memcpy(slots + tmp, ma_slots(node, mt), sizeof(void *) * end); | |
3217 | memcpy(pivs + tmp, ma_pivots(node, mt), sizeof(unsigned long) * end); | |
3218 | ||
3219 | l_mas.max = l_pivs[split]; | |
3220 | mas->min = l_mas.max + 1; | |
3221 | eparent = mt_mk_node(mte_parent(l_mas.node), | |
afc754c6 | 3222 | mas_parent_type(&l_mas, l_mas.node)); |
54a611b6 LH |
3223 | tmp += end; |
3224 | if (!in_rcu) { | |
3225 | unsigned char max_p = mt_pivots[mt]; | |
3226 | unsigned char max_s = mt_slots[mt]; | |
3227 | ||
3228 | if (tmp < max_p) | |
3229 | memset(pivs + tmp, 0, | |
fb20e99a | 3230 | sizeof(unsigned long) * (max_p - tmp)); |
54a611b6 LH |
3231 | |
3232 | if (tmp < mt_slots[mt]) | |
3233 | memset(slots + tmp, 0, sizeof(void *) * (max_s - tmp)); | |
3234 | ||
3235 | memcpy(node, newnode, sizeof(struct maple_node)); | |
3236 | ma_set_meta(node, mt, 0, tmp - 1); | |
3237 | mte_set_pivot(eparent, mte_parent_slot(l_mas.node), | |
3238 | l_pivs[split]); | |
3239 | ||
3240 | /* Remove data from l_pivs. */ | |
3241 | tmp = split + 1; | |
3242 | memset(l_pivs + tmp, 0, sizeof(unsigned long) * (max_p - tmp)); | |
3243 | memset(l_slots + tmp, 0, sizeof(void *) * (max_s - tmp)); | |
3244 | ma_set_meta(left, mt, 0, split); | |
3245 | ||
3246 | goto done; | |
3247 | } | |
3248 | ||
3249 | /* RCU requires replacing both l_mas, mas, and parent. */ | |
3250 | mas->node = mt_mk_node(newnode, mt); | |
3251 | ma_set_meta(newnode, mt, 0, tmp); | |
3252 | ||
3253 | new_left = mas_pop_node(mas); | |
3254 | new_left->parent = left->parent; | |
3255 | mt = mte_node_type(l_mas.node); | |
3256 | slots = ma_slots(new_left, mt); | |
3257 | pivs = ma_pivots(new_left, mt); | |
3258 | memcpy(slots, l_slots, sizeof(void *) * split); | |
3259 | memcpy(pivs, l_pivs, sizeof(unsigned long) * split); | |
3260 | ma_set_meta(new_left, mt, 0, split); | |
3261 | l_mas.node = mt_mk_node(new_left, mt); | |
3262 | ||
3263 | /* replace parent. */ | |
3264 | offset = mte_parent_slot(mas->node); | |
afc754c6 | 3265 | mt = mas_parent_type(&l_mas, l_mas.node); |
54a611b6 LH |
3266 | parent = mas_pop_node(mas); |
3267 | slots = ma_slots(parent, mt); | |
3268 | pivs = ma_pivots(parent, mt); | |
3269 | memcpy(parent, mte_to_node(eparent), sizeof(struct maple_node)); | |
3270 | rcu_assign_pointer(slots[offset], mas->node); | |
3271 | rcu_assign_pointer(slots[offset - 1], l_mas.node); | |
3272 | pivs[offset - 1] = l_mas.max; | |
3273 | eparent = mt_mk_node(parent, mt); | |
3274 | done: | |
3275 | gap = mas_leaf_max_gap(mas); | |
3276 | mte_set_gap(eparent, mte_parent_slot(mas->node), gap); | |
3277 | gap = mas_leaf_max_gap(&l_mas); | |
3278 | mte_set_gap(eparent, mte_parent_slot(l_mas.node), gap); | |
3279 | mas_ascend(mas); | |
3280 | ||
3281 | if (in_rcu) | |
3282 | mas_replace(mas, false); | |
3283 | ||
3284 | mas_update_gap(mas); | |
3285 | } | |
3286 | ||
3287 | /* | |
3288 | * mas_split_final_node() - Split the final node in a subtree operation. | |
3289 | * @mast: the maple subtree state | |
3290 | * @mas: The maple state | |
3291 | * @height: The height of the tree in case it's a new root. | |
3292 | */ | |
3293 | static inline bool mas_split_final_node(struct maple_subtree_state *mast, | |
3294 | struct ma_state *mas, int height) | |
3295 | { | |
3296 | struct maple_enode *ancestor; | |
3297 | ||
3298 | if (mte_is_root(mas->node)) { | |
3299 | if (mt_is_alloc(mas->tree)) | |
3300 | mast->bn->type = maple_arange_64; | |
3301 | else | |
3302 | mast->bn->type = maple_range_64; | |
3303 | mas->depth = height; | |
3304 | } | |
3305 | /* | |
3306 | * Only a single node is used here, could be root. | |
3307 | * The Big_node data should just fit in a single node. | |
3308 | */ | |
3309 | ancestor = mas_new_ma_node(mas, mast->bn); | |
bf96715e LH |
3310 | mas_set_parent(mas, mast->l->node, ancestor, mast->l->offset); |
3311 | mas_set_parent(mas, mast->r->node, ancestor, mast->r->offset); | |
54a611b6 LH |
3312 | mte_to_node(ancestor)->parent = mas_mn(mas)->parent; |
3313 | ||
3314 | mast->l->node = ancestor; | |
3315 | mab_mas_cp(mast->bn, 0, mt_slots[mast->bn->type] - 1, mast->l, true); | |
3316 | mas->offset = mast->bn->b_end - 1; | |
3317 | return true; | |
3318 | } | |
3319 | ||
3320 | /* | |
3321 | * mast_fill_bnode() - Copy data into the big node in the subtree state | |
3322 | * @mast: The maple subtree state | |
3323 | * @mas: the maple state | |
3324 | * @skip: The number of entries to skip for new nodes insertion. | |
3325 | */ | |
3326 | static inline void mast_fill_bnode(struct maple_subtree_state *mast, | |
3327 | struct ma_state *mas, | |
3328 | unsigned char skip) | |
3329 | { | |
3330 | bool cp = true; | |
3331 | struct maple_enode *old = mas->node; | |
3332 | unsigned char split; | |
3333 | ||
3334 | memset(mast->bn->gap, 0, sizeof(unsigned long) * ARRAY_SIZE(mast->bn->gap)); | |
3335 | memset(mast->bn->slot, 0, sizeof(unsigned long) * ARRAY_SIZE(mast->bn->slot)); | |
3336 | memset(mast->bn->pivot, 0, sizeof(unsigned long) * ARRAY_SIZE(mast->bn->pivot)); | |
3337 | mast->bn->b_end = 0; | |
3338 | ||
3339 | if (mte_is_root(mas->node)) { | |
3340 | cp = false; | |
3341 | } else { | |
3342 | mas_ascend(mas); | |
3343 | mat_add(mast->free, old); | |
3344 | mas->offset = mte_parent_slot(mas->node); | |
3345 | } | |
3346 | ||
3347 | if (cp && mast->l->offset) | |
3348 | mas_mab_cp(mas, 0, mast->l->offset - 1, mast->bn, 0); | |
3349 | ||
3350 | split = mast->bn->b_end; | |
3351 | mab_set_b_end(mast->bn, mast->l, mast->l->node); | |
3352 | mast->r->offset = mast->bn->b_end; | |
3353 | mab_set_b_end(mast->bn, mast->r, mast->r->node); | |
3354 | if (mast->bn->pivot[mast->bn->b_end - 1] == mas->max) | |
3355 | cp = false; | |
3356 | ||
3357 | if (cp) | |
3358 | mas_mab_cp(mas, split + skip, mt_slot_count(mas->node) - 1, | |
3359 | mast->bn, mast->bn->b_end); | |
3360 | ||
3361 | mast->bn->b_end--; | |
3362 | mast->bn->type = mte_node_type(mas->node); | |
3363 | } | |
3364 | ||
3365 | /* | |
3366 | * mast_split_data() - Split the data in the subtree state big node into regular | |
3367 | * nodes. | |
3368 | * @mast: The maple subtree state | |
3369 | * @mas: The maple state | |
3370 | * @split: The location to split the big node | |
3371 | */ | |
3372 | static inline void mast_split_data(struct maple_subtree_state *mast, | |
3373 | struct ma_state *mas, unsigned char split) | |
3374 | { | |
3375 | unsigned char p_slot; | |
3376 | ||
3377 | mab_mas_cp(mast->bn, 0, split, mast->l, true); | |
3378 | mte_set_pivot(mast->r->node, 0, mast->r->max); | |
3379 | mab_mas_cp(mast->bn, split + 1, mast->bn->b_end, mast->r, false); | |
3380 | mast->l->offset = mte_parent_slot(mas->node); | |
3381 | mast->l->max = mast->bn->pivot[split]; | |
3382 | mast->r->min = mast->l->max + 1; | |
3383 | if (mte_is_leaf(mas->node)) | |
3384 | return; | |
3385 | ||
3386 | p_slot = mast->orig_l->offset; | |
3387 | mas_set_split_parent(mast->orig_l, mast->l->node, mast->r->node, | |
3388 | &p_slot, split); | |
3389 | mas_set_split_parent(mast->orig_r, mast->l->node, mast->r->node, | |
3390 | &p_slot, split); | |
3391 | } | |
3392 | ||
3393 | /* | |
3394 | * mas_push_data() - Instead of splitting a node, it is beneficial to push the | |
3395 | * data to the right or left node if there is room. | |
3396 | * @mas: The maple state | |
3397 | * @height: The current height of the maple state | |
3398 | * @mast: The maple subtree state | |
3399 | * @left: Push left or not. | |
3400 | * | |
3401 | * Keeping the height of the tree low means faster lookups. | |
3402 | * | |
3403 | * Return: True if pushed, false otherwise. | |
3404 | */ | |
3405 | static inline bool mas_push_data(struct ma_state *mas, int height, | |
3406 | struct maple_subtree_state *mast, bool left) | |
3407 | { | |
3408 | unsigned char slot_total = mast->bn->b_end; | |
3409 | unsigned char end, space, split; | |
3410 | ||
3411 | MA_STATE(tmp_mas, mas->tree, mas->index, mas->last); | |
3412 | tmp_mas = *mas; | |
3413 | tmp_mas.depth = mast->l->depth; | |
3414 | ||
3415 | if (left && !mas_prev_sibling(&tmp_mas)) | |
3416 | return false; | |
3417 | else if (!left && !mas_next_sibling(&tmp_mas)) | |
3418 | return false; | |
3419 | ||
3420 | end = mas_data_end(&tmp_mas); | |
3421 | slot_total += end; | |
3422 | space = 2 * mt_slot_count(mas->node) - 2; | |
3423 | /* -2 instead of -1 to ensure there isn't a triple split */ | |
3424 | if (ma_is_leaf(mast->bn->type)) | |
3425 | space--; | |
3426 | ||
3427 | if (mas->max == ULONG_MAX) | |
3428 | space--; | |
3429 | ||
3430 | if (slot_total >= space) | |
3431 | return false; | |
3432 | ||
3433 | /* Get the data; Fill mast->bn */ | |
3434 | mast->bn->b_end++; | |
3435 | if (left) { | |
3436 | mab_shift_right(mast->bn, end + 1); | |
3437 | mas_mab_cp(&tmp_mas, 0, end, mast->bn, 0); | |
3438 | mast->bn->b_end = slot_total + 1; | |
3439 | } else { | |
3440 | mas_mab_cp(&tmp_mas, 0, end, mast->bn, mast->bn->b_end); | |
3441 | } | |
3442 | ||
3443 | /* Configure mast for splitting of mast->bn */ | |
3444 | split = mt_slots[mast->bn->type] - 2; | |
3445 | if (left) { | |
3446 | /* Switch mas to prev node */ | |
3447 | mat_add(mast->free, mas->node); | |
3448 | *mas = tmp_mas; | |
3449 | /* Start using mast->l for the left side. */ | |
3450 | tmp_mas.node = mast->l->node; | |
3451 | *mast->l = tmp_mas; | |
3452 | } else { | |
3453 | mat_add(mast->free, tmp_mas.node); | |
3454 | tmp_mas.node = mast->r->node; | |
3455 | *mast->r = tmp_mas; | |
3456 | split = slot_total - split; | |
3457 | } | |
3458 | split = mab_no_null_split(mast->bn, split, mt_slots[mast->bn->type]); | |
3459 | /* Update parent slot for split calculation. */ | |
3460 | if (left) | |
3461 | mast->orig_l->offset += end + 1; | |
3462 | ||
3463 | mast_split_data(mast, mas, split); | |
3464 | mast_fill_bnode(mast, mas, 2); | |
3465 | mas_split_final_node(mast, mas, height + 1); | |
3466 | return true; | |
3467 | } | |
3468 | ||
3469 | /* | |
3470 | * mas_split() - Split data that is too big for one node into two. | |
3471 | * @mas: The maple state | |
3472 | * @b_node: The maple big node | |
3473 | * Return: 1 on success, 0 on failure. | |
3474 | */ | |
3475 | static int mas_split(struct ma_state *mas, struct maple_big_node *b_node) | |
3476 | { | |
54a611b6 LH |
3477 | struct maple_subtree_state mast; |
3478 | int height = 0; | |
3479 | unsigned char mid_split, split = 0; | |
3480 | ||
3481 | /* | |
3482 | * Splitting is handled differently from any other B-tree; the Maple | |
3483 | * Tree splits upwards. Splitting up means that the split operation | |
3484 | * occurs when the walk of the tree hits the leaves and not on the way | |
3485 | * down. The reason for splitting up is that it is impossible to know | |
3486 | * how much space will be needed until the leaf is (or leaves are) | |
3487 | * reached. Since overwriting data is allowed and a range could | |
3488 | * overwrite more than one range or result in changing one entry into 3 | |
3489 | * entries, it is impossible to know if a split is required until the | |
3490 | * data is examined. | |
3491 | * | |
3492 | * Splitting is a balancing act between keeping allocations to a minimum | |
3493 | * and avoiding a 'jitter' event where a tree is expanded to make room | |
3494 | * for an entry followed by a contraction when the entry is removed. To | |
3495 | * accomplish the balance, there are empty slots remaining in both left | |
3496 | * and right nodes after a split. | |
3497 | */ | |
3498 | MA_STATE(l_mas, mas->tree, mas->index, mas->last); | |
3499 | MA_STATE(r_mas, mas->tree, mas->index, mas->last); | |
3500 | MA_STATE(prev_l_mas, mas->tree, mas->index, mas->last); | |
3501 | MA_STATE(prev_r_mas, mas->tree, mas->index, mas->last); | |
3502 | MA_TOPIARY(mat, mas->tree); | |
3503 | ||
3504 | trace_ma_op(__func__, mas); | |
3505 | mas->depth = mas_mt_height(mas); | |
3506 | /* Allocation failures will happen early. */ | |
3507 | mas_node_count(mas, 1 + mas->depth * 2); | |
3508 | if (mas_is_err(mas)) | |
3509 | return 0; | |
3510 | ||
3511 | mast.l = &l_mas; | |
3512 | mast.r = &r_mas; | |
3513 | mast.orig_l = &prev_l_mas; | |
3514 | mast.orig_r = &prev_r_mas; | |
3515 | mast.free = &mat; | |
3516 | mast.bn = b_node; | |
3517 | ||
3518 | while (height++ <= mas->depth) { | |
3519 | if (mt_slots[b_node->type] > b_node->b_end) { | |
3520 | mas_split_final_node(&mast, mas, height); | |
3521 | break; | |
3522 | } | |
3523 | ||
3524 | l_mas = r_mas = *mas; | |
3525 | l_mas.node = mas_new_ma_node(mas, b_node); | |
3526 | r_mas.node = mas_new_ma_node(mas, b_node); | |
3527 | /* | |
3528 | * Another way that 'jitter' is avoided is to terminate a split up early if the | |
3529 | * left or right node has space to spare. This is referred to as "pushing left" | |
3530 | * or "pushing right" and is similar to the B* tree, except the nodes left or | |
3531 | * right can rarely be reused due to RCU, but the ripple upwards is halted which | |
3532 | * is a significant savings. | |
3533 | */ | |
3534 | /* Try to push left. */ | |
3535 | if (mas_push_data(mas, height, &mast, true)) | |
3536 | break; | |
3537 | ||
3538 | /* Try to push right. */ | |
3539 | if (mas_push_data(mas, height, &mast, false)) | |
3540 | break; | |
3541 | ||
3542 | split = mab_calc_split(mas, b_node, &mid_split, prev_l_mas.min); | |
3543 | mast_split_data(&mast, mas, split); | |
3544 | /* | |
3545 | * Usually correct, mab_mas_cp in the above call overwrites | |
3546 | * r->max. | |
3547 | */ | |
3548 | mast.r->max = mas->max; | |
3549 | mast_fill_bnode(&mast, mas, 1); | |
3550 | prev_l_mas = *mast.l; | |
3551 | prev_r_mas = *mast.r; | |
3552 | } | |
3553 | ||
3554 | /* Set the original node as dead */ | |
3555 | mat_add(mast.free, mas->node); | |
3556 | mas->node = l_mas.node; | |
3557 | mas_wmb_replace(mas, mast.free, NULL); | |
3558 | mtree_range_walk(mas); | |
3559 | return 1; | |
3560 | } | |
3561 | ||
3562 | /* | |
3563 | * mas_reuse_node() - Reuse the node to store the data. | |
3564 | * @wr_mas: The maple write state | |
3565 | * @bn: The maple big node | |
3566 | * @end: The end of the data. | |
3567 | * | |
3568 | * Will always return false in RCU mode. | |
3569 | * | |
3570 | * Return: True if node was reused, false otherwise. | |
3571 | */ | |
3572 | static inline bool mas_reuse_node(struct ma_wr_state *wr_mas, | |
3573 | struct maple_big_node *bn, unsigned char end) | |
3574 | { | |
3575 | /* Need to be rcu safe. */ | |
3576 | if (mt_in_rcu(wr_mas->mas->tree)) | |
3577 | return false; | |
3578 | ||
3579 | if (end > bn->b_end) { | |
3580 | int clear = mt_slots[wr_mas->type] - bn->b_end; | |
3581 | ||
3582 | memset(wr_mas->slots + bn->b_end, 0, sizeof(void *) * clear--); | |
3583 | memset(wr_mas->pivots + bn->b_end, 0, sizeof(void *) * clear); | |
3584 | } | |
3585 | mab_mas_cp(bn, 0, bn->b_end, wr_mas->mas, false); | |
3586 | return true; | |
3587 | } | |
3588 | ||
3589 | /* | |
3590 | * mas_commit_b_node() - Commit the big node into the tree. | |
3591 | * @wr_mas: The maple write state | |
3592 | * @b_node: The maple big node | |
3593 | * @end: The end of the data. | |
3594 | */ | |
44081c77 | 3595 | static noinline_for_kasan int mas_commit_b_node(struct ma_wr_state *wr_mas, |
54a611b6 LH |
3596 | struct maple_big_node *b_node, unsigned char end) |
3597 | { | |
3598 | struct maple_node *node; | |
3599 | unsigned char b_end = b_node->b_end; | |
3600 | enum maple_type b_type = b_node->type; | |
3601 | ||
3602 | if ((b_end < mt_min_slots[b_type]) && | |
3603 | (!mte_is_root(wr_mas->mas->node)) && | |
3604 | (mas_mt_height(wr_mas->mas) > 1)) | |
3605 | return mas_rebalance(wr_mas->mas, b_node); | |
3606 | ||
3607 | if (b_end >= mt_slots[b_type]) | |
3608 | return mas_split(wr_mas->mas, b_node); | |
3609 | ||
3610 | if (mas_reuse_node(wr_mas, b_node, end)) | |
3611 | goto reuse_node; | |
3612 | ||
3613 | mas_node_count(wr_mas->mas, 1); | |
3614 | if (mas_is_err(wr_mas->mas)) | |
3615 | return 0; | |
3616 | ||
3617 | node = mas_pop_node(wr_mas->mas); | |
3618 | node->parent = mas_mn(wr_mas->mas)->parent; | |
3619 | wr_mas->mas->node = mt_mk_node(node, b_type); | |
7dc5ba62 | 3620 | mab_mas_cp(b_node, 0, b_end, wr_mas->mas, false); |
54a611b6 LH |
3621 | mas_replace(wr_mas->mas, false); |
3622 | reuse_node: | |
3623 | mas_update_gap(wr_mas->mas); | |
3624 | return 1; | |
3625 | } | |
3626 | ||
3627 | /* | |
3628 | * mas_root_expand() - Expand a root to a node | |
3629 | * @mas: The maple state | |
3630 | * @entry: The entry to store into the tree | |
3631 | */ | |
3632 | static inline int mas_root_expand(struct ma_state *mas, void *entry) | |
3633 | { | |
3634 | void *contents = mas_root_locked(mas); | |
3635 | enum maple_type type = maple_leaf_64; | |
3636 | struct maple_node *node; | |
3637 | void __rcu **slots; | |
3638 | unsigned long *pivots; | |
3639 | int slot = 0; | |
3640 | ||
3641 | mas_node_count(mas, 1); | |
3642 | if (unlikely(mas_is_err(mas))) | |
3643 | return 0; | |
3644 | ||
3645 | node = mas_pop_node(mas); | |
3646 | pivots = ma_pivots(node, type); | |
3647 | slots = ma_slots(node, type); | |
3648 | node->parent = ma_parent_ptr( | |
3649 | ((unsigned long)mas->tree | MA_ROOT_PARENT)); | |
3650 | mas->node = mt_mk_node(node, type); | |
3651 | ||
3652 | if (mas->index) { | |
3653 | if (contents) { | |
3654 | rcu_assign_pointer(slots[slot], contents); | |
3655 | if (likely(mas->index > 1)) | |
3656 | slot++; | |
3657 | } | |
3658 | pivots[slot++] = mas->index - 1; | |
3659 | } | |
3660 | ||
3661 | rcu_assign_pointer(slots[slot], entry); | |
3662 | mas->offset = slot; | |
3663 | pivots[slot] = mas->last; | |
3664 | if (mas->last != ULONG_MAX) | |
3c769fd8 PZ |
3665 | pivots[++slot] = ULONG_MAX; |
3666 | ||
54a611b6 LH |
3667 | mas->depth = 1; |
3668 | mas_set_height(mas); | |
c45ea315 | 3669 | ma_set_meta(node, maple_leaf_64, 0, slot); |
54a611b6 LH |
3670 | /* swap the new root into the tree */ |
3671 | rcu_assign_pointer(mas->tree->ma_root, mte_mk_root(mas->node)); | |
54a611b6 LH |
3672 | return slot; |
3673 | } | |
3674 | ||
3675 | static inline void mas_store_root(struct ma_state *mas, void *entry) | |
3676 | { | |
3677 | if (likely((mas->last != 0) || (mas->index != 0))) | |
3678 | mas_root_expand(mas, entry); | |
3679 | else if (((unsigned long) (entry) & 3) == 2) | |
3680 | mas_root_expand(mas, entry); | |
3681 | else { | |
3682 | rcu_assign_pointer(mas->tree->ma_root, entry); | |
3683 | mas->node = MAS_START; | |
3684 | } | |
3685 | } | |
3686 | ||
3687 | /* | |
3688 | * mas_is_span_wr() - Check if the write needs to be treated as a write that | |
3689 | * spans the node. | |
3690 | * @mas: The maple state | |
3691 | * @piv: The pivot value being written | |
3692 | * @type: The maple node type | |
3693 | * @entry: The data to write | |
3694 | * | |
3695 | * Spanning writes are writes that start in one node and end in another OR if | |
3696 | * the write of a %NULL will cause the node to end with a %NULL. | |
3697 | * | |
3698 | * Return: True if this is a spanning write, false otherwise. | |
3699 | */ | |
3700 | static bool mas_is_span_wr(struct ma_wr_state *wr_mas) | |
3701 | { | |
bc147f0f | 3702 | unsigned long max = wr_mas->r_max; |
54a611b6 | 3703 | unsigned long last = wr_mas->mas->last; |
54a611b6 LH |
3704 | enum maple_type type = wr_mas->type; |
3705 | void *entry = wr_mas->entry; | |
3706 | ||
bc147f0f PZ |
3707 | /* Contained in this pivot, fast path */ |
3708 | if (last < max) | |
54a611b6 LH |
3709 | return false; |
3710 | ||
bc147f0f PZ |
3711 | if (ma_is_leaf(type)) { |
3712 | max = wr_mas->mas->max; | |
54a611b6 LH |
3713 | if (last < max) |
3714 | return false; | |
bc147f0f | 3715 | } |
54a611b6 | 3716 | |
bc147f0f | 3717 | if (last == max) { |
54a611b6 | 3718 | /* |
bc147f0f PZ |
3719 | * The last entry of leaf node cannot be NULL unless it is the |
3720 | * rightmost node (writing ULONG_MAX), otherwise it spans slots. | |
54a611b6 | 3721 | */ |
bc147f0f | 3722 | if (entry || last == ULONG_MAX) |
54a611b6 LH |
3723 | return false; |
3724 | } | |
3725 | ||
bc147f0f | 3726 | trace_ma_write(__func__, wr_mas->mas, wr_mas->r_max, entry); |
54a611b6 LH |
3727 | return true; |
3728 | } | |
3729 | ||
3730 | static inline void mas_wr_walk_descend(struct ma_wr_state *wr_mas) | |
3731 | { | |
54a611b6 LH |
3732 | wr_mas->type = mte_node_type(wr_mas->mas->node); |
3733 | mas_wr_node_walk(wr_mas); | |
3734 | wr_mas->slots = ma_slots(wr_mas->node, wr_mas->type); | |
3735 | } | |
3736 | ||
3737 | static inline void mas_wr_walk_traverse(struct ma_wr_state *wr_mas) | |
3738 | { | |
3739 | wr_mas->mas->max = wr_mas->r_max; | |
3740 | wr_mas->mas->min = wr_mas->r_min; | |
3741 | wr_mas->mas->node = wr_mas->content; | |
3742 | wr_mas->mas->offset = 0; | |
9bbba563 | 3743 | wr_mas->mas->depth++; |
54a611b6 LH |
3744 | } |
3745 | /* | |
3746 | * mas_wr_walk() - Walk the tree for a write. | |
3747 | * @wr_mas: The maple write state | |
3748 | * | |
3749 | * Uses mas_slot_locked() and does not need to worry about dead nodes. | |
3750 | * | |
3751 | * Return: True if it's contained in a node, false on spanning write. | |
3752 | */ | |
3753 | static bool mas_wr_walk(struct ma_wr_state *wr_mas) | |
3754 | { | |
3755 | struct ma_state *mas = wr_mas->mas; | |
3756 | ||
3757 | while (true) { | |
3758 | mas_wr_walk_descend(wr_mas); | |
3759 | if (unlikely(mas_is_span_wr(wr_mas))) | |
3760 | return false; | |
3761 | ||
3762 | wr_mas->content = mas_slot_locked(mas, wr_mas->slots, | |
3763 | mas->offset); | |
3764 | if (ma_is_leaf(wr_mas->type)) | |
3765 | return true; | |
3766 | ||
3767 | mas_wr_walk_traverse(wr_mas); | |
3768 | } | |
3769 | ||
3770 | return true; | |
3771 | } | |
3772 | ||
3773 | static bool mas_wr_walk_index(struct ma_wr_state *wr_mas) | |
3774 | { | |
3775 | struct ma_state *mas = wr_mas->mas; | |
3776 | ||
3777 | while (true) { | |
3778 | mas_wr_walk_descend(wr_mas); | |
3779 | wr_mas->content = mas_slot_locked(mas, wr_mas->slots, | |
3780 | mas->offset); | |
3781 | if (ma_is_leaf(wr_mas->type)) | |
3782 | return true; | |
3783 | mas_wr_walk_traverse(wr_mas); | |
3784 | ||
3785 | } | |
3786 | return true; | |
3787 | } | |
3788 | /* | |
3789 | * mas_extend_spanning_null() - Extend a store of a %NULL to include surrounding %NULLs. | |
3790 | * @l_wr_mas: The left maple write state | |
3791 | * @r_wr_mas: The right maple write state | |
3792 | */ | |
3793 | static inline void mas_extend_spanning_null(struct ma_wr_state *l_wr_mas, | |
3794 | struct ma_wr_state *r_wr_mas) | |
3795 | { | |
3796 | struct ma_state *r_mas = r_wr_mas->mas; | |
3797 | struct ma_state *l_mas = l_wr_mas->mas; | |
3798 | unsigned char l_slot; | |
3799 | ||
3800 | l_slot = l_mas->offset; | |
3801 | if (!l_wr_mas->content) | |
3802 | l_mas->index = l_wr_mas->r_min; | |
3803 | ||
3804 | if ((l_mas->index == l_wr_mas->r_min) && | |
3805 | (l_slot && | |
3806 | !mas_slot_locked(l_mas, l_wr_mas->slots, l_slot - 1))) { | |
3807 | if (l_slot > 1) | |
3808 | l_mas->index = l_wr_mas->pivots[l_slot - 2] + 1; | |
3809 | else | |
3810 | l_mas->index = l_mas->min; | |
3811 | ||
3812 | l_mas->offset = l_slot - 1; | |
3813 | } | |
3814 | ||
3815 | if (!r_wr_mas->content) { | |
3816 | if (r_mas->last < r_wr_mas->r_max) | |
3817 | r_mas->last = r_wr_mas->r_max; | |
3818 | r_mas->offset++; | |
3819 | } else if ((r_mas->last == r_wr_mas->r_max) && | |
3820 | (r_mas->last < r_mas->max) && | |
3821 | !mas_slot_locked(r_mas, r_wr_mas->slots, r_mas->offset + 1)) { | |
3822 | r_mas->last = mas_safe_pivot(r_mas, r_wr_mas->pivots, | |
3823 | r_wr_mas->type, r_mas->offset + 1); | |
3824 | r_mas->offset++; | |
3825 | } | |
3826 | } | |
3827 | ||
3828 | static inline void *mas_state_walk(struct ma_state *mas) | |
3829 | { | |
3830 | void *entry; | |
3831 | ||
3832 | entry = mas_start(mas); | |
3833 | if (mas_is_none(mas)) | |
3834 | return NULL; | |
3835 | ||
3836 | if (mas_is_ptr(mas)) | |
3837 | return entry; | |
3838 | ||
3839 | return mtree_range_walk(mas); | |
3840 | } | |
3841 | ||
3842 | /* | |
3843 | * mtree_lookup_walk() - Internal quick lookup that does not keep maple state up | |
3844 | * to date. | |
3845 | * | |
3846 | * @mas: The maple state. | |
3847 | * | |
3848 | * Note: Leaves mas in undesirable state. | |
3849 | * Return: The entry for @mas->index or %NULL on dead node. | |
3850 | */ | |
3851 | static inline void *mtree_lookup_walk(struct ma_state *mas) | |
3852 | { | |
3853 | unsigned long *pivots; | |
3854 | unsigned char offset; | |
3855 | struct maple_node *node; | |
3856 | struct maple_enode *next; | |
3857 | enum maple_type type; | |
3858 | void __rcu **slots; | |
3859 | unsigned char end; | |
3860 | unsigned long max; | |
3861 | ||
3862 | next = mas->node; | |
3863 | max = ULONG_MAX; | |
3864 | do { | |
3865 | offset = 0; | |
3866 | node = mte_to_node(next); | |
3867 | type = mte_node_type(next); | |
3868 | pivots = ma_pivots(node, type); | |
3869 | end = ma_data_end(node, type, pivots, max); | |
3870 | if (unlikely(ma_dead_node(node))) | |
3871 | goto dead_node; | |
54a611b6 | 3872 | do { |
ec07967d PZ |
3873 | if (pivots[offset] >= mas->index) { |
3874 | max = pivots[offset]; | |
3875 | break; | |
3876 | } | |
3877 | } while (++offset < end); | |
54a611b6 | 3878 | |
54a611b6 LH |
3879 | slots = ma_slots(node, type); |
3880 | next = mt_slot(mas->tree, slots, offset); | |
3881 | if (unlikely(ma_dead_node(node))) | |
3882 | goto dead_node; | |
3883 | } while (!ma_is_leaf(type)); | |
3884 | ||
831978e3 | 3885 | return (void *)next; |
54a611b6 LH |
3886 | |
3887 | dead_node: | |
3888 | mas_reset(mas); | |
3889 | return NULL; | |
3890 | } | |
3891 | ||
3892 | /* | |
3893 | * mas_new_root() - Create a new root node that only contains the entry passed | |
3894 | * in. | |
3895 | * @mas: The maple state | |
3896 | * @entry: The entry to store. | |
3897 | * | |
3898 | * Only valid when the index == 0 and the last == ULONG_MAX | |
3899 | * | |
3900 | * Return 0 on error, 1 on success. | |
3901 | */ | |
3902 | static inline int mas_new_root(struct ma_state *mas, void *entry) | |
3903 | { | |
3904 | struct maple_enode *root = mas_root_locked(mas); | |
3905 | enum maple_type type = maple_leaf_64; | |
3906 | struct maple_node *node; | |
3907 | void __rcu **slots; | |
3908 | unsigned long *pivots; | |
3909 | ||
3910 | if (!entry && !mas->index && mas->last == ULONG_MAX) { | |
3911 | mas->depth = 0; | |
3912 | mas_set_height(mas); | |
3913 | rcu_assign_pointer(mas->tree->ma_root, entry); | |
3914 | mas->node = MAS_START; | |
3915 | goto done; | |
3916 | } | |
3917 | ||
3918 | mas_node_count(mas, 1); | |
3919 | if (mas_is_err(mas)) | |
3920 | return 0; | |
3921 | ||
3922 | node = mas_pop_node(mas); | |
3923 | pivots = ma_pivots(node, type); | |
3924 | slots = ma_slots(node, type); | |
3925 | node->parent = ma_parent_ptr( | |
3926 | ((unsigned long)mas->tree | MA_ROOT_PARENT)); | |
3927 | mas->node = mt_mk_node(node, type); | |
3928 | rcu_assign_pointer(slots[0], entry); | |
3929 | pivots[0] = mas->last; | |
3930 | mas->depth = 1; | |
3931 | mas_set_height(mas); | |
3932 | rcu_assign_pointer(mas->tree->ma_root, mte_mk_root(mas->node)); | |
3933 | ||
3934 | done: | |
3935 | if (xa_is_node(root)) | |
3936 | mte_destroy_walk(root, mas->tree); | |
3937 | ||
3938 | return 1; | |
3939 | } | |
3940 | /* | |
3941 | * mas_wr_spanning_store() - Create a subtree with the store operation completed | |
3942 | * and new nodes where necessary, then place the sub-tree in the actual tree. | |
3943 | * Note that mas is expected to point to the node which caused the store to | |
3944 | * span. | |
3945 | * @wr_mas: The maple write state | |
3946 | * | |
3947 | * Return: 0 on error, positive on success. | |
3948 | */ | |
3949 | static inline int mas_wr_spanning_store(struct ma_wr_state *wr_mas) | |
3950 | { | |
3951 | struct maple_subtree_state mast; | |
3952 | struct maple_big_node b_node; | |
3953 | struct ma_state *mas; | |
3954 | unsigned char height; | |
3955 | ||
3956 | /* Left and Right side of spanning store */ | |
3957 | MA_STATE(l_mas, NULL, 0, 0); | |
3958 | MA_STATE(r_mas, NULL, 0, 0); | |
3959 | ||
3960 | MA_WR_STATE(r_wr_mas, &r_mas, wr_mas->entry); | |
3961 | MA_WR_STATE(l_wr_mas, &l_mas, wr_mas->entry); | |
3962 | ||
3963 | /* | |
3964 | * A store operation that spans multiple nodes is called a spanning | |
3965 | * store and is handled early in the store call stack by the function | |
3966 | * mas_is_span_wr(). When a spanning store is identified, the maple | |
3967 | * state is duplicated. The first maple state walks the left tree path | |
3968 | * to ``index``, the duplicate walks the right tree path to ``last``. | |
3969 | * The data in the two nodes are combined into a single node, two nodes, | |
3970 | * or possibly three nodes (see the 3-way split above). A ``NULL`` | |
3971 | * written to the last entry of a node is considered a spanning store as | |
3972 | * a rebalance is required for the operation to complete and an overflow | |
3973 | * of data may happen. | |
3974 | */ | |
3975 | mas = wr_mas->mas; | |
3976 | trace_ma_op(__func__, mas); | |
3977 | ||
3978 | if (unlikely(!mas->index && mas->last == ULONG_MAX)) | |
3979 | return mas_new_root(mas, wr_mas->entry); | |
3980 | /* | |
3981 | * Node rebalancing may occur due to this store, so there may be three new | |
3982 | * entries per level plus a new root. | |
3983 | */ | |
3984 | height = mas_mt_height(mas); | |
3985 | mas_node_count(mas, 1 + height * 3); | |
3986 | if (mas_is_err(mas)) | |
3987 | return 0; | |
3988 | ||
3989 | /* | |
3990 | * Set up right side. Need to get to the next offset after the spanning | |
3991 | * store to ensure it's not NULL and to combine both the next node and | |
3992 | * the node with the start together. | |
3993 | */ | |
3994 | r_mas = *mas; | |
3995 | /* Avoid overflow, walk to next slot in the tree. */ | |
3996 | if (r_mas.last + 1) | |
3997 | r_mas.last++; | |
3998 | ||
3999 | r_mas.index = r_mas.last; | |
4000 | mas_wr_walk_index(&r_wr_mas); | |
4001 | r_mas.last = r_mas.index = mas->last; | |
4002 | ||
4003 | /* Set up left side. */ | |
4004 | l_mas = *mas; | |
4005 | mas_wr_walk_index(&l_wr_mas); | |
4006 | ||
4007 | if (!wr_mas->entry) { | |
4008 | mas_extend_spanning_null(&l_wr_mas, &r_wr_mas); | |
4009 | mas->offset = l_mas.offset; | |
4010 | mas->index = l_mas.index; | |
4011 | mas->last = l_mas.last = r_mas.last; | |
4012 | } | |
4013 | ||
4014 | /* expanding NULLs may make this cover the entire range */ | |
4015 | if (!l_mas.index && r_mas.last == ULONG_MAX) { | |
4016 | mas_set_range(mas, 0, ULONG_MAX); | |
4017 | return mas_new_root(mas, wr_mas->entry); | |
4018 | } | |
4019 | ||
4020 | memset(&b_node, 0, sizeof(struct maple_big_node)); | |
4021 | /* Copy l_mas and store the value in b_node. */ | |
4022 | mas_store_b_node(&l_wr_mas, &b_node, l_wr_mas.node_end); | |
4023 | /* Copy r_mas into b_node. */ | |
4024 | if (r_mas.offset <= r_wr_mas.node_end) | |
4025 | mas_mab_cp(&r_mas, r_mas.offset, r_wr_mas.node_end, | |
4026 | &b_node, b_node.b_end + 1); | |
4027 | else | |
4028 | b_node.b_end++; | |
4029 | ||
4030 | /* Stop spanning searches by searching for just index. */ | |
4031 | l_mas.index = l_mas.last = mas->index; | |
4032 | ||
4033 | mast.bn = &b_node; | |
4034 | mast.orig_l = &l_mas; | |
4035 | mast.orig_r = &r_mas; | |
4036 | /* Combine l_mas and r_mas and split them up evenly again. */ | |
4037 | return mas_spanning_rebalance(mas, &mast, height + 1); | |
4038 | } | |
4039 | ||
4040 | /* | |
4041 | * mas_wr_node_store() - Attempt to store the value in a node | |
4042 | * @wr_mas: The maple write state | |
4043 | * | |
4044 | * Attempts to reuse the node, but may allocate. | |
4045 | * | |
4046 | * Return: True if stored, false otherwise | |
4047 | */ | |
7a03ae39 PZ |
4048 | static inline bool mas_wr_node_store(struct ma_wr_state *wr_mas, |
4049 | unsigned char new_end) | |
54a611b6 LH |
4050 | { |
4051 | struct ma_state *mas = wr_mas->mas; | |
4052 | void __rcu **dst_slots; | |
4053 | unsigned long *dst_pivots; | |
7a03ae39 | 4054 | unsigned char dst_offset, offset_end = wr_mas->offset_end; |
54a611b6 | 4055 | struct maple_node reuse, *newnode; |
7a03ae39 | 4056 | unsigned char copy_size, node_pivots = mt_pivots[wr_mas->type]; |
54a611b6 LH |
4057 | bool in_rcu = mt_in_rcu(mas->tree); |
4058 | ||
7a03ae39 | 4059 | /* Check if there is enough data. The room is enough. */ |
54a611b6 LH |
4060 | if (!mte_is_root(mas->node) && (new_end <= mt_min_slots[wr_mas->type]) && |
4061 | !(mas->mas_flags & MA_STATE_BULK)) | |
4062 | return false; | |
4063 | ||
7a03ae39 PZ |
4064 | if (mas->last == wr_mas->end_piv) |
4065 | offset_end++; /* don't copy this offset */ | |
4066 | else if (unlikely(wr_mas->r_max == ULONG_MAX)) | |
4067 | mas_bulk_rebalance(mas, wr_mas->node_end, wr_mas->type); | |
4068 | ||
54a611b6 LH |
4069 | /* set up node. */ |
4070 | if (in_rcu) { | |
4071 | mas_node_count(mas, 1); | |
4072 | if (mas_is_err(mas)) | |
4073 | return false; | |
4074 | ||
4075 | newnode = mas_pop_node(mas); | |
4076 | } else { | |
4077 | memset(&reuse, 0, sizeof(struct maple_node)); | |
4078 | newnode = &reuse; | |
4079 | } | |
4080 | ||
4081 | newnode->parent = mas_mn(mas)->parent; | |
4082 | dst_pivots = ma_pivots(newnode, wr_mas->type); | |
4083 | dst_slots = ma_slots(newnode, wr_mas->type); | |
4084 | /* Copy from start to insert point */ | |
7a03ae39 PZ |
4085 | memcpy(dst_pivots, wr_mas->pivots, sizeof(unsigned long) * mas->offset); |
4086 | memcpy(dst_slots, wr_mas->slots, sizeof(void *) * mas->offset); | |
54a611b6 LH |
4087 | |
4088 | /* Handle insert of new range starting after old range */ | |
4089 | if (wr_mas->r_min < mas->index) { | |
7a03ae39 PZ |
4090 | rcu_assign_pointer(dst_slots[mas->offset], wr_mas->content); |
4091 | dst_pivots[mas->offset++] = mas->index - 1; | |
54a611b6 LH |
4092 | } |
4093 | ||
4094 | /* Store the new entry and range end. */ | |
7a03ae39 PZ |
4095 | if (mas->offset < node_pivots) |
4096 | dst_pivots[mas->offset] = mas->last; | |
4097 | rcu_assign_pointer(dst_slots[mas->offset], wr_mas->entry); | |
54a611b6 LH |
4098 | |
4099 | /* | |
4100 | * this range wrote to the end of the node or it overwrote the rest of | |
4101 | * the data | |
4102 | */ | |
7a03ae39 | 4103 | if (offset_end > wr_mas->node_end) |
54a611b6 | 4104 | goto done; |
54a611b6 | 4105 | |
7a03ae39 | 4106 | dst_offset = mas->offset + 1; |
54a611b6 | 4107 | /* Copy to the end of node if necessary. */ |
7a03ae39 PZ |
4108 | copy_size = wr_mas->node_end - offset_end + 1; |
4109 | memcpy(dst_slots + dst_offset, wr_mas->slots + offset_end, | |
54a611b6 | 4110 | sizeof(void *) * copy_size); |
7a03ae39 PZ |
4111 | memcpy(dst_pivots + dst_offset, wr_mas->pivots + offset_end, |
4112 | sizeof(unsigned long) * (copy_size - 1)); | |
54a611b6 | 4113 | |
7a03ae39 | 4114 | if (new_end < node_pivots) |
54a611b6 LH |
4115 | dst_pivots[new_end] = mas->max; |
4116 | ||
4117 | done: | |
4118 | mas_leaf_set_meta(mas, newnode, dst_pivots, maple_leaf_64, new_end); | |
4119 | if (in_rcu) { | |
c13af03d | 4120 | mte_set_node_dead(mas->node); |
54a611b6 LH |
4121 | mas->node = mt_mk_node(newnode, wr_mas->type); |
4122 | mas_replace(mas, false); | |
4123 | } else { | |
4124 | memcpy(wr_mas->node, newnode, sizeof(struct maple_node)); | |
4125 | } | |
4126 | trace_ma_write(__func__, mas, 0, wr_mas->entry); | |
4127 | mas_update_gap(mas); | |
4128 | return true; | |
4129 | } | |
4130 | ||
4131 | /* | |
4132 | * mas_wr_slot_store: Attempt to store a value in a slot. | |
4133 | * @wr_mas: the maple write state | |
4134 | * | |
4135 | * Return: True if stored, false otherwise | |
4136 | */ | |
4137 | static inline bool mas_wr_slot_store(struct ma_wr_state *wr_mas) | |
4138 | { | |
4139 | struct ma_state *mas = wr_mas->mas; | |
54a611b6 | 4140 | unsigned char offset = mas->offset; |
64891ba3 | 4141 | void __rcu **slots = wr_mas->slots; |
e6d1ffd6 | 4142 | bool gap = false; |
54a611b6 | 4143 | |
64891ba3 PZ |
4144 | gap |= !mt_slot_locked(mas->tree, slots, offset); |
4145 | gap |= !mt_slot_locked(mas->tree, slots, offset + 1); | |
54a611b6 | 4146 | |
64891ba3 PZ |
4147 | if (wr_mas->offset_end - offset == 1) { |
4148 | if (mas->index == wr_mas->r_min) { | |
4149 | /* Overwriting the range and a part of the next one */ | |
4150 | rcu_assign_pointer(slots[offset], wr_mas->entry); | |
4151 | wr_mas->pivots[offset] = mas->last; | |
4152 | } else { | |
4153 | /* Overwriting a part of the range and the next one */ | |
4154 | rcu_assign_pointer(slots[offset + 1], wr_mas->entry); | |
4155 | wr_mas->pivots[offset] = mas->index - 1; | |
4156 | mas->offset++; /* Keep mas accurate. */ | |
4157 | } | |
4158 | } else if (!mt_in_rcu(mas->tree)) { | |
4159 | /* | |
4160 | * Expand the range, only partially overwriting the previous and | |
4161 | * next ranges | |
4162 | */ | |
4163 | gap |= !mt_slot_locked(mas->tree, slots, offset + 2); | |
4164 | rcu_assign_pointer(slots[offset + 1], wr_mas->entry); | |
e6d1ffd6 | 4165 | wr_mas->pivots[offset] = mas->index - 1; |
64891ba3 | 4166 | wr_mas->pivots[offset + 1] = mas->last; |
e6d1ffd6 | 4167 | mas->offset++; /* Keep mas accurate. */ |
64891ba3 PZ |
4168 | } else { |
4169 | return false; | |
54a611b6 LH |
4170 | } |
4171 | ||
54a611b6 | 4172 | trace_ma_write(__func__, mas, 0, wr_mas->entry); |
e6d1ffd6 PZ |
4173 | /* |
4174 | * Only update gap when the new entry is empty or there is an empty | |
4175 | * entry in the original two ranges. | |
4176 | */ | |
4177 | if (!wr_mas->entry || gap) | |
4178 | mas_update_gap(mas); | |
4179 | ||
54a611b6 LH |
4180 | return true; |
4181 | } | |
4182 | ||
54a611b6 LH |
4183 | static inline void mas_wr_extend_null(struct ma_wr_state *wr_mas) |
4184 | { | |
4185 | struct ma_state *mas = wr_mas->mas; | |
4186 | ||
8c995a63 PZ |
4187 | if (!wr_mas->slots[wr_mas->offset_end]) { |
4188 | /* If this one is null, the next and prev are not */ | |
54a611b6 | 4189 | mas->last = wr_mas->end_piv; |
8c995a63 PZ |
4190 | } else { |
4191 | /* Check next slot(s) if we are overwriting the end */ | |
4192 | if ((mas->last == wr_mas->end_piv) && | |
4193 | (wr_mas->node_end != wr_mas->offset_end) && | |
4194 | !wr_mas->slots[wr_mas->offset_end + 1]) { | |
4195 | wr_mas->offset_end++; | |
4196 | if (wr_mas->offset_end == wr_mas->node_end) | |
4197 | mas->last = mas->max; | |
4198 | else | |
4199 | mas->last = wr_mas->pivots[wr_mas->offset_end]; | |
4200 | wr_mas->end_piv = mas->last; | |
4201 | } | |
54a611b6 LH |
4202 | } |
4203 | ||
4204 | if (!wr_mas->content) { | |
4205 | /* If this one is null, the next and prev are not */ | |
4206 | mas->index = wr_mas->r_min; | |
4207 | } else { | |
4208 | /* Check prev slot if we are overwriting the start */ | |
4209 | if (mas->index == wr_mas->r_min && mas->offset && | |
4210 | !wr_mas->slots[mas->offset - 1]) { | |
4211 | mas->offset--; | |
4212 | wr_mas->r_min = mas->index = | |
4213 | mas_safe_min(mas, wr_mas->pivots, mas->offset); | |
4214 | wr_mas->r_max = wr_mas->pivots[mas->offset]; | |
4215 | } | |
4216 | } | |
4217 | } | |
4218 | ||
a7496ad5 LH |
4219 | static inline void mas_wr_end_piv(struct ma_wr_state *wr_mas) |
4220 | { | |
4221 | while ((wr_mas->offset_end < wr_mas->node_end) && | |
4222 | (wr_mas->mas->last > wr_mas->pivots[wr_mas->offset_end])) | |
4223 | wr_mas->offset_end++; | |
4224 | ||
4225 | if (wr_mas->offset_end < wr_mas->node_end) | |
4226 | wr_mas->end_piv = wr_mas->pivots[wr_mas->offset_end]; | |
4227 | else | |
4228 | wr_mas->end_piv = wr_mas->mas->max; | |
4229 | ||
4230 | if (!wr_mas->entry) | |
4231 | mas_wr_extend_null(wr_mas); | |
4232 | } | |
4233 | ||
c6fc9e4a PZ |
4234 | static inline unsigned char mas_wr_new_end(struct ma_wr_state *wr_mas) |
4235 | { | |
4236 | struct ma_state *mas = wr_mas->mas; | |
4237 | unsigned char new_end = wr_mas->node_end + 2; | |
4238 | ||
4239 | new_end -= wr_mas->offset_end - mas->offset; | |
4240 | if (wr_mas->r_min == mas->index) | |
4241 | new_end--; | |
4242 | ||
4243 | if (wr_mas->end_piv == mas->last) | |
4244 | new_end--; | |
4245 | ||
4246 | return new_end; | |
4247 | } | |
4248 | ||
2e1da329 PZ |
4249 | /* |
4250 | * mas_wr_append: Attempt to append | |
4251 | * @wr_mas: the maple write state | |
4252 | * | |
4253 | * Return: True if appended, false otherwise | |
4254 | */ | |
23e9dde0 PZ |
4255 | static inline bool mas_wr_append(struct ma_wr_state *wr_mas, |
4256 | unsigned char new_end) | |
54a611b6 LH |
4257 | { |
4258 | unsigned char end = wr_mas->node_end; | |
54a611b6 LH |
4259 | struct ma_state *mas = wr_mas->mas; |
4260 | unsigned char node_pivots = mt_pivots[wr_mas->type]; | |
4261 | ||
2e1da329 PZ |
4262 | if (mas->offset != wr_mas->node_end) |
4263 | return false; | |
54a611b6 | 4264 | |
2e1da329 PZ |
4265 | if (new_end < node_pivots) { |
4266 | wr_mas->pivots[new_end] = wr_mas->pivots[end]; | |
4267 | ma_set_meta(wr_mas->node, maple_leaf_64, 0, new_end); | |
4268 | } | |
54a611b6 | 4269 | |
23e9dde0 PZ |
4270 | if (new_end == wr_mas->node_end + 1) { |
4271 | if (mas->last == wr_mas->r_max) { | |
4272 | /* Append to end of range */ | |
4273 | rcu_assign_pointer(wr_mas->slots[new_end], | |
4274 | wr_mas->entry); | |
4275 | wr_mas->pivots[end] = mas->index - 1; | |
4276 | mas->offset = new_end; | |
4277 | } else { | |
4278 | /* Append to start of range */ | |
4279 | rcu_assign_pointer(wr_mas->slots[new_end], | |
4280 | wr_mas->content); | |
4281 | wr_mas->pivots[end] = mas->last; | |
4282 | rcu_assign_pointer(wr_mas->slots[end], wr_mas->entry); | |
4283 | } | |
2e1da329 | 4284 | } else { |
23e9dde0 | 4285 | /* Append to the range without touching any boundaries. */ |
54a611b6 | 4286 | rcu_assign_pointer(wr_mas->slots[new_end], wr_mas->content); |
23e9dde0 PZ |
4287 | wr_mas->pivots[end + 1] = mas->last; |
4288 | rcu_assign_pointer(wr_mas->slots[end + 1], wr_mas->entry); | |
4289 | wr_mas->pivots[end] = mas->index - 1; | |
4290 | mas->offset = end + 1; | |
54a611b6 LH |
4291 | } |
4292 | ||
2e1da329 PZ |
4293 | if (!wr_mas->content || !wr_mas->entry) |
4294 | mas_update_gap(mas); | |
4295 | ||
4296 | return true; | |
54a611b6 LH |
4297 | } |
4298 | ||
4299 | /* | |
4300 | * mas_wr_bnode() - Slow path for a modification. | |
4301 | * @wr_mas: The write maple state | |
4302 | * | |
4303 | * This is where split, rebalance end up. | |
4304 | */ | |
4305 | static void mas_wr_bnode(struct ma_wr_state *wr_mas) | |
4306 | { | |
4307 | struct maple_big_node b_node; | |
4308 | ||
4309 | trace_ma_write(__func__, wr_mas->mas, 0, wr_mas->entry); | |
4310 | memset(&b_node, 0, sizeof(struct maple_big_node)); | |
4311 | mas_store_b_node(wr_mas, &b_node, wr_mas->offset_end); | |
4312 | mas_commit_b_node(wr_mas, &b_node, wr_mas->node_end); | |
4313 | } | |
4314 | ||
4315 | static inline void mas_wr_modify(struct ma_wr_state *wr_mas) | |
4316 | { | |
54a611b6 | 4317 | struct ma_state *mas = wr_mas->mas; |
c6fc9e4a | 4318 | unsigned char new_end; |
54a611b6 LH |
4319 | |
4320 | /* Direct replacement */ | |
4321 | if (wr_mas->r_min == mas->index && wr_mas->r_max == mas->last) { | |
4322 | rcu_assign_pointer(wr_mas->slots[mas->offset], wr_mas->entry); | |
4323 | if (!!wr_mas->entry ^ !!wr_mas->content) | |
4324 | mas_update_gap(mas); | |
4325 | return; | |
4326 | } | |
4327 | ||
c6fc9e4a PZ |
4328 | /* |
4329 | * new_end exceeds the size of the maple node and cannot enter the fast | |
4330 | * path. | |
4331 | */ | |
4332 | new_end = mas_wr_new_end(wr_mas); | |
4333 | if (new_end >= mt_slots[wr_mas->type]) | |
54a611b6 LH |
4334 | goto slow_path; |
4335 | ||
2e1da329 | 4336 | /* Attempt to append */ |
23e9dde0 | 4337 | if (mas_wr_append(wr_mas, new_end)) |
54a611b6 | 4338 | return; |
54a611b6 | 4339 | |
e6d1ffd6 | 4340 | if (new_end == wr_mas->node_end && mas_wr_slot_store(wr_mas)) |
54a611b6 | 4341 | return; |
7a03ae39 PZ |
4342 | |
4343 | if (mas_wr_node_store(wr_mas, new_end)) | |
54a611b6 LH |
4344 | return; |
4345 | ||
4346 | if (mas_is_err(mas)) | |
4347 | return; | |
4348 | ||
4349 | slow_path: | |
4350 | mas_wr_bnode(wr_mas); | |
4351 | } | |
4352 | ||
4353 | /* | |
4354 | * mas_wr_store_entry() - Internal call to store a value | |
4355 | * @mas: The maple state | |
4356 | * @entry: The entry to store. | |
4357 | * | |
4358 | * Return: The contents that was stored at the index. | |
4359 | */ | |
4360 | static inline void *mas_wr_store_entry(struct ma_wr_state *wr_mas) | |
4361 | { | |
4362 | struct ma_state *mas = wr_mas->mas; | |
4363 | ||
4364 | wr_mas->content = mas_start(mas); | |
4365 | if (mas_is_none(mas) || mas_is_ptr(mas)) { | |
4366 | mas_store_root(mas, wr_mas->entry); | |
4367 | return wr_mas->content; | |
4368 | } | |
4369 | ||
4370 | if (unlikely(!mas_wr_walk(wr_mas))) { | |
4371 | mas_wr_spanning_store(wr_mas); | |
4372 | return wr_mas->content; | |
4373 | } | |
4374 | ||
4375 | /* At this point, we are at the leaf node that needs to be altered. */ | |
54a611b6 | 4376 | mas_wr_end_piv(wr_mas); |
54a611b6 LH |
4377 | /* New root for a single pointer */ |
4378 | if (unlikely(!mas->index && mas->last == ULONG_MAX)) { | |
4379 | mas_new_root(mas, wr_mas->entry); | |
4380 | return wr_mas->content; | |
4381 | } | |
4382 | ||
4383 | mas_wr_modify(wr_mas); | |
4384 | return wr_mas->content; | |
4385 | } | |
4386 | ||
4387 | /** | |
4388 | * mas_insert() - Internal call to insert a value | |
4389 | * @mas: The maple state | |
4390 | * @entry: The entry to store | |
4391 | * | |
4392 | * Return: %NULL or the contents that already exists at the requested index | |
4393 | * otherwise. The maple state needs to be checked for error conditions. | |
4394 | */ | |
4395 | static inline void *mas_insert(struct ma_state *mas, void *entry) | |
4396 | { | |
4397 | MA_WR_STATE(wr_mas, mas, entry); | |
4398 | ||
4399 | /* | |
4400 | * Inserting a new range inserts either 0, 1, or 2 pivots within the | |
4401 | * tree. If the insert fits exactly into an existing gap with a value | |
4402 | * of NULL, then the slot only needs to be written with the new value. | |
4403 | * If the range being inserted is adjacent to another range, then only a | |
4404 | * single pivot needs to be inserted (as well as writing the entry). If | |
4405 | * the new range is within a gap but does not touch any other ranges, | |
4406 | * then two pivots need to be inserted: the start - 1, and the end. As | |
4407 | * usual, the entry must be written. Most operations require a new node | |
4408 | * to be allocated and replace an existing node to ensure RCU safety, | |
4409 | * when in RCU mode. The exception to requiring a newly allocated node | |
4410 | * is when inserting at the end of a node (appending). When done | |
4411 | * carefully, appending can reuse the node in place. | |
4412 | */ | |
4413 | wr_mas.content = mas_start(mas); | |
4414 | if (wr_mas.content) | |
4415 | goto exists; | |
4416 | ||
4417 | if (mas_is_none(mas) || mas_is_ptr(mas)) { | |
4418 | mas_store_root(mas, entry); | |
4419 | return NULL; | |
4420 | } | |
4421 | ||
4422 | /* spanning writes always overwrite something */ | |
4423 | if (!mas_wr_walk(&wr_mas)) | |
4424 | goto exists; | |
4425 | ||
4426 | /* At this point, we are at the leaf node that needs to be altered. */ | |
4427 | wr_mas.offset_end = mas->offset; | |
4428 | wr_mas.end_piv = wr_mas.r_max; | |
4429 | ||
4430 | if (wr_mas.content || (mas->last > wr_mas.r_max)) | |
4431 | goto exists; | |
4432 | ||
4433 | if (!entry) | |
4434 | return NULL; | |
4435 | ||
4436 | mas_wr_modify(&wr_mas); | |
4437 | return wr_mas.content; | |
4438 | ||
4439 | exists: | |
4440 | mas_set_err(mas, -EEXIST); | |
4441 | return wr_mas.content; | |
4442 | ||
4443 | } | |
4444 | ||
de6e386c LH |
4445 | static inline void mas_rewalk(struct ma_state *mas, unsigned long index) |
4446 | { | |
4447 | retry: | |
4448 | mas_set(mas, index); | |
4449 | mas_state_walk(mas); | |
4450 | if (mas_is_start(mas)) | |
4451 | goto retry; | |
4452 | } | |
4453 | ||
4454 | static inline bool mas_rewalk_if_dead(struct ma_state *mas, | |
4455 | struct maple_node *node, const unsigned long index) | |
4456 | { | |
4457 | if (unlikely(ma_dead_node(node))) { | |
4458 | mas_rewalk(mas, index); | |
4459 | return true; | |
4460 | } | |
4461 | return false; | |
4462 | } | |
4463 | ||
54a611b6 LH |
4464 | /* |
4465 | * mas_prev_node() - Find the prev non-null entry at the same level in the | |
4466 | * tree. The prev value will be mas->node[mas->offset] or MAS_NONE. | |
4467 | * @mas: The maple state | |
4468 | * @min: The lower limit to search | |
4469 | * | |
4470 | * The prev node value will be mas->node[mas->offset] or MAS_NONE. | |
4471 | * Return: 1 if the node is dead, 0 otherwise. | |
4472 | */ | |
4473 | static inline int mas_prev_node(struct ma_state *mas, unsigned long min) | |
4474 | { | |
4475 | enum maple_type mt; | |
4476 | int offset, level; | |
4477 | void __rcu **slots; | |
4478 | struct maple_node *node; | |
54a611b6 | 4479 | unsigned long *pivots; |
dd9a8513 | 4480 | unsigned long max; |
54a611b6 | 4481 | |
dd9a8513 LH |
4482 | node = mas_mn(mas); |
4483 | if (!mas->min) | |
4484 | goto no_entry; | |
4485 | ||
4486 | max = mas->min - 1; | |
4487 | if (max < min) | |
4488 | goto no_entry; | |
54a611b6 LH |
4489 | |
4490 | level = 0; | |
4491 | do { | |
54a611b6 LH |
4492 | if (ma_is_root(node)) |
4493 | goto no_entry; | |
4494 | ||
4495 | /* Walk up. */ | |
4496 | if (unlikely(mas_ascend(mas))) | |
4497 | return 1; | |
4498 | offset = mas->offset; | |
4499 | level++; | |
dd9a8513 | 4500 | node = mas_mn(mas); |
54a611b6 LH |
4501 | } while (!offset); |
4502 | ||
4503 | offset--; | |
4504 | mt = mte_node_type(mas->node); | |
54a611b6 LH |
4505 | while (level > 1) { |
4506 | level--; | |
dd9a8513 LH |
4507 | slots = ma_slots(node, mt); |
4508 | mas->node = mas_slot(mas, slots, offset); | |
54a611b6 LH |
4509 | if (unlikely(ma_dead_node(node))) |
4510 | return 1; | |
4511 | ||
54a611b6 LH |
4512 | mt = mte_node_type(mas->node); |
4513 | node = mas_mn(mas); | |
54a611b6 | 4514 | pivots = ma_pivots(node, mt); |
dd9a8513 | 4515 | offset = ma_data_end(node, mt, pivots, max); |
39d0bd86 LH |
4516 | if (unlikely(ma_dead_node(node))) |
4517 | return 1; | |
54a611b6 LH |
4518 | } |
4519 | ||
dd9a8513 | 4520 | slots = ma_slots(node, mt); |
54a611b6 | 4521 | mas->node = mas_slot(mas, slots, offset); |
dd9a8513 | 4522 | pivots = ma_pivots(node, mt); |
54a611b6 LH |
4523 | if (unlikely(ma_dead_node(node))) |
4524 | return 1; | |
4525 | ||
dd9a8513 LH |
4526 | if (likely(offset)) |
4527 | mas->min = pivots[offset - 1] + 1; | |
4528 | mas->max = max; | |
54a611b6 LH |
4529 | mas->offset = mas_data_end(mas); |
4530 | if (unlikely(mte_dead_node(mas->node))) | |
4531 | return 1; | |
4532 | ||
4533 | return 0; | |
4534 | ||
54a611b6 LH |
4535 | no_entry: |
4536 | if (unlikely(ma_dead_node(node))) | |
4537 | return 1; | |
4538 | ||
4539 | mas->node = MAS_NONE; | |
4540 | return 0; | |
4541 | } | |
4542 | ||
dd9a8513 LH |
4543 | /* |
4544 | * mas_prev_slot() - Get the entry in the previous slot | |
4545 | * | |
4546 | * @mas: The maple state | |
4547 | * @max: The minimum starting range | |
4548 | * | |
4549 | * Return: The entry in the previous slot which is possibly NULL | |
4550 | */ | |
4551 | static void *mas_prev_slot(struct ma_state *mas, unsigned long min, bool empty) | |
4552 | { | |
4553 | void *entry; | |
4554 | void __rcu **slots; | |
4555 | unsigned long pivot; | |
4556 | enum maple_type type; | |
4557 | unsigned long *pivots; | |
4558 | struct maple_node *node; | |
4559 | unsigned long save_point = mas->index; | |
4560 | ||
4561 | retry: | |
4562 | node = mas_mn(mas); | |
4563 | type = mte_node_type(mas->node); | |
4564 | pivots = ma_pivots(node, type); | |
4565 | if (unlikely(mas_rewalk_if_dead(mas, node, save_point))) | |
4566 | goto retry; | |
4567 | ||
4568 | again: | |
4569 | if (mas->min <= min) { | |
4570 | pivot = mas_safe_min(mas, pivots, mas->offset); | |
4571 | ||
4572 | if (unlikely(mas_rewalk_if_dead(mas, node, save_point))) | |
4573 | goto retry; | |
4574 | ||
4575 | if (pivot <= min) | |
4576 | return NULL; | |
4577 | } | |
4578 | ||
4579 | if (likely(mas->offset)) { | |
4580 | mas->offset--; | |
4581 | mas->last = mas->index - 1; | |
4582 | mas->index = mas_safe_min(mas, pivots, mas->offset); | |
4583 | } else { | |
4584 | if (mas_prev_node(mas, min)) { | |
4585 | mas_rewalk(mas, save_point); | |
4586 | goto retry; | |
4587 | } | |
4588 | ||
4589 | if (mas_is_none(mas)) | |
4590 | return NULL; | |
4591 | ||
4592 | mas->last = mas->max; | |
4593 | node = mas_mn(mas); | |
4594 | type = mte_node_type(mas->node); | |
4595 | pivots = ma_pivots(node, type); | |
4596 | mas->index = pivots[mas->offset - 1] + 1; | |
4597 | } | |
4598 | ||
4599 | slots = ma_slots(node, type); | |
4600 | entry = mas_slot(mas, slots, mas->offset); | |
4601 | if (unlikely(mas_rewalk_if_dead(mas, node, save_point))) | |
4602 | goto retry; | |
4603 | ||
4604 | if (likely(entry)) | |
4605 | return entry; | |
4606 | ||
4607 | if (!empty) | |
4608 | goto again; | |
4609 | ||
4610 | return entry; | |
4611 | } | |
4612 | ||
54a611b6 LH |
4613 | /* |
4614 | * mas_next_node() - Get the next node at the same level in the tree. | |
4615 | * @mas: The maple state | |
4616 | * @max: The maximum pivot value to check. | |
4617 | * | |
4618 | * The next value will be mas->node[mas->offset] or MAS_NONE. | |
4619 | * Return: 1 on dead node, 0 otherwise. | |
4620 | */ | |
4621 | static inline int mas_next_node(struct ma_state *mas, struct maple_node *node, | |
4622 | unsigned long max) | |
4623 | { | |
fff4a58c | 4624 | unsigned long min; |
54a611b6 LH |
4625 | unsigned long *pivots; |
4626 | struct maple_enode *enode; | |
4627 | int level = 0; | |
39d0bd86 | 4628 | unsigned char node_end; |
54a611b6 LH |
4629 | enum maple_type mt; |
4630 | void __rcu **slots; | |
4631 | ||
4632 | if (mas->max >= max) | |
4633 | goto no_entry; | |
4634 | ||
fff4a58c | 4635 | min = mas->max + 1; |
54a611b6 LH |
4636 | level = 0; |
4637 | do { | |
4638 | if (ma_is_root(node)) | |
4639 | goto no_entry; | |
4640 | ||
fff4a58c | 4641 | /* Walk up. */ |
54a611b6 LH |
4642 | if (unlikely(mas_ascend(mas))) |
4643 | return 1; | |
4644 | ||
54a611b6 LH |
4645 | level++; |
4646 | node = mas_mn(mas); | |
4647 | mt = mte_node_type(mas->node); | |
4648 | pivots = ma_pivots(node, mt); | |
39d0bd86 LH |
4649 | node_end = ma_data_end(node, mt, pivots, mas->max); |
4650 | if (unlikely(ma_dead_node(node))) | |
4651 | return 1; | |
4652 | ||
fff4a58c | 4653 | } while (unlikely(mas->offset == node_end)); |
54a611b6 LH |
4654 | |
4655 | slots = ma_slots(node, mt); | |
fff4a58c LH |
4656 | mas->offset++; |
4657 | enode = mas_slot(mas, slots, mas->offset); | |
4658 | if (unlikely(ma_dead_node(node))) | |
4659 | return 1; | |
54a611b6 | 4660 | |
fff4a58c LH |
4661 | if (level > 1) |
4662 | mas->offset = 0; | |
4663 | ||
4664 | while (unlikely(level > 1)) { | |
54a611b6 | 4665 | level--; |
fff4a58c | 4666 | mas->node = enode; |
54a611b6 LH |
4667 | node = mas_mn(mas); |
4668 | mt = mte_node_type(mas->node); | |
4669 | slots = ma_slots(node, mt); | |
fff4a58c | 4670 | enode = mas_slot(mas, slots, 0); |
39d0bd86 LH |
4671 | if (unlikely(ma_dead_node(node))) |
4672 | return 1; | |
54a611b6 LH |
4673 | } |
4674 | ||
fff4a58c LH |
4675 | if (!mas->offset) |
4676 | pivots = ma_pivots(node, mt); | |
4677 | ||
4678 | mas->max = mas_safe_pivot(mas, pivots, mas->offset, mt); | |
54a611b6 LH |
4679 | if (unlikely(ma_dead_node(node))) |
4680 | return 1; | |
4681 | ||
4682 | mas->node = enode; | |
4683 | mas->min = min; | |
54a611b6 LH |
4684 | return 0; |
4685 | ||
4686 | no_entry: | |
4687 | if (unlikely(ma_dead_node(node))) | |
4688 | return 1; | |
4689 | ||
4690 | mas->node = MAS_NONE; | |
4691 | return 0; | |
4692 | } | |
4693 | ||
4694 | /* | |
fff4a58c | 4695 | * mas_next_slot() - Get the entry in the next slot |
54a611b6 | 4696 | * |
fff4a58c LH |
4697 | * @mas: The maple state |
4698 | * @max: The maximum starting range | |
4699 | * @empty: Can be empty | |
54a611b6 | 4700 | * |
fff4a58c | 4701 | * Return: The entry in the next slot which is possibly NULL |
54a611b6 | 4702 | */ |
fff4a58c | 4703 | static void *mas_next_slot(struct ma_state *mas, unsigned long max, bool empty) |
54a611b6 | 4704 | { |
54a611b6 | 4705 | void __rcu **slots; |
fff4a58c LH |
4706 | unsigned long *pivots; |
4707 | unsigned long pivot; | |
4708 | enum maple_type type; | |
4709 | struct maple_node *node; | |
4710 | unsigned char data_end; | |
4711 | unsigned long save_point = mas->last; | |
54a611b6 LH |
4712 | void *entry; |
4713 | ||
fff4a58c LH |
4714 | retry: |
4715 | node = mas_mn(mas); | |
4716 | type = mte_node_type(mas->node); | |
39d0bd86 | 4717 | pivots = ma_pivots(node, type); |
fff4a58c LH |
4718 | data_end = ma_data_end(node, type, pivots, mas->max); |
4719 | if (unlikely(mas_rewalk_if_dead(mas, node, save_point))) | |
4720 | goto retry; | |
54a611b6 | 4721 | |
fff4a58c LH |
4722 | again: |
4723 | if (mas->max >= max) { | |
4724 | if (likely(mas->offset < data_end)) | |
4725 | pivot = pivots[mas->offset]; | |
4726 | else | |
4727 | return NULL; /* must be mas->max */ | |
54a611b6 | 4728 | |
fff4a58c LH |
4729 | if (unlikely(mas_rewalk_if_dead(mas, node, save_point))) |
4730 | goto retry; | |
54a611b6 LH |
4731 | |
4732 | if (pivot >= max) | |
4733 | return NULL; | |
fff4a58c | 4734 | } |
54a611b6 | 4735 | |
fff4a58c LH |
4736 | if (likely(mas->offset < data_end)) { |
4737 | mas->index = pivots[mas->offset] + 1; | |
4738 | mas->offset++; | |
4739 | if (likely(mas->offset < data_end)) | |
4740 | mas->last = pivots[mas->offset]; | |
4741 | else | |
4742 | mas->last = mas->max; | |
4743 | } else { | |
4744 | if (mas_next_node(mas, node, max)) { | |
4745 | mas_rewalk(mas, save_point); | |
4746 | goto retry; | |
4747 | } | |
4748 | ||
4749 | if (mas_is_none(mas)) | |
ca80f610 LH |
4750 | return NULL; |
4751 | ||
fff4a58c LH |
4752 | mas->offset = 0; |
4753 | mas->index = mas->min; | |
4754 | node = mas_mn(mas); | |
4755 | type = mte_node_type(mas->node); | |
4756 | pivots = ma_pivots(node, type); | |
4757 | mas->last = pivots[0]; | |
54a611b6 LH |
4758 | } |
4759 | ||
fff4a58c LH |
4760 | slots = ma_slots(node, type); |
4761 | entry = mt_slot(mas->tree, slots, mas->offset); | |
4762 | if (unlikely(mas_rewalk_if_dead(mas, node, save_point))) | |
4763 | goto retry; | |
54a611b6 | 4764 | |
fff4a58c LH |
4765 | if (entry) |
4766 | return entry; | |
54a611b6 | 4767 | |
fff4a58c LH |
4768 | if (!empty) { |
4769 | if (!mas->offset) | |
4770 | data_end = 2; | |
4771 | goto again; | |
4772 | } | |
4773 | ||
4774 | return entry; | |
54a611b6 LH |
4775 | } |
4776 | ||
4777 | /* | |
4778 | * mas_next_entry() - Internal function to get the next entry. | |
4779 | * @mas: The maple state | |
4780 | * @limit: The maximum range start. | |
4781 | * | |
4782 | * Set the @mas->node to the next entry and the range_start to | |
4783 | * the beginning value for the entry. Does not check beyond @limit. | |
4784 | * Sets @mas->index and @mas->last to the limit if it is hit. | |
4785 | * Restarts on dead nodes. | |
4786 | * | |
4787 | * Return: the next entry or %NULL. | |
4788 | */ | |
4789 | static inline void *mas_next_entry(struct ma_state *mas, unsigned long limit) | |
4790 | { | |
ca80f610 | 4791 | if (mas->last >= limit) |
50e81c82 | 4792 | return NULL; |
ca80f610 | 4793 | |
6169b553 | 4794 | return mas_next_slot(mas, limit, false); |
54a611b6 LH |
4795 | } |
4796 | ||
54a611b6 LH |
4797 | /* |
4798 | * mas_rev_awalk() - Internal function. Reverse allocation walk. Find the | |
4799 | * highest gap address of a given size in a given node and descend. | |
4800 | * @mas: The maple state | |
4801 | * @size: The needed size. | |
4802 | * | |
4803 | * Return: True if found in a leaf, false otherwise. | |
4804 | * | |
4805 | */ | |
fad8e429 LH |
4806 | static bool mas_rev_awalk(struct ma_state *mas, unsigned long size, |
4807 | unsigned long *gap_min, unsigned long *gap_max) | |
54a611b6 LH |
4808 | { |
4809 | enum maple_type type = mte_node_type(mas->node); | |
4810 | struct maple_node *node = mas_mn(mas); | |
4811 | unsigned long *pivots, *gaps; | |
4812 | void __rcu **slots; | |
4813 | unsigned long gap = 0; | |
7327e811 | 4814 | unsigned long max, min; |
54a611b6 LH |
4815 | unsigned char offset; |
4816 | ||
4817 | if (unlikely(mas_is_err(mas))) | |
4818 | return true; | |
4819 | ||
4820 | if (ma_is_dense(type)) { | |
4821 | /* dense nodes. */ | |
4822 | mas->offset = (unsigned char)(mas->index - mas->min); | |
4823 | return true; | |
4824 | } | |
4825 | ||
4826 | pivots = ma_pivots(node, type); | |
4827 | slots = ma_slots(node, type); | |
4828 | gaps = ma_gaps(node, type); | |
4829 | offset = mas->offset; | |
4830 | min = mas_safe_min(mas, pivots, offset); | |
4831 | /* Skip out of bounds. */ | |
4832 | while (mas->last < min) | |
4833 | min = mas_safe_min(mas, pivots, --offset); | |
4834 | ||
4835 | max = mas_safe_pivot(mas, pivots, offset, type); | |
7327e811 | 4836 | while (mas->index <= max) { |
54a611b6 LH |
4837 | gap = 0; |
4838 | if (gaps) | |
4839 | gap = gaps[offset]; | |
4840 | else if (!mas_slot(mas, slots, offset)) | |
4841 | gap = max - min + 1; | |
4842 | ||
4843 | if (gap) { | |
4844 | if ((size <= gap) && (size <= mas->last - min + 1)) | |
4845 | break; | |
4846 | ||
4847 | if (!gaps) { | |
4848 | /* Skip the next slot, it cannot be a gap. */ | |
4849 | if (offset < 2) | |
4850 | goto ascend; | |
4851 | ||
4852 | offset -= 2; | |
4853 | max = pivots[offset]; | |
4854 | min = mas_safe_min(mas, pivots, offset); | |
4855 | continue; | |
4856 | } | |
4857 | } | |
4858 | ||
4859 | if (!offset) | |
4860 | goto ascend; | |
4861 | ||
4862 | offset--; | |
4863 | max = min - 1; | |
4864 | min = mas_safe_min(mas, pivots, offset); | |
4865 | } | |
4866 | ||
7327e811 LH |
4867 | if (unlikely((mas->index > max) || (size - 1 > max - mas->index))) |
4868 | goto no_space; | |
54a611b6 LH |
4869 | |
4870 | if (unlikely(ma_is_leaf(type))) { | |
4871 | mas->offset = offset; | |
fad8e429 LH |
4872 | *gap_min = min; |
4873 | *gap_max = min + gap - 1; | |
54a611b6 LH |
4874 | return true; |
4875 | } | |
4876 | ||
4877 | /* descend, only happens under lock. */ | |
4878 | mas->node = mas_slot(mas, slots, offset); | |
4879 | mas->min = min; | |
4880 | mas->max = max; | |
4881 | mas->offset = mas_data_end(mas); | |
4882 | return false; | |
4883 | ||
4884 | ascend: | |
7327e811 LH |
4885 | if (!mte_is_root(mas->node)) |
4886 | return false; | |
54a611b6 | 4887 | |
7327e811 LH |
4888 | no_space: |
4889 | mas_set_err(mas, -EBUSY); | |
54a611b6 LH |
4890 | return false; |
4891 | } | |
4892 | ||
4893 | static inline bool mas_anode_descend(struct ma_state *mas, unsigned long size) | |
4894 | { | |
4895 | enum maple_type type = mte_node_type(mas->node); | |
4896 | unsigned long pivot, min, gap = 0; | |
06e8fd99 LH |
4897 | unsigned char offset, data_end; |
4898 | unsigned long *gaps, *pivots; | |
4899 | void __rcu **slots; | |
4900 | struct maple_node *node; | |
54a611b6 LH |
4901 | bool found = false; |
4902 | ||
4903 | if (ma_is_dense(type)) { | |
4904 | mas->offset = (unsigned char)(mas->index - mas->min); | |
4905 | return true; | |
4906 | } | |
4907 | ||
06e8fd99 LH |
4908 | node = mas_mn(mas); |
4909 | pivots = ma_pivots(node, type); | |
4910 | slots = ma_slots(node, type); | |
4911 | gaps = ma_gaps(node, type); | |
54a611b6 | 4912 | offset = mas->offset; |
54a611b6 | 4913 | min = mas_safe_min(mas, pivots, offset); |
06e8fd99 LH |
4914 | data_end = ma_data_end(node, type, pivots, mas->max); |
4915 | for (; offset <= data_end; offset++) { | |
29b2681f | 4916 | pivot = mas_safe_pivot(mas, pivots, offset, type); |
54a611b6 LH |
4917 | |
4918 | /* Not within lower bounds */ | |
4919 | if (mas->index > pivot) | |
4920 | goto next_slot; | |
4921 | ||
4922 | if (gaps) | |
4923 | gap = gaps[offset]; | |
4924 | else if (!mas_slot(mas, slots, offset)) | |
4925 | gap = min(pivot, mas->last) - max(mas->index, min) + 1; | |
4926 | else | |
4927 | goto next_slot; | |
4928 | ||
4929 | if (gap >= size) { | |
4930 | if (ma_is_leaf(type)) { | |
4931 | found = true; | |
4932 | goto done; | |
4933 | } | |
4934 | if (mas->index <= pivot) { | |
4935 | mas->node = mas_slot(mas, slots, offset); | |
4936 | mas->min = min; | |
4937 | mas->max = pivot; | |
4938 | offset = 0; | |
54a611b6 LH |
4939 | break; |
4940 | } | |
4941 | } | |
4942 | next_slot: | |
4943 | min = pivot + 1; | |
4944 | if (mas->last <= pivot) { | |
4945 | mas_set_err(mas, -EBUSY); | |
4946 | return true; | |
4947 | } | |
4948 | } | |
4949 | ||
4950 | if (mte_is_root(mas->node)) | |
4951 | found = true; | |
4952 | done: | |
4953 | mas->offset = offset; | |
4954 | return found; | |
4955 | } | |
4956 | ||
4957 | /** | |
4958 | * mas_walk() - Search for @mas->index in the tree. | |
4959 | * @mas: The maple state. | |
4960 | * | |
4961 | * mas->index and mas->last will be set to the range if there is a value. If | |
4962 | * mas->node is MAS_NONE, reset to MAS_START. | |
4963 | * | |
4964 | * Return: the entry at the location or %NULL. | |
4965 | */ | |
4966 | void *mas_walk(struct ma_state *mas) | |
4967 | { | |
4968 | void *entry; | |
4969 | ||
6b23a290 LH |
4970 | if (mas_is_none(mas) || mas_is_paused(mas) || mas_is_ptr(mas)) |
4971 | mas->node = MAS_START; | |
54a611b6 LH |
4972 | retry: |
4973 | entry = mas_state_walk(mas); | |
6b23a290 | 4974 | if (mas_is_start(mas)) { |
54a611b6 | 4975 | goto retry; |
6b23a290 LH |
4976 | } else if (mas_is_none(mas)) { |
4977 | mas->index = 0; | |
4978 | mas->last = ULONG_MAX; | |
4979 | } else if (mas_is_ptr(mas)) { | |
54a611b6 LH |
4980 | if (!mas->index) { |
4981 | mas->last = 0; | |
6b23a290 | 4982 | return entry; |
54a611b6 | 4983 | } |
54a611b6 | 4984 | |
6b23a290 | 4985 | mas->index = 1; |
54a611b6 | 4986 | mas->last = ULONG_MAX; |
6b23a290 LH |
4987 | mas->node = MAS_NONE; |
4988 | return NULL; | |
54a611b6 LH |
4989 | } |
4990 | ||
4991 | return entry; | |
4992 | } | |
120b1162 | 4993 | EXPORT_SYMBOL_GPL(mas_walk); |
54a611b6 LH |
4994 | |
4995 | static inline bool mas_rewind_node(struct ma_state *mas) | |
4996 | { | |
4997 | unsigned char slot; | |
4998 | ||
4999 | do { | |
5000 | if (mte_is_root(mas->node)) { | |
5001 | slot = mas->offset; | |
5002 | if (!slot) | |
5003 | return false; | |
5004 | } else { | |
5005 | mas_ascend(mas); | |
5006 | slot = mas->offset; | |
5007 | } | |
5008 | } while (!slot); | |
5009 | ||
5010 | mas->offset = --slot; | |
5011 | return true; | |
5012 | } | |
5013 | ||
5014 | /* | |
5015 | * mas_skip_node() - Internal function. Skip over a node. | |
5016 | * @mas: The maple state. | |
5017 | * | |
5018 | * Return: true if there is another node, false otherwise. | |
5019 | */ | |
5020 | static inline bool mas_skip_node(struct ma_state *mas) | |
5021 | { | |
0fa99fdf LH |
5022 | if (mas_is_err(mas)) |
5023 | return false; | |
54a611b6 | 5024 | |
54a611b6 LH |
5025 | do { |
5026 | if (mte_is_root(mas->node)) { | |
0fa99fdf | 5027 | if (mas->offset >= mas_data_end(mas)) { |
54a611b6 LH |
5028 | mas_set_err(mas, -EBUSY); |
5029 | return false; | |
5030 | } | |
5031 | } else { | |
5032 | mas_ascend(mas); | |
54a611b6 | 5033 | } |
0fa99fdf | 5034 | } while (mas->offset >= mas_data_end(mas)); |
54a611b6 | 5035 | |
0fa99fdf | 5036 | mas->offset++; |
54a611b6 LH |
5037 | return true; |
5038 | } | |
5039 | ||
5040 | /* | |
5041 | * mas_awalk() - Allocation walk. Search from low address to high, for a gap of | |
5042 | * @size | |
5043 | * @mas: The maple state | |
5044 | * @size: The size of the gap required | |
5045 | * | |
5046 | * Search between @mas->index and @mas->last for a gap of @size. | |
5047 | */ | |
5048 | static inline void mas_awalk(struct ma_state *mas, unsigned long size) | |
5049 | { | |
5050 | struct maple_enode *last = NULL; | |
5051 | ||
5052 | /* | |
5053 | * There are 4 options: | |
5054 | * go to child (descend) | |
5055 | * go back to parent (ascend) | |
5056 | * no gap found. (return, slot == MAPLE_NODE_SLOTS) | |
5057 | * found the gap. (return, slot != MAPLE_NODE_SLOTS) | |
5058 | */ | |
5059 | while (!mas_is_err(mas) && !mas_anode_descend(mas, size)) { | |
5060 | if (last == mas->node) | |
5061 | mas_skip_node(mas); | |
5062 | else | |
5063 | last = mas->node; | |
5064 | } | |
5065 | } | |
5066 | ||
54a611b6 LH |
5067 | /* |
5068 | * mas_sparse_area() - Internal function. Return upper or lower limit when | |
5069 | * searching for a gap in an empty tree. | |
5070 | * @mas: The maple state | |
5071 | * @min: the minimum range | |
5072 | * @max: The maximum range | |
5073 | * @size: The size of the gap | |
5074 | * @fwd: Searching forward or back | |
5075 | */ | |
29ad6bb3 | 5076 | static inline int mas_sparse_area(struct ma_state *mas, unsigned long min, |
54a611b6 LH |
5077 | unsigned long max, unsigned long size, bool fwd) |
5078 | { | |
29ad6bb3 PZ |
5079 | if (!unlikely(mas_is_none(mas)) && min == 0) { |
5080 | min++; | |
5081 | /* | |
5082 | * At this time, min is increased, we need to recheck whether | |
5083 | * the size is satisfied. | |
5084 | */ | |
5085 | if (min > max || max - min + 1 < size) | |
5086 | return -EBUSY; | |
5087 | } | |
54a611b6 LH |
5088 | /* mas_is_ptr */ |
5089 | ||
54a611b6 | 5090 | if (fwd) { |
29ad6bb3 PZ |
5091 | mas->index = min; |
5092 | mas->last = min + size - 1; | |
5093 | } else { | |
5094 | mas->last = max; | |
5095 | mas->index = max - size + 1; | |
54a611b6 | 5096 | } |
29ad6bb3 | 5097 | return 0; |
54a611b6 LH |
5098 | } |
5099 | ||
5100 | /* | |
5101 | * mas_empty_area() - Get the lowest address within the range that is | |
5102 | * sufficient for the size requested. | |
5103 | * @mas: The maple state | |
5104 | * @min: The lowest value of the range | |
5105 | * @max: The highest value of the range | |
5106 | * @size: The size needed | |
5107 | */ | |
5108 | int mas_empty_area(struct ma_state *mas, unsigned long min, | |
5109 | unsigned long max, unsigned long size) | |
5110 | { | |
5111 | unsigned char offset; | |
5112 | unsigned long *pivots; | |
5113 | enum maple_type mt; | |
5114 | ||
ba997212 LH |
5115 | if (min > max) |
5116 | return -EINVAL; | |
5117 | ||
5118 | if (size == 0 || max - min < size - 1) | |
fad8e429 LH |
5119 | return -EINVAL; |
5120 | ||
54a611b6 LH |
5121 | if (mas_is_start(mas)) |
5122 | mas_start(mas); | |
5123 | else if (mas->offset >= 2) | |
5124 | mas->offset -= 2; | |
5125 | else if (!mas_skip_node(mas)) | |
5126 | return -EBUSY; | |
5127 | ||
5128 | /* Empty set */ | |
29ad6bb3 PZ |
5129 | if (mas_is_none(mas) || mas_is_ptr(mas)) |
5130 | return mas_sparse_area(mas, min, max, size, true); | |
54a611b6 LH |
5131 | |
5132 | /* The start of the window can only be within these values */ | |
5133 | mas->index = min; | |
5134 | mas->last = max; | |
5135 | mas_awalk(mas, size); | |
5136 | ||
5137 | if (unlikely(mas_is_err(mas))) | |
5138 | return xa_err(mas->node); | |
5139 | ||
5140 | offset = mas->offset; | |
5141 | if (unlikely(offset == MAPLE_NODE_SLOTS)) | |
5142 | return -EBUSY; | |
5143 | ||
5144 | mt = mte_node_type(mas->node); | |
5145 | pivots = ma_pivots(mas_mn(mas), mt); | |
0257d990 PZ |
5146 | min = mas_safe_min(mas, pivots, offset); |
5147 | if (mas->index < min) | |
5148 | mas->index = min; | |
54a611b6 LH |
5149 | mas->last = mas->index + size - 1; |
5150 | return 0; | |
5151 | } | |
120b1162 | 5152 | EXPORT_SYMBOL_GPL(mas_empty_area); |
54a611b6 LH |
5153 | |
5154 | /* | |
5155 | * mas_empty_area_rev() - Get the highest address within the range that is | |
5156 | * sufficient for the size requested. | |
5157 | * @mas: The maple state | |
5158 | * @min: The lowest value of the range | |
5159 | * @max: The highest value of the range | |
5160 | * @size: The size needed | |
5161 | */ | |
5162 | int mas_empty_area_rev(struct ma_state *mas, unsigned long min, | |
5163 | unsigned long max, unsigned long size) | |
5164 | { | |
5165 | struct maple_enode *last = mas->node; | |
5166 | ||
ba997212 LH |
5167 | if (min > max) |
5168 | return -EINVAL; | |
5169 | ||
5170 | if (size == 0 || max - min < size - 1) | |
fad8e429 LH |
5171 | return -EINVAL; |
5172 | ||
54a611b6 LH |
5173 | if (mas_is_start(mas)) { |
5174 | mas_start(mas); | |
5175 | mas->offset = mas_data_end(mas); | |
5176 | } else if (mas->offset >= 2) { | |
5177 | mas->offset -= 2; | |
5178 | } else if (!mas_rewind_node(mas)) { | |
5179 | return -EBUSY; | |
5180 | } | |
5181 | ||
5182 | /* Empty set. */ | |
29ad6bb3 PZ |
5183 | if (mas_is_none(mas) || mas_is_ptr(mas)) |
5184 | return mas_sparse_area(mas, min, max, size, false); | |
54a611b6 LH |
5185 | |
5186 | /* The start of the window can only be within these values. */ | |
5187 | mas->index = min; | |
5188 | mas->last = max; | |
5189 | ||
fad8e429 | 5190 | while (!mas_rev_awalk(mas, size, &min, &max)) { |
54a611b6 LH |
5191 | if (last == mas->node) { |
5192 | if (!mas_rewind_node(mas)) | |
5193 | return -EBUSY; | |
5194 | } else { | |
5195 | last = mas->node; | |
5196 | } | |
5197 | } | |
5198 | ||
5199 | if (mas_is_err(mas)) | |
5200 | return xa_err(mas->node); | |
5201 | ||
5202 | if (unlikely(mas->offset == MAPLE_NODE_SLOTS)) | |
5203 | return -EBUSY; | |
5204 | ||
54a611b6 | 5205 | /* Trim the upper limit to the max. */ |
ba997212 | 5206 | if (max < mas->last) |
fad8e429 | 5207 | mas->last = max; |
54a611b6 LH |
5208 | |
5209 | mas->index = mas->last - size + 1; | |
5210 | return 0; | |
5211 | } | |
120b1162 | 5212 | EXPORT_SYMBOL_GPL(mas_empty_area_rev); |
54a611b6 | 5213 | |
54a611b6 | 5214 | /* |
790e1fa8 | 5215 | * mte_dead_leaves() - Mark all leaves of a node as dead. |
54a611b6 LH |
5216 | * @mas: The maple state |
5217 | * @slots: Pointer to the slot array | |
2e5b4921 | 5218 | * @type: The maple node type |
54a611b6 LH |
5219 | * |
5220 | * Must hold the write lock. | |
5221 | * | |
5222 | * Return: The number of leaves marked as dead. | |
5223 | */ | |
5224 | static inline | |
790e1fa8 LH |
5225 | unsigned char mte_dead_leaves(struct maple_enode *enode, struct maple_tree *mt, |
5226 | void __rcu **slots) | |
54a611b6 LH |
5227 | { |
5228 | struct maple_node *node; | |
5229 | enum maple_type type; | |
5230 | void *entry; | |
5231 | int offset; | |
5232 | ||
790e1fa8 LH |
5233 | for (offset = 0; offset < mt_slot_count(enode); offset++) { |
5234 | entry = mt_slot(mt, slots, offset); | |
54a611b6 LH |
5235 | type = mte_node_type(entry); |
5236 | node = mte_to_node(entry); | |
5237 | /* Use both node and type to catch LE & BE metadata */ | |
5238 | if (!node || !type) | |
5239 | break; | |
5240 | ||
5241 | mte_set_node_dead(entry); | |
54a611b6 LH |
5242 | node->type = type; |
5243 | rcu_assign_pointer(slots[offset], node); | |
5244 | } | |
5245 | ||
5246 | return offset; | |
5247 | } | |
5248 | ||
790e1fa8 LH |
5249 | /** |
5250 | * mte_dead_walk() - Walk down a dead tree to just before the leaves | |
5251 | * @enode: The maple encoded node | |
5252 | * @offset: The starting offset | |
5253 | * | |
5254 | * Note: This can only be used from the RCU callback context. | |
5255 | */ | |
5256 | static void __rcu **mte_dead_walk(struct maple_enode **enode, unsigned char offset) | |
54a611b6 LH |
5257 | { |
5258 | struct maple_node *node, *next; | |
5259 | void __rcu **slots = NULL; | |
5260 | ||
790e1fa8 | 5261 | next = mte_to_node(*enode); |
54a611b6 | 5262 | do { |
790e1fa8 LH |
5263 | *enode = ma_enode_ptr(next); |
5264 | node = mte_to_node(*enode); | |
54a611b6 | 5265 | slots = ma_slots(node, node->type); |
790e1fa8 LH |
5266 | next = rcu_dereference_protected(slots[offset], |
5267 | lock_is_held(&rcu_callback_map)); | |
54a611b6 LH |
5268 | offset = 0; |
5269 | } while (!ma_is_leaf(next->type)); | |
5270 | ||
5271 | return slots; | |
5272 | } | |
5273 | ||
790e1fa8 LH |
5274 | /** |
5275 | * mt_free_walk() - Walk & free a tree in the RCU callback context | |
5276 | * @head: The RCU head that's within the node. | |
5277 | * | |
5278 | * Note: This can only be used from the RCU callback context. | |
5279 | */ | |
54a611b6 LH |
5280 | static void mt_free_walk(struct rcu_head *head) |
5281 | { | |
5282 | void __rcu **slots; | |
5283 | struct maple_node *node, *start; | |
790e1fa8 | 5284 | struct maple_enode *enode; |
54a611b6 LH |
5285 | unsigned char offset; |
5286 | enum maple_type type; | |
54a611b6 LH |
5287 | |
5288 | node = container_of(head, struct maple_node, rcu); | |
5289 | ||
5290 | if (ma_is_leaf(node->type)) | |
5291 | goto free_leaf; | |
5292 | ||
54a611b6 | 5293 | start = node; |
790e1fa8 LH |
5294 | enode = mt_mk_node(node, node->type); |
5295 | slots = mte_dead_walk(&enode, 0); | |
5296 | node = mte_to_node(enode); | |
54a611b6 LH |
5297 | do { |
5298 | mt_free_bulk(node->slot_len, slots); | |
5299 | offset = node->parent_slot + 1; | |
790e1fa8 LH |
5300 | enode = node->piv_parent; |
5301 | if (mte_to_node(enode) == node) | |
5302 | goto free_leaf; | |
5303 | ||
5304 | type = mte_node_type(enode); | |
5305 | slots = ma_slots(mte_to_node(enode), type); | |
5306 | if ((offset < mt_slots[type]) && | |
5307 | rcu_dereference_protected(slots[offset], | |
5308 | lock_is_held(&rcu_callback_map))) | |
5309 | slots = mte_dead_walk(&enode, offset); | |
5310 | node = mte_to_node(enode); | |
54a611b6 LH |
5311 | } while ((node != start) || (node->slot_len < offset)); |
5312 | ||
5313 | slots = ma_slots(node, node->type); | |
5314 | mt_free_bulk(node->slot_len, slots); | |
5315 | ||
54a611b6 LH |
5316 | free_leaf: |
5317 | mt_free_rcu(&node->rcu); | |
5318 | } | |
5319 | ||
790e1fa8 LH |
5320 | static inline void __rcu **mte_destroy_descend(struct maple_enode **enode, |
5321 | struct maple_tree *mt, struct maple_enode *prev, unsigned char offset) | |
54a611b6 LH |
5322 | { |
5323 | struct maple_node *node; | |
790e1fa8 | 5324 | struct maple_enode *next = *enode; |
54a611b6 | 5325 | void __rcu **slots = NULL; |
790e1fa8 LH |
5326 | enum maple_type type; |
5327 | unsigned char next_offset = 0; | |
54a611b6 LH |
5328 | |
5329 | do { | |
790e1fa8 LH |
5330 | *enode = next; |
5331 | node = mte_to_node(*enode); | |
5332 | type = mte_node_type(*enode); | |
5333 | slots = ma_slots(node, type); | |
5334 | next = mt_slot_locked(mt, slots, next_offset); | |
54a611b6 | 5335 | if ((mte_dead_node(next))) |
790e1fa8 | 5336 | next = mt_slot_locked(mt, slots, ++next_offset); |
54a611b6 | 5337 | |
790e1fa8 LH |
5338 | mte_set_node_dead(*enode); |
5339 | node->type = type; | |
54a611b6 LH |
5340 | node->piv_parent = prev; |
5341 | node->parent_slot = offset; | |
790e1fa8 LH |
5342 | offset = next_offset; |
5343 | next_offset = 0; | |
5344 | prev = *enode; | |
54a611b6 LH |
5345 | } while (!mte_is_leaf(next)); |
5346 | ||
5347 | return slots; | |
5348 | } | |
5349 | ||
790e1fa8 | 5350 | static void mt_destroy_walk(struct maple_enode *enode, struct maple_tree *mt, |
54a611b6 LH |
5351 | bool free) |
5352 | { | |
5353 | void __rcu **slots; | |
5354 | struct maple_node *node = mte_to_node(enode); | |
5355 | struct maple_enode *start; | |
54a611b6 | 5356 | |
2e5b4921 LH |
5357 | if (mte_is_leaf(enode)) { |
5358 | node->type = mte_node_type(enode); | |
54a611b6 | 5359 | goto free_leaf; |
2e5b4921 | 5360 | } |
54a611b6 | 5361 | |
2e5b4921 | 5362 | start = enode; |
790e1fa8 LH |
5363 | slots = mte_destroy_descend(&enode, mt, start, 0); |
5364 | node = mte_to_node(enode); // Updated in the above call. | |
54a611b6 LH |
5365 | do { |
5366 | enum maple_type type; | |
5367 | unsigned char offset; | |
5368 | struct maple_enode *parent, *tmp; | |
5369 | ||
790e1fa8 | 5370 | node->slot_len = mte_dead_leaves(enode, mt, slots); |
54a611b6 LH |
5371 | if (free) |
5372 | mt_free_bulk(node->slot_len, slots); | |
5373 | offset = node->parent_slot + 1; | |
790e1fa8 LH |
5374 | enode = node->piv_parent; |
5375 | if (mte_to_node(enode) == node) | |
5376 | goto free_leaf; | |
54a611b6 | 5377 | |
790e1fa8 LH |
5378 | type = mte_node_type(enode); |
5379 | slots = ma_slots(mte_to_node(enode), type); | |
54a611b6 LH |
5380 | if (offset >= mt_slots[type]) |
5381 | goto next; | |
5382 | ||
790e1fa8 | 5383 | tmp = mt_slot_locked(mt, slots, offset); |
54a611b6 | 5384 | if (mte_node_type(tmp) && mte_to_node(tmp)) { |
790e1fa8 LH |
5385 | parent = enode; |
5386 | enode = tmp; | |
5387 | slots = mte_destroy_descend(&enode, mt, parent, offset); | |
54a611b6 LH |
5388 | } |
5389 | next: | |
790e1fa8 LH |
5390 | node = mte_to_node(enode); |
5391 | } while (start != enode); | |
54a611b6 | 5392 | |
790e1fa8 LH |
5393 | node = mte_to_node(enode); |
5394 | node->slot_len = mte_dead_leaves(enode, mt, slots); | |
54a611b6 LH |
5395 | if (free) |
5396 | mt_free_bulk(node->slot_len, slots); | |
5397 | ||
54a611b6 LH |
5398 | free_leaf: |
5399 | if (free) | |
5400 | mt_free_rcu(&node->rcu); | |
2e5b4921 | 5401 | else |
790e1fa8 | 5402 | mt_clear_meta(mt, node, node->type); |
54a611b6 LH |
5403 | } |
5404 | ||
5405 | /* | |
5406 | * mte_destroy_walk() - Free a tree or sub-tree. | |
f942b0f0 VY |
5407 | * @enode: the encoded maple node (maple_enode) to start |
5408 | * @mt: the tree to free - needed for node types. | |
54a611b6 LH |
5409 | * |
5410 | * Must hold the write lock. | |
5411 | */ | |
5412 | static inline void mte_destroy_walk(struct maple_enode *enode, | |
5413 | struct maple_tree *mt) | |
5414 | { | |
5415 | struct maple_node *node = mte_to_node(enode); | |
5416 | ||
5417 | if (mt_in_rcu(mt)) { | |
790e1fa8 | 5418 | mt_destroy_walk(enode, mt, false); |
54a611b6 LH |
5419 | call_rcu(&node->rcu, mt_free_walk); |
5420 | } else { | |
790e1fa8 | 5421 | mt_destroy_walk(enode, mt, true); |
54a611b6 LH |
5422 | } |
5423 | } | |
5424 | ||
5425 | static void mas_wr_store_setup(struct ma_wr_state *wr_mas) | |
5426 | { | |
fec29364 LH |
5427 | if (mas_is_start(wr_mas->mas)) |
5428 | return; | |
5429 | ||
1202700c | 5430 | if (unlikely(mas_is_paused(wr_mas->mas))) |
fec29364 | 5431 | goto reset; |
1202700c | 5432 | |
fec29364 LH |
5433 | if (unlikely(mas_is_none(wr_mas->mas))) |
5434 | goto reset; | |
5435 | ||
5436 | /* | |
5437 | * A less strict version of mas_is_span_wr() where we allow spanning | |
5438 | * writes within this node. This is to stop partial walks in | |
5439 | * mas_prealloc() from being reset. | |
5440 | */ | |
5441 | if (wr_mas->mas->last > wr_mas->mas->max) | |
5442 | goto reset; | |
5443 | ||
5444 | if (wr_mas->entry) | |
5445 | return; | |
5446 | ||
5447 | if (mte_is_leaf(wr_mas->mas->node) && | |
5448 | wr_mas->mas->last == wr_mas->mas->max) | |
5449 | goto reset; | |
5450 | ||
5451 | return; | |
5452 | ||
5453 | reset: | |
5454 | mas_reset(wr_mas->mas); | |
54a611b6 LH |
5455 | } |
5456 | ||
5457 | /* Interface */ | |
5458 | ||
5459 | /** | |
5460 | * mas_store() - Store an @entry. | |
5461 | * @mas: The maple state. | |
5462 | * @entry: The entry to store. | |
5463 | * | |
5464 | * The @mas->index and @mas->last is used to set the range for the @entry. | |
5465 | * Note: The @mas should have pre-allocated entries to ensure there is memory to | |
5466 | * store the entry. Please see mas_expected_entries()/mas_destroy() for more details. | |
5467 | * | |
5468 | * Return: the first entry between mas->index and mas->last or %NULL. | |
5469 | */ | |
5470 | void *mas_store(struct ma_state *mas, void *entry) | |
5471 | { | |
5472 | MA_WR_STATE(wr_mas, mas, entry); | |
5473 | ||
5474 | trace_ma_write(__func__, mas, 0, entry); | |
5475 | #ifdef CONFIG_DEBUG_MAPLE_TREE | |
e6d6792a | 5476 | if (MAS_WARN_ON(mas, mas->index > mas->last)) |
89f499f3 | 5477 | pr_err("Error %lX > %lX %p\n", mas->index, mas->last, entry); |
e6d6792a | 5478 | |
54a611b6 LH |
5479 | if (mas->index > mas->last) { |
5480 | mas_set_err(mas, -EINVAL); | |
5481 | return NULL; | |
5482 | } | |
5483 | ||
5484 | #endif | |
5485 | ||
5486 | /* | |
5487 | * Storing is the same operation as insert with the added caveat that it | |
5488 | * can overwrite entries. Although this seems simple enough, one may | |
5489 | * want to examine what happens if a single store operation was to | |
5490 | * overwrite multiple entries within a self-balancing B-Tree. | |
5491 | */ | |
5492 | mas_wr_store_setup(&wr_mas); | |
5493 | mas_wr_store_entry(&wr_mas); | |
5494 | return wr_mas.content; | |
5495 | } | |
120b1162 | 5496 | EXPORT_SYMBOL_GPL(mas_store); |
54a611b6 LH |
5497 | |
5498 | /** | |
5499 | * mas_store_gfp() - Store a value into the tree. | |
5500 | * @mas: The maple state | |
5501 | * @entry: The entry to store | |
5502 | * @gfp: The GFP_FLAGS to use for allocations if necessary. | |
5503 | * | |
5504 | * Return: 0 on success, -EINVAL on invalid request, -ENOMEM if memory could not | |
5505 | * be allocated. | |
5506 | */ | |
5507 | int mas_store_gfp(struct ma_state *mas, void *entry, gfp_t gfp) | |
5508 | { | |
5509 | MA_WR_STATE(wr_mas, mas, entry); | |
5510 | ||
5511 | mas_wr_store_setup(&wr_mas); | |
5512 | trace_ma_write(__func__, mas, 0, entry); | |
5513 | retry: | |
5514 | mas_wr_store_entry(&wr_mas); | |
5515 | if (unlikely(mas_nomem(mas, gfp))) | |
5516 | goto retry; | |
5517 | ||
5518 | if (unlikely(mas_is_err(mas))) | |
5519 | return xa_err(mas->node); | |
5520 | ||
5521 | return 0; | |
5522 | } | |
120b1162 | 5523 | EXPORT_SYMBOL_GPL(mas_store_gfp); |
54a611b6 LH |
5524 | |
5525 | /** | |
5526 | * mas_store_prealloc() - Store a value into the tree using memory | |
5527 | * preallocated in the maple state. | |
5528 | * @mas: The maple state | |
5529 | * @entry: The entry to store. | |
5530 | */ | |
5531 | void mas_store_prealloc(struct ma_state *mas, void *entry) | |
5532 | { | |
5533 | MA_WR_STATE(wr_mas, mas, entry); | |
5534 | ||
5535 | mas_wr_store_setup(&wr_mas); | |
5536 | trace_ma_write(__func__, mas, 0, entry); | |
5537 | mas_wr_store_entry(&wr_mas); | |
1c414c6a | 5538 | MAS_WR_BUG_ON(&wr_mas, mas_is_err(mas)); |
54a611b6 LH |
5539 | mas_destroy(mas); |
5540 | } | |
120b1162 | 5541 | EXPORT_SYMBOL_GPL(mas_store_prealloc); |
54a611b6 LH |
5542 | |
5543 | /** | |
5544 | * mas_preallocate() - Preallocate enough nodes for a store operation | |
5545 | * @mas: The maple state | |
da089254 | 5546 | * @entry: The entry that will be stored |
54a611b6 LH |
5547 | * @gfp: The GFP_FLAGS to use for allocations. |
5548 | * | |
5549 | * Return: 0 on success, -ENOMEM if memory could not be allocated. | |
5550 | */ | |
da089254 | 5551 | int mas_preallocate(struct ma_state *mas, void *entry, gfp_t gfp) |
54a611b6 | 5552 | { |
17983dc6 LH |
5553 | MA_WR_STATE(wr_mas, mas, entry); |
5554 | unsigned char node_size; | |
5555 | int request = 1; | |
54a611b6 LH |
5556 | int ret; |
5557 | ||
17983dc6 LH |
5558 | |
5559 | if (unlikely(!mas->index && mas->last == ULONG_MAX)) | |
5560 | goto ask_now; | |
5561 | ||
5562 | mas_wr_store_setup(&wr_mas); | |
5563 | wr_mas.content = mas_start(mas); | |
5564 | /* Root expand */ | |
5565 | if (unlikely(mas_is_none(mas) || mas_is_ptr(mas))) | |
5566 | goto ask_now; | |
5567 | ||
5568 | if (unlikely(!mas_wr_walk(&wr_mas))) { | |
5569 | /* Spanning store, use worst case for now */ | |
5570 | request = 1 + mas_mt_height(mas) * 3; | |
5571 | goto ask_now; | |
5572 | } | |
5573 | ||
5574 | /* At this point, we are at the leaf node that needs to be altered. */ | |
5575 | /* Exact fit, no nodes needed. */ | |
5576 | if (wr_mas.r_min == mas->index && wr_mas.r_max == mas->last) | |
5577 | return 0; | |
5578 | ||
5579 | mas_wr_end_piv(&wr_mas); | |
5580 | node_size = mas_wr_new_end(&wr_mas); | |
5581 | if (node_size >= mt_slots[wr_mas.type]) { | |
5582 | /* Split, worst case for now. */ | |
5583 | request = 1 + mas_mt_height(mas) * 2; | |
5584 | goto ask_now; | |
5585 | } | |
5586 | ||
5587 | /* New root needs a singe node */ | |
5588 | if (unlikely(mte_is_root(mas->node))) | |
5589 | goto ask_now; | |
5590 | ||
5591 | /* Potential spanning rebalance collapsing a node, use worst-case */ | |
5592 | if (node_size - 1 <= mt_min_slots[wr_mas.type]) | |
5593 | request = mas_mt_height(mas) * 2 - 1; | |
5594 | ||
5595 | /* node store, slot store needs one node */ | |
5596 | ask_now: | |
5597 | mas_node_count_gfp(mas, request, gfp); | |
54a611b6 LH |
5598 | mas->mas_flags |= MA_STATE_PREALLOC; |
5599 | if (likely(!mas_is_err(mas))) | |
5600 | return 0; | |
5601 | ||
5602 | mas_set_alloc_req(mas, 0); | |
5603 | ret = xa_err(mas->node); | |
5604 | mas_reset(mas); | |
5605 | mas_destroy(mas); | |
5606 | mas_reset(mas); | |
5607 | return ret; | |
5608 | } | |
5c63a7c3 | 5609 | EXPORT_SYMBOL_GPL(mas_preallocate); |
54a611b6 LH |
5610 | |
5611 | /* | |
5612 | * mas_destroy() - destroy a maple state. | |
5613 | * @mas: The maple state | |
5614 | * | |
5615 | * Upon completion, check the left-most node and rebalance against the node to | |
5616 | * the right if necessary. Frees any allocated nodes associated with this maple | |
5617 | * state. | |
5618 | */ | |
5619 | void mas_destroy(struct ma_state *mas) | |
5620 | { | |
5621 | struct maple_alloc *node; | |
541e06b7 | 5622 | unsigned long total; |
54a611b6 LH |
5623 | |
5624 | /* | |
5625 | * When using mas_for_each() to insert an expected number of elements, | |
5626 | * it is possible that the number inserted is less than the expected | |
5627 | * number. To fix an invalid final node, a check is performed here to | |
5628 | * rebalance the previous node with the final node. | |
5629 | */ | |
5630 | if (mas->mas_flags & MA_STATE_REBALANCE) { | |
5631 | unsigned char end; | |
5632 | ||
23e734ec | 5633 | mas_start(mas); |
54a611b6 LH |
5634 | mtree_range_walk(mas); |
5635 | end = mas_data_end(mas) + 1; | |
5636 | if (end < mt_min_slot_count(mas->node) - 1) | |
5637 | mas_destroy_rebalance(mas, end); | |
5638 | ||
5639 | mas->mas_flags &= ~MA_STATE_REBALANCE; | |
5640 | } | |
5641 | mas->mas_flags &= ~(MA_STATE_BULK|MA_STATE_PREALLOC); | |
5642 | ||
541e06b7 LH |
5643 | total = mas_allocated(mas); |
5644 | while (total) { | |
54a611b6 LH |
5645 | node = mas->alloc; |
5646 | mas->alloc = node->slot[0]; | |
541e06b7 LH |
5647 | if (node->node_count > 1) { |
5648 | size_t count = node->node_count - 1; | |
5649 | ||
5650 | mt_free_bulk(count, (void __rcu **)&node->slot[1]); | |
5651 | total -= count; | |
5652 | } | |
54a611b6 | 5653 | kmem_cache_free(maple_node_cache, node); |
541e06b7 | 5654 | total--; |
54a611b6 | 5655 | } |
541e06b7 | 5656 | |
54a611b6 LH |
5657 | mas->alloc = NULL; |
5658 | } | |
120b1162 | 5659 | EXPORT_SYMBOL_GPL(mas_destroy); |
54a611b6 LH |
5660 | |
5661 | /* | |
5662 | * mas_expected_entries() - Set the expected number of entries that will be inserted. | |
5663 | * @mas: The maple state | |
5664 | * @nr_entries: The number of expected entries. | |
5665 | * | |
5666 | * This will attempt to pre-allocate enough nodes to store the expected number | |
5667 | * of entries. The allocations will occur using the bulk allocator interface | |
5668 | * for speed. Please call mas_destroy() on the @mas after inserting the entries | |
5669 | * to ensure any unused nodes are freed. | |
5670 | * | |
5671 | * Return: 0 on success, -ENOMEM if memory could not be allocated. | |
5672 | */ | |
5673 | int mas_expected_entries(struct ma_state *mas, unsigned long nr_entries) | |
5674 | { | |
5675 | int nonleaf_cap = MAPLE_ARANGE64_SLOTS - 2; | |
5676 | struct maple_enode *enode = mas->node; | |
5677 | int nr_nodes; | |
5678 | int ret; | |
5679 | ||
5680 | /* | |
5681 | * Sometimes it is necessary to duplicate a tree to a new tree, such as | |
5682 | * forking a process and duplicating the VMAs from one tree to a new | |
5683 | * tree. When such a situation arises, it is known that the new tree is | |
5684 | * not going to be used until the entire tree is populated. For | |
5685 | * performance reasons, it is best to use a bulk load with RCU disabled. | |
5686 | * This allows for optimistic splitting that favours the left and reuse | |
5687 | * of nodes during the operation. | |
5688 | */ | |
5689 | ||
5690 | /* Optimize splitting for bulk insert in-order */ | |
5691 | mas->mas_flags |= MA_STATE_BULK; | |
5692 | ||
5693 | /* | |
5694 | * Avoid overflow, assume a gap between each entry and a trailing null. | |
5695 | * If this is wrong, it just means allocation can happen during | |
5696 | * insertion of entries. | |
5697 | */ | |
5698 | nr_nodes = max(nr_entries, nr_entries * 2 + 1); | |
5699 | if (!mt_is_alloc(mas->tree)) | |
5700 | nonleaf_cap = MAPLE_RANGE64_SLOTS - 2; | |
5701 | ||
5702 | /* Leaves; reduce slots to keep space for expansion */ | |
5703 | nr_nodes = DIV_ROUND_UP(nr_nodes, MAPLE_RANGE64_SLOTS - 2); | |
5704 | /* Internal nodes */ | |
5705 | nr_nodes += DIV_ROUND_UP(nr_nodes, nonleaf_cap); | |
5706 | /* Add working room for split (2 nodes) + new parents */ | |
5707 | mas_node_count(mas, nr_nodes + 3); | |
5708 | ||
5709 | /* Detect if allocations run out */ | |
5710 | mas->mas_flags |= MA_STATE_PREALLOC; | |
5711 | ||
5712 | if (!mas_is_err(mas)) | |
5713 | return 0; | |
5714 | ||
5715 | ret = xa_err(mas->node); | |
5716 | mas->node = enode; | |
5717 | mas_destroy(mas); | |
5718 | return ret; | |
5719 | ||
5720 | } | |
120b1162 | 5721 | EXPORT_SYMBOL_GPL(mas_expected_entries); |
54a611b6 | 5722 | |
6169b553 LH |
5723 | static inline bool mas_next_setup(struct ma_state *mas, unsigned long max, |
5724 | void **entry) | |
54a611b6 | 5725 | { |
ca80f610 LH |
5726 | bool was_none = mas_is_none(mas); |
5727 | ||
54a611b6 LH |
5728 | if (mas_is_none(mas) || mas_is_paused(mas)) |
5729 | mas->node = MAS_START; | |
5730 | ||
5731 | if (mas_is_start(mas)) | |
6169b553 | 5732 | *entry = mas_walk(mas); /* Retries on dead nodes handled by mas_walk */ |
54a611b6 LH |
5733 | |
5734 | if (mas_is_ptr(mas)) { | |
6169b553 | 5735 | *entry = NULL; |
ca80f610 LH |
5736 | if (was_none && mas->index == 0) { |
5737 | mas->index = mas->last = 0; | |
6169b553 | 5738 | return true; |
54a611b6 | 5739 | } |
ca80f610 LH |
5740 | mas->index = 1; |
5741 | mas->last = ULONG_MAX; | |
5742 | mas->node = MAS_NONE; | |
6169b553 | 5743 | return true; |
54a611b6 LH |
5744 | } |
5745 | ||
6169b553 LH |
5746 | if (mas_is_none(mas)) |
5747 | return true; | |
5748 | return false; | |
5749 | } | |
5750 | ||
5751 | /** | |
5752 | * mas_next() - Get the next entry. | |
5753 | * @mas: The maple state | |
5754 | * @max: The maximum index to check. | |
5755 | * | |
5756 | * Returns the next entry after @mas->index. | |
5757 | * Must hold rcu_read_lock or the write lock. | |
5758 | * Can return the zero entry. | |
5759 | * | |
5760 | * Return: The next entry or %NULL | |
5761 | */ | |
5762 | void *mas_next(struct ma_state *mas, unsigned long max) | |
5763 | { | |
5764 | void *entry = NULL; | |
5765 | ||
5766 | if (mas_next_setup(mas, max, &entry)) | |
5767 | return entry; | |
5768 | ||
5769 | /* Retries on dead nodes handled by mas_next_slot */ | |
5770 | return mas_next_slot(mas, max, false); | |
54a611b6 LH |
5771 | } |
5772 | EXPORT_SYMBOL_GPL(mas_next); | |
5773 | ||
6169b553 LH |
5774 | /** |
5775 | * mas_next_range() - Advance the maple state to the next range | |
5776 | * @mas: The maple state | |
5777 | * @max: The maximum index to check. | |
5778 | * | |
5779 | * Sets @mas->index and @mas->last to the range. | |
5780 | * Must hold rcu_read_lock or the write lock. | |
5781 | * Can return the zero entry. | |
5782 | * | |
5783 | * Return: The next entry or %NULL | |
5784 | */ | |
5785 | void *mas_next_range(struct ma_state *mas, unsigned long max) | |
5786 | { | |
5787 | void *entry = NULL; | |
5788 | ||
5789 | if (mas_next_setup(mas, max, &entry)) | |
5790 | return entry; | |
5791 | ||
5792 | /* Retries on dead nodes handled by mas_next_slot */ | |
5793 | return mas_next_slot(mas, max, true); | |
5794 | } | |
5795 | EXPORT_SYMBOL_GPL(mas_next_range); | |
5796 | ||
54a611b6 LH |
5797 | /** |
5798 | * mt_next() - get the next value in the maple tree | |
5799 | * @mt: The maple tree | |
5800 | * @index: The start index | |
5801 | * @max: The maximum index to check | |
5802 | * | |
fad9c80e TG |
5803 | * Takes RCU read lock internally to protect the search, which does not |
5804 | * protect the returned pointer after dropping RCU read lock. | |
5805 | * See also: Documentation/core-api/maple_tree.rst | |
5806 | * | |
5807 | * Return: The entry higher than @index or %NULL if nothing is found. | |
54a611b6 LH |
5808 | */ |
5809 | void *mt_next(struct maple_tree *mt, unsigned long index, unsigned long max) | |
5810 | { | |
5811 | void *entry = NULL; | |
5812 | MA_STATE(mas, mt, index, index); | |
5813 | ||
5814 | rcu_read_lock(); | |
5815 | entry = mas_next(&mas, max); | |
5816 | rcu_read_unlock(); | |
5817 | return entry; | |
5818 | } | |
5819 | EXPORT_SYMBOL_GPL(mt_next); | |
5820 | ||
6b9e93e0 LH |
5821 | static inline bool mas_prev_setup(struct ma_state *mas, unsigned long min, |
5822 | void **entry) | |
54a611b6 | 5823 | { |
39193685 LH |
5824 | if (mas->index <= min) |
5825 | goto none; | |
54a611b6 LH |
5826 | |
5827 | if (mas_is_none(mas) || mas_is_paused(mas)) | |
5828 | mas->node = MAS_START; | |
5829 | ||
5830 | if (mas_is_start(mas)) { | |
5831 | mas_walk(mas); | |
5832 | if (!mas->index) | |
39193685 | 5833 | goto none; |
54a611b6 LH |
5834 | } |
5835 | ||
39193685 LH |
5836 | if (unlikely(mas_is_ptr(mas))) { |
5837 | if (!mas->index) | |
5838 | goto none; | |
54a611b6 | 5839 | mas->index = mas->last = 0; |
6b9e93e0 LH |
5840 | *entry = mas_root(mas); |
5841 | return true; | |
39193685 LH |
5842 | } |
5843 | ||
5844 | if (mas_is_none(mas)) { | |
5845 | if (mas->index) { | |
5846 | /* Walked to out-of-range pointer? */ | |
5847 | mas->index = mas->last = 0; | |
5848 | mas->node = MAS_ROOT; | |
6b9e93e0 LH |
5849 | *entry = mas_root(mas); |
5850 | return true; | |
39193685 | 5851 | } |
6b9e93e0 | 5852 | return true; |
54a611b6 | 5853 | } |
6b9e93e0 LH |
5854 | |
5855 | return false; | |
39193685 LH |
5856 | |
5857 | none: | |
5858 | mas->node = MAS_NONE; | |
6b9e93e0 LH |
5859 | return true; |
5860 | } | |
5861 | ||
5862 | /** | |
5863 | * mas_prev() - Get the previous entry | |
5864 | * @mas: The maple state | |
5865 | * @min: The minimum value to check. | |
5866 | * | |
5867 | * Must hold rcu_read_lock or the write lock. | |
5868 | * Will reset mas to MAS_START if the node is MAS_NONE. Will stop on not | |
5869 | * searchable nodes. | |
5870 | * | |
5871 | * Return: the previous value or %NULL. | |
5872 | */ | |
5873 | void *mas_prev(struct ma_state *mas, unsigned long min) | |
5874 | { | |
5875 | void *entry = NULL; | |
5876 | ||
5877 | if (mas_prev_setup(mas, min, &entry)) | |
5878 | return entry; | |
5879 | ||
5880 | return mas_prev_slot(mas, min, false); | |
54a611b6 LH |
5881 | } |
5882 | EXPORT_SYMBOL_GPL(mas_prev); | |
5883 | ||
6b9e93e0 LH |
5884 | /** |
5885 | * mas_prev_range() - Advance to the previous range | |
5886 | * @mas: The maple state | |
5887 | * @min: The minimum value to check. | |
5888 | * | |
5889 | * Sets @mas->index and @mas->last to the range. | |
5890 | * Must hold rcu_read_lock or the write lock. | |
5891 | * Will reset mas to MAS_START if the node is MAS_NONE. Will stop on not | |
5892 | * searchable nodes. | |
5893 | * | |
5894 | * Return: the previous value or %NULL. | |
5895 | */ | |
5896 | void *mas_prev_range(struct ma_state *mas, unsigned long min) | |
5897 | { | |
5898 | void *entry = NULL; | |
5899 | ||
5900 | if (mas_prev_setup(mas, min, &entry)) | |
5901 | return entry; | |
5902 | ||
5903 | return mas_prev_slot(mas, min, true); | |
5904 | } | |
5905 | EXPORT_SYMBOL_GPL(mas_prev_range); | |
5906 | ||
54a611b6 LH |
5907 | /** |
5908 | * mt_prev() - get the previous value in the maple tree | |
5909 | * @mt: The maple tree | |
5910 | * @index: The start index | |
5911 | * @min: The minimum index to check | |
5912 | * | |
fad9c80e TG |
5913 | * Takes RCU read lock internally to protect the search, which does not |
5914 | * protect the returned pointer after dropping RCU read lock. | |
5915 | * See also: Documentation/core-api/maple_tree.rst | |
5916 | * | |
5917 | * Return: The entry before @index or %NULL if nothing is found. | |
54a611b6 LH |
5918 | */ |
5919 | void *mt_prev(struct maple_tree *mt, unsigned long index, unsigned long min) | |
5920 | { | |
5921 | void *entry = NULL; | |
5922 | MA_STATE(mas, mt, index, index); | |
5923 | ||
5924 | rcu_read_lock(); | |
5925 | entry = mas_prev(&mas, min); | |
5926 | rcu_read_unlock(); | |
5927 | return entry; | |
5928 | } | |
5929 | EXPORT_SYMBOL_GPL(mt_prev); | |
5930 | ||
5931 | /** | |
5932 | * mas_pause() - Pause a mas_find/mas_for_each to drop the lock. | |
5933 | * @mas: The maple state to pause | |
5934 | * | |
5935 | * Some users need to pause a walk and drop the lock they're holding in | |
5936 | * order to yield to a higher priority thread or carry out an operation | |
5937 | * on an entry. Those users should call this function before they drop | |
5938 | * the lock. It resets the @mas to be suitable for the next iteration | |
5939 | * of the loop after the user has reacquired the lock. If most entries | |
5940 | * found during a walk require you to call mas_pause(), the mt_for_each() | |
5941 | * iterator may be more appropriate. | |
5942 | * | |
5943 | */ | |
5944 | void mas_pause(struct ma_state *mas) | |
5945 | { | |
5946 | mas->node = MAS_PAUSE; | |
5947 | } | |
5948 | EXPORT_SYMBOL_GPL(mas_pause); | |
5949 | ||
5950 | /** | |
6169b553 | 5951 | * mas_find_setup() - Internal function to set up mas_find*(). |
54a611b6 | 5952 | * @mas: The maple state |
6169b553 LH |
5953 | * @max: The maximum index |
5954 | * @entry: Pointer to the entry | |
54a611b6 | 5955 | * |
6169b553 | 5956 | * Returns: True if entry is the answer, false otherwise. |
54a611b6 | 5957 | */ |
6169b553 LH |
5958 | static inline bool mas_find_setup(struct ma_state *mas, unsigned long max, |
5959 | void **entry) | |
54a611b6 | 5960 | { |
6169b553 LH |
5961 | *entry = NULL; |
5962 | ||
ca80f610 LH |
5963 | if (unlikely(mas_is_none(mas))) { |
5964 | if (unlikely(mas->last >= max)) | |
6169b553 | 5965 | return true; |
ca80f610 LH |
5966 | |
5967 | mas->index = mas->last; | |
5968 | mas->node = MAS_START; | |
6169b553 | 5969 | } else if (unlikely(mas_is_paused(mas))) { |
ca80f610 | 5970 | if (unlikely(mas->last >= max)) |
6169b553 | 5971 | return true; |
ca80f610 | 5972 | |
54a611b6 LH |
5973 | mas->node = MAS_START; |
5974 | mas->index = ++mas->last; | |
6169b553 | 5975 | } else if (unlikely(mas_is_ptr(mas))) |
ca80f610 | 5976 | goto ptr_out_of_range; |
17dc622c | 5977 | |
54a611b6 LH |
5978 | if (unlikely(mas_is_start(mas))) { |
5979 | /* First run or continue */ | |
54a611b6 | 5980 | if (mas->index > max) |
6169b553 | 5981 | return true; |
54a611b6 | 5982 | |
6169b553 LH |
5983 | *entry = mas_walk(mas); |
5984 | if (*entry) | |
5985 | return true; | |
ca80f610 | 5986 | |
54a611b6 LH |
5987 | } |
5988 | ||
ca80f610 LH |
5989 | if (unlikely(!mas_searchable(mas))) { |
5990 | if (unlikely(mas_is_ptr(mas))) | |
5991 | goto ptr_out_of_range; | |
5992 | ||
6169b553 | 5993 | return true; |
ca80f610 LH |
5994 | } |
5995 | ||
5996 | if (mas->index == max) | |
6169b553 | 5997 | return true; |
54a611b6 | 5998 | |
6169b553 | 5999 | return false; |
ca80f610 LH |
6000 | |
6001 | ptr_out_of_range: | |
6002 | mas->node = MAS_NONE; | |
6003 | mas->index = 1; | |
6004 | mas->last = ULONG_MAX; | |
6169b553 LH |
6005 | return true; |
6006 | } | |
6007 | ||
6008 | /** | |
6009 | * mas_find() - On the first call, find the entry at or after mas->index up to | |
6010 | * %max. Otherwise, find the entry after mas->index. | |
6011 | * @mas: The maple state | |
6012 | * @max: The maximum value to check. | |
6013 | * | |
6014 | * Must hold rcu_read_lock or the write lock. | |
6015 | * If an entry exists, last and index are updated accordingly. | |
6016 | * May set @mas->node to MAS_NONE. | |
6017 | * | |
6018 | * Return: The entry or %NULL. | |
6019 | */ | |
6020 | void *mas_find(struct ma_state *mas, unsigned long max) | |
6021 | { | |
6022 | void *entry = NULL; | |
6023 | ||
6024 | if (mas_find_setup(mas, max, &entry)) | |
6025 | return entry; | |
6026 | ||
6027 | /* Retries on dead nodes handled by mas_next_slot */ | |
6028 | return mas_next_slot(mas, max, false); | |
54a611b6 | 6029 | } |
120b1162 | 6030 | EXPORT_SYMBOL_GPL(mas_find); |
54a611b6 | 6031 | |
6169b553 LH |
6032 | /** |
6033 | * mas_find_range() - On the first call, find the entry at or after | |
6034 | * mas->index up to %max. Otherwise, advance to the next slot mas->index. | |
6035 | * @mas: The maple state | |
6036 | * @max: The maximum value to check. | |
6037 | * | |
6038 | * Must hold rcu_read_lock or the write lock. | |
6039 | * If an entry exists, last and index are updated accordingly. | |
6040 | * May set @mas->node to MAS_NONE. | |
6041 | * | |
6042 | * Return: The entry or %NULL. | |
6043 | */ | |
6044 | void *mas_find_range(struct ma_state *mas, unsigned long max) | |
6045 | { | |
6046 | void *entry; | |
6047 | ||
6048 | if (mas_find_setup(mas, max, &entry)) | |
6049 | return entry; | |
6050 | ||
6051 | /* Retries on dead nodes handled by mas_next_slot */ | |
6052 | return mas_next_slot(mas, max, true); | |
6053 | } | |
6054 | EXPORT_SYMBOL_GPL(mas_find_range); | |
6055 | ||
54a611b6 | 6056 | /** |
6b9e93e0 | 6057 | * mas_find_rev_setup() - Internal function to set up mas_find_*_rev() |
54a611b6 | 6058 | * @mas: The maple state |
6b9e93e0 LH |
6059 | * @min: The minimum index |
6060 | * @entry: Pointer to the entry | |
54a611b6 | 6061 | * |
6b9e93e0 | 6062 | * Returns: True if entry is the answer, false otherwise. |
54a611b6 | 6063 | */ |
6b9e93e0 LH |
6064 | static inline bool mas_find_rev_setup(struct ma_state *mas, unsigned long min, |
6065 | void **entry) | |
54a611b6 | 6066 | { |
6b9e93e0 LH |
6067 | *entry = NULL; |
6068 | ||
39193685 LH |
6069 | if (unlikely(mas_is_none(mas))) { |
6070 | if (mas->index <= min) | |
6071 | goto none; | |
6072 | ||
6073 | mas->last = mas->index; | |
6074 | mas->node = MAS_START; | |
6075 | } | |
6076 | ||
54a611b6 | 6077 | if (unlikely(mas_is_paused(mas))) { |
39193685 | 6078 | if (unlikely(mas->index <= min)) { |
54a611b6 | 6079 | mas->node = MAS_NONE; |
6b9e93e0 | 6080 | return true; |
54a611b6 LH |
6081 | } |
6082 | mas->node = MAS_START; | |
6083 | mas->last = --mas->index; | |
6084 | } | |
6085 | ||
6086 | if (unlikely(mas_is_start(mas))) { | |
6087 | /* First run or continue */ | |
54a611b6 | 6088 | if (mas->index < min) |
6b9e93e0 | 6089 | return true; |
54a611b6 | 6090 | |
6b9e93e0 LH |
6091 | *entry = mas_walk(mas); |
6092 | if (*entry) | |
6093 | return true; | |
54a611b6 LH |
6094 | } |
6095 | ||
39193685 LH |
6096 | if (unlikely(!mas_searchable(mas))) { |
6097 | if (mas_is_ptr(mas)) | |
6098 | goto none; | |
6099 | ||
6100 | if (mas_is_none(mas)) { | |
6101 | /* | |
6102 | * Walked to the location, and there was nothing so the | |
6103 | * previous location is 0. | |
6104 | */ | |
6105 | mas->last = mas->index = 0; | |
6106 | mas->node = MAS_ROOT; | |
6b9e93e0 LH |
6107 | *entry = mas_root(mas); |
6108 | return true; | |
39193685 LH |
6109 | } |
6110 | } | |
54a611b6 LH |
6111 | |
6112 | if (mas->index < min) | |
6b9e93e0 | 6113 | return true; |
54a611b6 | 6114 | |
6b9e93e0 | 6115 | return false; |
39193685 LH |
6116 | |
6117 | none: | |
6118 | mas->node = MAS_NONE; | |
6b9e93e0 LH |
6119 | return true; |
6120 | } | |
6121 | ||
6122 | /** | |
6123 | * mas_find_rev: On the first call, find the first non-null entry at or below | |
6124 | * mas->index down to %min. Otherwise find the first non-null entry below | |
6125 | * mas->index down to %min. | |
6126 | * @mas: The maple state | |
6127 | * @min: The minimum value to check. | |
6128 | * | |
6129 | * Must hold rcu_read_lock or the write lock. | |
6130 | * If an entry exists, last and index are updated accordingly. | |
6131 | * May set @mas->node to MAS_NONE. | |
6132 | * | |
6133 | * Return: The entry or %NULL. | |
6134 | */ | |
6135 | void *mas_find_rev(struct ma_state *mas, unsigned long min) | |
6136 | { | |
6137 | void *entry; | |
6138 | ||
6139 | if (mas_find_rev_setup(mas, min, &entry)) | |
6140 | return entry; | |
6141 | ||
6142 | /* Retries on dead nodes handled by mas_prev_slot */ | |
6143 | return mas_prev_slot(mas, min, false); | |
6144 | ||
54a611b6 | 6145 | } |
120b1162 | 6146 | EXPORT_SYMBOL_GPL(mas_find_rev); |
54a611b6 | 6147 | |
6b9e93e0 LH |
6148 | /** |
6149 | * mas_find_range_rev: On the first call, find the first non-null entry at or | |
6150 | * below mas->index down to %min. Otherwise advance to the previous slot after | |
6151 | * mas->index down to %min. | |
6152 | * @mas: The maple state | |
6153 | * @min: The minimum value to check. | |
6154 | * | |
6155 | * Must hold rcu_read_lock or the write lock. | |
6156 | * If an entry exists, last and index are updated accordingly. | |
6157 | * May set @mas->node to MAS_NONE. | |
6158 | * | |
6159 | * Return: The entry or %NULL. | |
6160 | */ | |
6161 | void *mas_find_range_rev(struct ma_state *mas, unsigned long min) | |
6162 | { | |
6163 | void *entry; | |
6164 | ||
6165 | if (mas_find_rev_setup(mas, min, &entry)) | |
6166 | return entry; | |
6167 | ||
6168 | /* Retries on dead nodes handled by mas_prev_slot */ | |
6169 | return mas_prev_slot(mas, min, true); | |
6170 | } | |
6171 | EXPORT_SYMBOL_GPL(mas_find_range_rev); | |
6172 | ||
54a611b6 LH |
6173 | /** |
6174 | * mas_erase() - Find the range in which index resides and erase the entire | |
6175 | * range. | |
6176 | * @mas: The maple state | |
6177 | * | |
6178 | * Must hold the write lock. | |
6179 | * Searches for @mas->index, sets @mas->index and @mas->last to the range and | |
6180 | * erases that range. | |
6181 | * | |
6182 | * Return: the entry that was erased or %NULL, @mas->index and @mas->last are updated. | |
6183 | */ | |
6184 | void *mas_erase(struct ma_state *mas) | |
6185 | { | |
6186 | void *entry; | |
6187 | MA_WR_STATE(wr_mas, mas, NULL); | |
6188 | ||
6189 | if (mas_is_none(mas) || mas_is_paused(mas)) | |
6190 | mas->node = MAS_START; | |
6191 | ||
6192 | /* Retry unnecessary when holding the write lock. */ | |
6193 | entry = mas_state_walk(mas); | |
6194 | if (!entry) | |
6195 | return NULL; | |
6196 | ||
6197 | write_retry: | |
6198 | /* Must reset to ensure spanning writes of last slot are detected */ | |
6199 | mas_reset(mas); | |
6200 | mas_wr_store_setup(&wr_mas); | |
6201 | mas_wr_store_entry(&wr_mas); | |
6202 | if (mas_nomem(mas, GFP_KERNEL)) | |
6203 | goto write_retry; | |
6204 | ||
6205 | return entry; | |
6206 | } | |
6207 | EXPORT_SYMBOL_GPL(mas_erase); | |
6208 | ||
6209 | /** | |
6210 | * mas_nomem() - Check if there was an error allocating and do the allocation | |
6211 | * if necessary If there are allocations, then free them. | |
6212 | * @mas: The maple state | |
6213 | * @gfp: The GFP_FLAGS to use for allocations | |
6214 | * Return: true on allocation, false otherwise. | |
6215 | */ | |
6216 | bool mas_nomem(struct ma_state *mas, gfp_t gfp) | |
14c4b5ab | 6217 | __must_hold(mas->tree->ma_lock) |
54a611b6 LH |
6218 | { |
6219 | if (likely(mas->node != MA_ERROR(-ENOMEM))) { | |
6220 | mas_destroy(mas); | |
6221 | return false; | |
6222 | } | |
6223 | ||
6224 | if (gfpflags_allow_blocking(gfp) && !mt_external_lock(mas->tree)) { | |
6225 | mtree_unlock(mas->tree); | |
6226 | mas_alloc_nodes(mas, gfp); | |
6227 | mtree_lock(mas->tree); | |
6228 | } else { | |
6229 | mas_alloc_nodes(mas, gfp); | |
6230 | } | |
6231 | ||
6232 | if (!mas_allocated(mas)) | |
6233 | return false; | |
6234 | ||
6235 | mas->node = MAS_START; | |
6236 | return true; | |
6237 | } | |
6238 | ||
6239 | void __init maple_tree_init(void) | |
6240 | { | |
6241 | maple_node_cache = kmem_cache_create("maple_node", | |
6242 | sizeof(struct maple_node), sizeof(struct maple_node), | |
6243 | SLAB_PANIC, NULL); | |
6244 | } | |
6245 | ||
6246 | /** | |
6247 | * mtree_load() - Load a value stored in a maple tree | |
6248 | * @mt: The maple tree | |
6249 | * @index: The index to load | |
6250 | * | |
6251 | * Return: the entry or %NULL | |
6252 | */ | |
6253 | void *mtree_load(struct maple_tree *mt, unsigned long index) | |
6254 | { | |
6255 | MA_STATE(mas, mt, index, index); | |
6256 | void *entry; | |
6257 | ||
6258 | trace_ma_read(__func__, &mas); | |
6259 | rcu_read_lock(); | |
6260 | retry: | |
6261 | entry = mas_start(&mas); | |
6262 | if (unlikely(mas_is_none(&mas))) | |
6263 | goto unlock; | |
6264 | ||
6265 | if (unlikely(mas_is_ptr(&mas))) { | |
6266 | if (index) | |
6267 | entry = NULL; | |
6268 | ||
6269 | goto unlock; | |
6270 | } | |
6271 | ||
6272 | entry = mtree_lookup_walk(&mas); | |
6273 | if (!entry && unlikely(mas_is_start(&mas))) | |
6274 | goto retry; | |
6275 | unlock: | |
6276 | rcu_read_unlock(); | |
6277 | if (xa_is_zero(entry)) | |
6278 | return NULL; | |
6279 | ||
6280 | return entry; | |
6281 | } | |
6282 | EXPORT_SYMBOL(mtree_load); | |
6283 | ||
6284 | /** | |
6285 | * mtree_store_range() - Store an entry at a given range. | |
6286 | * @mt: The maple tree | |
6287 | * @index: The start of the range | |
6288 | * @last: The end of the range | |
6289 | * @entry: The entry to store | |
6290 | * @gfp: The GFP_FLAGS to use for allocations | |
6291 | * | |
6292 | * Return: 0 on success, -EINVAL on invalid request, -ENOMEM if memory could not | |
6293 | * be allocated. | |
6294 | */ | |
6295 | int mtree_store_range(struct maple_tree *mt, unsigned long index, | |
6296 | unsigned long last, void *entry, gfp_t gfp) | |
6297 | { | |
6298 | MA_STATE(mas, mt, index, last); | |
6299 | MA_WR_STATE(wr_mas, &mas, entry); | |
6300 | ||
6301 | trace_ma_write(__func__, &mas, 0, entry); | |
6302 | if (WARN_ON_ONCE(xa_is_advanced(entry))) | |
6303 | return -EINVAL; | |
6304 | ||
6305 | if (index > last) | |
6306 | return -EINVAL; | |
6307 | ||
6308 | mtree_lock(mt); | |
6309 | retry: | |
6310 | mas_wr_store_entry(&wr_mas); | |
6311 | if (mas_nomem(&mas, gfp)) | |
6312 | goto retry; | |
6313 | ||
6314 | mtree_unlock(mt); | |
6315 | if (mas_is_err(&mas)) | |
6316 | return xa_err(mas.node); | |
6317 | ||
6318 | return 0; | |
6319 | } | |
6320 | EXPORT_SYMBOL(mtree_store_range); | |
6321 | ||
6322 | /** | |
6323 | * mtree_store() - Store an entry at a given index. | |
6324 | * @mt: The maple tree | |
6325 | * @index: The index to store the value | |
6326 | * @entry: The entry to store | |
6327 | * @gfp: The GFP_FLAGS to use for allocations | |
6328 | * | |
6329 | * Return: 0 on success, -EINVAL on invalid request, -ENOMEM if memory could not | |
6330 | * be allocated. | |
6331 | */ | |
6332 | int mtree_store(struct maple_tree *mt, unsigned long index, void *entry, | |
6333 | gfp_t gfp) | |
6334 | { | |
6335 | return mtree_store_range(mt, index, index, entry, gfp); | |
6336 | } | |
6337 | EXPORT_SYMBOL(mtree_store); | |
6338 | ||
6339 | /** | |
4445e582 | 6340 | * mtree_insert_range() - Insert an entry at a given range if there is no value. |
54a611b6 LH |
6341 | * @mt: The maple tree |
6342 | * @first: The start of the range | |
6343 | * @last: The end of the range | |
6344 | * @entry: The entry to store | |
6345 | * @gfp: The GFP_FLAGS to use for allocations. | |
6346 | * | |
6347 | * Return: 0 on success, -EEXISTS if the range is occupied, -EINVAL on invalid | |
6348 | * request, -ENOMEM if memory could not be allocated. | |
6349 | */ | |
6350 | int mtree_insert_range(struct maple_tree *mt, unsigned long first, | |
6351 | unsigned long last, void *entry, gfp_t gfp) | |
6352 | { | |
6353 | MA_STATE(ms, mt, first, last); | |
6354 | ||
6355 | if (WARN_ON_ONCE(xa_is_advanced(entry))) | |
6356 | return -EINVAL; | |
6357 | ||
6358 | if (first > last) | |
6359 | return -EINVAL; | |
6360 | ||
6361 | mtree_lock(mt); | |
6362 | retry: | |
6363 | mas_insert(&ms, entry); | |
6364 | if (mas_nomem(&ms, gfp)) | |
6365 | goto retry; | |
6366 | ||
6367 | mtree_unlock(mt); | |
6368 | if (mas_is_err(&ms)) | |
6369 | return xa_err(ms.node); | |
6370 | ||
6371 | return 0; | |
6372 | } | |
6373 | EXPORT_SYMBOL(mtree_insert_range); | |
6374 | ||
6375 | /** | |
4445e582 | 6376 | * mtree_insert() - Insert an entry at a given index if there is no value. |
54a611b6 LH |
6377 | * @mt: The maple tree |
6378 | * @index : The index to store the value | |
6379 | * @entry: The entry to store | |
4ae6944d | 6380 | * @gfp: The GFP_FLAGS to use for allocations. |
54a611b6 LH |
6381 | * |
6382 | * Return: 0 on success, -EEXISTS if the range is occupied, -EINVAL on invalid | |
6383 | * request, -ENOMEM if memory could not be allocated. | |
6384 | */ | |
6385 | int mtree_insert(struct maple_tree *mt, unsigned long index, void *entry, | |
6386 | gfp_t gfp) | |
6387 | { | |
6388 | return mtree_insert_range(mt, index, index, entry, gfp); | |
6389 | } | |
6390 | EXPORT_SYMBOL(mtree_insert); | |
6391 | ||
6392 | int mtree_alloc_range(struct maple_tree *mt, unsigned long *startp, | |
6393 | void *entry, unsigned long size, unsigned long min, | |
6394 | unsigned long max, gfp_t gfp) | |
6395 | { | |
6396 | int ret = 0; | |
6397 | ||
52371677 | 6398 | MA_STATE(mas, mt, 0, 0); |
54a611b6 LH |
6399 | if (!mt_is_alloc(mt)) |
6400 | return -EINVAL; | |
6401 | ||
6402 | if (WARN_ON_ONCE(mt_is_reserved(entry))) | |
6403 | return -EINVAL; | |
6404 | ||
54a611b6 LH |
6405 | mtree_lock(mt); |
6406 | retry: | |
52371677 PZ |
6407 | ret = mas_empty_area(&mas, min, max, size); |
6408 | if (ret) | |
6409 | goto unlock; | |
6410 | ||
6411 | mas_insert(&mas, entry); | |
6412 | /* | |
6413 | * mas_nomem() may release the lock, causing the allocated area | |
6414 | * to be unavailable, so try to allocate a free area again. | |
6415 | */ | |
54a611b6 LH |
6416 | if (mas_nomem(&mas, gfp)) |
6417 | goto retry; | |
6418 | ||
52371677 PZ |
6419 | if (mas_is_err(&mas)) |
6420 | ret = xa_err(mas.node); | |
6421 | else | |
6422 | *startp = mas.index; | |
6423 | ||
6424 | unlock: | |
54a611b6 LH |
6425 | mtree_unlock(mt); |
6426 | return ret; | |
6427 | } | |
6428 | EXPORT_SYMBOL(mtree_alloc_range); | |
6429 | ||
6430 | int mtree_alloc_rrange(struct maple_tree *mt, unsigned long *startp, | |
6431 | void *entry, unsigned long size, unsigned long min, | |
6432 | unsigned long max, gfp_t gfp) | |
6433 | { | |
6434 | int ret = 0; | |
6435 | ||
52371677 | 6436 | MA_STATE(mas, mt, 0, 0); |
54a611b6 LH |
6437 | if (!mt_is_alloc(mt)) |
6438 | return -EINVAL; | |
6439 | ||
6440 | if (WARN_ON_ONCE(mt_is_reserved(entry))) | |
6441 | return -EINVAL; | |
6442 | ||
54a611b6 LH |
6443 | mtree_lock(mt); |
6444 | retry: | |
52371677 PZ |
6445 | ret = mas_empty_area_rev(&mas, min, max, size); |
6446 | if (ret) | |
6447 | goto unlock; | |
6448 | ||
6449 | mas_insert(&mas, entry); | |
6450 | /* | |
6451 | * mas_nomem() may release the lock, causing the allocated area | |
6452 | * to be unavailable, so try to allocate a free area again. | |
6453 | */ | |
54a611b6 LH |
6454 | if (mas_nomem(&mas, gfp)) |
6455 | goto retry; | |
6456 | ||
52371677 PZ |
6457 | if (mas_is_err(&mas)) |
6458 | ret = xa_err(mas.node); | |
6459 | else | |
6460 | *startp = mas.index; | |
6461 | ||
6462 | unlock: | |
54a611b6 LH |
6463 | mtree_unlock(mt); |
6464 | return ret; | |
6465 | } | |
6466 | EXPORT_SYMBOL(mtree_alloc_rrange); | |
6467 | ||
6468 | /** | |
6469 | * mtree_erase() - Find an index and erase the entire range. | |
6470 | * @mt: The maple tree | |
6471 | * @index: The index to erase | |
6472 | * | |
6473 | * Erasing is the same as a walk to an entry then a store of a NULL to that | |
6474 | * ENTIRE range. In fact, it is implemented as such using the advanced API. | |
6475 | * | |
6476 | * Return: The entry stored at the @index or %NULL | |
6477 | */ | |
6478 | void *mtree_erase(struct maple_tree *mt, unsigned long index) | |
6479 | { | |
6480 | void *entry = NULL; | |
6481 | ||
6482 | MA_STATE(mas, mt, index, index); | |
6483 | trace_ma_op(__func__, &mas); | |
6484 | ||
6485 | mtree_lock(mt); | |
6486 | entry = mas_erase(&mas); | |
6487 | mtree_unlock(mt); | |
6488 | ||
6489 | return entry; | |
6490 | } | |
6491 | EXPORT_SYMBOL(mtree_erase); | |
6492 | ||
6493 | /** | |
6494 | * __mt_destroy() - Walk and free all nodes of a locked maple tree. | |
6495 | * @mt: The maple tree | |
6496 | * | |
6497 | * Note: Does not handle locking. | |
6498 | */ | |
6499 | void __mt_destroy(struct maple_tree *mt) | |
6500 | { | |
6501 | void *root = mt_root_locked(mt); | |
6502 | ||
6503 | rcu_assign_pointer(mt->ma_root, NULL); | |
6504 | if (xa_is_node(root)) | |
6505 | mte_destroy_walk(root, mt); | |
6506 | ||
6507 | mt->ma_flags = 0; | |
6508 | } | |
6509 | EXPORT_SYMBOL_GPL(__mt_destroy); | |
6510 | ||
6511 | /** | |
6512 | * mtree_destroy() - Destroy a maple tree | |
6513 | * @mt: The maple tree | |
6514 | * | |
6515 | * Frees all resources used by the tree. Handles locking. | |
6516 | */ | |
6517 | void mtree_destroy(struct maple_tree *mt) | |
6518 | { | |
6519 | mtree_lock(mt); | |
6520 | __mt_destroy(mt); | |
6521 | mtree_unlock(mt); | |
6522 | } | |
6523 | EXPORT_SYMBOL(mtree_destroy); | |
6524 | ||
6525 | /** | |
6526 | * mt_find() - Search from the start up until an entry is found. | |
6527 | * @mt: The maple tree | |
6528 | * @index: Pointer which contains the start location of the search | |
fad9c80e TG |
6529 | * @max: The maximum value of the search range |
6530 | * | |
6531 | * Takes RCU read lock internally to protect the search, which does not | |
6532 | * protect the returned pointer after dropping RCU read lock. | |
6533 | * See also: Documentation/core-api/maple_tree.rst | |
54a611b6 | 6534 | * |
fad9c80e TG |
6535 | * In case that an entry is found @index is updated to point to the next |
6536 | * possible entry independent whether the found entry is occupying a | |
6537 | * single index or a range if indices. | |
54a611b6 LH |
6538 | * |
6539 | * Return: The entry at or after the @index or %NULL | |
6540 | */ | |
6541 | void *mt_find(struct maple_tree *mt, unsigned long *index, unsigned long max) | |
6542 | { | |
6543 | MA_STATE(mas, mt, *index, *index); | |
6544 | void *entry; | |
6545 | #ifdef CONFIG_DEBUG_MAPLE_TREE | |
6546 | unsigned long copy = *index; | |
6547 | #endif | |
6548 | ||
6549 | trace_ma_read(__func__, &mas); | |
6550 | ||
6551 | if ((*index) > max) | |
6552 | return NULL; | |
6553 | ||
6554 | rcu_read_lock(); | |
6555 | retry: | |
6556 | entry = mas_state_walk(&mas); | |
6557 | if (mas_is_start(&mas)) | |
6558 | goto retry; | |
6559 | ||
6560 | if (unlikely(xa_is_zero(entry))) | |
6561 | entry = NULL; | |
6562 | ||
6563 | if (entry) | |
6564 | goto unlock; | |
6565 | ||
ca80f610 | 6566 | while (mas_searchable(&mas) && (mas.last < max)) { |
54a611b6 LH |
6567 | entry = mas_next_entry(&mas, max); |
6568 | if (likely(entry && !xa_is_zero(entry))) | |
6569 | break; | |
6570 | } | |
6571 | ||
6572 | if (unlikely(xa_is_zero(entry))) | |
6573 | entry = NULL; | |
6574 | unlock: | |
6575 | rcu_read_unlock(); | |
6576 | if (likely(entry)) { | |
6577 | *index = mas.last + 1; | |
6578 | #ifdef CONFIG_DEBUG_MAPLE_TREE | |
e6d6792a | 6579 | if (MT_WARN_ON(mt, (*index) && ((*index) <= copy))) |
54a611b6 LH |
6580 | pr_err("index not increased! %lx <= %lx\n", |
6581 | *index, copy); | |
54a611b6 LH |
6582 | #endif |
6583 | } | |
6584 | ||
6585 | return entry; | |
6586 | } | |
6587 | EXPORT_SYMBOL(mt_find); | |
6588 | ||
6589 | /** | |
6590 | * mt_find_after() - Search from the start up until an entry is found. | |
6591 | * @mt: The maple tree | |
6592 | * @index: Pointer which contains the start location of the search | |
6593 | * @max: The maximum value to check | |
6594 | * | |
fad9c80e TG |
6595 | * Same as mt_find() except that it checks @index for 0 before |
6596 | * searching. If @index == 0, the search is aborted. This covers a wrap | |
6597 | * around of @index to 0 in an iterator loop. | |
54a611b6 LH |
6598 | * |
6599 | * Return: The entry at or after the @index or %NULL | |
6600 | */ | |
6601 | void *mt_find_after(struct maple_tree *mt, unsigned long *index, | |
6602 | unsigned long max) | |
6603 | { | |
6604 | if (!(*index)) | |
6605 | return NULL; | |
6606 | ||
6607 | return mt_find(mt, index, max); | |
6608 | } | |
6609 | EXPORT_SYMBOL(mt_find_after); | |
6610 | ||
6611 | #ifdef CONFIG_DEBUG_MAPLE_TREE | |
6612 | atomic_t maple_tree_tests_run; | |
6613 | EXPORT_SYMBOL_GPL(maple_tree_tests_run); | |
6614 | atomic_t maple_tree_tests_passed; | |
6615 | EXPORT_SYMBOL_GPL(maple_tree_tests_passed); | |
6616 | ||
6617 | #ifndef __KERNEL__ | |
6618 | extern void kmem_cache_set_non_kernel(struct kmem_cache *, unsigned int); | |
6619 | void mt_set_non_kernel(unsigned int val) | |
6620 | { | |
6621 | kmem_cache_set_non_kernel(maple_node_cache, val); | |
6622 | } | |
6623 | ||
6624 | extern unsigned long kmem_cache_get_alloc(struct kmem_cache *); | |
6625 | unsigned long mt_get_alloc_size(void) | |
6626 | { | |
6627 | return kmem_cache_get_alloc(maple_node_cache); | |
6628 | } | |
6629 | ||
6630 | extern void kmem_cache_zero_nr_tallocated(struct kmem_cache *); | |
6631 | void mt_zero_nr_tallocated(void) | |
6632 | { | |
6633 | kmem_cache_zero_nr_tallocated(maple_node_cache); | |
6634 | } | |
6635 | ||
6636 | extern unsigned int kmem_cache_nr_tallocated(struct kmem_cache *); | |
6637 | unsigned int mt_nr_tallocated(void) | |
6638 | { | |
6639 | return kmem_cache_nr_tallocated(maple_node_cache); | |
6640 | } | |
6641 | ||
6642 | extern unsigned int kmem_cache_nr_allocated(struct kmem_cache *); | |
6643 | unsigned int mt_nr_allocated(void) | |
6644 | { | |
6645 | return kmem_cache_nr_allocated(maple_node_cache); | |
6646 | } | |
6647 | ||
6648 | /* | |
6649 | * mas_dead_node() - Check if the maple state is pointing to a dead node. | |
6650 | * @mas: The maple state | |
6651 | * @index: The index to restore in @mas. | |
6652 | * | |
6653 | * Used in test code. | |
6654 | * Return: 1 if @mas has been reset to MAS_START, 0 otherwise. | |
6655 | */ | |
6656 | static inline int mas_dead_node(struct ma_state *mas, unsigned long index) | |
6657 | { | |
6658 | if (unlikely(!mas_searchable(mas) || mas_is_start(mas))) | |
6659 | return 0; | |
6660 | ||
6661 | if (likely(!mte_dead_node(mas->node))) | |
6662 | return 0; | |
6663 | ||
6664 | mas_rewalk(mas, index); | |
6665 | return 1; | |
6666 | } | |
54a611b6 | 6667 | |
120b1162 LH |
6668 | void mt_cache_shrink(void) |
6669 | { | |
6670 | } | |
6671 | #else | |
6672 | /* | |
6673 | * mt_cache_shrink() - For testing, don't use this. | |
6674 | * | |
6675 | * Certain testcases can trigger an OOM when combined with other memory | |
6676 | * debugging configuration options. This function is used to reduce the | |
6677 | * possibility of an out of memory even due to kmem_cache objects remaining | |
6678 | * around for longer than usual. | |
6679 | */ | |
6680 | void mt_cache_shrink(void) | |
6681 | { | |
6682 | kmem_cache_shrink(maple_node_cache); | |
6683 | ||
6684 | } | |
6685 | EXPORT_SYMBOL_GPL(mt_cache_shrink); | |
6686 | ||
6687 | #endif /* not defined __KERNEL__ */ | |
54a611b6 LH |
6688 | /* |
6689 | * mas_get_slot() - Get the entry in the maple state node stored at @offset. | |
6690 | * @mas: The maple state | |
6691 | * @offset: The offset into the slot array to fetch. | |
6692 | * | |
6693 | * Return: The entry stored at @offset. | |
6694 | */ | |
6695 | static inline struct maple_enode *mas_get_slot(struct ma_state *mas, | |
6696 | unsigned char offset) | |
6697 | { | |
6698 | return mas_slot(mas, ma_slots(mas_mn(mas), mte_node_type(mas->node)), | |
6699 | offset); | |
6700 | } | |
6701 | ||
54a611b6 LH |
6702 | /* Depth first search, post-order */ |
6703 | static void mas_dfs_postorder(struct ma_state *mas, unsigned long max) | |
6704 | { | |
6705 | ||
6706 | struct maple_enode *p = MAS_NONE, *mn = mas->node; | |
6707 | unsigned long p_min, p_max; | |
6708 | ||
6709 | mas_next_node(mas, mas_mn(mas), max); | |
6710 | if (!mas_is_none(mas)) | |
6711 | return; | |
6712 | ||
6713 | if (mte_is_root(mn)) | |
6714 | return; | |
6715 | ||
6716 | mas->node = mn; | |
6717 | mas_ascend(mas); | |
c3eb787e | 6718 | do { |
54a611b6 LH |
6719 | p = mas->node; |
6720 | p_min = mas->min; | |
6721 | p_max = mas->max; | |
6722 | mas_prev_node(mas, 0); | |
c3eb787e | 6723 | } while (!mas_is_none(mas)); |
54a611b6 LH |
6724 | |
6725 | mas->node = p; | |
6726 | mas->max = p_max; | |
6727 | mas->min = p_min; | |
6728 | } | |
6729 | ||
6730 | /* Tree validations */ | |
6731 | static void mt_dump_node(const struct maple_tree *mt, void *entry, | |
89f499f3 LH |
6732 | unsigned long min, unsigned long max, unsigned int depth, |
6733 | enum mt_dump_format format); | |
54a611b6 | 6734 | static void mt_dump_range(unsigned long min, unsigned long max, |
89f499f3 | 6735 | unsigned int depth, enum mt_dump_format format) |
54a611b6 LH |
6736 | { |
6737 | static const char spaces[] = " "; | |
6738 | ||
89f499f3 LH |
6739 | switch(format) { |
6740 | case mt_dump_hex: | |
6741 | if (min == max) | |
6742 | pr_info("%.*s%lx: ", depth * 2, spaces, min); | |
6743 | else | |
6744 | pr_info("%.*s%lx-%lx: ", depth * 2, spaces, min, max); | |
6745 | break; | |
6746 | default: | |
6747 | case mt_dump_dec: | |
6748 | if (min == max) | |
6749 | pr_info("%.*s%lu: ", depth * 2, spaces, min); | |
6750 | else | |
6751 | pr_info("%.*s%lu-%lu: ", depth * 2, spaces, min, max); | |
6752 | } | |
54a611b6 LH |
6753 | } |
6754 | ||
6755 | static void mt_dump_entry(void *entry, unsigned long min, unsigned long max, | |
89f499f3 | 6756 | unsigned int depth, enum mt_dump_format format) |
54a611b6 | 6757 | { |
89f499f3 | 6758 | mt_dump_range(min, max, depth, format); |
54a611b6 LH |
6759 | |
6760 | if (xa_is_value(entry)) | |
6761 | pr_cont("value %ld (0x%lx) [%p]\n", xa_to_value(entry), | |
6762 | xa_to_value(entry), entry); | |
6763 | else if (xa_is_zero(entry)) | |
6764 | pr_cont("zero (%ld)\n", xa_to_internal(entry)); | |
6765 | else if (mt_is_reserved(entry)) | |
6766 | pr_cont("UNKNOWN ENTRY (%p)\n", entry); | |
6767 | else | |
6768 | pr_cont("%p\n", entry); | |
6769 | } | |
6770 | ||
6771 | static void mt_dump_range64(const struct maple_tree *mt, void *entry, | |
89f499f3 LH |
6772 | unsigned long min, unsigned long max, unsigned int depth, |
6773 | enum mt_dump_format format) | |
54a611b6 LH |
6774 | { |
6775 | struct maple_range_64 *node = &mte_to_node(entry)->mr64; | |
6776 | bool leaf = mte_is_leaf(entry); | |
6777 | unsigned long first = min; | |
6778 | int i; | |
6779 | ||
6780 | pr_cont(" contents: "); | |
89f499f3 LH |
6781 | for (i = 0; i < MAPLE_RANGE64_SLOTS - 1; i++) { |
6782 | switch(format) { | |
6783 | case mt_dump_hex: | |
6784 | pr_cont("%p %lX ", node->slot[i], node->pivot[i]); | |
6785 | break; | |
6786 | default: | |
6787 | case mt_dump_dec: | |
6788 | pr_cont("%p %lu ", node->slot[i], node->pivot[i]); | |
6789 | } | |
6790 | } | |
54a611b6 LH |
6791 | pr_cont("%p\n", node->slot[i]); |
6792 | for (i = 0; i < MAPLE_RANGE64_SLOTS; i++) { | |
6793 | unsigned long last = max; | |
6794 | ||
6795 | if (i < (MAPLE_RANGE64_SLOTS - 1)) | |
6796 | last = node->pivot[i]; | |
bd592703 | 6797 | else if (!node->slot[i] && max != mt_node_max(entry)) |
54a611b6 LH |
6798 | break; |
6799 | if (last == 0 && i > 0) | |
6800 | break; | |
6801 | if (leaf) | |
6802 | mt_dump_entry(mt_slot(mt, node->slot, i), | |
89f499f3 | 6803 | first, last, depth + 1, format); |
54a611b6 LH |
6804 | else if (node->slot[i]) |
6805 | mt_dump_node(mt, mt_slot(mt, node->slot, i), | |
89f499f3 | 6806 | first, last, depth + 1, format); |
54a611b6 LH |
6807 | |
6808 | if (last == max) | |
6809 | break; | |
6810 | if (last > max) { | |
89f499f3 LH |
6811 | switch(format) { |
6812 | case mt_dump_hex: | |
6813 | pr_err("node %p last (%lx) > max (%lx) at pivot %d!\n", | |
54a611b6 | 6814 | node, last, max, i); |
89f499f3 LH |
6815 | break; |
6816 | default: | |
6817 | case mt_dump_dec: | |
6818 | pr_err("node %p last (%lu) > max (%lu) at pivot %d!\n", | |
6819 | node, last, max, i); | |
6820 | } | |
54a611b6 LH |
6821 | } |
6822 | first = last + 1; | |
6823 | } | |
6824 | } | |
6825 | ||
6826 | static void mt_dump_arange64(const struct maple_tree *mt, void *entry, | |
89f499f3 LH |
6827 | unsigned long min, unsigned long max, unsigned int depth, |
6828 | enum mt_dump_format format) | |
54a611b6 LH |
6829 | { |
6830 | struct maple_arange_64 *node = &mte_to_node(entry)->ma64; | |
6831 | bool leaf = mte_is_leaf(entry); | |
6832 | unsigned long first = min; | |
6833 | int i; | |
6834 | ||
6835 | pr_cont(" contents: "); | |
6836 | for (i = 0; i < MAPLE_ARANGE64_SLOTS; i++) | |
6837 | pr_cont("%lu ", node->gap[i]); | |
6838 | pr_cont("| %02X %02X| ", node->meta.end, node->meta.gap); | |
6839 | for (i = 0; i < MAPLE_ARANGE64_SLOTS - 1; i++) | |
6840 | pr_cont("%p %lu ", node->slot[i], node->pivot[i]); | |
6841 | pr_cont("%p\n", node->slot[i]); | |
6842 | for (i = 0; i < MAPLE_ARANGE64_SLOTS; i++) { | |
6843 | unsigned long last = max; | |
6844 | ||
6845 | if (i < (MAPLE_ARANGE64_SLOTS - 1)) | |
6846 | last = node->pivot[i]; | |
6847 | else if (!node->slot[i]) | |
6848 | break; | |
6849 | if (last == 0 && i > 0) | |
6850 | break; | |
6851 | if (leaf) | |
6852 | mt_dump_entry(mt_slot(mt, node->slot, i), | |
89f499f3 | 6853 | first, last, depth + 1, format); |
54a611b6 LH |
6854 | else if (node->slot[i]) |
6855 | mt_dump_node(mt, mt_slot(mt, node->slot, i), | |
89f499f3 | 6856 | first, last, depth + 1, format); |
54a611b6 LH |
6857 | |
6858 | if (last == max) | |
6859 | break; | |
6860 | if (last > max) { | |
6861 | pr_err("node %p last (%lu) > max (%lu) at pivot %d!\n", | |
6862 | node, last, max, i); | |
6863 | break; | |
6864 | } | |
6865 | first = last + 1; | |
6866 | } | |
6867 | } | |
6868 | ||
6869 | static void mt_dump_node(const struct maple_tree *mt, void *entry, | |
89f499f3 LH |
6870 | unsigned long min, unsigned long max, unsigned int depth, |
6871 | enum mt_dump_format format) | |
54a611b6 LH |
6872 | { |
6873 | struct maple_node *node = mte_to_node(entry); | |
6874 | unsigned int type = mte_node_type(entry); | |
6875 | unsigned int i; | |
6876 | ||
89f499f3 | 6877 | mt_dump_range(min, max, depth, format); |
54a611b6 LH |
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), | |
89f499f3 | 6888 | min + i, min + i, depth, format); |
54a611b6 LH |
6889 | } |
6890 | break; | |
6891 | case maple_leaf_64: | |
6892 | case maple_range_64: | |
89f499f3 | 6893 | mt_dump_range64(mt, entry, min, max, depth, format); |
54a611b6 LH |
6894 | break; |
6895 | case maple_arange_64: | |
89f499f3 | 6896 | mt_dump_arange64(mt, entry, min, max, depth, format); |
54a611b6 LH |
6897 | break; |
6898 | ||
6899 | default: | |
6900 | pr_cont(" UNKNOWN TYPE\n"); | |
6901 | } | |
6902 | } | |
6903 | ||
89f499f3 | 6904 | void mt_dump(const struct maple_tree *mt, enum mt_dump_format format) |
54a611b6 LH |
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)) | |
89f499f3 | 6911 | mt_dump_entry(entry, 0, 0, 0, format); |
54a611b6 | 6912 | else if (entry) |
89f499f3 | 6913 | mt_dump_node(mt, entry, 0, mt_node_max(entry), 0, format); |
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; | |
f8e5eac8 PZ |
6924 | struct maple_node *p_mn, *node = mte_to_node(mte); |
6925 | enum maple_type mt = mte_node_type(mas->node); | |
54a611b6 LH |
6926 | unsigned long gap = 0, max_gap = 0; |
6927 | unsigned long p_end, p_start = mas->min; | |
f8e5eac8 | 6928 | unsigned char p_slot, offset; |
54a611b6 | 6929 | unsigned long *gaps = NULL; |
f8e5eac8 PZ |
6930 | unsigned long *pivots = ma_pivots(node, mt); |
6931 | unsigned int i; | |
54a611b6 | 6932 | |
f8e5eac8 | 6933 | if (ma_is_dense(mt)) { |
54a611b6 LH |
6934 | for (i = 0; i < mt_slot_count(mte); i++) { |
6935 | if (mas_get_slot(mas, i)) { | |
6936 | if (gap > max_gap) | |
6937 | max_gap = gap; | |
6938 | gap = 0; | |
6939 | continue; | |
6940 | } | |
6941 | gap++; | |
6942 | } | |
6943 | goto counted; | |
6944 | } | |
6945 | ||
f8e5eac8 | 6946 | gaps = ma_gaps(node, mt); |
54a611b6 | 6947 | for (i = 0; i < mt_slot_count(mte); i++) { |
29b2681f | 6948 | p_end = mas_safe_pivot(mas, pivots, i, mt); |
54a611b6 LH |
6949 | |
6950 | if (!gaps) { | |
f8e5eac8 PZ |
6951 | if (!mas_get_slot(mas, i)) |
6952 | gap = p_end - p_start + 1; | |
54a611b6 LH |
6953 | } else { |
6954 | void *entry = mas_get_slot(mas, i); | |
6955 | ||
6956 | gap = gaps[i]; | |
f8e5eac8 PZ |
6957 | MT_BUG_ON(mas->tree, !entry); |
6958 | ||
6959 | if (gap > p_end - p_start + 1) { | |
6960 | pr_err("%p[%u] %lu >= %lu - %lu + 1 (%lu)\n", | |
6961 | mas_mn(mas), i, gap, p_end, p_start, | |
6962 | p_end - p_start + 1); | |
6963 | MT_BUG_ON(mas->tree, gap > p_end - p_start + 1); | |
54a611b6 LH |
6964 | } |
6965 | } | |
6966 | ||
6967 | if (gap > max_gap) | |
6968 | max_gap = gap; | |
f8e5eac8 | 6969 | |
54a611b6 LH |
6970 | p_start = p_end + 1; |
6971 | if (p_end >= mas->max) | |
6972 | break; | |
6973 | } | |
6974 | ||
6975 | counted: | |
f8e5eac8 PZ |
6976 | if (mt == maple_arange_64) { |
6977 | offset = ma_meta_gap(node, mt); | |
6978 | if (offset > i) { | |
6979 | pr_err("gap offset %p[%u] is invalid\n", node, offset); | |
6980 | MT_BUG_ON(mas->tree, 1); | |
6981 | } | |
6982 | ||
6983 | if (gaps[offset] != max_gap) { | |
6984 | pr_err("gap %p[%u] is not the largest gap %lu\n", | |
6985 | node, offset, max_gap); | |
6986 | MT_BUG_ON(mas->tree, 1); | |
6987 | } | |
6988 | ||
6989 | MT_BUG_ON(mas->tree, !gaps); | |
6990 | for (i++ ; i < mt_slot_count(mte); i++) { | |
6991 | if (gaps[i] != 0) { | |
6992 | pr_err("gap %p[%u] beyond node limit != 0\n", | |
6993 | node, i); | |
6994 | MT_BUG_ON(mas->tree, 1); | |
6995 | } | |
6996 | } | |
6997 | } | |
6998 | ||
54a611b6 LH |
6999 | if (mte_is_root(mte)) |
7000 | return; | |
7001 | ||
7002 | p_slot = mte_parent_slot(mas->node); | |
7003 | p_mn = mte_parent(mte); | |
7004 | MT_BUG_ON(mas->tree, max_gap > mas->max); | |
afc754c6 | 7005 | if (ma_gaps(p_mn, mas_parent_type(mas, mte))[p_slot] != max_gap) { |
54a611b6 | 7006 | pr_err("gap %p[%u] != %lu\n", p_mn, p_slot, max_gap); |
89f499f3 | 7007 | mt_dump(mas->tree, mt_dump_hex); |
f8e5eac8 | 7008 | MT_BUG_ON(mas->tree, 1); |
54a611b6 | 7009 | } |
54a611b6 LH |
7010 | } |
7011 | ||
7012 | static void mas_validate_parent_slot(struct ma_state *mas) | |
7013 | { | |
7014 | struct maple_node *parent; | |
7015 | struct maple_enode *node; | |
afc754c6 LH |
7016 | enum maple_type p_type; |
7017 | unsigned char p_slot; | |
54a611b6 LH |
7018 | void __rcu **slots; |
7019 | int i; | |
7020 | ||
7021 | if (mte_is_root(mas->node)) | |
7022 | return; | |
7023 | ||
afc754c6 LH |
7024 | p_slot = mte_parent_slot(mas->node); |
7025 | p_type = mas_parent_type(mas, mas->node); | |
54a611b6 LH |
7026 | parent = mte_parent(mas->node); |
7027 | slots = ma_slots(parent, p_type); | |
7028 | MT_BUG_ON(mas->tree, mas_mn(mas) == parent); | |
7029 | ||
7030 | /* Check prev/next parent slot for duplicate node entry */ | |
7031 | ||
7032 | for (i = 0; i < mt_slots[p_type]; i++) { | |
7033 | node = mas_slot(mas, slots, i); | |
7034 | if (i == p_slot) { | |
7035 | if (node != mas->node) | |
7036 | pr_err("parent %p[%u] does not have %p\n", | |
7037 | parent, i, mas_mn(mas)); | |
7038 | MT_BUG_ON(mas->tree, node != mas->node); | |
7039 | } else if (node == mas->node) { | |
7040 | pr_err("Invalid child %p at parent %p[%u] p_slot %u\n", | |
7041 | mas_mn(mas), parent, i, p_slot); | |
7042 | MT_BUG_ON(mas->tree, node == mas->node); | |
7043 | } | |
7044 | } | |
7045 | } | |
7046 | ||
7047 | static void mas_validate_child_slot(struct ma_state *mas) | |
7048 | { | |
7049 | enum maple_type type = mte_node_type(mas->node); | |
7050 | void __rcu **slots = ma_slots(mte_to_node(mas->node), type); | |
7051 | unsigned long *pivots = ma_pivots(mte_to_node(mas->node), type); | |
7052 | struct maple_enode *child; | |
7053 | unsigned char i; | |
7054 | ||
7055 | if (mte_is_leaf(mas->node)) | |
7056 | return; | |
7057 | ||
7058 | for (i = 0; i < mt_slots[type]; i++) { | |
7059 | child = mas_slot(mas, slots, i); | |
54a611b6 | 7060 | |
e93fda5a PZ |
7061 | if (!child) { |
7062 | pr_err("Non-leaf node lacks child at %p[%u]\n", | |
7063 | mas_mn(mas), i); | |
7064 | MT_BUG_ON(mas->tree, 1); | |
7065 | } | |
54a611b6 LH |
7066 | |
7067 | if (mte_parent_slot(child) != i) { | |
7068 | pr_err("Slot error at %p[%u]: child %p has pslot %u\n", | |
7069 | mas_mn(mas), i, mte_to_node(child), | |
7070 | mte_parent_slot(child)); | |
7071 | MT_BUG_ON(mas->tree, 1); | |
7072 | } | |
7073 | ||
7074 | if (mte_parent(child) != mte_to_node(mas->node)) { | |
7075 | pr_err("child %p has parent %p not %p\n", | |
7076 | mte_to_node(child), mte_parent(child), | |
7077 | mte_to_node(mas->node)); | |
7078 | MT_BUG_ON(mas->tree, 1); | |
7079 | } | |
e93fda5a PZ |
7080 | |
7081 | if (i < mt_pivots[type] && pivots[i] == mas->max) | |
7082 | break; | |
54a611b6 LH |
7083 | } |
7084 | } | |
7085 | ||
7086 | /* | |
33af39d0 PZ |
7087 | * Validate all pivots are within mas->min and mas->max, check metadata ends |
7088 | * where the maximum ends and ensure there is no slots or pivots set outside of | |
7089 | * the end of the data. | |
54a611b6 LH |
7090 | */ |
7091 | static void mas_validate_limits(struct ma_state *mas) | |
7092 | { | |
7093 | int i; | |
7094 | unsigned long prev_piv = 0; | |
7095 | enum maple_type type = mte_node_type(mas->node); | |
7096 | void __rcu **slots = ma_slots(mte_to_node(mas->node), type); | |
7097 | unsigned long *pivots = ma_pivots(mas_mn(mas), type); | |
7098 | ||
54a611b6 LH |
7099 | for (i = 0; i < mt_slots[type]; i++) { |
7100 | unsigned long piv; | |
7101 | ||
7102 | piv = mas_safe_pivot(mas, pivots, i, type); | |
7103 | ||
33af39d0 PZ |
7104 | if (!piv && (i != 0)) { |
7105 | pr_err("Missing node limit pivot at %p[%u]", | |
7106 | mas_mn(mas), i); | |
7107 | MAS_WARN_ON(mas, 1); | |
54a611b6 LH |
7108 | } |
7109 | ||
7110 | if (prev_piv > piv) { | |
7111 | pr_err("%p[%u] piv %lu < prev_piv %lu\n", | |
7112 | mas_mn(mas), i, piv, prev_piv); | |
e6d6792a | 7113 | MAS_WARN_ON(mas, piv < prev_piv); |
54a611b6 LH |
7114 | } |
7115 | ||
7116 | if (piv < mas->min) { | |
7117 | pr_err("%p[%u] %lu < %lu\n", mas_mn(mas), i, | |
7118 | piv, mas->min); | |
e6d6792a | 7119 | MAS_WARN_ON(mas, piv < mas->min); |
54a611b6 LH |
7120 | } |
7121 | if (piv > mas->max) { | |
7122 | pr_err("%p[%u] %lu > %lu\n", mas_mn(mas), i, | |
7123 | piv, mas->max); | |
e6d6792a | 7124 | MAS_WARN_ON(mas, piv > mas->max); |
54a611b6 LH |
7125 | } |
7126 | prev_piv = piv; | |
7127 | if (piv == mas->max) | |
7128 | break; | |
7129 | } | |
33af39d0 PZ |
7130 | |
7131 | if (mas_data_end(mas) != i) { | |
7132 | pr_err("node%p: data_end %u != the last slot offset %u\n", | |
7133 | mas_mn(mas), mas_data_end(mas), i); | |
7134 | MT_BUG_ON(mas->tree, 1); | |
7135 | } | |
7136 | ||
54a611b6 LH |
7137 | for (i += 1; i < mt_slots[type]; i++) { |
7138 | void *entry = mas_slot(mas, slots, i); | |
7139 | ||
7140 | if (entry && (i != mt_slots[type] - 1)) { | |
7141 | pr_err("%p[%u] should not have entry %p\n", mas_mn(mas), | |
7142 | i, entry); | |
7143 | MT_BUG_ON(mas->tree, entry != NULL); | |
7144 | } | |
7145 | ||
7146 | if (i < mt_pivots[type]) { | |
7147 | unsigned long piv = pivots[i]; | |
7148 | ||
7149 | if (!piv) | |
7150 | continue; | |
7151 | ||
7152 | pr_err("%p[%u] should not have piv %lu\n", | |
7153 | mas_mn(mas), i, piv); | |
e6d6792a | 7154 | MAS_WARN_ON(mas, i < mt_pivots[type] - 1); |
54a611b6 LH |
7155 | } |
7156 | } | |
7157 | } | |
7158 | ||
7159 | static void mt_validate_nulls(struct maple_tree *mt) | |
7160 | { | |
7161 | void *entry, *last = (void *)1; | |
7162 | unsigned char offset = 0; | |
7163 | void __rcu **slots; | |
7164 | MA_STATE(mas, mt, 0, 0); | |
7165 | ||
7166 | mas_start(&mas); | |
7167 | if (mas_is_none(&mas) || (mas.node == MAS_ROOT)) | |
7168 | return; | |
7169 | ||
7170 | while (!mte_is_leaf(mas.node)) | |
7171 | mas_descend(&mas); | |
7172 | ||
7173 | slots = ma_slots(mte_to_node(mas.node), mte_node_type(mas.node)); | |
7174 | do { | |
7175 | entry = mas_slot(&mas, slots, offset); | |
7176 | if (!last && !entry) { | |
7177 | pr_err("Sequential nulls end at %p[%u]\n", | |
7178 | mas_mn(&mas), offset); | |
7179 | } | |
7180 | MT_BUG_ON(mt, !last && !entry); | |
7181 | last = entry; | |
7182 | if (offset == mas_data_end(&mas)) { | |
7183 | mas_next_node(&mas, mas_mn(&mas), ULONG_MAX); | |
7184 | if (mas_is_none(&mas)) | |
7185 | return; | |
7186 | offset = 0; | |
7187 | slots = ma_slots(mte_to_node(mas.node), | |
7188 | mte_node_type(mas.node)); | |
7189 | } else { | |
7190 | offset++; | |
7191 | } | |
7192 | ||
7193 | } while (!mas_is_none(&mas)); | |
7194 | } | |
7195 | ||
7196 | /* | |
7197 | * validate a maple tree by checking: | |
7198 | * 1. The limits (pivots are within mas->min to mas->max) | |
7199 | * 2. The gap is correctly set in the parents | |
7200 | */ | |
7201 | void mt_validate(struct maple_tree *mt) | |
7202 | { | |
7203 | unsigned char end; | |
7204 | ||
7205 | MA_STATE(mas, mt, 0, 0); | |
7206 | rcu_read_lock(); | |
7207 | mas_start(&mas); | |
7208 | if (!mas_searchable(&mas)) | |
7209 | goto done; | |
7210 | ||
a489539e PZ |
7211 | while (!mte_is_leaf(mas.node)) |
7212 | mas_descend(&mas); | |
7213 | ||
54a611b6 | 7214 | while (!mas_is_none(&mas)) { |
e6d6792a | 7215 | MAS_WARN_ON(&mas, mte_dead_node(mas.node)); |
a489539e PZ |
7216 | end = mas_data_end(&mas); |
7217 | if (MAS_WARN_ON(&mas, (end < mt_min_slot_count(mas.node)) && | |
7218 | (mas.max != ULONG_MAX))) { | |
7219 | pr_err("Invalid size %u of %p\n", end, mas_mn(&mas)); | |
54a611b6 | 7220 | } |
a489539e | 7221 | |
54a611b6 | 7222 | mas_validate_parent_slot(&mas); |
54a611b6 | 7223 | mas_validate_limits(&mas); |
a489539e | 7224 | mas_validate_child_slot(&mas); |
54a611b6 LH |
7225 | if (mt_is_alloc(mt)) |
7226 | mas_validate_gaps(&mas); | |
7227 | mas_dfs_postorder(&mas, ULONG_MAX); | |
7228 | } | |
7229 | mt_validate_nulls(mt); | |
7230 | done: | |
7231 | rcu_read_unlock(); | |
7232 | ||
7233 | } | |
120b1162 | 7234 | EXPORT_SYMBOL_GPL(mt_validate); |
54a611b6 | 7235 | |
f0a1f866 LH |
7236 | void mas_dump(const struct ma_state *mas) |
7237 | { | |
7238 | pr_err("MAS: tree=%p enode=%p ", mas->tree, mas->node); | |
7239 | if (mas_is_none(mas)) | |
7240 | pr_err("(MAS_NONE) "); | |
7241 | else if (mas_is_ptr(mas)) | |
7242 | pr_err("(MAS_ROOT) "); | |
7243 | else if (mas_is_start(mas)) | |
7244 | pr_err("(MAS_START) "); | |
7245 | else if (mas_is_paused(mas)) | |
7246 | pr_err("(MAS_PAUSED) "); | |
7247 | ||
7248 | pr_err("[%u] index=%lx last=%lx\n", mas->offset, mas->index, mas->last); | |
7249 | pr_err(" min=%lx max=%lx alloc=%p, depth=%u, flags=%x\n", | |
7250 | mas->min, mas->max, mas->alloc, mas->depth, mas->mas_flags); | |
7251 | if (mas->index > mas->last) | |
7252 | pr_err("Check index & last\n"); | |
7253 | } | |
7254 | EXPORT_SYMBOL_GPL(mas_dump); | |
7255 | ||
7256 | void mas_wr_dump(const struct ma_wr_state *wr_mas) | |
7257 | { | |
7258 | pr_err("WR_MAS: node=%p r_min=%lx r_max=%lx\n", | |
7259 | wr_mas->node, wr_mas->r_min, wr_mas->r_max); | |
7260 | pr_err(" type=%u off_end=%u, node_end=%u, end_piv=%lx\n", | |
7261 | wr_mas->type, wr_mas->offset_end, wr_mas->node_end, | |
7262 | wr_mas->end_piv); | |
7263 | } | |
7264 | EXPORT_SYMBOL_GPL(mas_wr_dump); | |
7265 | ||
54a611b6 | 7266 | #endif /* CONFIG_DEBUG_MAPLE_TREE */ |