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 | ||
780 | static inline bool mt_locked(const struct maple_tree *mt) | |
781 | { | |
782 | return mt_external_lock(mt) ? mt_lock_is_held(mt) : | |
783 | lockdep_is_held(&mt->ma_lock); | |
784 | } | |
785 | ||
786 | static inline void *mt_slot(const struct maple_tree *mt, | |
787 | void __rcu **slots, unsigned char offset) | |
788 | { | |
789 | return rcu_dereference_check(slots[offset], mt_locked(mt)); | |
790 | } | |
791 | ||
790e1fa8 LH |
792 | static inline void *mt_slot_locked(struct maple_tree *mt, void __rcu **slots, |
793 | unsigned char offset) | |
794 | { | |
795 | return rcu_dereference_protected(slots[offset], mt_locked(mt)); | |
796 | } | |
54a611b6 LH |
797 | /* |
798 | * mas_slot_locked() - Get the slot value when holding the maple tree lock. | |
799 | * @mas: The maple state | |
800 | * @slots: The pointer to the slots | |
801 | * @offset: The offset into the slots array to fetch | |
802 | * | |
803 | * Return: The entry stored in @slots at the @offset. | |
804 | */ | |
805 | static inline void *mas_slot_locked(struct ma_state *mas, void __rcu **slots, | |
806 | unsigned char offset) | |
807 | { | |
790e1fa8 | 808 | return mt_slot_locked(mas->tree, slots, offset); |
54a611b6 LH |
809 | } |
810 | ||
811 | /* | |
812 | * mas_slot() - Get the slot value when not holding the maple tree lock. | |
813 | * @mas: The maple state | |
814 | * @slots: The pointer to the slots | |
815 | * @offset: The offset into the slots array to fetch | |
816 | * | |
817 | * Return: The entry stored in @slots at the @offset | |
818 | */ | |
819 | static inline void *mas_slot(struct ma_state *mas, void __rcu **slots, | |
820 | unsigned char offset) | |
821 | { | |
822 | return mt_slot(mas->tree, slots, offset); | |
823 | } | |
824 | ||
825 | /* | |
826 | * mas_root() - Get the maple tree root. | |
827 | * @mas: The maple state. | |
828 | * | |
829 | * Return: The pointer to the root of the tree | |
830 | */ | |
831 | static inline void *mas_root(struct ma_state *mas) | |
832 | { | |
833 | return rcu_dereference_check(mas->tree->ma_root, mt_locked(mas->tree)); | |
834 | } | |
835 | ||
836 | static inline void *mt_root_locked(struct maple_tree *mt) | |
837 | { | |
838 | return rcu_dereference_protected(mt->ma_root, mt_locked(mt)); | |
839 | } | |
840 | ||
841 | /* | |
842 | * mas_root_locked() - Get the maple tree root when holding the maple tree lock. | |
843 | * @mas: The maple state. | |
844 | * | |
845 | * Return: The pointer to the root of the tree | |
846 | */ | |
847 | static inline void *mas_root_locked(struct ma_state *mas) | |
848 | { | |
849 | return mt_root_locked(mas->tree); | |
850 | } | |
851 | ||
852 | static inline struct maple_metadata *ma_meta(struct maple_node *mn, | |
853 | enum maple_type mt) | |
854 | { | |
855 | switch (mt) { | |
856 | case maple_arange_64: | |
857 | return &mn->ma64.meta; | |
858 | default: | |
859 | return &mn->mr64.meta; | |
860 | } | |
861 | } | |
862 | ||
863 | /* | |
864 | * ma_set_meta() - Set the metadata information of a node. | |
865 | * @mn: The maple node | |
866 | * @mt: The maple node type | |
867 | * @offset: The offset of the highest sub-gap in this node. | |
868 | * @end: The end of the data in this node. | |
869 | */ | |
870 | static inline void ma_set_meta(struct maple_node *mn, enum maple_type mt, | |
871 | unsigned char offset, unsigned char end) | |
872 | { | |
873 | struct maple_metadata *meta = ma_meta(mn, mt); | |
874 | ||
875 | meta->gap = offset; | |
876 | meta->end = end; | |
877 | } | |
878 | ||
2e5b4921 | 879 | /* |
790e1fa8 LH |
880 | * mt_clear_meta() - clear the metadata information of a node, if it exists |
881 | * @mt: The maple tree | |
2e5b4921 | 882 | * @mn: The maple node |
790e1fa8 | 883 | * @type: The maple node type |
2e5b4921 LH |
884 | * @offset: The offset of the highest sub-gap in this node. |
885 | * @end: The end of the data in this node. | |
886 | */ | |
790e1fa8 LH |
887 | static inline void mt_clear_meta(struct maple_tree *mt, struct maple_node *mn, |
888 | enum maple_type type) | |
2e5b4921 LH |
889 | { |
890 | struct maple_metadata *meta; | |
891 | unsigned long *pivots; | |
892 | void __rcu **slots; | |
893 | void *next; | |
894 | ||
790e1fa8 | 895 | switch (type) { |
2e5b4921 LH |
896 | case maple_range_64: |
897 | pivots = mn->mr64.pivot; | |
898 | if (unlikely(pivots[MAPLE_RANGE64_SLOTS - 2])) { | |
899 | slots = mn->mr64.slot; | |
790e1fa8 LH |
900 | next = mt_slot_locked(mt, slots, |
901 | MAPLE_RANGE64_SLOTS - 1); | |
902 | if (unlikely((mte_to_node(next) && | |
903 | mte_node_type(next)))) | |
904 | return; /* no metadata, could be node */ | |
2e5b4921 LH |
905 | } |
906 | fallthrough; | |
907 | case maple_arange_64: | |
790e1fa8 | 908 | meta = ma_meta(mn, type); |
2e5b4921 LH |
909 | break; |
910 | default: | |
911 | return; | |
912 | } | |
913 | ||
914 | meta->gap = 0; | |
915 | meta->end = 0; | |
916 | } | |
917 | ||
54a611b6 LH |
918 | /* |
919 | * ma_meta_end() - Get the data end of a node from the metadata | |
920 | * @mn: The maple node | |
921 | * @mt: The maple node type | |
922 | */ | |
923 | static inline unsigned char ma_meta_end(struct maple_node *mn, | |
924 | enum maple_type mt) | |
925 | { | |
926 | struct maple_metadata *meta = ma_meta(mn, mt); | |
927 | ||
928 | return meta->end; | |
929 | } | |
930 | ||
931 | /* | |
932 | * ma_meta_gap() - Get the largest gap location of a node from the metadata | |
933 | * @mn: The maple node | |
934 | * @mt: The maple node type | |
935 | */ | |
936 | static inline unsigned char ma_meta_gap(struct maple_node *mn, | |
937 | enum maple_type mt) | |
938 | { | |
54a611b6 LH |
939 | return mn->ma64.meta.gap; |
940 | } | |
941 | ||
942 | /* | |
943 | * ma_set_meta_gap() - Set the largest gap location in a nodes metadata | |
944 | * @mn: The maple node | |
945 | * @mn: The maple node type | |
946 | * @offset: The location of the largest gap. | |
947 | */ | |
948 | static inline void ma_set_meta_gap(struct maple_node *mn, enum maple_type mt, | |
949 | unsigned char offset) | |
950 | { | |
951 | ||
952 | struct maple_metadata *meta = ma_meta(mn, mt); | |
953 | ||
954 | meta->gap = offset; | |
955 | } | |
956 | ||
957 | /* | |
958 | * mat_add() - Add a @dead_enode to the ma_topiary of a list of dead nodes. | |
959 | * @mat - the ma_topiary, a linked list of dead nodes. | |
960 | * @dead_enode - the node to be marked as dead and added to the tail of the list | |
961 | * | |
962 | * Add the @dead_enode to the linked list in @mat. | |
963 | */ | |
964 | static inline void mat_add(struct ma_topiary *mat, | |
965 | struct maple_enode *dead_enode) | |
966 | { | |
967 | mte_set_node_dead(dead_enode); | |
968 | mte_to_mat(dead_enode)->next = NULL; | |
969 | if (!mat->tail) { | |
970 | mat->tail = mat->head = dead_enode; | |
971 | return; | |
972 | } | |
973 | ||
974 | mte_to_mat(mat->tail)->next = dead_enode; | |
975 | mat->tail = dead_enode; | |
976 | } | |
977 | ||
978 | static void mte_destroy_walk(struct maple_enode *, struct maple_tree *); | |
979 | static inline void mas_free(struct ma_state *mas, struct maple_enode *used); | |
980 | ||
981 | /* | |
982 | * mas_mat_free() - Free all nodes in a dead list. | |
983 | * @mas - the maple state | |
984 | * @mat - the ma_topiary linked list of dead nodes to free. | |
985 | * | |
986 | * Free walk a dead list. | |
987 | */ | |
988 | static void mas_mat_free(struct ma_state *mas, struct ma_topiary *mat) | |
989 | { | |
990 | struct maple_enode *next; | |
991 | ||
992 | while (mat->head) { | |
993 | next = mte_to_mat(mat->head)->next; | |
994 | mas_free(mas, mat->head); | |
995 | mat->head = next; | |
996 | } | |
997 | } | |
998 | ||
999 | /* | |
1000 | * mas_mat_destroy() - Free all nodes and subtrees in a dead list. | |
1001 | * @mas - the maple state | |
1002 | * @mat - the ma_topiary linked list of dead nodes to free. | |
1003 | * | |
1004 | * Destroy walk a dead list. | |
1005 | */ | |
1006 | static void mas_mat_destroy(struct ma_state *mas, struct ma_topiary *mat) | |
1007 | { | |
1008 | struct maple_enode *next; | |
1009 | ||
1010 | while (mat->head) { | |
1011 | next = mte_to_mat(mat->head)->next; | |
1012 | mte_destroy_walk(mat->head, mat->mtree); | |
1013 | mat->head = next; | |
1014 | } | |
1015 | } | |
1016 | /* | |
1017 | * mas_descend() - Descend into the slot stored in the ma_state. | |
1018 | * @mas - the maple state. | |
1019 | * | |
1020 | * Note: Not RCU safe, only use in write side or debug code. | |
1021 | */ | |
1022 | static inline void mas_descend(struct ma_state *mas) | |
1023 | { | |
1024 | enum maple_type type; | |
1025 | unsigned long *pivots; | |
1026 | struct maple_node *node; | |
1027 | void __rcu **slots; | |
1028 | ||
1029 | node = mas_mn(mas); | |
1030 | type = mte_node_type(mas->node); | |
1031 | pivots = ma_pivots(node, type); | |
1032 | slots = ma_slots(node, type); | |
1033 | ||
1034 | if (mas->offset) | |
1035 | mas->min = pivots[mas->offset - 1] + 1; | |
1036 | mas->max = mas_safe_pivot(mas, pivots, mas->offset, type); | |
1037 | mas->node = mas_slot(mas, slots, mas->offset); | |
1038 | } | |
1039 | ||
1040 | /* | |
1041 | * mte_set_gap() - Set a maple node gap. | |
1042 | * @mn: The encoded maple node | |
1043 | * @gap: The offset of the gap to set | |
1044 | * @val: The gap value | |
1045 | */ | |
1046 | static inline void mte_set_gap(const struct maple_enode *mn, | |
1047 | unsigned char gap, unsigned long val) | |
1048 | { | |
1049 | switch (mte_node_type(mn)) { | |
1050 | default: | |
1051 | break; | |
1052 | case maple_arange_64: | |
1053 | mte_to_node(mn)->ma64.gap[gap] = val; | |
1054 | break; | |
1055 | } | |
1056 | } | |
1057 | ||
1058 | /* | |
1059 | * mas_ascend() - Walk up a level of the tree. | |
1060 | * @mas: The maple state | |
1061 | * | |
1062 | * Sets the @mas->max and @mas->min to the correct values when walking up. This | |
1063 | * may cause several levels of walking up to find the correct min and max. | |
1064 | * May find a dead node which will cause a premature return. | |
1065 | * Return: 1 on dead node, 0 otherwise | |
1066 | */ | |
1067 | static int mas_ascend(struct ma_state *mas) | |
1068 | { | |
1069 | struct maple_enode *p_enode; /* parent enode. */ | |
1070 | struct maple_enode *a_enode; /* ancestor enode. */ | |
1071 | struct maple_node *a_node; /* ancestor node. */ | |
1072 | struct maple_node *p_node; /* parent node. */ | |
1073 | unsigned char a_slot; | |
1074 | enum maple_type a_type; | |
1075 | unsigned long min, max; | |
1076 | unsigned long *pivots; | |
54a611b6 LH |
1077 | bool set_max = false, set_min = false; |
1078 | ||
1079 | a_node = mas_mn(mas); | |
1080 | if (ma_is_root(a_node)) { | |
1081 | mas->offset = 0; | |
1082 | return 0; | |
1083 | } | |
1084 | ||
1085 | p_node = mte_parent(mas->node); | |
1086 | if (unlikely(a_node == p_node)) | |
1087 | return 1; | |
633769c9 | 1088 | |
afc754c6 | 1089 | a_type = mas_parent_type(mas, mas->node); |
633769c9 | 1090 | mas->offset = mte_parent_slot(mas->node); |
54a611b6 LH |
1091 | a_enode = mt_mk_node(p_node, a_type); |
1092 | ||
1093 | /* Check to make sure all parent information is still accurate */ | |
1094 | if (p_node != mte_parent(mas->node)) | |
1095 | return 1; | |
1096 | ||
1097 | mas->node = a_enode; | |
54a611b6 LH |
1098 | |
1099 | if (mte_is_root(a_enode)) { | |
1100 | mas->max = ULONG_MAX; | |
1101 | mas->min = 0; | |
1102 | return 0; | |
1103 | } | |
1104 | ||
633769c9 LH |
1105 | if (!mas->min) |
1106 | set_min = true; | |
1107 | ||
1108 | if (mas->max == ULONG_MAX) | |
1109 | set_max = true; | |
1110 | ||
54a611b6 LH |
1111 | min = 0; |
1112 | max = ULONG_MAX; | |
1113 | do { | |
1114 | p_enode = a_enode; | |
afc754c6 | 1115 | a_type = mas_parent_type(mas, p_enode); |
54a611b6 LH |
1116 | a_node = mte_parent(p_enode); |
1117 | a_slot = mte_parent_slot(p_enode); | |
54a611b6 | 1118 | a_enode = mt_mk_node(a_node, a_type); |
39d0bd86 LH |
1119 | pivots = ma_pivots(a_node, a_type); |
1120 | ||
1121 | if (unlikely(ma_dead_node(a_node))) | |
1122 | return 1; | |
54a611b6 LH |
1123 | |
1124 | if (!set_min && a_slot) { | |
1125 | set_min = true; | |
1126 | min = pivots[a_slot - 1] + 1; | |
1127 | } | |
1128 | ||
1129 | if (!set_max && a_slot < mt_pivots[a_type]) { | |
1130 | set_max = true; | |
1131 | max = pivots[a_slot]; | |
1132 | } | |
1133 | ||
1134 | if (unlikely(ma_dead_node(a_node))) | |
1135 | return 1; | |
1136 | ||
1137 | if (unlikely(ma_is_root(a_node))) | |
1138 | break; | |
1139 | ||
1140 | } while (!set_min || !set_max); | |
1141 | ||
1142 | mas->max = max; | |
1143 | mas->min = min; | |
1144 | return 0; | |
1145 | } | |
1146 | ||
1147 | /* | |
1148 | * mas_pop_node() - Get a previously allocated maple node from the maple state. | |
1149 | * @mas: The maple state | |
1150 | * | |
1151 | * Return: A pointer to a maple node. | |
1152 | */ | |
1153 | static inline struct maple_node *mas_pop_node(struct ma_state *mas) | |
1154 | { | |
1155 | struct maple_alloc *ret, *node = mas->alloc; | |
1156 | unsigned long total = mas_allocated(mas); | |
541e06b7 | 1157 | unsigned int req = mas_alloc_req(mas); |
54a611b6 LH |
1158 | |
1159 | /* nothing or a request pending. */ | |
541e06b7 | 1160 | if (WARN_ON(!total)) |
54a611b6 LH |
1161 | return NULL; |
1162 | ||
1163 | if (total == 1) { | |
1164 | /* single allocation in this ma_state */ | |
1165 | mas->alloc = NULL; | |
1166 | ret = node; | |
1167 | goto single_node; | |
1168 | } | |
1169 | ||
541e06b7 | 1170 | if (node->node_count == 1) { |
54a611b6 LH |
1171 | /* Single allocation in this node. */ |
1172 | mas->alloc = node->slot[0]; | |
54a611b6 LH |
1173 | mas->alloc->total = node->total - 1; |
1174 | ret = node; | |
1175 | goto new_head; | |
1176 | } | |
54a611b6 | 1177 | node->total--; |
541e06b7 LH |
1178 | ret = node->slot[--node->node_count]; |
1179 | node->slot[node->node_count] = NULL; | |
54a611b6 LH |
1180 | |
1181 | single_node: | |
1182 | new_head: | |
541e06b7 LH |
1183 | if (req) { |
1184 | req++; | |
1185 | mas_set_alloc_req(mas, req); | |
54a611b6 | 1186 | } |
541e06b7 LH |
1187 | |
1188 | memset(ret, 0, sizeof(*ret)); | |
54a611b6 LH |
1189 | return (struct maple_node *)ret; |
1190 | } | |
1191 | ||
1192 | /* | |
1193 | * mas_push_node() - Push a node back on the maple state allocation. | |
1194 | * @mas: The maple state | |
1195 | * @used: The used maple node | |
1196 | * | |
1197 | * Stores the maple node back into @mas->alloc for reuse. Updates allocated and | |
1198 | * requested node count as necessary. | |
1199 | */ | |
1200 | static inline void mas_push_node(struct ma_state *mas, struct maple_node *used) | |
1201 | { | |
1202 | struct maple_alloc *reuse = (struct maple_alloc *)used; | |
1203 | struct maple_alloc *head = mas->alloc; | |
1204 | unsigned long count; | |
1205 | unsigned int requested = mas_alloc_req(mas); | |
1206 | ||
54a611b6 LH |
1207 | count = mas_allocated(mas); |
1208 | ||
541e06b7 LH |
1209 | reuse->request_count = 0; |
1210 | reuse->node_count = 0; | |
1211 | if (count && (head->node_count < MAPLE_ALLOC_SLOTS)) { | |
1212 | head->slot[head->node_count++] = reuse; | |
54a611b6 LH |
1213 | head->total++; |
1214 | goto done; | |
1215 | } | |
1216 | ||
1217 | reuse->total = 1; | |
1218 | if ((head) && !((unsigned long)head & 0x1)) { | |
54a611b6 | 1219 | reuse->slot[0] = head; |
541e06b7 | 1220 | reuse->node_count = 1; |
54a611b6 LH |
1221 | reuse->total += head->total; |
1222 | } | |
1223 | ||
1224 | mas->alloc = reuse; | |
1225 | done: | |
1226 | if (requested > 1) | |
1227 | mas_set_alloc_req(mas, requested - 1); | |
1228 | } | |
1229 | ||
1230 | /* | |
1231 | * mas_alloc_nodes() - Allocate nodes into a maple state | |
1232 | * @mas: The maple state | |
1233 | * @gfp: The GFP Flags | |
1234 | */ | |
1235 | static inline void mas_alloc_nodes(struct ma_state *mas, gfp_t gfp) | |
1236 | { | |
1237 | struct maple_alloc *node; | |
54a611b6 | 1238 | unsigned long allocated = mas_allocated(mas); |
54a611b6 LH |
1239 | unsigned int requested = mas_alloc_req(mas); |
1240 | unsigned int count; | |
1241 | void **slots = NULL; | |
1242 | unsigned int max_req = 0; | |
1243 | ||
1244 | if (!requested) | |
1245 | return; | |
1246 | ||
1247 | mas_set_alloc_req(mas, 0); | |
1248 | if (mas->mas_flags & MA_STATE_PREALLOC) { | |
1249 | if (allocated) | |
1250 | return; | |
1251 | WARN_ON(!allocated); | |
1252 | } | |
1253 | ||
541e06b7 | 1254 | if (!allocated || mas->alloc->node_count == MAPLE_ALLOC_SLOTS) { |
54a611b6 LH |
1255 | node = (struct maple_alloc *)mt_alloc_one(gfp); |
1256 | if (!node) | |
1257 | goto nomem_one; | |
1258 | ||
541e06b7 | 1259 | if (allocated) { |
54a611b6 | 1260 | node->slot[0] = mas->alloc; |
541e06b7 LH |
1261 | node->node_count = 1; |
1262 | } else { | |
1263 | node->node_count = 0; | |
1264 | } | |
54a611b6 | 1265 | |
54a611b6 | 1266 | mas->alloc = node; |
541e06b7 | 1267 | node->total = ++allocated; |
54a611b6 LH |
1268 | requested--; |
1269 | } | |
1270 | ||
1271 | node = mas->alloc; | |
541e06b7 | 1272 | node->request_count = 0; |
54a611b6 | 1273 | while (requested) { |
1f5f12ec PZ |
1274 | max_req = MAPLE_ALLOC_SLOTS - node->node_count; |
1275 | slots = (void **)&node->slot[node->node_count]; | |
54a611b6 LH |
1276 | max_req = min(requested, max_req); |
1277 | count = mt_alloc_bulk(gfp, max_req, slots); | |
1278 | if (!count) | |
1279 | goto nomem_bulk; | |
1280 | ||
1f5f12ec PZ |
1281 | if (node->node_count == 0) { |
1282 | node->slot[0]->node_count = 0; | |
1283 | node->slot[0]->request_count = 0; | |
1284 | } | |
1285 | ||
54a611b6 | 1286 | node->node_count += count; |
541e06b7 | 1287 | allocated += count; |
c61b3a2b | 1288 | node = node->slot[0]; |
54a611b6 LH |
1289 | requested -= count; |
1290 | } | |
541e06b7 | 1291 | mas->alloc->total = allocated; |
54a611b6 LH |
1292 | return; |
1293 | ||
1294 | nomem_bulk: | |
1295 | /* Clean up potential freed allocations on bulk failure */ | |
1296 | memset(slots, 0, max_req * sizeof(unsigned long)); | |
1297 | nomem_one: | |
1298 | mas_set_alloc_req(mas, requested); | |
1299 | if (mas->alloc && !(((unsigned long)mas->alloc & 0x1))) | |
541e06b7 | 1300 | mas->alloc->total = allocated; |
54a611b6 | 1301 | mas_set_err(mas, -ENOMEM); |
54a611b6 LH |
1302 | } |
1303 | ||
1304 | /* | |
1305 | * mas_free() - Free an encoded maple node | |
1306 | * @mas: The maple state | |
1307 | * @used: The encoded maple node to free. | |
1308 | * | |
1309 | * Uses rcu free if necessary, pushes @used back on the maple state allocations | |
1310 | * otherwise. | |
1311 | */ | |
1312 | static inline void mas_free(struct ma_state *mas, struct maple_enode *used) | |
1313 | { | |
1314 | struct maple_node *tmp = mte_to_node(used); | |
1315 | ||
1316 | if (mt_in_rcu(mas->tree)) | |
1317 | ma_free_rcu(tmp); | |
1318 | else | |
1319 | mas_push_node(mas, tmp); | |
1320 | } | |
1321 | ||
1322 | /* | |
1323 | * mas_node_count() - Check if enough nodes are allocated and request more if | |
1324 | * there is not enough nodes. | |
1325 | * @mas: The maple state | |
1326 | * @count: The number of nodes needed | |
1327 | * @gfp: the gfp flags | |
1328 | */ | |
1329 | static void mas_node_count_gfp(struct ma_state *mas, int count, gfp_t gfp) | |
1330 | { | |
1331 | unsigned long allocated = mas_allocated(mas); | |
1332 | ||
1333 | if (allocated < count) { | |
1334 | mas_set_alloc_req(mas, count - allocated); | |
1335 | mas_alloc_nodes(mas, gfp); | |
1336 | } | |
1337 | } | |
1338 | ||
1339 | /* | |
1340 | * mas_node_count() - Check if enough nodes are allocated and request more if | |
1341 | * there is not enough nodes. | |
1342 | * @mas: The maple state | |
1343 | * @count: The number of nodes needed | |
1344 | * | |
1345 | * Note: Uses GFP_NOWAIT | __GFP_NOWARN for gfp flags. | |
1346 | */ | |
1347 | static void mas_node_count(struct ma_state *mas, int count) | |
1348 | { | |
1349 | return mas_node_count_gfp(mas, count, GFP_NOWAIT | __GFP_NOWARN); | |
1350 | } | |
1351 | ||
1352 | /* | |
1353 | * mas_start() - Sets up maple state for operations. | |
1354 | * @mas: The maple state. | |
1355 | * | |
46b34584 | 1356 | * If mas->node == MAS_START, then set the min, max and depth to |
54a611b6 LH |
1357 | * defaults. |
1358 | * | |
1359 | * Return: | |
1360 | * - If mas->node is an error or not MAS_START, return NULL. | |
1361 | * - If it's an empty tree: NULL & mas->node == MAS_NONE | |
1362 | * - If it's a single entry: The entry & mas->node == MAS_ROOT | |
1363 | * - If it's a tree: NULL & mas->node == safe root node. | |
1364 | */ | |
1365 | static inline struct maple_enode *mas_start(struct ma_state *mas) | |
1366 | { | |
1367 | if (likely(mas_is_start(mas))) { | |
1368 | struct maple_enode *root; | |
1369 | ||
54a611b6 LH |
1370 | mas->min = 0; |
1371 | mas->max = ULONG_MAX; | |
54a611b6 | 1372 | |
a7b92d59 | 1373 | retry: |
d0411860 | 1374 | mas->depth = 0; |
54a611b6 LH |
1375 | root = mas_root(mas); |
1376 | /* Tree with nodes */ | |
1377 | if (likely(xa_is_node(root))) { | |
9bbba563 | 1378 | mas->depth = 1; |
54a611b6 | 1379 | mas->node = mte_safe_root(root); |
46b34584 | 1380 | mas->offset = 0; |
a7b92d59 LH |
1381 | if (mte_dead_node(mas->node)) |
1382 | goto retry; | |
1383 | ||
54a611b6 LH |
1384 | return NULL; |
1385 | } | |
1386 | ||
1387 | /* empty tree */ | |
1388 | if (unlikely(!root)) { | |
46b34584 | 1389 | mas->node = MAS_NONE; |
54a611b6 LH |
1390 | mas->offset = MAPLE_NODE_SLOTS; |
1391 | return NULL; | |
1392 | } | |
1393 | ||
1394 | /* Single entry tree */ | |
1395 | mas->node = MAS_ROOT; | |
1396 | mas->offset = MAPLE_NODE_SLOTS; | |
1397 | ||
1398 | /* Single entry tree. */ | |
1399 | if (mas->index > 0) | |
1400 | return NULL; | |
1401 | ||
1402 | return root; | |
1403 | } | |
1404 | ||
1405 | return NULL; | |
1406 | } | |
1407 | ||
1408 | /* | |
1409 | * ma_data_end() - Find the end of the data in a node. | |
1410 | * @node: The maple node | |
1411 | * @type: The maple node type | |
1412 | * @pivots: The array of pivots in the node | |
1413 | * @max: The maximum value in the node | |
1414 | * | |
1415 | * Uses metadata to find the end of the data when possible. | |
1416 | * Return: The zero indexed last slot with data (may be null). | |
1417 | */ | |
1418 | static inline unsigned char ma_data_end(struct maple_node *node, | |
1419 | enum maple_type type, | |
1420 | unsigned long *pivots, | |
1421 | unsigned long max) | |
1422 | { | |
1423 | unsigned char offset; | |
1424 | ||
39d0bd86 LH |
1425 | if (!pivots) |
1426 | return 0; | |
1427 | ||
54a611b6 LH |
1428 | if (type == maple_arange_64) |
1429 | return ma_meta_end(node, type); | |
1430 | ||
1431 | offset = mt_pivots[type] - 1; | |
1432 | if (likely(!pivots[offset])) | |
1433 | return ma_meta_end(node, type); | |
1434 | ||
1435 | if (likely(pivots[offset] == max)) | |
1436 | return offset; | |
1437 | ||
1438 | return mt_pivots[type]; | |
1439 | } | |
1440 | ||
1441 | /* | |
1442 | * mas_data_end() - Find the end of the data (slot). | |
1443 | * @mas: the maple state | |
1444 | * | |
1445 | * This method is optimized to check the metadata of a node if the node type | |
1446 | * supports data end metadata. | |
1447 | * | |
1448 | * Return: The zero indexed last slot with data (may be null). | |
1449 | */ | |
1450 | static inline unsigned char mas_data_end(struct ma_state *mas) | |
1451 | { | |
1452 | enum maple_type type; | |
1453 | struct maple_node *node; | |
1454 | unsigned char offset; | |
1455 | unsigned long *pivots; | |
1456 | ||
1457 | type = mte_node_type(mas->node); | |
1458 | node = mas_mn(mas); | |
1459 | if (type == maple_arange_64) | |
1460 | return ma_meta_end(node, type); | |
1461 | ||
1462 | pivots = ma_pivots(node, type); | |
39d0bd86 LH |
1463 | if (unlikely(ma_dead_node(node))) |
1464 | return 0; | |
1465 | ||
54a611b6 LH |
1466 | offset = mt_pivots[type] - 1; |
1467 | if (likely(!pivots[offset])) | |
1468 | return ma_meta_end(node, type); | |
1469 | ||
1470 | if (likely(pivots[offset] == mas->max)) | |
1471 | return offset; | |
1472 | ||
1473 | return mt_pivots[type]; | |
1474 | } | |
1475 | ||
1476 | /* | |
1477 | * mas_leaf_max_gap() - Returns the largest gap in a leaf node | |
1478 | * @mas - the maple state | |
1479 | * | |
1480 | * Return: The maximum gap in the leaf. | |
1481 | */ | |
1482 | static unsigned long mas_leaf_max_gap(struct ma_state *mas) | |
1483 | { | |
1484 | enum maple_type mt; | |
1485 | unsigned long pstart, gap, max_gap; | |
1486 | struct maple_node *mn; | |
1487 | unsigned long *pivots; | |
1488 | void __rcu **slots; | |
1489 | unsigned char i; | |
1490 | unsigned char max_piv; | |
1491 | ||
1492 | mt = mte_node_type(mas->node); | |
1493 | mn = mas_mn(mas); | |
1494 | slots = ma_slots(mn, mt); | |
1495 | max_gap = 0; | |
1496 | if (unlikely(ma_is_dense(mt))) { | |
1497 | gap = 0; | |
1498 | for (i = 0; i < mt_slots[mt]; i++) { | |
1499 | if (slots[i]) { | |
1500 | if (gap > max_gap) | |
1501 | max_gap = gap; | |
1502 | gap = 0; | |
1503 | } else { | |
1504 | gap++; | |
1505 | } | |
1506 | } | |
1507 | if (gap > max_gap) | |
1508 | max_gap = gap; | |
1509 | return max_gap; | |
1510 | } | |
1511 | ||
1512 | /* | |
1513 | * Check the first implied pivot optimizes the loop below and slot 1 may | |
1514 | * be skipped if there is a gap in slot 0. | |
1515 | */ | |
1516 | pivots = ma_pivots(mn, mt); | |
1517 | if (likely(!slots[0])) { | |
1518 | max_gap = pivots[0] - mas->min + 1; | |
1519 | i = 2; | |
1520 | } else { | |
1521 | i = 1; | |
1522 | } | |
1523 | ||
1524 | /* reduce max_piv as the special case is checked before the loop */ | |
1525 | max_piv = ma_data_end(mn, mt, pivots, mas->max) - 1; | |
1526 | /* | |
1527 | * Check end implied pivot which can only be a gap on the right most | |
1528 | * node. | |
1529 | */ | |
1530 | if (unlikely(mas->max == ULONG_MAX) && !slots[max_piv + 1]) { | |
1531 | gap = ULONG_MAX - pivots[max_piv]; | |
1532 | if (gap > max_gap) | |
1533 | max_gap = gap; | |
1534 | } | |
1535 | ||
1536 | for (; i <= max_piv; i++) { | |
1537 | /* data == no gap. */ | |
1538 | if (likely(slots[i])) | |
1539 | continue; | |
1540 | ||
1541 | pstart = pivots[i - 1]; | |
1542 | gap = pivots[i] - pstart; | |
1543 | if (gap > max_gap) | |
1544 | max_gap = gap; | |
1545 | ||
1546 | /* There cannot be two gaps in a row. */ | |
1547 | i++; | |
1548 | } | |
1549 | return max_gap; | |
1550 | } | |
1551 | ||
1552 | /* | |
1553 | * ma_max_gap() - Get the maximum gap in a maple node (non-leaf) | |
1554 | * @node: The maple node | |
1555 | * @gaps: The pointer to the gaps | |
1556 | * @mt: The maple node type | |
1557 | * @*off: Pointer to store the offset location of the gap. | |
1558 | * | |
1559 | * Uses the metadata data end to scan backwards across set gaps. | |
1560 | * | |
1561 | * Return: The maximum gap value | |
1562 | */ | |
1563 | static inline unsigned long | |
1564 | ma_max_gap(struct maple_node *node, unsigned long *gaps, enum maple_type mt, | |
1565 | unsigned char *off) | |
1566 | { | |
1567 | unsigned char offset, i; | |
1568 | unsigned long max_gap = 0; | |
1569 | ||
1570 | i = offset = ma_meta_end(node, mt); | |
1571 | do { | |
1572 | if (gaps[i] > max_gap) { | |
1573 | max_gap = gaps[i]; | |
1574 | offset = i; | |
1575 | } | |
1576 | } while (i--); | |
1577 | ||
1578 | *off = offset; | |
1579 | return max_gap; | |
1580 | } | |
1581 | ||
1582 | /* | |
1583 | * mas_max_gap() - find the largest gap in a non-leaf node and set the slot. | |
1584 | * @mas: The maple state. | |
1585 | * | |
54a611b6 LH |
1586 | * Return: The gap value. |
1587 | */ | |
1588 | static inline unsigned long mas_max_gap(struct ma_state *mas) | |
1589 | { | |
1590 | unsigned long *gaps; | |
1591 | unsigned char offset; | |
1592 | enum maple_type mt; | |
1593 | struct maple_node *node; | |
1594 | ||
1595 | mt = mte_node_type(mas->node); | |
1596 | if (ma_is_leaf(mt)) | |
1597 | return mas_leaf_max_gap(mas); | |
1598 | ||
1599 | node = mas_mn(mas); | |
bec1b51e | 1600 | MAS_BUG_ON(mas, mt != maple_arange_64); |
54a611b6 | 1601 | offset = ma_meta_gap(node, mt); |
54a611b6 LH |
1602 | gaps = ma_gaps(node, mt); |
1603 | return gaps[offset]; | |
1604 | } | |
1605 | ||
1606 | /* | |
1607 | * mas_parent_gap() - Set the parent gap and any gaps above, as needed | |
1608 | * @mas: The maple state | |
1609 | * @offset: The gap offset in the parent to set | |
1610 | * @new: The new gap value. | |
1611 | * | |
1612 | * Set the parent gap then continue to set the gap upwards, using the metadata | |
1613 | * of the parent to see if it is necessary to check the node above. | |
1614 | */ | |
1615 | static inline void mas_parent_gap(struct ma_state *mas, unsigned char offset, | |
1616 | unsigned long new) | |
1617 | { | |
1618 | unsigned long meta_gap = 0; | |
1619 | struct maple_node *pnode; | |
1620 | struct maple_enode *penode; | |
1621 | unsigned long *pgaps; | |
1622 | unsigned char meta_offset; | |
1623 | enum maple_type pmt; | |
1624 | ||
1625 | pnode = mte_parent(mas->node); | |
afc754c6 | 1626 | pmt = mas_parent_type(mas, mas->node); |
54a611b6 LH |
1627 | penode = mt_mk_node(pnode, pmt); |
1628 | pgaps = ma_gaps(pnode, pmt); | |
1629 | ||
1630 | ascend: | |
bec1b51e | 1631 | MAS_BUG_ON(mas, pmt != maple_arange_64); |
54a611b6 | 1632 | meta_offset = ma_meta_gap(pnode, pmt); |
d695c30a | 1633 | meta_gap = pgaps[meta_offset]; |
54a611b6 LH |
1634 | |
1635 | pgaps[offset] = new; | |
1636 | ||
1637 | if (meta_gap == new) | |
1638 | return; | |
1639 | ||
1640 | if (offset != meta_offset) { | |
1641 | if (meta_gap > new) | |
1642 | return; | |
1643 | ||
1644 | ma_set_meta_gap(pnode, pmt, offset); | |
1645 | } else if (new < meta_gap) { | |
54a611b6 LH |
1646 | new = ma_max_gap(pnode, pgaps, pmt, &meta_offset); |
1647 | ma_set_meta_gap(pnode, pmt, meta_offset); | |
1648 | } | |
1649 | ||
1650 | if (ma_is_root(pnode)) | |
1651 | return; | |
1652 | ||
1653 | /* Go to the parent node. */ | |
1654 | pnode = mte_parent(penode); | |
afc754c6 | 1655 | pmt = mas_parent_type(mas, penode); |
54a611b6 LH |
1656 | pgaps = ma_gaps(pnode, pmt); |
1657 | offset = mte_parent_slot(penode); | |
1658 | penode = mt_mk_node(pnode, pmt); | |
1659 | goto ascend; | |
1660 | } | |
1661 | ||
1662 | /* | |
1663 | * mas_update_gap() - Update a nodes gaps and propagate up if necessary. | |
1664 | * @mas - the maple state. | |
1665 | */ | |
1666 | static inline void mas_update_gap(struct ma_state *mas) | |
1667 | { | |
1668 | unsigned char pslot; | |
1669 | unsigned long p_gap; | |
1670 | unsigned long max_gap; | |
1671 | ||
1672 | if (!mt_is_alloc(mas->tree)) | |
1673 | return; | |
1674 | ||
1675 | if (mte_is_root(mas->node)) | |
1676 | return; | |
1677 | ||
1678 | max_gap = mas_max_gap(mas); | |
1679 | ||
1680 | pslot = mte_parent_slot(mas->node); | |
1681 | p_gap = ma_gaps(mte_parent(mas->node), | |
afc754c6 | 1682 | mas_parent_type(mas, mas->node))[pslot]; |
54a611b6 LH |
1683 | |
1684 | if (p_gap != max_gap) | |
1685 | mas_parent_gap(mas, pslot, max_gap); | |
1686 | } | |
1687 | ||
1688 | /* | |
1689 | * mas_adopt_children() - Set the parent pointer of all nodes in @parent to | |
1690 | * @parent with the slot encoded. | |
1691 | * @mas - the maple state (for the tree) | |
1692 | * @parent - the maple encoded node containing the children. | |
1693 | */ | |
1694 | static inline void mas_adopt_children(struct ma_state *mas, | |
1695 | struct maple_enode *parent) | |
1696 | { | |
1697 | enum maple_type type = mte_node_type(parent); | |
1698 | struct maple_node *node = mas_mn(mas); | |
1699 | void __rcu **slots = ma_slots(node, type); | |
1700 | unsigned long *pivots = ma_pivots(node, type); | |
1701 | struct maple_enode *child; | |
1702 | unsigned char offset; | |
1703 | ||
1704 | offset = ma_data_end(node, type, pivots, mas->max); | |
1705 | do { | |
1706 | child = mas_slot_locked(mas, slots, offset); | |
bf96715e | 1707 | mas_set_parent(mas, child, parent, offset); |
54a611b6 LH |
1708 | } while (offset--); |
1709 | } | |
1710 | ||
1711 | /* | |
1712 | * mas_replace() - Replace a maple node in the tree with mas->node. Uses the | |
1713 | * parent encoding to locate the maple node in the tree. | |
1714 | * @mas - the ma_state to use for operations. | |
1715 | * @advanced - boolean to adopt the child nodes and free the old node (false) or | |
1716 | * leave the node (true) and handle the adoption and free elsewhere. | |
1717 | */ | |
1718 | static inline void mas_replace(struct ma_state *mas, bool advanced) | |
14c4b5ab | 1719 | __must_hold(mas->tree->ma_lock) |
54a611b6 LH |
1720 | { |
1721 | struct maple_node *mn = mas_mn(mas); | |
1722 | struct maple_enode *old_enode; | |
1723 | unsigned char offset = 0; | |
1724 | void __rcu **slots = NULL; | |
1725 | ||
1726 | if (ma_is_root(mn)) { | |
1727 | old_enode = mas_root_locked(mas); | |
1728 | } else { | |
1729 | offset = mte_parent_slot(mas->node); | |
1730 | slots = ma_slots(mte_parent(mas->node), | |
afc754c6 | 1731 | mas_parent_type(mas, mas->node)); |
54a611b6 LH |
1732 | old_enode = mas_slot_locked(mas, slots, offset); |
1733 | } | |
1734 | ||
1735 | if (!advanced && !mte_is_leaf(mas->node)) | |
1736 | mas_adopt_children(mas, mas->node); | |
1737 | ||
1738 | if (mte_is_root(mas->node)) { | |
1739 | mn->parent = ma_parent_ptr( | |
1740 | ((unsigned long)mas->tree | MA_ROOT_PARENT)); | |
1741 | rcu_assign_pointer(mas->tree->ma_root, mte_mk_root(mas->node)); | |
1742 | mas_set_height(mas); | |
1743 | } else { | |
1744 | rcu_assign_pointer(slots[offset], mas->node); | |
1745 | } | |
1746 | ||
c13af03d LH |
1747 | if (!advanced) { |
1748 | mte_set_node_dead(old_enode); | |
54a611b6 | 1749 | mas_free(mas, old_enode); |
c13af03d | 1750 | } |
54a611b6 LH |
1751 | } |
1752 | ||
1753 | /* | |
1754 | * mas_new_child() - Find the new child of a node. | |
1755 | * @mas: the maple state | |
1756 | * @child: the maple state to store the child. | |
1757 | */ | |
1758 | static inline bool mas_new_child(struct ma_state *mas, struct ma_state *child) | |
14c4b5ab | 1759 | __must_hold(mas->tree->ma_lock) |
54a611b6 LH |
1760 | { |
1761 | enum maple_type mt; | |
1762 | unsigned char offset; | |
1763 | unsigned char end; | |
1764 | unsigned long *pivots; | |
1765 | struct maple_enode *entry; | |
1766 | struct maple_node *node; | |
1767 | void __rcu **slots; | |
1768 | ||
1769 | mt = mte_node_type(mas->node); | |
1770 | node = mas_mn(mas); | |
1771 | slots = ma_slots(node, mt); | |
1772 | pivots = ma_pivots(node, mt); | |
1773 | end = ma_data_end(node, mt, pivots, mas->max); | |
1774 | for (offset = mas->offset; offset <= end; offset++) { | |
1775 | entry = mas_slot_locked(mas, slots, offset); | |
1776 | if (mte_parent(entry) == node) { | |
1777 | *child = *mas; | |
1778 | mas->offset = offset + 1; | |
1779 | child->offset = offset; | |
1780 | mas_descend(child); | |
1781 | child->offset = 0; | |
1782 | return true; | |
1783 | } | |
1784 | } | |
1785 | return false; | |
1786 | } | |
1787 | ||
1788 | /* | |
1789 | * mab_shift_right() - Shift the data in mab right. Note, does not clean out the | |
1790 | * old data or set b_node->b_end. | |
1791 | * @b_node: the maple_big_node | |
1792 | * @shift: the shift count | |
1793 | */ | |
1794 | static inline void mab_shift_right(struct maple_big_node *b_node, | |
1795 | unsigned char shift) | |
1796 | { | |
1797 | unsigned long size = b_node->b_end * sizeof(unsigned long); | |
1798 | ||
1799 | memmove(b_node->pivot + shift, b_node->pivot, size); | |
1800 | memmove(b_node->slot + shift, b_node->slot, size); | |
1801 | if (b_node->type == maple_arange_64) | |
1802 | memmove(b_node->gap + shift, b_node->gap, size); | |
1803 | } | |
1804 | ||
1805 | /* | |
1806 | * mab_middle_node() - Check if a middle node is needed (unlikely) | |
1807 | * @b_node: the maple_big_node that contains the data. | |
1808 | * @size: the amount of data in the b_node | |
1809 | * @split: the potential split location | |
1810 | * @slot_count: the size that can be stored in a single node being considered. | |
1811 | * | |
1812 | * Return: true if a middle node is required. | |
1813 | */ | |
1814 | static inline bool mab_middle_node(struct maple_big_node *b_node, int split, | |
1815 | unsigned char slot_count) | |
1816 | { | |
1817 | unsigned char size = b_node->b_end; | |
1818 | ||
1819 | if (size >= 2 * slot_count) | |
1820 | return true; | |
1821 | ||
1822 | if (!b_node->slot[split] && (size >= 2 * slot_count - 1)) | |
1823 | return true; | |
1824 | ||
1825 | return false; | |
1826 | } | |
1827 | ||
1828 | /* | |
1829 | * mab_no_null_split() - ensure the split doesn't fall on a NULL | |
1830 | * @b_node: the maple_big_node with the data | |
1831 | * @split: the suggested split location | |
1832 | * @slot_count: the number of slots in the node being considered. | |
1833 | * | |
1834 | * Return: the split location. | |
1835 | */ | |
1836 | static inline int mab_no_null_split(struct maple_big_node *b_node, | |
1837 | unsigned char split, unsigned char slot_count) | |
1838 | { | |
1839 | if (!b_node->slot[split]) { | |
1840 | /* | |
1841 | * If the split is less than the max slot && the right side will | |
1842 | * still be sufficient, then increment the split on NULL. | |
1843 | */ | |
1844 | if ((split < slot_count - 1) && | |
1845 | (b_node->b_end - split) > (mt_min_slots[b_node->type])) | |
1846 | split++; | |
1847 | else | |
1848 | split--; | |
1849 | } | |
1850 | return split; | |
1851 | } | |
1852 | ||
1853 | /* | |
1854 | * mab_calc_split() - Calculate the split location and if there needs to be two | |
1855 | * splits. | |
1856 | * @bn: The maple_big_node with the data | |
1857 | * @mid_split: The second split, if required. 0 otherwise. | |
1858 | * | |
1859 | * Return: The first split location. The middle split is set in @mid_split. | |
1860 | */ | |
1861 | static inline int mab_calc_split(struct ma_state *mas, | |
1862 | struct maple_big_node *bn, unsigned char *mid_split, unsigned long min) | |
1863 | { | |
1864 | unsigned char b_end = bn->b_end; | |
1865 | int split = b_end / 2; /* Assume equal split. */ | |
1866 | unsigned char slot_min, slot_count = mt_slots[bn->type]; | |
1867 | ||
1868 | /* | |
1869 | * To support gap tracking, all NULL entries are kept together and a node cannot | |
1870 | * end on a NULL entry, with the exception of the left-most leaf. The | |
1871 | * limitation means that the split of a node must be checked for this condition | |
1872 | * and be able to put more data in one direction or the other. | |
1873 | */ | |
1874 | if (unlikely((mas->mas_flags & MA_STATE_BULK))) { | |
1875 | *mid_split = 0; | |
1876 | split = b_end - mt_min_slots[bn->type]; | |
1877 | ||
1878 | if (!ma_is_leaf(bn->type)) | |
1879 | return split; | |
1880 | ||
1881 | mas->mas_flags |= MA_STATE_REBALANCE; | |
1882 | if (!bn->slot[split]) | |
1883 | split--; | |
1884 | return split; | |
1885 | } | |
1886 | ||
1887 | /* | |
1888 | * Although extremely rare, it is possible to enter what is known as the 3-way | |
1889 | * split scenario. The 3-way split comes about by means of a store of a range | |
1890 | * that overwrites the end and beginning of two full nodes. The result is a set | |
1891 | * of entries that cannot be stored in 2 nodes. Sometimes, these two nodes can | |
1892 | * also be located in different parent nodes which are also full. This can | |
1893 | * carry upwards all the way to the root in the worst case. | |
1894 | */ | |
1895 | if (unlikely(mab_middle_node(bn, split, slot_count))) { | |
1896 | split = b_end / 3; | |
1897 | *mid_split = split * 2; | |
1898 | } else { | |
1899 | slot_min = mt_min_slots[bn->type]; | |
1900 | ||
1901 | *mid_split = 0; | |
1902 | /* | |
1903 | * Avoid having a range less than the slot count unless it | |
1904 | * causes one node to be deficient. | |
1905 | * NOTE: mt_min_slots is 1 based, b_end and split are zero. | |
1906 | */ | |
5729e06c LH |
1907 | while ((split < slot_count - 1) && |
1908 | ((bn->pivot[split] - min) < slot_count - 1) && | |
1909 | (b_end - split > slot_min)) | |
54a611b6 LH |
1910 | split++; |
1911 | } | |
1912 | ||
1913 | /* Avoid ending a node on a NULL entry */ | |
1914 | split = mab_no_null_split(bn, split, slot_count); | |
54a611b6 | 1915 | |
e11cb683 VY |
1916 | if (unlikely(*mid_split)) |
1917 | *mid_split = mab_no_null_split(bn, *mid_split, slot_count); | |
54a611b6 LH |
1918 | |
1919 | return split; | |
1920 | } | |
1921 | ||
1922 | /* | |
1923 | * mas_mab_cp() - Copy data from a maple state inclusively to a maple_big_node | |
1924 | * and set @b_node->b_end to the next free slot. | |
1925 | * @mas: The maple state | |
1926 | * @mas_start: The starting slot to copy | |
1927 | * @mas_end: The end slot to copy (inclusively) | |
1928 | * @b_node: The maple_big_node to place the data | |
1929 | * @mab_start: The starting location in maple_big_node to store the data. | |
1930 | */ | |
1931 | static inline void mas_mab_cp(struct ma_state *mas, unsigned char mas_start, | |
1932 | unsigned char mas_end, struct maple_big_node *b_node, | |
1933 | unsigned char mab_start) | |
1934 | { | |
1935 | enum maple_type mt; | |
1936 | struct maple_node *node; | |
1937 | void __rcu **slots; | |
1938 | unsigned long *pivots, *gaps; | |
1939 | int i = mas_start, j = mab_start; | |
1940 | unsigned char piv_end; | |
1941 | ||
1942 | node = mas_mn(mas); | |
1943 | mt = mte_node_type(mas->node); | |
1944 | pivots = ma_pivots(node, mt); | |
1945 | if (!i) { | |
1946 | b_node->pivot[j] = pivots[i++]; | |
1947 | if (unlikely(i > mas_end)) | |
1948 | goto complete; | |
1949 | j++; | |
1950 | } | |
1951 | ||
1952 | piv_end = min(mas_end, mt_pivots[mt]); | |
1953 | for (; i < piv_end; i++, j++) { | |
1954 | b_node->pivot[j] = pivots[i]; | |
1955 | if (unlikely(!b_node->pivot[j])) | |
1956 | break; | |
1957 | ||
1958 | if (unlikely(mas->max == b_node->pivot[j])) | |
1959 | goto complete; | |
1960 | } | |
1961 | ||
1962 | if (likely(i <= mas_end)) | |
1963 | b_node->pivot[j] = mas_safe_pivot(mas, pivots, i, mt); | |
1964 | ||
1965 | complete: | |
1966 | b_node->b_end = ++j; | |
1967 | j -= mab_start; | |
1968 | slots = ma_slots(node, mt); | |
1969 | memcpy(b_node->slot + mab_start, slots + mas_start, sizeof(void *) * j); | |
1970 | if (!ma_is_leaf(mt) && mt_is_alloc(mas->tree)) { | |
1971 | gaps = ma_gaps(node, mt); | |
1972 | memcpy(b_node->gap + mab_start, gaps + mas_start, | |
1973 | sizeof(unsigned long) * j); | |
1974 | } | |
1975 | } | |
1976 | ||
1977 | /* | |
1978 | * mas_leaf_set_meta() - Set the metadata of a leaf if possible. | |
1979 | * @mas: The maple state | |
1980 | * @node: The maple node | |
1981 | * @pivots: pointer to the maple node pivots | |
1982 | * @mt: The maple type | |
1983 | * @end: The assumed end | |
1984 | * | |
1985 | * Note, end may be incremented within this function but not modified at the | |
1986 | * source. This is fine since the metadata is the last thing to be stored in a | |
1987 | * node during a write. | |
1988 | */ | |
1989 | static inline void mas_leaf_set_meta(struct ma_state *mas, | |
1990 | struct maple_node *node, unsigned long *pivots, | |
1991 | enum maple_type mt, unsigned char end) | |
1992 | { | |
1993 | /* There is no room for metadata already */ | |
1994 | if (mt_pivots[mt] <= end) | |
1995 | return; | |
1996 | ||
1997 | if (pivots[end] && pivots[end] < mas->max) | |
1998 | end++; | |
1999 | ||
2000 | if (end < mt_slots[mt] - 1) | |
2001 | ma_set_meta(node, mt, 0, end); | |
2002 | } | |
2003 | ||
2004 | /* | |
2005 | * mab_mas_cp() - Copy data from maple_big_node to a maple encoded node. | |
2006 | * @b_node: the maple_big_node that has the data | |
2007 | * @mab_start: the start location in @b_node. | |
2008 | * @mab_end: The end location in @b_node (inclusively) | |
2009 | * @mas: The maple state with the maple encoded node. | |
2010 | */ | |
2011 | static inline void mab_mas_cp(struct maple_big_node *b_node, | |
2012 | unsigned char mab_start, unsigned char mab_end, | |
2013 | struct ma_state *mas, bool new_max) | |
2014 | { | |
2015 | int i, j = 0; | |
2016 | enum maple_type mt = mte_node_type(mas->node); | |
2017 | struct maple_node *node = mte_to_node(mas->node); | |
2018 | void __rcu **slots = ma_slots(node, mt); | |
2019 | unsigned long *pivots = ma_pivots(node, mt); | |
2020 | unsigned long *gaps = NULL; | |
2021 | unsigned char end; | |
2022 | ||
2023 | if (mab_end - mab_start > mt_pivots[mt]) | |
2024 | mab_end--; | |
2025 | ||
2026 | if (!pivots[mt_pivots[mt] - 1]) | |
2027 | slots[mt_pivots[mt]] = NULL; | |
2028 | ||
2029 | i = mab_start; | |
2030 | do { | |
2031 | pivots[j++] = b_node->pivot[i++]; | |
2032 | } while (i <= mab_end && likely(b_node->pivot[i])); | |
2033 | ||
2034 | memcpy(slots, b_node->slot + mab_start, | |
2035 | sizeof(void *) * (i - mab_start)); | |
2036 | ||
2037 | if (new_max) | |
2038 | mas->max = b_node->pivot[i - 1]; | |
2039 | ||
2040 | end = j - 1; | |
2041 | if (likely(!ma_is_leaf(mt) && mt_is_alloc(mas->tree))) { | |
2042 | unsigned long max_gap = 0; | |
d695c30a | 2043 | unsigned char offset = 0; |
54a611b6 LH |
2044 | |
2045 | gaps = ma_gaps(node, mt); | |
2046 | do { | |
2047 | gaps[--j] = b_node->gap[--i]; | |
2048 | if (gaps[j] > max_gap) { | |
2049 | offset = j; | |
2050 | max_gap = gaps[j]; | |
2051 | } | |
2052 | } while (j); | |
2053 | ||
2054 | ma_set_meta(node, mt, offset, end); | |
2055 | } else { | |
2056 | mas_leaf_set_meta(mas, node, pivots, mt, end); | |
2057 | } | |
2058 | } | |
2059 | ||
2060 | /* | |
2061 | * mas_descend_adopt() - Descend through a sub-tree and adopt children. | |
2062 | * @mas: the maple state with the maple encoded node of the sub-tree. | |
2063 | * | |
2064 | * Descend through a sub-tree and adopt children who do not have the correct | |
2065 | * parents set. Follow the parents which have the correct parents as they are | |
2066 | * the new entries which need to be followed to find other incorrectly set | |
2067 | * parents. | |
2068 | */ | |
2069 | static inline void mas_descend_adopt(struct ma_state *mas) | |
2070 | { | |
2071 | struct ma_state list[3], next[3]; | |
2072 | int i, n; | |
2073 | ||
2074 | /* | |
2075 | * At each level there may be up to 3 correct parent pointers which indicates | |
2076 | * the new nodes which need to be walked to find any new nodes at a lower level. | |
2077 | */ | |
2078 | ||
2079 | for (i = 0; i < 3; i++) { | |
2080 | list[i] = *mas; | |
2081 | list[i].offset = 0; | |
2082 | next[i].offset = 0; | |
2083 | } | |
2084 | next[0] = *mas; | |
2085 | ||
2086 | while (!mte_is_leaf(list[0].node)) { | |
2087 | n = 0; | |
2088 | for (i = 0; i < 3; i++) { | |
2089 | if (mas_is_none(&list[i])) | |
2090 | continue; | |
2091 | ||
2092 | if (i && list[i-1].node == list[i].node) | |
2093 | continue; | |
2094 | ||
2095 | while ((n < 3) && (mas_new_child(&list[i], &next[n]))) | |
2096 | n++; | |
2097 | ||
2098 | mas_adopt_children(&list[i], list[i].node); | |
2099 | } | |
2100 | ||
2101 | while (n < 3) | |
2102 | next[n++].node = MAS_NONE; | |
2103 | ||
2104 | /* descend by setting the list to the children */ | |
2105 | for (i = 0; i < 3; i++) | |
2106 | list[i] = next[i]; | |
2107 | } | |
2108 | } | |
2109 | ||
2110 | /* | |
2111 | * mas_bulk_rebalance() - Rebalance the end of a tree after a bulk insert. | |
2112 | * @mas: The maple state | |
2113 | * @end: The maple node end | |
2114 | * @mt: The maple node type | |
2115 | */ | |
2116 | static inline void mas_bulk_rebalance(struct ma_state *mas, unsigned char end, | |
2117 | enum maple_type mt) | |
2118 | { | |
2119 | if (!(mas->mas_flags & MA_STATE_BULK)) | |
2120 | return; | |
2121 | ||
2122 | if (mte_is_root(mas->node)) | |
2123 | return; | |
2124 | ||
2125 | if (end > mt_min_slots[mt]) { | |
2126 | mas->mas_flags &= ~MA_STATE_REBALANCE; | |
2127 | return; | |
2128 | } | |
2129 | } | |
2130 | ||
2131 | /* | |
2132 | * mas_store_b_node() - Store an @entry into the b_node while also copying the | |
2133 | * data from a maple encoded node. | |
2134 | * @wr_mas: the maple write state | |
2135 | * @b_node: the maple_big_node to fill with data | |
2136 | * @offset_end: the offset to end copying | |
2137 | * | |
2138 | * Return: The actual end of the data stored in @b_node | |
2139 | */ | |
44081c77 | 2140 | static noinline_for_kasan void mas_store_b_node(struct ma_wr_state *wr_mas, |
54a611b6 LH |
2141 | struct maple_big_node *b_node, unsigned char offset_end) |
2142 | { | |
2143 | unsigned char slot; | |
2144 | unsigned char b_end; | |
2145 | /* Possible underflow of piv will wrap back to 0 before use. */ | |
2146 | unsigned long piv; | |
2147 | struct ma_state *mas = wr_mas->mas; | |
2148 | ||
2149 | b_node->type = wr_mas->type; | |
2150 | b_end = 0; | |
2151 | slot = mas->offset; | |
2152 | if (slot) { | |
2153 | /* Copy start data up to insert. */ | |
2154 | mas_mab_cp(mas, 0, slot - 1, b_node, 0); | |
2155 | b_end = b_node->b_end; | |
2156 | piv = b_node->pivot[b_end - 1]; | |
2157 | } else | |
2158 | piv = mas->min - 1; | |
2159 | ||
2160 | if (piv + 1 < mas->index) { | |
2161 | /* Handle range starting after old range */ | |
2162 | b_node->slot[b_end] = wr_mas->content; | |
2163 | if (!wr_mas->content) | |
2164 | b_node->gap[b_end] = mas->index - 1 - piv; | |
2165 | b_node->pivot[b_end++] = mas->index - 1; | |
2166 | } | |
2167 | ||
2168 | /* Store the new entry. */ | |
2169 | mas->offset = b_end; | |
2170 | b_node->slot[b_end] = wr_mas->entry; | |
2171 | b_node->pivot[b_end] = mas->last; | |
2172 | ||
2173 | /* Appended. */ | |
2174 | if (mas->last >= mas->max) | |
2175 | goto b_end; | |
2176 | ||
2177 | /* Handle new range ending before old range ends */ | |
29b2681f | 2178 | piv = mas_safe_pivot(mas, wr_mas->pivots, offset_end, wr_mas->type); |
54a611b6 LH |
2179 | if (piv > mas->last) { |
2180 | if (piv == ULONG_MAX) | |
2181 | mas_bulk_rebalance(mas, b_node->b_end, wr_mas->type); | |
2182 | ||
2183 | if (offset_end != slot) | |
2184 | wr_mas->content = mas_slot_locked(mas, wr_mas->slots, | |
2185 | offset_end); | |
2186 | ||
2187 | b_node->slot[++b_end] = wr_mas->content; | |
2188 | if (!wr_mas->content) | |
2189 | b_node->gap[b_end] = piv - mas->last + 1; | |
2190 | b_node->pivot[b_end] = piv; | |
2191 | } | |
2192 | ||
2193 | slot = offset_end + 1; | |
2194 | if (slot > wr_mas->node_end) | |
2195 | goto b_end; | |
2196 | ||
2197 | /* Copy end data to the end of the node. */ | |
2198 | mas_mab_cp(mas, slot, wr_mas->node_end + 1, b_node, ++b_end); | |
2199 | b_node->b_end--; | |
2200 | return; | |
2201 | ||
2202 | b_end: | |
2203 | b_node->b_end = b_end; | |
2204 | } | |
2205 | ||
2206 | /* | |
2207 | * mas_prev_sibling() - Find the previous node with the same parent. | |
2208 | * @mas: the maple state | |
2209 | * | |
2210 | * Return: True if there is a previous sibling, false otherwise. | |
2211 | */ | |
2212 | static inline bool mas_prev_sibling(struct ma_state *mas) | |
2213 | { | |
2214 | unsigned int p_slot = mte_parent_slot(mas->node); | |
2215 | ||
2216 | if (mte_is_root(mas->node)) | |
2217 | return false; | |
2218 | ||
2219 | if (!p_slot) | |
2220 | return false; | |
2221 | ||
2222 | mas_ascend(mas); | |
2223 | mas->offset = p_slot - 1; | |
2224 | mas_descend(mas); | |
2225 | return true; | |
2226 | } | |
2227 | ||
2228 | /* | |
2229 | * mas_next_sibling() - Find the next node with the same parent. | |
2230 | * @mas: the maple state | |
2231 | * | |
2232 | * Return: true if there is a next sibling, false otherwise. | |
2233 | */ | |
2234 | static inline bool mas_next_sibling(struct ma_state *mas) | |
2235 | { | |
2236 | MA_STATE(parent, mas->tree, mas->index, mas->last); | |
2237 | ||
2238 | if (mte_is_root(mas->node)) | |
2239 | return false; | |
2240 | ||
2241 | parent = *mas; | |
2242 | mas_ascend(&parent); | |
2243 | parent.offset = mte_parent_slot(mas->node) + 1; | |
2244 | if (parent.offset > mas_data_end(&parent)) | |
2245 | return false; | |
2246 | ||
2247 | *mas = parent; | |
2248 | mas_descend(mas); | |
2249 | return true; | |
2250 | } | |
2251 | ||
2252 | /* | |
2253 | * mte_node_or_node() - Return the encoded node or MAS_NONE. | |
2254 | * @enode: The encoded maple node. | |
2255 | * | |
2256 | * Shorthand to avoid setting %NULLs in the tree or maple_subtree_state. | |
2257 | * | |
2258 | * Return: @enode or MAS_NONE | |
2259 | */ | |
2260 | static inline struct maple_enode *mte_node_or_none(struct maple_enode *enode) | |
2261 | { | |
2262 | if (enode) | |
2263 | return enode; | |
2264 | ||
2265 | return ma_enode_ptr(MAS_NONE); | |
2266 | } | |
2267 | ||
2268 | /* | |
2269 | * mas_wr_node_walk() - Find the correct offset for the index in the @mas. | |
2270 | * @wr_mas: The maple write state | |
2271 | * | |
2272 | * Uses mas_slot_locked() and does not need to worry about dead nodes. | |
2273 | */ | |
2274 | static inline void mas_wr_node_walk(struct ma_wr_state *wr_mas) | |
2275 | { | |
2276 | struct ma_state *mas = wr_mas->mas; | |
97f7e094 | 2277 | unsigned char count, offset; |
54a611b6 LH |
2278 | |
2279 | if (unlikely(ma_is_dense(wr_mas->type))) { | |
2280 | wr_mas->r_max = wr_mas->r_min = mas->index; | |
2281 | mas->offset = mas->index = mas->min; | |
2282 | return; | |
2283 | } | |
2284 | ||
2285 | wr_mas->node = mas_mn(wr_mas->mas); | |
2286 | wr_mas->pivots = ma_pivots(wr_mas->node, wr_mas->type); | |
2287 | count = wr_mas->node_end = ma_data_end(wr_mas->node, wr_mas->type, | |
2288 | wr_mas->pivots, mas->max); | |
2289 | offset = mas->offset; | |
54a611b6 | 2290 | |
97f7e094 PZ |
2291 | while (offset < count && mas->index > wr_mas->pivots[offset]) |
2292 | offset++; | |
54a611b6 | 2293 | |
97f7e094 PZ |
2294 | wr_mas->r_max = offset < count ? wr_mas->pivots[offset] : mas->max; |
2295 | wr_mas->r_min = mas_safe_min(mas, wr_mas->pivots, offset); | |
54a611b6 LH |
2296 | wr_mas->offset_end = mas->offset = offset; |
2297 | } | |
2298 | ||
2299 | /* | |
2300 | * mas_topiary_range() - Add a range of slots to the topiary. | |
2301 | * @mas: The maple state | |
2302 | * @destroy: The topiary to add the slots (usually destroy) | |
2303 | * @start: The starting slot inclusively | |
2304 | * @end: The end slot inclusively | |
2305 | */ | |
2306 | static inline void mas_topiary_range(struct ma_state *mas, | |
2307 | struct ma_topiary *destroy, unsigned char start, unsigned char end) | |
2308 | { | |
2309 | void __rcu **slots; | |
2310 | unsigned char offset; | |
2311 | ||
4bbd1748 LH |
2312 | MAS_BUG_ON(mas, mte_is_leaf(mas->node)); |
2313 | ||
54a611b6 LH |
2314 | slots = ma_slots(mas_mn(mas), mte_node_type(mas->node)); |
2315 | for (offset = start; offset <= end; offset++) { | |
2316 | struct maple_enode *enode = mas_slot_locked(mas, slots, offset); | |
2317 | ||
2318 | if (mte_dead_node(enode)) | |
2319 | continue; | |
2320 | ||
2321 | mat_add(destroy, enode); | |
2322 | } | |
2323 | } | |
2324 | ||
2325 | /* | |
2326 | * mast_topiary() - Add the portions of the tree to the removal list; either to | |
2327 | * be freed or discarded (destroy walk). | |
2328 | * @mast: The maple_subtree_state. | |
2329 | */ | |
2330 | static inline void mast_topiary(struct maple_subtree_state *mast) | |
2331 | { | |
2332 | MA_WR_STATE(wr_mas, mast->orig_l, NULL); | |
2333 | unsigned char r_start, r_end; | |
2334 | unsigned char l_start, l_end; | |
2335 | void __rcu **l_slots, **r_slots; | |
2336 | ||
2337 | wr_mas.type = mte_node_type(mast->orig_l->node); | |
2338 | mast->orig_l->index = mast->orig_l->last; | |
2339 | mas_wr_node_walk(&wr_mas); | |
2340 | l_start = mast->orig_l->offset + 1; | |
2341 | l_end = mas_data_end(mast->orig_l); | |
2342 | r_start = 0; | |
2343 | r_end = mast->orig_r->offset; | |
2344 | ||
2345 | if (r_end) | |
2346 | r_end--; | |
2347 | ||
2348 | l_slots = ma_slots(mas_mn(mast->orig_l), | |
2349 | mte_node_type(mast->orig_l->node)); | |
2350 | ||
2351 | r_slots = ma_slots(mas_mn(mast->orig_r), | |
2352 | mte_node_type(mast->orig_r->node)); | |
2353 | ||
2354 | if ((l_start < l_end) && | |
2355 | mte_dead_node(mas_slot_locked(mast->orig_l, l_slots, l_start))) { | |
2356 | l_start++; | |
2357 | } | |
2358 | ||
2359 | if (mte_dead_node(mas_slot_locked(mast->orig_r, r_slots, r_end))) { | |
2360 | if (r_end) | |
2361 | r_end--; | |
2362 | } | |
2363 | ||
2364 | if ((l_start > r_end) && (mast->orig_l->node == mast->orig_r->node)) | |
2365 | return; | |
2366 | ||
2367 | /* At the node where left and right sides meet, add the parts between */ | |
2368 | if (mast->orig_l->node == mast->orig_r->node) { | |
2369 | return mas_topiary_range(mast->orig_l, mast->destroy, | |
2370 | l_start, r_end); | |
2371 | } | |
2372 | ||
2373 | /* mast->orig_r is different and consumed. */ | |
2374 | if (mte_is_leaf(mast->orig_r->node)) | |
2375 | return; | |
2376 | ||
2377 | if (mte_dead_node(mas_slot_locked(mast->orig_l, l_slots, l_end))) | |
2378 | l_end--; | |
2379 | ||
2380 | ||
2381 | if (l_start <= l_end) | |
2382 | mas_topiary_range(mast->orig_l, mast->destroy, l_start, l_end); | |
2383 | ||
2384 | if (mte_dead_node(mas_slot_locked(mast->orig_r, r_slots, r_start))) | |
2385 | r_start++; | |
2386 | ||
2387 | if (r_start <= r_end) | |
2388 | mas_topiary_range(mast->orig_r, mast->destroy, 0, r_end); | |
2389 | } | |
2390 | ||
2391 | /* | |
2392 | * mast_rebalance_next() - Rebalance against the next node | |
2393 | * @mast: The maple subtree state | |
2394 | * @old_r: The encoded maple node to the right (next node). | |
2395 | */ | |
2396 | static inline void mast_rebalance_next(struct maple_subtree_state *mast) | |
2397 | { | |
2398 | unsigned char b_end = mast->bn->b_end; | |
2399 | ||
2400 | mas_mab_cp(mast->orig_r, 0, mt_slot_count(mast->orig_r->node), | |
2401 | mast->bn, b_end); | |
2402 | mast->orig_r->last = mast->orig_r->max; | |
2403 | } | |
2404 | ||
2405 | /* | |
2406 | * mast_rebalance_prev() - Rebalance against the previous node | |
2407 | * @mast: The maple subtree state | |
2408 | * @old_l: The encoded maple node to the left (previous node) | |
2409 | */ | |
2410 | static inline void mast_rebalance_prev(struct maple_subtree_state *mast) | |
2411 | { | |
2412 | unsigned char end = mas_data_end(mast->orig_l) + 1; | |
2413 | unsigned char b_end = mast->bn->b_end; | |
2414 | ||
2415 | mab_shift_right(mast->bn, end); | |
2416 | mas_mab_cp(mast->orig_l, 0, end - 1, mast->bn, 0); | |
2417 | mast->l->min = mast->orig_l->min; | |
2418 | mast->orig_l->index = mast->orig_l->min; | |
2419 | mast->bn->b_end = end + b_end; | |
2420 | mast->l->offset += end; | |
2421 | } | |
2422 | ||
2423 | /* | |
2424 | * mast_spanning_rebalance() - Rebalance nodes with nearest neighbour favouring | |
2425 | * the node to the right. Checking the nodes to the right then the left at each | |
2426 | * level upwards until root is reached. Free and destroy as needed. | |
2427 | * Data is copied into the @mast->bn. | |
2428 | * @mast: The maple_subtree_state. | |
2429 | */ | |
2430 | static inline | |
2431 | bool mast_spanning_rebalance(struct maple_subtree_state *mast) | |
2432 | { | |
2433 | struct ma_state r_tmp = *mast->orig_r; | |
2434 | struct ma_state l_tmp = *mast->orig_l; | |
2435 | struct maple_enode *ancestor = NULL; | |
2436 | unsigned char start, end; | |
2437 | unsigned char depth = 0; | |
2438 | ||
2439 | r_tmp = *mast->orig_r; | |
2440 | l_tmp = *mast->orig_l; | |
2441 | do { | |
2442 | mas_ascend(mast->orig_r); | |
2443 | mas_ascend(mast->orig_l); | |
2444 | depth++; | |
2445 | if (!ancestor && | |
2446 | (mast->orig_r->node == mast->orig_l->node)) { | |
2447 | ancestor = mast->orig_r->node; | |
2448 | end = mast->orig_r->offset - 1; | |
2449 | start = mast->orig_l->offset + 1; | |
2450 | } | |
2451 | ||
2452 | if (mast->orig_r->offset < mas_data_end(mast->orig_r)) { | |
2453 | if (!ancestor) { | |
2454 | ancestor = mast->orig_r->node; | |
2455 | start = 0; | |
2456 | } | |
2457 | ||
2458 | mast->orig_r->offset++; | |
2459 | do { | |
2460 | mas_descend(mast->orig_r); | |
2461 | mast->orig_r->offset = 0; | |
2462 | depth--; | |
2463 | } while (depth); | |
2464 | ||
2465 | mast_rebalance_next(mast); | |
2466 | do { | |
2467 | unsigned char l_off = 0; | |
2468 | struct maple_enode *child = r_tmp.node; | |
2469 | ||
2470 | mas_ascend(&r_tmp); | |
2471 | if (ancestor == r_tmp.node) | |
2472 | l_off = start; | |
2473 | ||
2474 | if (r_tmp.offset) | |
2475 | r_tmp.offset--; | |
2476 | ||
2477 | if (l_off < r_tmp.offset) | |
2478 | mas_topiary_range(&r_tmp, mast->destroy, | |
2479 | l_off, r_tmp.offset); | |
2480 | ||
2481 | if (l_tmp.node != child) | |
2482 | mat_add(mast->free, child); | |
2483 | ||
2484 | } while (r_tmp.node != ancestor); | |
2485 | ||
2486 | *mast->orig_l = l_tmp; | |
2487 | return true; | |
2488 | ||
2489 | } else if (mast->orig_l->offset != 0) { | |
2490 | if (!ancestor) { | |
2491 | ancestor = mast->orig_l->node; | |
2492 | end = mas_data_end(mast->orig_l); | |
2493 | } | |
2494 | ||
2495 | mast->orig_l->offset--; | |
2496 | do { | |
2497 | mas_descend(mast->orig_l); | |
2498 | mast->orig_l->offset = | |
2499 | mas_data_end(mast->orig_l); | |
2500 | depth--; | |
2501 | } while (depth); | |
2502 | ||
2503 | mast_rebalance_prev(mast); | |
2504 | do { | |
2505 | unsigned char r_off; | |
2506 | struct maple_enode *child = l_tmp.node; | |
2507 | ||
2508 | mas_ascend(&l_tmp); | |
2509 | if (ancestor == l_tmp.node) | |
2510 | r_off = end; | |
2511 | else | |
2512 | r_off = mas_data_end(&l_tmp); | |
2513 | ||
2514 | if (l_tmp.offset < r_off) | |
2515 | l_tmp.offset++; | |
2516 | ||
2517 | if (l_tmp.offset < r_off) | |
2518 | mas_topiary_range(&l_tmp, mast->destroy, | |
2519 | l_tmp.offset, r_off); | |
2520 | ||
2521 | if (r_tmp.node != child) | |
2522 | mat_add(mast->free, child); | |
2523 | ||
2524 | } while (l_tmp.node != ancestor); | |
2525 | ||
2526 | *mast->orig_r = r_tmp; | |
2527 | return true; | |
2528 | } | |
2529 | } while (!mte_is_root(mast->orig_r->node)); | |
2530 | ||
2531 | *mast->orig_r = r_tmp; | |
2532 | *mast->orig_l = l_tmp; | |
2533 | return false; | |
2534 | } | |
2535 | ||
2536 | /* | |
2537 | * mast_ascend_free() - Add current original maple state nodes to the free list | |
2538 | * and ascend. | |
2539 | * @mast: the maple subtree state. | |
2540 | * | |
2541 | * Ascend the original left and right sides and add the previous nodes to the | |
2542 | * free list. Set the slots to point to the correct location in the new nodes. | |
2543 | */ | |
2544 | static inline void | |
2545 | mast_ascend_free(struct maple_subtree_state *mast) | |
2546 | { | |
2547 | MA_WR_STATE(wr_mas, mast->orig_r, NULL); | |
2548 | struct maple_enode *left = mast->orig_l->node; | |
2549 | struct maple_enode *right = mast->orig_r->node; | |
2550 | ||
2551 | mas_ascend(mast->orig_l); | |
2552 | mas_ascend(mast->orig_r); | |
2553 | mat_add(mast->free, left); | |
2554 | ||
2555 | if (left != right) | |
2556 | mat_add(mast->free, right); | |
2557 | ||
2558 | mast->orig_r->offset = 0; | |
2559 | mast->orig_r->index = mast->r->max; | |
2560 | /* last should be larger than or equal to index */ | |
2561 | if (mast->orig_r->last < mast->orig_r->index) | |
2562 | mast->orig_r->last = mast->orig_r->index; | |
2563 | /* | |
2564 | * The node may not contain the value so set slot to ensure all | |
2565 | * of the nodes contents are freed or destroyed. | |
2566 | */ | |
2567 | wr_mas.type = mte_node_type(mast->orig_r->node); | |
2568 | mas_wr_node_walk(&wr_mas); | |
2569 | /* Set up the left side of things */ | |
2570 | mast->orig_l->offset = 0; | |
2571 | mast->orig_l->index = mast->l->min; | |
2572 | wr_mas.mas = mast->orig_l; | |
2573 | wr_mas.type = mte_node_type(mast->orig_l->node); | |
2574 | mas_wr_node_walk(&wr_mas); | |
2575 | ||
2576 | mast->bn->type = wr_mas.type; | |
2577 | } | |
2578 | ||
2579 | /* | |
2580 | * mas_new_ma_node() - Create and return a new maple node. Helper function. | |
2581 | * @mas: the maple state with the allocations. | |
2582 | * @b_node: the maple_big_node with the type encoding. | |
2583 | * | |
2584 | * Use the node type from the maple_big_node to allocate a new node from the | |
2585 | * ma_state. This function exists mainly for code readability. | |
2586 | * | |
2587 | * Return: A new maple encoded node | |
2588 | */ | |
2589 | static inline struct maple_enode | |
2590 | *mas_new_ma_node(struct ma_state *mas, struct maple_big_node *b_node) | |
2591 | { | |
2592 | return mt_mk_node(ma_mnode_ptr(mas_pop_node(mas)), b_node->type); | |
2593 | } | |
2594 | ||
2595 | /* | |
2596 | * mas_mab_to_node() - Set up right and middle nodes | |
2597 | * | |
2598 | * @mas: the maple state that contains the allocations. | |
2599 | * @b_node: the node which contains the data. | |
2600 | * @left: The pointer which will have the left node | |
2601 | * @right: The pointer which may have the right node | |
2602 | * @middle: the pointer which may have the middle node (rare) | |
2603 | * @mid_split: the split location for the middle node | |
2604 | * | |
2605 | * Return: the split of left. | |
2606 | */ | |
2607 | static inline unsigned char mas_mab_to_node(struct ma_state *mas, | |
2608 | struct maple_big_node *b_node, struct maple_enode **left, | |
2609 | struct maple_enode **right, struct maple_enode **middle, | |
2610 | unsigned char *mid_split, unsigned long min) | |
2611 | { | |
2612 | unsigned char split = 0; | |
2613 | unsigned char slot_count = mt_slots[b_node->type]; | |
2614 | ||
2615 | *left = mas_new_ma_node(mas, b_node); | |
2616 | *right = NULL; | |
2617 | *middle = NULL; | |
2618 | *mid_split = 0; | |
2619 | ||
2620 | if (b_node->b_end < slot_count) { | |
2621 | split = b_node->b_end; | |
2622 | } else { | |
2623 | split = mab_calc_split(mas, b_node, mid_split, min); | |
2624 | *right = mas_new_ma_node(mas, b_node); | |
2625 | } | |
2626 | ||
2627 | if (*mid_split) | |
2628 | *middle = mas_new_ma_node(mas, b_node); | |
2629 | ||
2630 | return split; | |
2631 | ||
2632 | } | |
2633 | ||
2634 | /* | |
2635 | * mab_set_b_end() - Add entry to b_node at b_node->b_end and increment the end | |
2636 | * pointer. | |
2637 | * @b_node - the big node to add the entry | |
2638 | * @mas - the maple state to get the pivot (mas->max) | |
2639 | * @entry - the entry to add, if NULL nothing happens. | |
2640 | */ | |
2641 | static inline void mab_set_b_end(struct maple_big_node *b_node, | |
2642 | struct ma_state *mas, | |
2643 | void *entry) | |
2644 | { | |
2645 | if (!entry) | |
2646 | return; | |
2647 | ||
2648 | b_node->slot[b_node->b_end] = entry; | |
2649 | if (mt_is_alloc(mas->tree)) | |
2650 | b_node->gap[b_node->b_end] = mas_max_gap(mas); | |
2651 | b_node->pivot[b_node->b_end++] = mas->max; | |
2652 | } | |
2653 | ||
2654 | /* | |
2655 | * mas_set_split_parent() - combine_then_separate helper function. Sets the parent | |
2656 | * of @mas->node to either @left or @right, depending on @slot and @split | |
2657 | * | |
2658 | * @mas - the maple state with the node that needs a parent | |
2659 | * @left - possible parent 1 | |
2660 | * @right - possible parent 2 | |
2661 | * @slot - the slot the mas->node was placed | |
2662 | * @split - the split location between @left and @right | |
2663 | */ | |
2664 | static inline void mas_set_split_parent(struct ma_state *mas, | |
2665 | struct maple_enode *left, | |
2666 | struct maple_enode *right, | |
2667 | unsigned char *slot, unsigned char split) | |
2668 | { | |
2669 | if (mas_is_none(mas)) | |
2670 | return; | |
2671 | ||
2672 | if ((*slot) <= split) | |
bf96715e | 2673 | mas_set_parent(mas, mas->node, left, *slot); |
54a611b6 | 2674 | else if (right) |
bf96715e | 2675 | mas_set_parent(mas, mas->node, right, (*slot) - split - 1); |
54a611b6 LH |
2676 | |
2677 | (*slot)++; | |
2678 | } | |
2679 | ||
2680 | /* | |
2681 | * mte_mid_split_check() - Check if the next node passes the mid-split | |
2682 | * @**l: Pointer to left encoded maple node. | |
2683 | * @**m: Pointer to middle encoded maple node. | |
2684 | * @**r: Pointer to right encoded maple node. | |
2685 | * @slot: The offset | |
2686 | * @*split: The split location. | |
2687 | * @mid_split: The middle split. | |
2688 | */ | |
2689 | static inline void mte_mid_split_check(struct maple_enode **l, | |
2690 | struct maple_enode **r, | |
2691 | struct maple_enode *right, | |
2692 | unsigned char slot, | |
2693 | unsigned char *split, | |
2694 | unsigned char mid_split) | |
2695 | { | |
2696 | if (*r == right) | |
2697 | return; | |
2698 | ||
2699 | if (slot < mid_split) | |
2700 | return; | |
2701 | ||
2702 | *l = *r; | |
2703 | *r = right; | |
2704 | *split = mid_split; | |
2705 | } | |
2706 | ||
2707 | /* | |
2708 | * mast_set_split_parents() - Helper function to set three nodes parents. Slot | |
2709 | * is taken from @mast->l. | |
2710 | * @mast - the maple subtree state | |
2711 | * @left - the left node | |
2712 | * @right - the right node | |
2713 | * @split - the split location. | |
2714 | */ | |
2715 | static inline void mast_set_split_parents(struct maple_subtree_state *mast, | |
2716 | struct maple_enode *left, | |
2717 | struct maple_enode *middle, | |
2718 | struct maple_enode *right, | |
2719 | unsigned char split, | |
2720 | unsigned char mid_split) | |
2721 | { | |
2722 | unsigned char slot; | |
2723 | struct maple_enode *l = left; | |
2724 | struct maple_enode *r = right; | |
2725 | ||
2726 | if (mas_is_none(mast->l)) | |
2727 | return; | |
2728 | ||
2729 | if (middle) | |
2730 | r = middle; | |
2731 | ||
2732 | slot = mast->l->offset; | |
2733 | ||
2734 | mte_mid_split_check(&l, &r, right, slot, &split, mid_split); | |
2735 | mas_set_split_parent(mast->l, l, r, &slot, split); | |
2736 | ||
2737 | mte_mid_split_check(&l, &r, right, slot, &split, mid_split); | |
2738 | mas_set_split_parent(mast->m, l, r, &slot, split); | |
2739 | ||
2740 | mte_mid_split_check(&l, &r, right, slot, &split, mid_split); | |
2741 | mas_set_split_parent(mast->r, l, r, &slot, split); | |
2742 | } | |
2743 | ||
2744 | /* | |
2745 | * mas_wmb_replace() - Write memory barrier and replace | |
2746 | * @mas: The maple state | |
2747 | * @free: the maple topiary list of nodes to free | |
2748 | * @destroy: The maple topiary list of nodes to destroy (walk and free) | |
2749 | * | |
2750 | * Updates gap as necessary. | |
2751 | */ | |
2752 | static inline void mas_wmb_replace(struct ma_state *mas, | |
2753 | struct ma_topiary *free, | |
2754 | struct ma_topiary *destroy) | |
2755 | { | |
2756 | /* All nodes must see old data as dead prior to replacing that data */ | |
2757 | smp_wmb(); /* Needed for RCU */ | |
2758 | ||
2759 | /* Insert the new data in the tree */ | |
2760 | mas_replace(mas, true); | |
2761 | ||
2762 | if (!mte_is_leaf(mas->node)) | |
2763 | mas_descend_adopt(mas); | |
2764 | ||
2765 | mas_mat_free(mas, free); | |
2766 | ||
2767 | if (destroy) | |
2768 | mas_mat_destroy(mas, destroy); | |
2769 | ||
2770 | if (mte_is_leaf(mas->node)) | |
2771 | return; | |
2772 | ||
2773 | mas_update_gap(mas); | |
2774 | } | |
2775 | ||
2776 | /* | |
2777 | * mast_new_root() - Set a new tree root during subtree creation | |
2778 | * @mast: The maple subtree state | |
2779 | * @mas: The maple state | |
2780 | */ | |
2781 | static inline void mast_new_root(struct maple_subtree_state *mast, | |
2782 | struct ma_state *mas) | |
2783 | { | |
2784 | mas_mn(mast->l)->parent = | |
2785 | ma_parent_ptr(((unsigned long)mas->tree | MA_ROOT_PARENT)); | |
2786 | if (!mte_dead_node(mast->orig_l->node) && | |
2787 | !mte_is_root(mast->orig_l->node)) { | |
2788 | do { | |
2789 | mast_ascend_free(mast); | |
2790 | mast_topiary(mast); | |
2791 | } while (!mte_is_root(mast->orig_l->node)); | |
2792 | } | |
2793 | if ((mast->orig_l->node != mas->node) && | |
2794 | (mast->l->depth > mas_mt_height(mas))) { | |
2795 | mat_add(mast->free, mas->node); | |
2796 | } | |
2797 | } | |
2798 | ||
2799 | /* | |
2800 | * mast_cp_to_nodes() - Copy data out to nodes. | |
2801 | * @mast: The maple subtree state | |
2802 | * @left: The left encoded maple node | |
2803 | * @middle: The middle encoded maple node | |
2804 | * @right: The right encoded maple node | |
2805 | * @split: The location to split between left and (middle ? middle : right) | |
2806 | * @mid_split: The location to split between middle and right. | |
2807 | */ | |
2808 | static inline void mast_cp_to_nodes(struct maple_subtree_state *mast, | |
2809 | struct maple_enode *left, struct maple_enode *middle, | |
2810 | struct maple_enode *right, unsigned char split, unsigned char mid_split) | |
2811 | { | |
2812 | bool new_lmax = true; | |
2813 | ||
2814 | mast->l->node = mte_node_or_none(left); | |
2815 | mast->m->node = mte_node_or_none(middle); | |
2816 | mast->r->node = mte_node_or_none(right); | |
2817 | ||
2818 | mast->l->min = mast->orig_l->min; | |
2819 | if (split == mast->bn->b_end) { | |
2820 | mast->l->max = mast->orig_r->max; | |
2821 | new_lmax = false; | |
2822 | } | |
2823 | ||
2824 | mab_mas_cp(mast->bn, 0, split, mast->l, new_lmax); | |
2825 | ||
2826 | if (middle) { | |
2827 | mab_mas_cp(mast->bn, 1 + split, mid_split, mast->m, true); | |
2828 | mast->m->min = mast->bn->pivot[split] + 1; | |
2829 | split = mid_split; | |
2830 | } | |
2831 | ||
2832 | mast->r->max = mast->orig_r->max; | |
2833 | if (right) { | |
2834 | mab_mas_cp(mast->bn, 1 + split, mast->bn->b_end, mast->r, false); | |
2835 | mast->r->min = mast->bn->pivot[split] + 1; | |
2836 | } | |
2837 | } | |
2838 | ||
2839 | /* | |
2840 | * mast_combine_cp_left - Copy in the original left side of the tree into the | |
2841 | * combined data set in the maple subtree state big node. | |
2842 | * @mast: The maple subtree state | |
2843 | */ | |
2844 | static inline void mast_combine_cp_left(struct maple_subtree_state *mast) | |
2845 | { | |
2846 | unsigned char l_slot = mast->orig_l->offset; | |
2847 | ||
2848 | if (!l_slot) | |
2849 | return; | |
2850 | ||
2851 | mas_mab_cp(mast->orig_l, 0, l_slot - 1, mast->bn, 0); | |
2852 | } | |
2853 | ||
2854 | /* | |
2855 | * mast_combine_cp_right: Copy in the original right side of the tree into the | |
2856 | * combined data set in the maple subtree state big node. | |
2857 | * @mast: The maple subtree state | |
2858 | */ | |
2859 | static inline void mast_combine_cp_right(struct maple_subtree_state *mast) | |
2860 | { | |
2861 | if (mast->bn->pivot[mast->bn->b_end - 1] >= mast->orig_r->max) | |
2862 | return; | |
2863 | ||
2864 | mas_mab_cp(mast->orig_r, mast->orig_r->offset + 1, | |
2865 | mt_slot_count(mast->orig_r->node), mast->bn, | |
2866 | mast->bn->b_end); | |
2867 | mast->orig_r->last = mast->orig_r->max; | |
2868 | } | |
2869 | ||
2870 | /* | |
2871 | * mast_sufficient: Check if the maple subtree state has enough data in the big | |
2872 | * node to create at least one sufficient node | |
2873 | * @mast: the maple subtree state | |
2874 | */ | |
2875 | static inline bool mast_sufficient(struct maple_subtree_state *mast) | |
2876 | { | |
2877 | if (mast->bn->b_end > mt_min_slot_count(mast->orig_l->node)) | |
2878 | return true; | |
2879 | ||
2880 | return false; | |
2881 | } | |
2882 | ||
2883 | /* | |
2884 | * mast_overflow: Check if there is too much data in the subtree state for a | |
2885 | * single node. | |
2886 | * @mast: The maple subtree state | |
2887 | */ | |
2888 | static inline bool mast_overflow(struct maple_subtree_state *mast) | |
2889 | { | |
2890 | if (mast->bn->b_end >= mt_slot_count(mast->orig_l->node)) | |
2891 | return true; | |
2892 | ||
2893 | return false; | |
2894 | } | |
2895 | ||
2896 | static inline void *mtree_range_walk(struct ma_state *mas) | |
2897 | { | |
2898 | unsigned long *pivots; | |
2899 | unsigned char offset; | |
2900 | struct maple_node *node; | |
2901 | struct maple_enode *next, *last; | |
2902 | enum maple_type type; | |
2903 | void __rcu **slots; | |
2904 | unsigned char end; | |
2905 | unsigned long max, min; | |
2906 | unsigned long prev_max, prev_min; | |
2907 | ||
1b9c9183 LB |
2908 | next = mas->node; |
2909 | min = mas->min; | |
54a611b6 LH |
2910 | max = mas->max; |
2911 | do { | |
2912 | offset = 0; | |
2913 | last = next; | |
2914 | node = mte_to_node(next); | |
2915 | type = mte_node_type(next); | |
2916 | pivots = ma_pivots(node, type); | |
2917 | end = ma_data_end(node, type, pivots, max); | |
2918 | if (unlikely(ma_dead_node(node))) | |
2919 | goto dead_node; | |
2920 | ||
2921 | if (pivots[offset] >= mas->index) { | |
2922 | prev_max = max; | |
2923 | prev_min = min; | |
2924 | max = pivots[offset]; | |
2925 | goto next; | |
2926 | } | |
2927 | ||
2928 | do { | |
2929 | offset++; | |
2930 | } while ((offset < end) && (pivots[offset] < mas->index)); | |
2931 | ||
2932 | prev_min = min; | |
2933 | min = pivots[offset - 1] + 1; | |
2934 | prev_max = max; | |
2935 | if (likely(offset < end && pivots[offset])) | |
2936 | max = pivots[offset]; | |
2937 | ||
2938 | next: | |
2939 | slots = ma_slots(node, type); | |
2940 | next = mt_slot(mas->tree, slots, offset); | |
2941 | if (unlikely(ma_dead_node(node))) | |
2942 | goto dead_node; | |
2943 | } while (!ma_is_leaf(type)); | |
2944 | ||
2945 | mas->offset = offset; | |
2946 | mas->index = min; | |
2947 | mas->last = max; | |
2948 | mas->min = prev_min; | |
2949 | mas->max = prev_max; | |
2950 | mas->node = last; | |
831978e3 | 2951 | return (void *)next; |
54a611b6 LH |
2952 | |
2953 | dead_node: | |
2954 | mas_reset(mas); | |
2955 | return NULL; | |
2956 | } | |
2957 | ||
2958 | /* | |
2959 | * mas_spanning_rebalance() - Rebalance across two nodes which may not be peers. | |
2960 | * @mas: The starting maple state | |
2961 | * @mast: The maple_subtree_state, keeps track of 4 maple states. | |
2962 | * @count: The estimated count of iterations needed. | |
2963 | * | |
2964 | * Follow the tree upwards from @l_mas and @r_mas for @count, or until the root | |
2965 | * is hit. First @b_node is split into two entries which are inserted into the | |
2966 | * next iteration of the loop. @b_node is returned populated with the final | |
2967 | * iteration. @mas is used to obtain allocations. orig_l_mas keeps track of the | |
2968 | * nodes that will remain active by using orig_l_mas->index and orig_l_mas->last | |
2969 | * to account of what has been copied into the new sub-tree. The update of | |
2970 | * orig_l_mas->last is used in mas_consume to find the slots that will need to | |
2971 | * be either freed or destroyed. orig_l_mas->depth keeps track of the height of | |
2972 | * the new sub-tree in case the sub-tree becomes the full tree. | |
2973 | * | |
2974 | * Return: the number of elements in b_node during the last loop. | |
2975 | */ | |
2976 | static int mas_spanning_rebalance(struct ma_state *mas, | |
2977 | struct maple_subtree_state *mast, unsigned char count) | |
2978 | { | |
2979 | unsigned char split, mid_split; | |
2980 | unsigned char slot = 0; | |
2981 | struct maple_enode *left = NULL, *middle = NULL, *right = NULL; | |
2982 | ||
2983 | MA_STATE(l_mas, mas->tree, mas->index, mas->index); | |
2984 | MA_STATE(r_mas, mas->tree, mas->index, mas->last); | |
2985 | MA_STATE(m_mas, mas->tree, mas->index, mas->index); | |
2986 | MA_TOPIARY(free, mas->tree); | |
2987 | MA_TOPIARY(destroy, mas->tree); | |
2988 | ||
2989 | /* | |
2990 | * The tree needs to be rebalanced and leaves need to be kept at the same level. | |
2991 | * Rebalancing is done by use of the ``struct maple_topiary``. | |
2992 | */ | |
2993 | mast->l = &l_mas; | |
2994 | mast->m = &m_mas; | |
2995 | mast->r = &r_mas; | |
2996 | mast->free = &free; | |
2997 | mast->destroy = &destroy; | |
2998 | l_mas.node = r_mas.node = m_mas.node = MAS_NONE; | |
0abb964a LH |
2999 | |
3000 | /* Check if this is not root and has sufficient data. */ | |
3001 | if (((mast->orig_l->min != 0) || (mast->orig_r->max != ULONG_MAX)) && | |
54a611b6 LH |
3002 | unlikely(mast->bn->b_end <= mt_min_slots[mast->bn->type])) |
3003 | mast_spanning_rebalance(mast); | |
3004 | ||
3005 | mast->orig_l->depth = 0; | |
3006 | ||
3007 | /* | |
3008 | * Each level of the tree is examined and balanced, pushing data to the left or | |
3009 | * right, or rebalancing against left or right nodes is employed to avoid | |
3010 | * rippling up the tree to limit the amount of churn. Once a new sub-section of | |
3011 | * the tree is created, there may be a mix of new and old nodes. The old nodes | |
3012 | * will have the incorrect parent pointers and currently be in two trees: the | |
3013 | * original tree and the partially new tree. To remedy the parent pointers in | |
3014 | * the old tree, the new data is swapped into the active tree and a walk down | |
3015 | * the tree is performed and the parent pointers are updated. | |
3016 | * See mas_descend_adopt() for more information.. | |
3017 | */ | |
3018 | while (count--) { | |
3019 | mast->bn->b_end--; | |
3020 | mast->bn->type = mte_node_type(mast->orig_l->node); | |
3021 | split = mas_mab_to_node(mas, mast->bn, &left, &right, &middle, | |
3022 | &mid_split, mast->orig_l->min); | |
3023 | mast_set_split_parents(mast, left, middle, right, split, | |
3024 | mid_split); | |
3025 | mast_cp_to_nodes(mast, left, middle, right, split, mid_split); | |
3026 | ||
3027 | /* | |
3028 | * Copy data from next level in the tree to mast->bn from next | |
3029 | * iteration | |
3030 | */ | |
3031 | memset(mast->bn, 0, sizeof(struct maple_big_node)); | |
3032 | mast->bn->type = mte_node_type(left); | |
3033 | mast->orig_l->depth++; | |
3034 | ||
3035 | /* Root already stored in l->node. */ | |
3036 | if (mas_is_root_limits(mast->l)) | |
3037 | goto new_root; | |
3038 | ||
3039 | mast_ascend_free(mast); | |
3040 | mast_combine_cp_left(mast); | |
3041 | l_mas.offset = mast->bn->b_end; | |
3042 | mab_set_b_end(mast->bn, &l_mas, left); | |
3043 | mab_set_b_end(mast->bn, &m_mas, middle); | |
3044 | mab_set_b_end(mast->bn, &r_mas, right); | |
3045 | ||
3046 | /* Copy anything necessary out of the right node. */ | |
3047 | mast_combine_cp_right(mast); | |
3048 | mast_topiary(mast); | |
3049 | mast->orig_l->last = mast->orig_l->max; | |
3050 | ||
3051 | if (mast_sufficient(mast)) | |
3052 | continue; | |
3053 | ||
3054 | if (mast_overflow(mast)) | |
3055 | continue; | |
3056 | ||
3057 | /* May be a new root stored in mast->bn */ | |
3058 | if (mas_is_root_limits(mast->orig_l)) | |
3059 | break; | |
3060 | ||
3061 | mast_spanning_rebalance(mast); | |
3062 | ||
3063 | /* rebalancing from other nodes may require another loop. */ | |
3064 | if (!count) | |
3065 | count++; | |
3066 | } | |
3067 | ||
3068 | l_mas.node = mt_mk_node(ma_mnode_ptr(mas_pop_node(mas)), | |
3069 | mte_node_type(mast->orig_l->node)); | |
3070 | mast->orig_l->depth++; | |
3071 | mab_mas_cp(mast->bn, 0, mt_slots[mast->bn->type] - 1, &l_mas, true); | |
bf96715e | 3072 | mas_set_parent(mas, left, l_mas.node, slot); |
54a611b6 | 3073 | if (middle) |
bf96715e | 3074 | mas_set_parent(mas, middle, l_mas.node, ++slot); |
54a611b6 LH |
3075 | |
3076 | if (right) | |
bf96715e | 3077 | mas_set_parent(mas, right, l_mas.node, ++slot); |
54a611b6 LH |
3078 | |
3079 | if (mas_is_root_limits(mast->l)) { | |
3080 | new_root: | |
3081 | mast_new_root(mast, mas); | |
3082 | } else { | |
3083 | mas_mn(&l_mas)->parent = mas_mn(mast->orig_l)->parent; | |
3084 | } | |
3085 | ||
3086 | if (!mte_dead_node(mast->orig_l->node)) | |
3087 | mat_add(&free, mast->orig_l->node); | |
3088 | ||
3089 | mas->depth = mast->orig_l->depth; | |
3090 | *mast->orig_l = l_mas; | |
3091 | mte_set_node_dead(mas->node); | |
3092 | ||
3093 | /* Set up mas for insertion. */ | |
3094 | mast->orig_l->depth = mas->depth; | |
3095 | mast->orig_l->alloc = mas->alloc; | |
3096 | *mas = *mast->orig_l; | |
3097 | mas_wmb_replace(mas, &free, &destroy); | |
3098 | mtree_range_walk(mas); | |
3099 | return mast->bn->b_end; | |
3100 | } | |
3101 | ||
3102 | /* | |
3103 | * mas_rebalance() - Rebalance a given node. | |
3104 | * @mas: The maple state | |
3105 | * @b_node: The big maple node. | |
3106 | * | |
3107 | * Rebalance two nodes into a single node or two new nodes that are sufficient. | |
3108 | * Continue upwards until tree is sufficient. | |
3109 | * | |
3110 | * Return: the number of elements in b_node during the last loop. | |
3111 | */ | |
3112 | static inline int mas_rebalance(struct ma_state *mas, | |
3113 | struct maple_big_node *b_node) | |
3114 | { | |
3115 | char empty_count = mas_mt_height(mas); | |
3116 | struct maple_subtree_state mast; | |
3117 | unsigned char shift, b_end = ++b_node->b_end; | |
3118 | ||
3119 | MA_STATE(l_mas, mas->tree, mas->index, mas->last); | |
3120 | MA_STATE(r_mas, mas->tree, mas->index, mas->last); | |
3121 | ||
3122 | trace_ma_op(__func__, mas); | |
3123 | ||
3124 | /* | |
3125 | * Rebalancing occurs if a node is insufficient. Data is rebalanced | |
3126 | * against the node to the right if it exists, otherwise the node to the | |
3127 | * left of this node is rebalanced against this node. If rebalancing | |
3128 | * causes just one node to be produced instead of two, then the parent | |
3129 | * is also examined and rebalanced if it is insufficient. Every level | |
3130 | * tries to combine the data in the same way. If one node contains the | |
3131 | * entire range of the tree, then that node is used as a new root node. | |
3132 | */ | |
3133 | mas_node_count(mas, 1 + empty_count * 3); | |
3134 | if (mas_is_err(mas)) | |
3135 | return 0; | |
3136 | ||
3137 | mast.orig_l = &l_mas; | |
3138 | mast.orig_r = &r_mas; | |
3139 | mast.bn = b_node; | |
3140 | mast.bn->type = mte_node_type(mas->node); | |
3141 | ||
3142 | l_mas = r_mas = *mas; | |
3143 | ||
3144 | if (mas_next_sibling(&r_mas)) { | |
3145 | mas_mab_cp(&r_mas, 0, mt_slot_count(r_mas.node), b_node, b_end); | |
3146 | r_mas.last = r_mas.index = r_mas.max; | |
3147 | } else { | |
3148 | mas_prev_sibling(&l_mas); | |
3149 | shift = mas_data_end(&l_mas) + 1; | |
3150 | mab_shift_right(b_node, shift); | |
3151 | mas->offset += shift; | |
3152 | mas_mab_cp(&l_mas, 0, shift - 1, b_node, 0); | |
3153 | b_node->b_end = shift + b_end; | |
3154 | l_mas.index = l_mas.last = l_mas.min; | |
3155 | } | |
3156 | ||
3157 | return mas_spanning_rebalance(mas, &mast, empty_count); | |
3158 | } | |
3159 | ||
3160 | /* | |
3161 | * mas_destroy_rebalance() - Rebalance left-most node while destroying the maple | |
3162 | * state. | |
3163 | * @mas: The maple state | |
3164 | * @end: The end of the left-most node. | |
3165 | * | |
3166 | * During a mass-insert event (such as forking), it may be necessary to | |
3167 | * rebalance the left-most node when it is not sufficient. | |
3168 | */ | |
3169 | static inline void mas_destroy_rebalance(struct ma_state *mas, unsigned char end) | |
3170 | { | |
3171 | enum maple_type mt = mte_node_type(mas->node); | |
3172 | struct maple_node reuse, *newnode, *parent, *new_left, *left, *node; | |
3173 | struct maple_enode *eparent; | |
3174 | unsigned char offset, tmp, split = mt_slots[mt] / 2; | |
3175 | void __rcu **l_slots, **slots; | |
3176 | unsigned long *l_pivs, *pivs, gap; | |
3177 | bool in_rcu = mt_in_rcu(mas->tree); | |
3178 | ||
3179 | MA_STATE(l_mas, mas->tree, mas->index, mas->last); | |
3180 | ||
3181 | l_mas = *mas; | |
3182 | mas_prev_sibling(&l_mas); | |
3183 | ||
3184 | /* set up node. */ | |
3185 | if (in_rcu) { | |
3186 | /* Allocate for both left and right as well as parent. */ | |
3187 | mas_node_count(mas, 3); | |
3188 | if (mas_is_err(mas)) | |
3189 | return; | |
3190 | ||
3191 | newnode = mas_pop_node(mas); | |
3192 | } else { | |
3193 | newnode = &reuse; | |
3194 | } | |
3195 | ||
3196 | node = mas_mn(mas); | |
3197 | newnode->parent = node->parent; | |
3198 | slots = ma_slots(newnode, mt); | |
3199 | pivs = ma_pivots(newnode, mt); | |
3200 | left = mas_mn(&l_mas); | |
3201 | l_slots = ma_slots(left, mt); | |
3202 | l_pivs = ma_pivots(left, mt); | |
3203 | if (!l_slots[split]) | |
3204 | split++; | |
3205 | tmp = mas_data_end(&l_mas) - split; | |
3206 | ||
3207 | memcpy(slots, l_slots + split + 1, sizeof(void *) * tmp); | |
3208 | memcpy(pivs, l_pivs + split + 1, sizeof(unsigned long) * tmp); | |
3209 | pivs[tmp] = l_mas.max; | |
3210 | memcpy(slots + tmp, ma_slots(node, mt), sizeof(void *) * end); | |
3211 | memcpy(pivs + tmp, ma_pivots(node, mt), sizeof(unsigned long) * end); | |
3212 | ||
3213 | l_mas.max = l_pivs[split]; | |
3214 | mas->min = l_mas.max + 1; | |
3215 | eparent = mt_mk_node(mte_parent(l_mas.node), | |
afc754c6 | 3216 | mas_parent_type(&l_mas, l_mas.node)); |
54a611b6 LH |
3217 | tmp += end; |
3218 | if (!in_rcu) { | |
3219 | unsigned char max_p = mt_pivots[mt]; | |
3220 | unsigned char max_s = mt_slots[mt]; | |
3221 | ||
3222 | if (tmp < max_p) | |
3223 | memset(pivs + tmp, 0, | |
fb20e99a | 3224 | sizeof(unsigned long) * (max_p - tmp)); |
54a611b6 LH |
3225 | |
3226 | if (tmp < mt_slots[mt]) | |
3227 | memset(slots + tmp, 0, sizeof(void *) * (max_s - tmp)); | |
3228 | ||
3229 | memcpy(node, newnode, sizeof(struct maple_node)); | |
3230 | ma_set_meta(node, mt, 0, tmp - 1); | |
3231 | mte_set_pivot(eparent, mte_parent_slot(l_mas.node), | |
3232 | l_pivs[split]); | |
3233 | ||
3234 | /* Remove data from l_pivs. */ | |
3235 | tmp = split + 1; | |
3236 | memset(l_pivs + tmp, 0, sizeof(unsigned long) * (max_p - tmp)); | |
3237 | memset(l_slots + tmp, 0, sizeof(void *) * (max_s - tmp)); | |
3238 | ma_set_meta(left, mt, 0, split); | |
3239 | ||
3240 | goto done; | |
3241 | } | |
3242 | ||
3243 | /* RCU requires replacing both l_mas, mas, and parent. */ | |
3244 | mas->node = mt_mk_node(newnode, mt); | |
3245 | ma_set_meta(newnode, mt, 0, tmp); | |
3246 | ||
3247 | new_left = mas_pop_node(mas); | |
3248 | new_left->parent = left->parent; | |
3249 | mt = mte_node_type(l_mas.node); | |
3250 | slots = ma_slots(new_left, mt); | |
3251 | pivs = ma_pivots(new_left, mt); | |
3252 | memcpy(slots, l_slots, sizeof(void *) * split); | |
3253 | memcpy(pivs, l_pivs, sizeof(unsigned long) * split); | |
3254 | ma_set_meta(new_left, mt, 0, split); | |
3255 | l_mas.node = mt_mk_node(new_left, mt); | |
3256 | ||
3257 | /* replace parent. */ | |
3258 | offset = mte_parent_slot(mas->node); | |
afc754c6 | 3259 | mt = mas_parent_type(&l_mas, l_mas.node); |
54a611b6 LH |
3260 | parent = mas_pop_node(mas); |
3261 | slots = ma_slots(parent, mt); | |
3262 | pivs = ma_pivots(parent, mt); | |
3263 | memcpy(parent, mte_to_node(eparent), sizeof(struct maple_node)); | |
3264 | rcu_assign_pointer(slots[offset], mas->node); | |
3265 | rcu_assign_pointer(slots[offset - 1], l_mas.node); | |
3266 | pivs[offset - 1] = l_mas.max; | |
3267 | eparent = mt_mk_node(parent, mt); | |
3268 | done: | |
3269 | gap = mas_leaf_max_gap(mas); | |
3270 | mte_set_gap(eparent, mte_parent_slot(mas->node), gap); | |
3271 | gap = mas_leaf_max_gap(&l_mas); | |
3272 | mte_set_gap(eparent, mte_parent_slot(l_mas.node), gap); | |
3273 | mas_ascend(mas); | |
3274 | ||
3275 | if (in_rcu) | |
3276 | mas_replace(mas, false); | |
3277 | ||
3278 | mas_update_gap(mas); | |
3279 | } | |
3280 | ||
3281 | /* | |
3282 | * mas_split_final_node() - Split the final node in a subtree operation. | |
3283 | * @mast: the maple subtree state | |
3284 | * @mas: The maple state | |
3285 | * @height: The height of the tree in case it's a new root. | |
3286 | */ | |
3287 | static inline bool mas_split_final_node(struct maple_subtree_state *mast, | |
3288 | struct ma_state *mas, int height) | |
3289 | { | |
3290 | struct maple_enode *ancestor; | |
3291 | ||
3292 | if (mte_is_root(mas->node)) { | |
3293 | if (mt_is_alloc(mas->tree)) | |
3294 | mast->bn->type = maple_arange_64; | |
3295 | else | |
3296 | mast->bn->type = maple_range_64; | |
3297 | mas->depth = height; | |
3298 | } | |
3299 | /* | |
3300 | * Only a single node is used here, could be root. | |
3301 | * The Big_node data should just fit in a single node. | |
3302 | */ | |
3303 | ancestor = mas_new_ma_node(mas, mast->bn); | |
bf96715e LH |
3304 | mas_set_parent(mas, mast->l->node, ancestor, mast->l->offset); |
3305 | mas_set_parent(mas, mast->r->node, ancestor, mast->r->offset); | |
54a611b6 LH |
3306 | mte_to_node(ancestor)->parent = mas_mn(mas)->parent; |
3307 | ||
3308 | mast->l->node = ancestor; | |
3309 | mab_mas_cp(mast->bn, 0, mt_slots[mast->bn->type] - 1, mast->l, true); | |
3310 | mas->offset = mast->bn->b_end - 1; | |
3311 | return true; | |
3312 | } | |
3313 | ||
3314 | /* | |
3315 | * mast_fill_bnode() - Copy data into the big node in the subtree state | |
3316 | * @mast: The maple subtree state | |
3317 | * @mas: the maple state | |
3318 | * @skip: The number of entries to skip for new nodes insertion. | |
3319 | */ | |
3320 | static inline void mast_fill_bnode(struct maple_subtree_state *mast, | |
3321 | struct ma_state *mas, | |
3322 | unsigned char skip) | |
3323 | { | |
3324 | bool cp = true; | |
3325 | struct maple_enode *old = mas->node; | |
3326 | unsigned char split; | |
3327 | ||
3328 | memset(mast->bn->gap, 0, sizeof(unsigned long) * ARRAY_SIZE(mast->bn->gap)); | |
3329 | memset(mast->bn->slot, 0, sizeof(unsigned long) * ARRAY_SIZE(mast->bn->slot)); | |
3330 | memset(mast->bn->pivot, 0, sizeof(unsigned long) * ARRAY_SIZE(mast->bn->pivot)); | |
3331 | mast->bn->b_end = 0; | |
3332 | ||
3333 | if (mte_is_root(mas->node)) { | |
3334 | cp = false; | |
3335 | } else { | |
3336 | mas_ascend(mas); | |
3337 | mat_add(mast->free, old); | |
3338 | mas->offset = mte_parent_slot(mas->node); | |
3339 | } | |
3340 | ||
3341 | if (cp && mast->l->offset) | |
3342 | mas_mab_cp(mas, 0, mast->l->offset - 1, mast->bn, 0); | |
3343 | ||
3344 | split = mast->bn->b_end; | |
3345 | mab_set_b_end(mast->bn, mast->l, mast->l->node); | |
3346 | mast->r->offset = mast->bn->b_end; | |
3347 | mab_set_b_end(mast->bn, mast->r, mast->r->node); | |
3348 | if (mast->bn->pivot[mast->bn->b_end - 1] == mas->max) | |
3349 | cp = false; | |
3350 | ||
3351 | if (cp) | |
3352 | mas_mab_cp(mas, split + skip, mt_slot_count(mas->node) - 1, | |
3353 | mast->bn, mast->bn->b_end); | |
3354 | ||
3355 | mast->bn->b_end--; | |
3356 | mast->bn->type = mte_node_type(mas->node); | |
3357 | } | |
3358 | ||
3359 | /* | |
3360 | * mast_split_data() - Split the data in the subtree state big node into regular | |
3361 | * nodes. | |
3362 | * @mast: The maple subtree state | |
3363 | * @mas: The maple state | |
3364 | * @split: The location to split the big node | |
3365 | */ | |
3366 | static inline void mast_split_data(struct maple_subtree_state *mast, | |
3367 | struct ma_state *mas, unsigned char split) | |
3368 | { | |
3369 | unsigned char p_slot; | |
3370 | ||
3371 | mab_mas_cp(mast->bn, 0, split, mast->l, true); | |
3372 | mte_set_pivot(mast->r->node, 0, mast->r->max); | |
3373 | mab_mas_cp(mast->bn, split + 1, mast->bn->b_end, mast->r, false); | |
3374 | mast->l->offset = mte_parent_slot(mas->node); | |
3375 | mast->l->max = mast->bn->pivot[split]; | |
3376 | mast->r->min = mast->l->max + 1; | |
3377 | if (mte_is_leaf(mas->node)) | |
3378 | return; | |
3379 | ||
3380 | p_slot = mast->orig_l->offset; | |
3381 | mas_set_split_parent(mast->orig_l, mast->l->node, mast->r->node, | |
3382 | &p_slot, split); | |
3383 | mas_set_split_parent(mast->orig_r, mast->l->node, mast->r->node, | |
3384 | &p_slot, split); | |
3385 | } | |
3386 | ||
3387 | /* | |
3388 | * mas_push_data() - Instead of splitting a node, it is beneficial to push the | |
3389 | * data to the right or left node if there is room. | |
3390 | * @mas: The maple state | |
3391 | * @height: The current height of the maple state | |
3392 | * @mast: The maple subtree state | |
3393 | * @left: Push left or not. | |
3394 | * | |
3395 | * Keeping the height of the tree low means faster lookups. | |
3396 | * | |
3397 | * Return: True if pushed, false otherwise. | |
3398 | */ | |
3399 | static inline bool mas_push_data(struct ma_state *mas, int height, | |
3400 | struct maple_subtree_state *mast, bool left) | |
3401 | { | |
3402 | unsigned char slot_total = mast->bn->b_end; | |
3403 | unsigned char end, space, split; | |
3404 | ||
3405 | MA_STATE(tmp_mas, mas->tree, mas->index, mas->last); | |
3406 | tmp_mas = *mas; | |
3407 | tmp_mas.depth = mast->l->depth; | |
3408 | ||
3409 | if (left && !mas_prev_sibling(&tmp_mas)) | |
3410 | return false; | |
3411 | else if (!left && !mas_next_sibling(&tmp_mas)) | |
3412 | return false; | |
3413 | ||
3414 | end = mas_data_end(&tmp_mas); | |
3415 | slot_total += end; | |
3416 | space = 2 * mt_slot_count(mas->node) - 2; | |
3417 | /* -2 instead of -1 to ensure there isn't a triple split */ | |
3418 | if (ma_is_leaf(mast->bn->type)) | |
3419 | space--; | |
3420 | ||
3421 | if (mas->max == ULONG_MAX) | |
3422 | space--; | |
3423 | ||
3424 | if (slot_total >= space) | |
3425 | return false; | |
3426 | ||
3427 | /* Get the data; Fill mast->bn */ | |
3428 | mast->bn->b_end++; | |
3429 | if (left) { | |
3430 | mab_shift_right(mast->bn, end + 1); | |
3431 | mas_mab_cp(&tmp_mas, 0, end, mast->bn, 0); | |
3432 | mast->bn->b_end = slot_total + 1; | |
3433 | } else { | |
3434 | mas_mab_cp(&tmp_mas, 0, end, mast->bn, mast->bn->b_end); | |
3435 | } | |
3436 | ||
3437 | /* Configure mast for splitting of mast->bn */ | |
3438 | split = mt_slots[mast->bn->type] - 2; | |
3439 | if (left) { | |
3440 | /* Switch mas to prev node */ | |
3441 | mat_add(mast->free, mas->node); | |
3442 | *mas = tmp_mas; | |
3443 | /* Start using mast->l for the left side. */ | |
3444 | tmp_mas.node = mast->l->node; | |
3445 | *mast->l = tmp_mas; | |
3446 | } else { | |
3447 | mat_add(mast->free, tmp_mas.node); | |
3448 | tmp_mas.node = mast->r->node; | |
3449 | *mast->r = tmp_mas; | |
3450 | split = slot_total - split; | |
3451 | } | |
3452 | split = mab_no_null_split(mast->bn, split, mt_slots[mast->bn->type]); | |
3453 | /* Update parent slot for split calculation. */ | |
3454 | if (left) | |
3455 | mast->orig_l->offset += end + 1; | |
3456 | ||
3457 | mast_split_data(mast, mas, split); | |
3458 | mast_fill_bnode(mast, mas, 2); | |
3459 | mas_split_final_node(mast, mas, height + 1); | |
3460 | return true; | |
3461 | } | |
3462 | ||
3463 | /* | |
3464 | * mas_split() - Split data that is too big for one node into two. | |
3465 | * @mas: The maple state | |
3466 | * @b_node: The maple big node | |
3467 | * Return: 1 on success, 0 on failure. | |
3468 | */ | |
3469 | static int mas_split(struct ma_state *mas, struct maple_big_node *b_node) | |
3470 | { | |
54a611b6 LH |
3471 | struct maple_subtree_state mast; |
3472 | int height = 0; | |
3473 | unsigned char mid_split, split = 0; | |
3474 | ||
3475 | /* | |
3476 | * Splitting is handled differently from any other B-tree; the Maple | |
3477 | * Tree splits upwards. Splitting up means that the split operation | |
3478 | * occurs when the walk of the tree hits the leaves and not on the way | |
3479 | * down. The reason for splitting up is that it is impossible to know | |
3480 | * how much space will be needed until the leaf is (or leaves are) | |
3481 | * reached. Since overwriting data is allowed and a range could | |
3482 | * overwrite more than one range or result in changing one entry into 3 | |
3483 | * entries, it is impossible to know if a split is required until the | |
3484 | * data is examined. | |
3485 | * | |
3486 | * Splitting is a balancing act between keeping allocations to a minimum | |
3487 | * and avoiding a 'jitter' event where a tree is expanded to make room | |
3488 | * for an entry followed by a contraction when the entry is removed. To | |
3489 | * accomplish the balance, there are empty slots remaining in both left | |
3490 | * and right nodes after a split. | |
3491 | */ | |
3492 | MA_STATE(l_mas, mas->tree, mas->index, mas->last); | |
3493 | MA_STATE(r_mas, mas->tree, mas->index, mas->last); | |
3494 | MA_STATE(prev_l_mas, mas->tree, mas->index, mas->last); | |
3495 | MA_STATE(prev_r_mas, mas->tree, mas->index, mas->last); | |
3496 | MA_TOPIARY(mat, mas->tree); | |
3497 | ||
3498 | trace_ma_op(__func__, mas); | |
3499 | mas->depth = mas_mt_height(mas); | |
3500 | /* Allocation failures will happen early. */ | |
3501 | mas_node_count(mas, 1 + mas->depth * 2); | |
3502 | if (mas_is_err(mas)) | |
3503 | return 0; | |
3504 | ||
3505 | mast.l = &l_mas; | |
3506 | mast.r = &r_mas; | |
3507 | mast.orig_l = &prev_l_mas; | |
3508 | mast.orig_r = &prev_r_mas; | |
3509 | mast.free = &mat; | |
3510 | mast.bn = b_node; | |
3511 | ||
3512 | while (height++ <= mas->depth) { | |
3513 | if (mt_slots[b_node->type] > b_node->b_end) { | |
3514 | mas_split_final_node(&mast, mas, height); | |
3515 | break; | |
3516 | } | |
3517 | ||
3518 | l_mas = r_mas = *mas; | |
3519 | l_mas.node = mas_new_ma_node(mas, b_node); | |
3520 | r_mas.node = mas_new_ma_node(mas, b_node); | |
3521 | /* | |
3522 | * Another way that 'jitter' is avoided is to terminate a split up early if the | |
3523 | * left or right node has space to spare. This is referred to as "pushing left" | |
3524 | * or "pushing right" and is similar to the B* tree, except the nodes left or | |
3525 | * right can rarely be reused due to RCU, but the ripple upwards is halted which | |
3526 | * is a significant savings. | |
3527 | */ | |
3528 | /* Try to push left. */ | |
3529 | if (mas_push_data(mas, height, &mast, true)) | |
3530 | break; | |
3531 | ||
3532 | /* Try to push right. */ | |
3533 | if (mas_push_data(mas, height, &mast, false)) | |
3534 | break; | |
3535 | ||
3536 | split = mab_calc_split(mas, b_node, &mid_split, prev_l_mas.min); | |
3537 | mast_split_data(&mast, mas, split); | |
3538 | /* | |
3539 | * Usually correct, mab_mas_cp in the above call overwrites | |
3540 | * r->max. | |
3541 | */ | |
3542 | mast.r->max = mas->max; | |
3543 | mast_fill_bnode(&mast, mas, 1); | |
3544 | prev_l_mas = *mast.l; | |
3545 | prev_r_mas = *mast.r; | |
3546 | } | |
3547 | ||
3548 | /* Set the original node as dead */ | |
3549 | mat_add(mast.free, mas->node); | |
3550 | mas->node = l_mas.node; | |
3551 | mas_wmb_replace(mas, mast.free, NULL); | |
3552 | mtree_range_walk(mas); | |
3553 | return 1; | |
3554 | } | |
3555 | ||
3556 | /* | |
3557 | * mas_reuse_node() - Reuse the node to store the data. | |
3558 | * @wr_mas: The maple write state | |
3559 | * @bn: The maple big node | |
3560 | * @end: The end of the data. | |
3561 | * | |
3562 | * Will always return false in RCU mode. | |
3563 | * | |
3564 | * Return: True if node was reused, false otherwise. | |
3565 | */ | |
3566 | static inline bool mas_reuse_node(struct ma_wr_state *wr_mas, | |
3567 | struct maple_big_node *bn, unsigned char end) | |
3568 | { | |
3569 | /* Need to be rcu safe. */ | |
3570 | if (mt_in_rcu(wr_mas->mas->tree)) | |
3571 | return false; | |
3572 | ||
3573 | if (end > bn->b_end) { | |
3574 | int clear = mt_slots[wr_mas->type] - bn->b_end; | |
3575 | ||
3576 | memset(wr_mas->slots + bn->b_end, 0, sizeof(void *) * clear--); | |
3577 | memset(wr_mas->pivots + bn->b_end, 0, sizeof(void *) * clear); | |
3578 | } | |
3579 | mab_mas_cp(bn, 0, bn->b_end, wr_mas->mas, false); | |
3580 | return true; | |
3581 | } | |
3582 | ||
3583 | /* | |
3584 | * mas_commit_b_node() - Commit the big node into the tree. | |
3585 | * @wr_mas: The maple write state | |
3586 | * @b_node: The maple big node | |
3587 | * @end: The end of the data. | |
3588 | */ | |
44081c77 | 3589 | static noinline_for_kasan int mas_commit_b_node(struct ma_wr_state *wr_mas, |
54a611b6 LH |
3590 | struct maple_big_node *b_node, unsigned char end) |
3591 | { | |
3592 | struct maple_node *node; | |
3593 | unsigned char b_end = b_node->b_end; | |
3594 | enum maple_type b_type = b_node->type; | |
3595 | ||
3596 | if ((b_end < mt_min_slots[b_type]) && | |
3597 | (!mte_is_root(wr_mas->mas->node)) && | |
3598 | (mas_mt_height(wr_mas->mas) > 1)) | |
3599 | return mas_rebalance(wr_mas->mas, b_node); | |
3600 | ||
3601 | if (b_end >= mt_slots[b_type]) | |
3602 | return mas_split(wr_mas->mas, b_node); | |
3603 | ||
3604 | if (mas_reuse_node(wr_mas, b_node, end)) | |
3605 | goto reuse_node; | |
3606 | ||
3607 | mas_node_count(wr_mas->mas, 1); | |
3608 | if (mas_is_err(wr_mas->mas)) | |
3609 | return 0; | |
3610 | ||
3611 | node = mas_pop_node(wr_mas->mas); | |
3612 | node->parent = mas_mn(wr_mas->mas)->parent; | |
3613 | wr_mas->mas->node = mt_mk_node(node, b_type); | |
7dc5ba62 | 3614 | mab_mas_cp(b_node, 0, b_end, wr_mas->mas, false); |
54a611b6 LH |
3615 | mas_replace(wr_mas->mas, false); |
3616 | reuse_node: | |
3617 | mas_update_gap(wr_mas->mas); | |
3618 | return 1; | |
3619 | } | |
3620 | ||
3621 | /* | |
3622 | * mas_root_expand() - Expand a root to a node | |
3623 | * @mas: The maple state | |
3624 | * @entry: The entry to store into the tree | |
3625 | */ | |
3626 | static inline int mas_root_expand(struct ma_state *mas, void *entry) | |
3627 | { | |
3628 | void *contents = mas_root_locked(mas); | |
3629 | enum maple_type type = maple_leaf_64; | |
3630 | struct maple_node *node; | |
3631 | void __rcu **slots; | |
3632 | unsigned long *pivots; | |
3633 | int slot = 0; | |
3634 | ||
3635 | mas_node_count(mas, 1); | |
3636 | if (unlikely(mas_is_err(mas))) | |
3637 | return 0; | |
3638 | ||
3639 | node = mas_pop_node(mas); | |
3640 | pivots = ma_pivots(node, type); | |
3641 | slots = ma_slots(node, type); | |
3642 | node->parent = ma_parent_ptr( | |
3643 | ((unsigned long)mas->tree | MA_ROOT_PARENT)); | |
3644 | mas->node = mt_mk_node(node, type); | |
3645 | ||
3646 | if (mas->index) { | |
3647 | if (contents) { | |
3648 | rcu_assign_pointer(slots[slot], contents); | |
3649 | if (likely(mas->index > 1)) | |
3650 | slot++; | |
3651 | } | |
3652 | pivots[slot++] = mas->index - 1; | |
3653 | } | |
3654 | ||
3655 | rcu_assign_pointer(slots[slot], entry); | |
3656 | mas->offset = slot; | |
3657 | pivots[slot] = mas->last; | |
3658 | if (mas->last != ULONG_MAX) | |
3c769fd8 PZ |
3659 | pivots[++slot] = ULONG_MAX; |
3660 | ||
54a611b6 LH |
3661 | mas->depth = 1; |
3662 | mas_set_height(mas); | |
c45ea315 | 3663 | ma_set_meta(node, maple_leaf_64, 0, slot); |
54a611b6 LH |
3664 | /* swap the new root into the tree */ |
3665 | rcu_assign_pointer(mas->tree->ma_root, mte_mk_root(mas->node)); | |
54a611b6 LH |
3666 | return slot; |
3667 | } | |
3668 | ||
3669 | static inline void mas_store_root(struct ma_state *mas, void *entry) | |
3670 | { | |
3671 | if (likely((mas->last != 0) || (mas->index != 0))) | |
3672 | mas_root_expand(mas, entry); | |
3673 | else if (((unsigned long) (entry) & 3) == 2) | |
3674 | mas_root_expand(mas, entry); | |
3675 | else { | |
3676 | rcu_assign_pointer(mas->tree->ma_root, entry); | |
3677 | mas->node = MAS_START; | |
3678 | } | |
3679 | } | |
3680 | ||
3681 | /* | |
3682 | * mas_is_span_wr() - Check if the write needs to be treated as a write that | |
3683 | * spans the node. | |
3684 | * @mas: The maple state | |
3685 | * @piv: The pivot value being written | |
3686 | * @type: The maple node type | |
3687 | * @entry: The data to write | |
3688 | * | |
3689 | * Spanning writes are writes that start in one node and end in another OR if | |
3690 | * the write of a %NULL will cause the node to end with a %NULL. | |
3691 | * | |
3692 | * Return: True if this is a spanning write, false otherwise. | |
3693 | */ | |
3694 | static bool mas_is_span_wr(struct ma_wr_state *wr_mas) | |
3695 | { | |
bc147f0f | 3696 | unsigned long max = wr_mas->r_max; |
54a611b6 | 3697 | unsigned long last = wr_mas->mas->last; |
54a611b6 LH |
3698 | enum maple_type type = wr_mas->type; |
3699 | void *entry = wr_mas->entry; | |
3700 | ||
bc147f0f PZ |
3701 | /* Contained in this pivot, fast path */ |
3702 | if (last < max) | |
54a611b6 LH |
3703 | return false; |
3704 | ||
bc147f0f PZ |
3705 | if (ma_is_leaf(type)) { |
3706 | max = wr_mas->mas->max; | |
54a611b6 LH |
3707 | if (last < max) |
3708 | return false; | |
bc147f0f | 3709 | } |
54a611b6 | 3710 | |
bc147f0f | 3711 | if (last == max) { |
54a611b6 | 3712 | /* |
bc147f0f PZ |
3713 | * The last entry of leaf node cannot be NULL unless it is the |
3714 | * rightmost node (writing ULONG_MAX), otherwise it spans slots. | |
54a611b6 | 3715 | */ |
bc147f0f | 3716 | if (entry || last == ULONG_MAX) |
54a611b6 LH |
3717 | return false; |
3718 | } | |
3719 | ||
bc147f0f | 3720 | trace_ma_write(__func__, wr_mas->mas, wr_mas->r_max, entry); |
54a611b6 LH |
3721 | return true; |
3722 | } | |
3723 | ||
3724 | static inline void mas_wr_walk_descend(struct ma_wr_state *wr_mas) | |
3725 | { | |
54a611b6 LH |
3726 | wr_mas->type = mte_node_type(wr_mas->mas->node); |
3727 | mas_wr_node_walk(wr_mas); | |
3728 | wr_mas->slots = ma_slots(wr_mas->node, wr_mas->type); | |
3729 | } | |
3730 | ||
3731 | static inline void mas_wr_walk_traverse(struct ma_wr_state *wr_mas) | |
3732 | { | |
3733 | wr_mas->mas->max = wr_mas->r_max; | |
3734 | wr_mas->mas->min = wr_mas->r_min; | |
3735 | wr_mas->mas->node = wr_mas->content; | |
3736 | wr_mas->mas->offset = 0; | |
9bbba563 | 3737 | wr_mas->mas->depth++; |
54a611b6 LH |
3738 | } |
3739 | /* | |
3740 | * mas_wr_walk() - Walk the tree for a write. | |
3741 | * @wr_mas: The maple write state | |
3742 | * | |
3743 | * Uses mas_slot_locked() and does not need to worry about dead nodes. | |
3744 | * | |
3745 | * Return: True if it's contained in a node, false on spanning write. | |
3746 | */ | |
3747 | static bool mas_wr_walk(struct ma_wr_state *wr_mas) | |
3748 | { | |
3749 | struct ma_state *mas = wr_mas->mas; | |
3750 | ||
3751 | while (true) { | |
3752 | mas_wr_walk_descend(wr_mas); | |
3753 | if (unlikely(mas_is_span_wr(wr_mas))) | |
3754 | return false; | |
3755 | ||
3756 | wr_mas->content = mas_slot_locked(mas, wr_mas->slots, | |
3757 | mas->offset); | |
3758 | if (ma_is_leaf(wr_mas->type)) | |
3759 | return true; | |
3760 | ||
3761 | mas_wr_walk_traverse(wr_mas); | |
3762 | } | |
3763 | ||
3764 | return true; | |
3765 | } | |
3766 | ||
3767 | static bool mas_wr_walk_index(struct ma_wr_state *wr_mas) | |
3768 | { | |
3769 | struct ma_state *mas = wr_mas->mas; | |
3770 | ||
3771 | while (true) { | |
3772 | mas_wr_walk_descend(wr_mas); | |
3773 | wr_mas->content = mas_slot_locked(mas, wr_mas->slots, | |
3774 | mas->offset); | |
3775 | if (ma_is_leaf(wr_mas->type)) | |
3776 | return true; | |
3777 | mas_wr_walk_traverse(wr_mas); | |
3778 | ||
3779 | } | |
3780 | return true; | |
3781 | } | |
3782 | /* | |
3783 | * mas_extend_spanning_null() - Extend a store of a %NULL to include surrounding %NULLs. | |
3784 | * @l_wr_mas: The left maple write state | |
3785 | * @r_wr_mas: The right maple write state | |
3786 | */ | |
3787 | static inline void mas_extend_spanning_null(struct ma_wr_state *l_wr_mas, | |
3788 | struct ma_wr_state *r_wr_mas) | |
3789 | { | |
3790 | struct ma_state *r_mas = r_wr_mas->mas; | |
3791 | struct ma_state *l_mas = l_wr_mas->mas; | |
3792 | unsigned char l_slot; | |
3793 | ||
3794 | l_slot = l_mas->offset; | |
3795 | if (!l_wr_mas->content) | |
3796 | l_mas->index = l_wr_mas->r_min; | |
3797 | ||
3798 | if ((l_mas->index == l_wr_mas->r_min) && | |
3799 | (l_slot && | |
3800 | !mas_slot_locked(l_mas, l_wr_mas->slots, l_slot - 1))) { | |
3801 | if (l_slot > 1) | |
3802 | l_mas->index = l_wr_mas->pivots[l_slot - 2] + 1; | |
3803 | else | |
3804 | l_mas->index = l_mas->min; | |
3805 | ||
3806 | l_mas->offset = l_slot - 1; | |
3807 | } | |
3808 | ||
3809 | if (!r_wr_mas->content) { | |
3810 | if (r_mas->last < r_wr_mas->r_max) | |
3811 | r_mas->last = r_wr_mas->r_max; | |
3812 | r_mas->offset++; | |
3813 | } else if ((r_mas->last == r_wr_mas->r_max) && | |
3814 | (r_mas->last < r_mas->max) && | |
3815 | !mas_slot_locked(r_mas, r_wr_mas->slots, r_mas->offset + 1)) { | |
3816 | r_mas->last = mas_safe_pivot(r_mas, r_wr_mas->pivots, | |
3817 | r_wr_mas->type, r_mas->offset + 1); | |
3818 | r_mas->offset++; | |
3819 | } | |
3820 | } | |
3821 | ||
3822 | static inline void *mas_state_walk(struct ma_state *mas) | |
3823 | { | |
3824 | void *entry; | |
3825 | ||
3826 | entry = mas_start(mas); | |
3827 | if (mas_is_none(mas)) | |
3828 | return NULL; | |
3829 | ||
3830 | if (mas_is_ptr(mas)) | |
3831 | return entry; | |
3832 | ||
3833 | return mtree_range_walk(mas); | |
3834 | } | |
3835 | ||
3836 | /* | |
3837 | * mtree_lookup_walk() - Internal quick lookup that does not keep maple state up | |
3838 | * to date. | |
3839 | * | |
3840 | * @mas: The maple state. | |
3841 | * | |
3842 | * Note: Leaves mas in undesirable state. | |
3843 | * Return: The entry for @mas->index or %NULL on dead node. | |
3844 | */ | |
3845 | static inline void *mtree_lookup_walk(struct ma_state *mas) | |
3846 | { | |
3847 | unsigned long *pivots; | |
3848 | unsigned char offset; | |
3849 | struct maple_node *node; | |
3850 | struct maple_enode *next; | |
3851 | enum maple_type type; | |
3852 | void __rcu **slots; | |
3853 | unsigned char end; | |
3854 | unsigned long max; | |
3855 | ||
3856 | next = mas->node; | |
3857 | max = ULONG_MAX; | |
3858 | do { | |
3859 | offset = 0; | |
3860 | node = mte_to_node(next); | |
3861 | type = mte_node_type(next); | |
3862 | pivots = ma_pivots(node, type); | |
3863 | end = ma_data_end(node, type, pivots, max); | |
3864 | if (unlikely(ma_dead_node(node))) | |
3865 | goto dead_node; | |
54a611b6 | 3866 | do { |
ec07967d PZ |
3867 | if (pivots[offset] >= mas->index) { |
3868 | max = pivots[offset]; | |
3869 | break; | |
3870 | } | |
3871 | } while (++offset < end); | |
54a611b6 | 3872 | |
54a611b6 LH |
3873 | slots = ma_slots(node, type); |
3874 | next = mt_slot(mas->tree, slots, offset); | |
3875 | if (unlikely(ma_dead_node(node))) | |
3876 | goto dead_node; | |
3877 | } while (!ma_is_leaf(type)); | |
3878 | ||
831978e3 | 3879 | return (void *)next; |
54a611b6 LH |
3880 | |
3881 | dead_node: | |
3882 | mas_reset(mas); | |
3883 | return NULL; | |
3884 | } | |
3885 | ||
3886 | /* | |
3887 | * mas_new_root() - Create a new root node that only contains the entry passed | |
3888 | * in. | |
3889 | * @mas: The maple state | |
3890 | * @entry: The entry to store. | |
3891 | * | |
3892 | * Only valid when the index == 0 and the last == ULONG_MAX | |
3893 | * | |
3894 | * Return 0 on error, 1 on success. | |
3895 | */ | |
3896 | static inline int mas_new_root(struct ma_state *mas, void *entry) | |
3897 | { | |
3898 | struct maple_enode *root = mas_root_locked(mas); | |
3899 | enum maple_type type = maple_leaf_64; | |
3900 | struct maple_node *node; | |
3901 | void __rcu **slots; | |
3902 | unsigned long *pivots; | |
3903 | ||
3904 | if (!entry && !mas->index && mas->last == ULONG_MAX) { | |
3905 | mas->depth = 0; | |
3906 | mas_set_height(mas); | |
3907 | rcu_assign_pointer(mas->tree->ma_root, entry); | |
3908 | mas->node = MAS_START; | |
3909 | goto done; | |
3910 | } | |
3911 | ||
3912 | mas_node_count(mas, 1); | |
3913 | if (mas_is_err(mas)) | |
3914 | return 0; | |
3915 | ||
3916 | node = mas_pop_node(mas); | |
3917 | pivots = ma_pivots(node, type); | |
3918 | slots = ma_slots(node, type); | |
3919 | node->parent = ma_parent_ptr( | |
3920 | ((unsigned long)mas->tree | MA_ROOT_PARENT)); | |
3921 | mas->node = mt_mk_node(node, type); | |
3922 | rcu_assign_pointer(slots[0], entry); | |
3923 | pivots[0] = mas->last; | |
3924 | mas->depth = 1; | |
3925 | mas_set_height(mas); | |
3926 | rcu_assign_pointer(mas->tree->ma_root, mte_mk_root(mas->node)); | |
3927 | ||
3928 | done: | |
3929 | if (xa_is_node(root)) | |
3930 | mte_destroy_walk(root, mas->tree); | |
3931 | ||
3932 | return 1; | |
3933 | } | |
3934 | /* | |
3935 | * mas_wr_spanning_store() - Create a subtree with the store operation completed | |
3936 | * and new nodes where necessary, then place the sub-tree in the actual tree. | |
3937 | * Note that mas is expected to point to the node which caused the store to | |
3938 | * span. | |
3939 | * @wr_mas: The maple write state | |
3940 | * | |
3941 | * Return: 0 on error, positive on success. | |
3942 | */ | |
3943 | static inline int mas_wr_spanning_store(struct ma_wr_state *wr_mas) | |
3944 | { | |
3945 | struct maple_subtree_state mast; | |
3946 | struct maple_big_node b_node; | |
3947 | struct ma_state *mas; | |
3948 | unsigned char height; | |
3949 | ||
3950 | /* Left and Right side of spanning store */ | |
3951 | MA_STATE(l_mas, NULL, 0, 0); | |
3952 | MA_STATE(r_mas, NULL, 0, 0); | |
3953 | ||
3954 | MA_WR_STATE(r_wr_mas, &r_mas, wr_mas->entry); | |
3955 | MA_WR_STATE(l_wr_mas, &l_mas, wr_mas->entry); | |
3956 | ||
3957 | /* | |
3958 | * A store operation that spans multiple nodes is called a spanning | |
3959 | * store and is handled early in the store call stack by the function | |
3960 | * mas_is_span_wr(). When a spanning store is identified, the maple | |
3961 | * state is duplicated. The first maple state walks the left tree path | |
3962 | * to ``index``, the duplicate walks the right tree path to ``last``. | |
3963 | * The data in the two nodes are combined into a single node, two nodes, | |
3964 | * or possibly three nodes (see the 3-way split above). A ``NULL`` | |
3965 | * written to the last entry of a node is considered a spanning store as | |
3966 | * a rebalance is required for the operation to complete and an overflow | |
3967 | * of data may happen. | |
3968 | */ | |
3969 | mas = wr_mas->mas; | |
3970 | trace_ma_op(__func__, mas); | |
3971 | ||
3972 | if (unlikely(!mas->index && mas->last == ULONG_MAX)) | |
3973 | return mas_new_root(mas, wr_mas->entry); | |
3974 | /* | |
3975 | * Node rebalancing may occur due to this store, so there may be three new | |
3976 | * entries per level plus a new root. | |
3977 | */ | |
3978 | height = mas_mt_height(mas); | |
3979 | mas_node_count(mas, 1 + height * 3); | |
3980 | if (mas_is_err(mas)) | |
3981 | return 0; | |
3982 | ||
3983 | /* | |
3984 | * Set up right side. Need to get to the next offset after the spanning | |
3985 | * store to ensure it's not NULL and to combine both the next node and | |
3986 | * the node with the start together. | |
3987 | */ | |
3988 | r_mas = *mas; | |
3989 | /* Avoid overflow, walk to next slot in the tree. */ | |
3990 | if (r_mas.last + 1) | |
3991 | r_mas.last++; | |
3992 | ||
3993 | r_mas.index = r_mas.last; | |
3994 | mas_wr_walk_index(&r_wr_mas); | |
3995 | r_mas.last = r_mas.index = mas->last; | |
3996 | ||
3997 | /* Set up left side. */ | |
3998 | l_mas = *mas; | |
3999 | mas_wr_walk_index(&l_wr_mas); | |
4000 | ||
4001 | if (!wr_mas->entry) { | |
4002 | mas_extend_spanning_null(&l_wr_mas, &r_wr_mas); | |
4003 | mas->offset = l_mas.offset; | |
4004 | mas->index = l_mas.index; | |
4005 | mas->last = l_mas.last = r_mas.last; | |
4006 | } | |
4007 | ||
4008 | /* expanding NULLs may make this cover the entire range */ | |
4009 | if (!l_mas.index && r_mas.last == ULONG_MAX) { | |
4010 | mas_set_range(mas, 0, ULONG_MAX); | |
4011 | return mas_new_root(mas, wr_mas->entry); | |
4012 | } | |
4013 | ||
4014 | memset(&b_node, 0, sizeof(struct maple_big_node)); | |
4015 | /* Copy l_mas and store the value in b_node. */ | |
4016 | mas_store_b_node(&l_wr_mas, &b_node, l_wr_mas.node_end); | |
4017 | /* Copy r_mas into b_node. */ | |
4018 | if (r_mas.offset <= r_wr_mas.node_end) | |
4019 | mas_mab_cp(&r_mas, r_mas.offset, r_wr_mas.node_end, | |
4020 | &b_node, b_node.b_end + 1); | |
4021 | else | |
4022 | b_node.b_end++; | |
4023 | ||
4024 | /* Stop spanning searches by searching for just index. */ | |
4025 | l_mas.index = l_mas.last = mas->index; | |
4026 | ||
4027 | mast.bn = &b_node; | |
4028 | mast.orig_l = &l_mas; | |
4029 | mast.orig_r = &r_mas; | |
4030 | /* Combine l_mas and r_mas and split them up evenly again. */ | |
4031 | return mas_spanning_rebalance(mas, &mast, height + 1); | |
4032 | } | |
4033 | ||
4034 | /* | |
4035 | * mas_wr_node_store() - Attempt to store the value in a node | |
4036 | * @wr_mas: The maple write state | |
4037 | * | |
4038 | * Attempts to reuse the node, but may allocate. | |
4039 | * | |
4040 | * Return: True if stored, false otherwise | |
4041 | */ | |
7a03ae39 PZ |
4042 | static inline bool mas_wr_node_store(struct ma_wr_state *wr_mas, |
4043 | unsigned char new_end) | |
54a611b6 LH |
4044 | { |
4045 | struct ma_state *mas = wr_mas->mas; | |
4046 | void __rcu **dst_slots; | |
4047 | unsigned long *dst_pivots; | |
7a03ae39 | 4048 | unsigned char dst_offset, offset_end = wr_mas->offset_end; |
54a611b6 | 4049 | struct maple_node reuse, *newnode; |
7a03ae39 | 4050 | unsigned char copy_size, node_pivots = mt_pivots[wr_mas->type]; |
54a611b6 LH |
4051 | bool in_rcu = mt_in_rcu(mas->tree); |
4052 | ||
7a03ae39 | 4053 | /* Check if there is enough data. The room is enough. */ |
54a611b6 LH |
4054 | if (!mte_is_root(mas->node) && (new_end <= mt_min_slots[wr_mas->type]) && |
4055 | !(mas->mas_flags & MA_STATE_BULK)) | |
4056 | return false; | |
4057 | ||
7a03ae39 PZ |
4058 | if (mas->last == wr_mas->end_piv) |
4059 | offset_end++; /* don't copy this offset */ | |
4060 | else if (unlikely(wr_mas->r_max == ULONG_MAX)) | |
4061 | mas_bulk_rebalance(mas, wr_mas->node_end, wr_mas->type); | |
4062 | ||
54a611b6 LH |
4063 | /* set up node. */ |
4064 | if (in_rcu) { | |
4065 | mas_node_count(mas, 1); | |
4066 | if (mas_is_err(mas)) | |
4067 | return false; | |
4068 | ||
4069 | newnode = mas_pop_node(mas); | |
4070 | } else { | |
4071 | memset(&reuse, 0, sizeof(struct maple_node)); | |
4072 | newnode = &reuse; | |
4073 | } | |
4074 | ||
4075 | newnode->parent = mas_mn(mas)->parent; | |
4076 | dst_pivots = ma_pivots(newnode, wr_mas->type); | |
4077 | dst_slots = ma_slots(newnode, wr_mas->type); | |
4078 | /* Copy from start to insert point */ | |
7a03ae39 PZ |
4079 | memcpy(dst_pivots, wr_mas->pivots, sizeof(unsigned long) * mas->offset); |
4080 | memcpy(dst_slots, wr_mas->slots, sizeof(void *) * mas->offset); | |
54a611b6 LH |
4081 | |
4082 | /* Handle insert of new range starting after old range */ | |
4083 | if (wr_mas->r_min < mas->index) { | |
7a03ae39 PZ |
4084 | rcu_assign_pointer(dst_slots[mas->offset], wr_mas->content); |
4085 | dst_pivots[mas->offset++] = mas->index - 1; | |
54a611b6 LH |
4086 | } |
4087 | ||
4088 | /* Store the new entry and range end. */ | |
7a03ae39 PZ |
4089 | if (mas->offset < node_pivots) |
4090 | dst_pivots[mas->offset] = mas->last; | |
4091 | rcu_assign_pointer(dst_slots[mas->offset], wr_mas->entry); | |
54a611b6 LH |
4092 | |
4093 | /* | |
4094 | * this range wrote to the end of the node or it overwrote the rest of | |
4095 | * the data | |
4096 | */ | |
7a03ae39 | 4097 | if (offset_end > wr_mas->node_end) |
54a611b6 | 4098 | goto done; |
54a611b6 | 4099 | |
7a03ae39 | 4100 | dst_offset = mas->offset + 1; |
54a611b6 | 4101 | /* Copy to the end of node if necessary. */ |
7a03ae39 PZ |
4102 | copy_size = wr_mas->node_end - offset_end + 1; |
4103 | memcpy(dst_slots + dst_offset, wr_mas->slots + offset_end, | |
54a611b6 | 4104 | sizeof(void *) * copy_size); |
7a03ae39 PZ |
4105 | memcpy(dst_pivots + dst_offset, wr_mas->pivots + offset_end, |
4106 | sizeof(unsigned long) * (copy_size - 1)); | |
54a611b6 | 4107 | |
7a03ae39 | 4108 | if (new_end < node_pivots) |
54a611b6 LH |
4109 | dst_pivots[new_end] = mas->max; |
4110 | ||
4111 | done: | |
4112 | mas_leaf_set_meta(mas, newnode, dst_pivots, maple_leaf_64, new_end); | |
4113 | if (in_rcu) { | |
c13af03d | 4114 | mte_set_node_dead(mas->node); |
54a611b6 LH |
4115 | mas->node = mt_mk_node(newnode, wr_mas->type); |
4116 | mas_replace(mas, false); | |
4117 | } else { | |
4118 | memcpy(wr_mas->node, newnode, sizeof(struct maple_node)); | |
4119 | } | |
4120 | trace_ma_write(__func__, mas, 0, wr_mas->entry); | |
4121 | mas_update_gap(mas); | |
4122 | return true; | |
4123 | } | |
4124 | ||
4125 | /* | |
4126 | * mas_wr_slot_store: Attempt to store a value in a slot. | |
4127 | * @wr_mas: the maple write state | |
4128 | * | |
4129 | * Return: True if stored, false otherwise | |
4130 | */ | |
4131 | static inline bool mas_wr_slot_store(struct ma_wr_state *wr_mas) | |
4132 | { | |
4133 | struct ma_state *mas = wr_mas->mas; | |
54a611b6 | 4134 | unsigned char offset = mas->offset; |
64891ba3 | 4135 | void __rcu **slots = wr_mas->slots; |
e6d1ffd6 | 4136 | bool gap = false; |
54a611b6 | 4137 | |
64891ba3 PZ |
4138 | gap |= !mt_slot_locked(mas->tree, slots, offset); |
4139 | gap |= !mt_slot_locked(mas->tree, slots, offset + 1); | |
54a611b6 | 4140 | |
64891ba3 PZ |
4141 | if (wr_mas->offset_end - offset == 1) { |
4142 | if (mas->index == wr_mas->r_min) { | |
4143 | /* Overwriting the range and a part of the next one */ | |
4144 | rcu_assign_pointer(slots[offset], wr_mas->entry); | |
4145 | wr_mas->pivots[offset] = mas->last; | |
4146 | } else { | |
4147 | /* Overwriting a part of the range and the next one */ | |
4148 | rcu_assign_pointer(slots[offset + 1], wr_mas->entry); | |
4149 | wr_mas->pivots[offset] = mas->index - 1; | |
4150 | mas->offset++; /* Keep mas accurate. */ | |
4151 | } | |
4152 | } else if (!mt_in_rcu(mas->tree)) { | |
4153 | /* | |
4154 | * Expand the range, only partially overwriting the previous and | |
4155 | * next ranges | |
4156 | */ | |
4157 | gap |= !mt_slot_locked(mas->tree, slots, offset + 2); | |
4158 | rcu_assign_pointer(slots[offset + 1], wr_mas->entry); | |
e6d1ffd6 | 4159 | wr_mas->pivots[offset] = mas->index - 1; |
64891ba3 | 4160 | wr_mas->pivots[offset + 1] = mas->last; |
e6d1ffd6 | 4161 | mas->offset++; /* Keep mas accurate. */ |
64891ba3 PZ |
4162 | } else { |
4163 | return false; | |
54a611b6 LH |
4164 | } |
4165 | ||
54a611b6 | 4166 | trace_ma_write(__func__, mas, 0, wr_mas->entry); |
e6d1ffd6 PZ |
4167 | /* |
4168 | * Only update gap when the new entry is empty or there is an empty | |
4169 | * entry in the original two ranges. | |
4170 | */ | |
4171 | if (!wr_mas->entry || gap) | |
4172 | mas_update_gap(mas); | |
4173 | ||
54a611b6 LH |
4174 | return true; |
4175 | } | |
4176 | ||
4177 | static inline void mas_wr_end_piv(struct ma_wr_state *wr_mas) | |
4178 | { | |
cd00dd25 PZ |
4179 | while ((wr_mas->offset_end < wr_mas->node_end) && |
4180 | (wr_mas->mas->last > wr_mas->pivots[wr_mas->offset_end])) | |
4181 | wr_mas->offset_end++; | |
54a611b6 | 4182 | |
cd00dd25 PZ |
4183 | if (wr_mas->offset_end < wr_mas->node_end) |
4184 | wr_mas->end_piv = wr_mas->pivots[wr_mas->offset_end]; | |
4185 | else | |
54a611b6 LH |
4186 | wr_mas->end_piv = wr_mas->mas->max; |
4187 | } | |
4188 | ||
4189 | static inline void mas_wr_extend_null(struct ma_wr_state *wr_mas) | |
4190 | { | |
4191 | struct ma_state *mas = wr_mas->mas; | |
4192 | ||
8c995a63 PZ |
4193 | if (!wr_mas->slots[wr_mas->offset_end]) { |
4194 | /* If this one is null, the next and prev are not */ | |
54a611b6 | 4195 | mas->last = wr_mas->end_piv; |
8c995a63 PZ |
4196 | } else { |
4197 | /* Check next slot(s) if we are overwriting the end */ | |
4198 | if ((mas->last == wr_mas->end_piv) && | |
4199 | (wr_mas->node_end != wr_mas->offset_end) && | |
4200 | !wr_mas->slots[wr_mas->offset_end + 1]) { | |
4201 | wr_mas->offset_end++; | |
4202 | if (wr_mas->offset_end == wr_mas->node_end) | |
4203 | mas->last = mas->max; | |
4204 | else | |
4205 | mas->last = wr_mas->pivots[wr_mas->offset_end]; | |
4206 | wr_mas->end_piv = mas->last; | |
4207 | } | |
54a611b6 LH |
4208 | } |
4209 | ||
4210 | if (!wr_mas->content) { | |
4211 | /* If this one is null, the next and prev are not */ | |
4212 | mas->index = wr_mas->r_min; | |
4213 | } else { | |
4214 | /* Check prev slot if we are overwriting the start */ | |
4215 | if (mas->index == wr_mas->r_min && mas->offset && | |
4216 | !wr_mas->slots[mas->offset - 1]) { | |
4217 | mas->offset--; | |
4218 | wr_mas->r_min = mas->index = | |
4219 | mas_safe_min(mas, wr_mas->pivots, mas->offset); | |
4220 | wr_mas->r_max = wr_mas->pivots[mas->offset]; | |
4221 | } | |
4222 | } | |
4223 | } | |
4224 | ||
c6fc9e4a PZ |
4225 | static inline unsigned char mas_wr_new_end(struct ma_wr_state *wr_mas) |
4226 | { | |
4227 | struct ma_state *mas = wr_mas->mas; | |
4228 | unsigned char new_end = wr_mas->node_end + 2; | |
4229 | ||
4230 | new_end -= wr_mas->offset_end - mas->offset; | |
4231 | if (wr_mas->r_min == mas->index) | |
4232 | new_end--; | |
4233 | ||
4234 | if (wr_mas->end_piv == mas->last) | |
4235 | new_end--; | |
4236 | ||
4237 | return new_end; | |
4238 | } | |
4239 | ||
2e1da329 PZ |
4240 | /* |
4241 | * mas_wr_append: Attempt to append | |
4242 | * @wr_mas: the maple write state | |
4243 | * | |
4244 | * Return: True if appended, false otherwise | |
4245 | */ | |
23e9dde0 PZ |
4246 | static inline bool mas_wr_append(struct ma_wr_state *wr_mas, |
4247 | unsigned char new_end) | |
54a611b6 LH |
4248 | { |
4249 | unsigned char end = wr_mas->node_end; | |
54a611b6 LH |
4250 | struct ma_state *mas = wr_mas->mas; |
4251 | unsigned char node_pivots = mt_pivots[wr_mas->type]; | |
4252 | ||
2e1da329 PZ |
4253 | if (mas->offset != wr_mas->node_end) |
4254 | return false; | |
54a611b6 | 4255 | |
2e1da329 PZ |
4256 | if (new_end < node_pivots) { |
4257 | wr_mas->pivots[new_end] = wr_mas->pivots[end]; | |
4258 | ma_set_meta(wr_mas->node, maple_leaf_64, 0, new_end); | |
4259 | } | |
54a611b6 | 4260 | |
23e9dde0 PZ |
4261 | if (new_end == wr_mas->node_end + 1) { |
4262 | if (mas->last == wr_mas->r_max) { | |
4263 | /* Append to end of range */ | |
4264 | rcu_assign_pointer(wr_mas->slots[new_end], | |
4265 | wr_mas->entry); | |
4266 | wr_mas->pivots[end] = mas->index - 1; | |
4267 | mas->offset = new_end; | |
4268 | } else { | |
4269 | /* Append to start of range */ | |
4270 | rcu_assign_pointer(wr_mas->slots[new_end], | |
4271 | wr_mas->content); | |
4272 | wr_mas->pivots[end] = mas->last; | |
4273 | rcu_assign_pointer(wr_mas->slots[end], wr_mas->entry); | |
4274 | } | |
2e1da329 | 4275 | } else { |
23e9dde0 | 4276 | /* Append to the range without touching any boundaries. */ |
54a611b6 | 4277 | rcu_assign_pointer(wr_mas->slots[new_end], wr_mas->content); |
23e9dde0 PZ |
4278 | wr_mas->pivots[end + 1] = mas->last; |
4279 | rcu_assign_pointer(wr_mas->slots[end + 1], wr_mas->entry); | |
4280 | wr_mas->pivots[end] = mas->index - 1; | |
4281 | mas->offset = end + 1; | |
54a611b6 LH |
4282 | } |
4283 | ||
2e1da329 PZ |
4284 | if (!wr_mas->content || !wr_mas->entry) |
4285 | mas_update_gap(mas); | |
4286 | ||
4287 | return true; | |
54a611b6 LH |
4288 | } |
4289 | ||
4290 | /* | |
4291 | * mas_wr_bnode() - Slow path for a modification. | |
4292 | * @wr_mas: The write maple state | |
4293 | * | |
4294 | * This is where split, rebalance end up. | |
4295 | */ | |
4296 | static void mas_wr_bnode(struct ma_wr_state *wr_mas) | |
4297 | { | |
4298 | struct maple_big_node b_node; | |
4299 | ||
4300 | trace_ma_write(__func__, wr_mas->mas, 0, wr_mas->entry); | |
4301 | memset(&b_node, 0, sizeof(struct maple_big_node)); | |
4302 | mas_store_b_node(wr_mas, &b_node, wr_mas->offset_end); | |
4303 | mas_commit_b_node(wr_mas, &b_node, wr_mas->node_end); | |
4304 | } | |
4305 | ||
4306 | static inline void mas_wr_modify(struct ma_wr_state *wr_mas) | |
4307 | { | |
54a611b6 | 4308 | struct ma_state *mas = wr_mas->mas; |
c6fc9e4a | 4309 | unsigned char new_end; |
54a611b6 LH |
4310 | |
4311 | /* Direct replacement */ | |
4312 | if (wr_mas->r_min == mas->index && wr_mas->r_max == mas->last) { | |
4313 | rcu_assign_pointer(wr_mas->slots[mas->offset], wr_mas->entry); | |
4314 | if (!!wr_mas->entry ^ !!wr_mas->content) | |
4315 | mas_update_gap(mas); | |
4316 | return; | |
4317 | } | |
4318 | ||
c6fc9e4a PZ |
4319 | /* |
4320 | * new_end exceeds the size of the maple node and cannot enter the fast | |
4321 | * path. | |
4322 | */ | |
4323 | new_end = mas_wr_new_end(wr_mas); | |
4324 | if (new_end >= mt_slots[wr_mas->type]) | |
54a611b6 LH |
4325 | goto slow_path; |
4326 | ||
2e1da329 | 4327 | /* Attempt to append */ |
23e9dde0 | 4328 | if (mas_wr_append(wr_mas, new_end)) |
54a611b6 | 4329 | return; |
54a611b6 | 4330 | |
e6d1ffd6 | 4331 | if (new_end == wr_mas->node_end && mas_wr_slot_store(wr_mas)) |
54a611b6 | 4332 | return; |
7a03ae39 PZ |
4333 | |
4334 | if (mas_wr_node_store(wr_mas, new_end)) | |
54a611b6 LH |
4335 | return; |
4336 | ||
4337 | if (mas_is_err(mas)) | |
4338 | return; | |
4339 | ||
4340 | slow_path: | |
4341 | mas_wr_bnode(wr_mas); | |
4342 | } | |
4343 | ||
4344 | /* | |
4345 | * mas_wr_store_entry() - Internal call to store a value | |
4346 | * @mas: The maple state | |
4347 | * @entry: The entry to store. | |
4348 | * | |
4349 | * Return: The contents that was stored at the index. | |
4350 | */ | |
4351 | static inline void *mas_wr_store_entry(struct ma_wr_state *wr_mas) | |
4352 | { | |
4353 | struct ma_state *mas = wr_mas->mas; | |
4354 | ||
4355 | wr_mas->content = mas_start(mas); | |
4356 | if (mas_is_none(mas) || mas_is_ptr(mas)) { | |
4357 | mas_store_root(mas, wr_mas->entry); | |
4358 | return wr_mas->content; | |
4359 | } | |
4360 | ||
4361 | if (unlikely(!mas_wr_walk(wr_mas))) { | |
4362 | mas_wr_spanning_store(wr_mas); | |
4363 | return wr_mas->content; | |
4364 | } | |
4365 | ||
4366 | /* At this point, we are at the leaf node that needs to be altered. */ | |
54a611b6 LH |
4367 | mas_wr_end_piv(wr_mas); |
4368 | ||
4369 | if (!wr_mas->entry) | |
4370 | mas_wr_extend_null(wr_mas); | |
4371 | ||
4372 | /* New root for a single pointer */ | |
4373 | if (unlikely(!mas->index && mas->last == ULONG_MAX)) { | |
4374 | mas_new_root(mas, wr_mas->entry); | |
4375 | return wr_mas->content; | |
4376 | } | |
4377 | ||
4378 | mas_wr_modify(wr_mas); | |
4379 | return wr_mas->content; | |
4380 | } | |
4381 | ||
4382 | /** | |
4383 | * mas_insert() - Internal call to insert a value | |
4384 | * @mas: The maple state | |
4385 | * @entry: The entry to store | |
4386 | * | |
4387 | * Return: %NULL or the contents that already exists at the requested index | |
4388 | * otherwise. The maple state needs to be checked for error conditions. | |
4389 | */ | |
4390 | static inline void *mas_insert(struct ma_state *mas, void *entry) | |
4391 | { | |
4392 | MA_WR_STATE(wr_mas, mas, entry); | |
4393 | ||
4394 | /* | |
4395 | * Inserting a new range inserts either 0, 1, or 2 pivots within the | |
4396 | * tree. If the insert fits exactly into an existing gap with a value | |
4397 | * of NULL, then the slot only needs to be written with the new value. | |
4398 | * If the range being inserted is adjacent to another range, then only a | |
4399 | * single pivot needs to be inserted (as well as writing the entry). If | |
4400 | * the new range is within a gap but does not touch any other ranges, | |
4401 | * then two pivots need to be inserted: the start - 1, and the end. As | |
4402 | * usual, the entry must be written. Most operations require a new node | |
4403 | * to be allocated and replace an existing node to ensure RCU safety, | |
4404 | * when in RCU mode. The exception to requiring a newly allocated node | |
4405 | * is when inserting at the end of a node (appending). When done | |
4406 | * carefully, appending can reuse the node in place. | |
4407 | */ | |
4408 | wr_mas.content = mas_start(mas); | |
4409 | if (wr_mas.content) | |
4410 | goto exists; | |
4411 | ||
4412 | if (mas_is_none(mas) || mas_is_ptr(mas)) { | |
4413 | mas_store_root(mas, entry); | |
4414 | return NULL; | |
4415 | } | |
4416 | ||
4417 | /* spanning writes always overwrite something */ | |
4418 | if (!mas_wr_walk(&wr_mas)) | |
4419 | goto exists; | |
4420 | ||
4421 | /* At this point, we are at the leaf node that needs to be altered. */ | |
4422 | wr_mas.offset_end = mas->offset; | |
4423 | wr_mas.end_piv = wr_mas.r_max; | |
4424 | ||
4425 | if (wr_mas.content || (mas->last > wr_mas.r_max)) | |
4426 | goto exists; | |
4427 | ||
4428 | if (!entry) | |
4429 | return NULL; | |
4430 | ||
4431 | mas_wr_modify(&wr_mas); | |
4432 | return wr_mas.content; | |
4433 | ||
4434 | exists: | |
4435 | mas_set_err(mas, -EEXIST); | |
4436 | return wr_mas.content; | |
4437 | ||
4438 | } | |
4439 | ||
de6e386c LH |
4440 | static inline void mas_rewalk(struct ma_state *mas, unsigned long index) |
4441 | { | |
4442 | retry: | |
4443 | mas_set(mas, index); | |
4444 | mas_state_walk(mas); | |
4445 | if (mas_is_start(mas)) | |
4446 | goto retry; | |
4447 | } | |
4448 | ||
4449 | static inline bool mas_rewalk_if_dead(struct ma_state *mas, | |
4450 | struct maple_node *node, const unsigned long index) | |
4451 | { | |
4452 | if (unlikely(ma_dead_node(node))) { | |
4453 | mas_rewalk(mas, index); | |
4454 | return true; | |
4455 | } | |
4456 | return false; | |
4457 | } | |
4458 | ||
54a611b6 LH |
4459 | /* |
4460 | * mas_prev_node() - Find the prev non-null entry at the same level in the | |
4461 | * tree. The prev value will be mas->node[mas->offset] or MAS_NONE. | |
4462 | * @mas: The maple state | |
4463 | * @min: The lower limit to search | |
4464 | * | |
4465 | * The prev node value will be mas->node[mas->offset] or MAS_NONE. | |
4466 | * Return: 1 if the node is dead, 0 otherwise. | |
4467 | */ | |
4468 | static inline int mas_prev_node(struct ma_state *mas, unsigned long min) | |
4469 | { | |
4470 | enum maple_type mt; | |
4471 | int offset, level; | |
4472 | void __rcu **slots; | |
4473 | struct maple_node *node; | |
54a611b6 | 4474 | unsigned long *pivots; |
dd9a8513 | 4475 | unsigned long max; |
54a611b6 | 4476 | |
dd9a8513 LH |
4477 | node = mas_mn(mas); |
4478 | if (!mas->min) | |
4479 | goto no_entry; | |
4480 | ||
4481 | max = mas->min - 1; | |
4482 | if (max < min) | |
4483 | goto no_entry; | |
54a611b6 LH |
4484 | |
4485 | level = 0; | |
4486 | do { | |
54a611b6 LH |
4487 | if (ma_is_root(node)) |
4488 | goto no_entry; | |
4489 | ||
4490 | /* Walk up. */ | |
4491 | if (unlikely(mas_ascend(mas))) | |
4492 | return 1; | |
4493 | offset = mas->offset; | |
4494 | level++; | |
dd9a8513 | 4495 | node = mas_mn(mas); |
54a611b6 LH |
4496 | } while (!offset); |
4497 | ||
4498 | offset--; | |
4499 | mt = mte_node_type(mas->node); | |
54a611b6 LH |
4500 | while (level > 1) { |
4501 | level--; | |
dd9a8513 LH |
4502 | slots = ma_slots(node, mt); |
4503 | mas->node = mas_slot(mas, slots, offset); | |
54a611b6 LH |
4504 | if (unlikely(ma_dead_node(node))) |
4505 | return 1; | |
4506 | ||
54a611b6 LH |
4507 | mt = mte_node_type(mas->node); |
4508 | node = mas_mn(mas); | |
54a611b6 | 4509 | pivots = ma_pivots(node, mt); |
dd9a8513 | 4510 | offset = ma_data_end(node, mt, pivots, max); |
39d0bd86 LH |
4511 | if (unlikely(ma_dead_node(node))) |
4512 | return 1; | |
54a611b6 LH |
4513 | } |
4514 | ||
dd9a8513 | 4515 | slots = ma_slots(node, mt); |
54a611b6 | 4516 | mas->node = mas_slot(mas, slots, offset); |
dd9a8513 | 4517 | pivots = ma_pivots(node, mt); |
54a611b6 LH |
4518 | if (unlikely(ma_dead_node(node))) |
4519 | return 1; | |
4520 | ||
dd9a8513 LH |
4521 | if (likely(offset)) |
4522 | mas->min = pivots[offset - 1] + 1; | |
4523 | mas->max = max; | |
54a611b6 LH |
4524 | mas->offset = mas_data_end(mas); |
4525 | if (unlikely(mte_dead_node(mas->node))) | |
4526 | return 1; | |
4527 | ||
4528 | return 0; | |
4529 | ||
54a611b6 LH |
4530 | no_entry: |
4531 | if (unlikely(ma_dead_node(node))) | |
4532 | return 1; | |
4533 | ||
4534 | mas->node = MAS_NONE; | |
4535 | return 0; | |
4536 | } | |
4537 | ||
dd9a8513 LH |
4538 | /* |
4539 | * mas_prev_slot() - Get the entry in the previous slot | |
4540 | * | |
4541 | * @mas: The maple state | |
4542 | * @max: The minimum starting range | |
4543 | * | |
4544 | * Return: The entry in the previous slot which is possibly NULL | |
4545 | */ | |
4546 | static void *mas_prev_slot(struct ma_state *mas, unsigned long min, bool empty) | |
4547 | { | |
4548 | void *entry; | |
4549 | void __rcu **slots; | |
4550 | unsigned long pivot; | |
4551 | enum maple_type type; | |
4552 | unsigned long *pivots; | |
4553 | struct maple_node *node; | |
4554 | unsigned long save_point = mas->index; | |
4555 | ||
4556 | retry: | |
4557 | node = mas_mn(mas); | |
4558 | type = mte_node_type(mas->node); | |
4559 | pivots = ma_pivots(node, type); | |
4560 | if (unlikely(mas_rewalk_if_dead(mas, node, save_point))) | |
4561 | goto retry; | |
4562 | ||
4563 | again: | |
4564 | if (mas->min <= min) { | |
4565 | pivot = mas_safe_min(mas, pivots, mas->offset); | |
4566 | ||
4567 | if (unlikely(mas_rewalk_if_dead(mas, node, save_point))) | |
4568 | goto retry; | |
4569 | ||
4570 | if (pivot <= min) | |
4571 | return NULL; | |
4572 | } | |
4573 | ||
4574 | if (likely(mas->offset)) { | |
4575 | mas->offset--; | |
4576 | mas->last = mas->index - 1; | |
4577 | mas->index = mas_safe_min(mas, pivots, mas->offset); | |
4578 | } else { | |
4579 | if (mas_prev_node(mas, min)) { | |
4580 | mas_rewalk(mas, save_point); | |
4581 | goto retry; | |
4582 | } | |
4583 | ||
4584 | if (mas_is_none(mas)) | |
4585 | return NULL; | |
4586 | ||
4587 | mas->last = mas->max; | |
4588 | node = mas_mn(mas); | |
4589 | type = mte_node_type(mas->node); | |
4590 | pivots = ma_pivots(node, type); | |
4591 | mas->index = pivots[mas->offset - 1] + 1; | |
4592 | } | |
4593 | ||
4594 | slots = ma_slots(node, type); | |
4595 | entry = mas_slot(mas, slots, mas->offset); | |
4596 | if (unlikely(mas_rewalk_if_dead(mas, node, save_point))) | |
4597 | goto retry; | |
4598 | ||
4599 | if (likely(entry)) | |
4600 | return entry; | |
4601 | ||
4602 | if (!empty) | |
4603 | goto again; | |
4604 | ||
4605 | return entry; | |
4606 | } | |
4607 | ||
54a611b6 LH |
4608 | /* |
4609 | * mas_next_node() - Get the next node at the same level in the tree. | |
4610 | * @mas: The maple state | |
4611 | * @max: The maximum pivot value to check. | |
4612 | * | |
4613 | * The next value will be mas->node[mas->offset] or MAS_NONE. | |
4614 | * Return: 1 on dead node, 0 otherwise. | |
4615 | */ | |
4616 | static inline int mas_next_node(struct ma_state *mas, struct maple_node *node, | |
4617 | unsigned long max) | |
4618 | { | |
fff4a58c | 4619 | unsigned long min; |
54a611b6 LH |
4620 | unsigned long *pivots; |
4621 | struct maple_enode *enode; | |
4622 | int level = 0; | |
39d0bd86 | 4623 | unsigned char node_end; |
54a611b6 LH |
4624 | enum maple_type mt; |
4625 | void __rcu **slots; | |
4626 | ||
4627 | if (mas->max >= max) | |
4628 | goto no_entry; | |
4629 | ||
fff4a58c | 4630 | min = mas->max + 1; |
54a611b6 LH |
4631 | level = 0; |
4632 | do { | |
4633 | if (ma_is_root(node)) | |
4634 | goto no_entry; | |
4635 | ||
fff4a58c | 4636 | /* Walk up. */ |
54a611b6 LH |
4637 | if (unlikely(mas_ascend(mas))) |
4638 | return 1; | |
4639 | ||
54a611b6 LH |
4640 | level++; |
4641 | node = mas_mn(mas); | |
4642 | mt = mte_node_type(mas->node); | |
4643 | pivots = ma_pivots(node, mt); | |
39d0bd86 LH |
4644 | node_end = ma_data_end(node, mt, pivots, mas->max); |
4645 | if (unlikely(ma_dead_node(node))) | |
4646 | return 1; | |
4647 | ||
fff4a58c | 4648 | } while (unlikely(mas->offset == node_end)); |
54a611b6 LH |
4649 | |
4650 | slots = ma_slots(node, mt); | |
fff4a58c LH |
4651 | mas->offset++; |
4652 | enode = mas_slot(mas, slots, mas->offset); | |
4653 | if (unlikely(ma_dead_node(node))) | |
4654 | return 1; | |
54a611b6 | 4655 | |
fff4a58c LH |
4656 | if (level > 1) |
4657 | mas->offset = 0; | |
4658 | ||
4659 | while (unlikely(level > 1)) { | |
54a611b6 | 4660 | level--; |
fff4a58c | 4661 | mas->node = enode; |
54a611b6 LH |
4662 | node = mas_mn(mas); |
4663 | mt = mte_node_type(mas->node); | |
4664 | slots = ma_slots(node, mt); | |
fff4a58c | 4665 | enode = mas_slot(mas, slots, 0); |
39d0bd86 LH |
4666 | if (unlikely(ma_dead_node(node))) |
4667 | return 1; | |
54a611b6 LH |
4668 | } |
4669 | ||
fff4a58c LH |
4670 | if (!mas->offset) |
4671 | pivots = ma_pivots(node, mt); | |
4672 | ||
4673 | mas->max = mas_safe_pivot(mas, pivots, mas->offset, mt); | |
54a611b6 LH |
4674 | if (unlikely(ma_dead_node(node))) |
4675 | return 1; | |
4676 | ||
4677 | mas->node = enode; | |
4678 | mas->min = min; | |
54a611b6 LH |
4679 | return 0; |
4680 | ||
4681 | no_entry: | |
4682 | if (unlikely(ma_dead_node(node))) | |
4683 | return 1; | |
4684 | ||
4685 | mas->node = MAS_NONE; | |
4686 | return 0; | |
4687 | } | |
4688 | ||
4689 | /* | |
fff4a58c | 4690 | * mas_next_slot() - Get the entry in the next slot |
54a611b6 | 4691 | * |
fff4a58c LH |
4692 | * @mas: The maple state |
4693 | * @max: The maximum starting range | |
4694 | * @empty: Can be empty | |
54a611b6 | 4695 | * |
fff4a58c | 4696 | * Return: The entry in the next slot which is possibly NULL |
54a611b6 | 4697 | */ |
fff4a58c | 4698 | static void *mas_next_slot(struct ma_state *mas, unsigned long max, bool empty) |
54a611b6 | 4699 | { |
54a611b6 | 4700 | void __rcu **slots; |
fff4a58c LH |
4701 | unsigned long *pivots; |
4702 | unsigned long pivot; | |
4703 | enum maple_type type; | |
4704 | struct maple_node *node; | |
4705 | unsigned char data_end; | |
4706 | unsigned long save_point = mas->last; | |
54a611b6 LH |
4707 | void *entry; |
4708 | ||
fff4a58c LH |
4709 | retry: |
4710 | node = mas_mn(mas); | |
4711 | type = mte_node_type(mas->node); | |
39d0bd86 | 4712 | pivots = ma_pivots(node, type); |
fff4a58c LH |
4713 | data_end = ma_data_end(node, type, pivots, mas->max); |
4714 | if (unlikely(mas_rewalk_if_dead(mas, node, save_point))) | |
4715 | goto retry; | |
54a611b6 | 4716 | |
fff4a58c LH |
4717 | again: |
4718 | if (mas->max >= max) { | |
4719 | if (likely(mas->offset < data_end)) | |
4720 | pivot = pivots[mas->offset]; | |
4721 | else | |
4722 | return NULL; /* must be mas->max */ | |
54a611b6 | 4723 | |
fff4a58c LH |
4724 | if (unlikely(mas_rewalk_if_dead(mas, node, save_point))) |
4725 | goto retry; | |
54a611b6 LH |
4726 | |
4727 | if (pivot >= max) | |
4728 | return NULL; | |
fff4a58c | 4729 | } |
54a611b6 | 4730 | |
fff4a58c LH |
4731 | if (likely(mas->offset < data_end)) { |
4732 | mas->index = pivots[mas->offset] + 1; | |
4733 | mas->offset++; | |
4734 | if (likely(mas->offset < data_end)) | |
4735 | mas->last = pivots[mas->offset]; | |
4736 | else | |
4737 | mas->last = mas->max; | |
4738 | } else { | |
4739 | if (mas_next_node(mas, node, max)) { | |
4740 | mas_rewalk(mas, save_point); | |
4741 | goto retry; | |
4742 | } | |
4743 | ||
4744 | if (mas_is_none(mas)) | |
ca80f610 LH |
4745 | return NULL; |
4746 | ||
fff4a58c LH |
4747 | mas->offset = 0; |
4748 | mas->index = mas->min; | |
4749 | node = mas_mn(mas); | |
4750 | type = mte_node_type(mas->node); | |
4751 | pivots = ma_pivots(node, type); | |
4752 | mas->last = pivots[0]; | |
54a611b6 LH |
4753 | } |
4754 | ||
fff4a58c LH |
4755 | slots = ma_slots(node, type); |
4756 | entry = mt_slot(mas->tree, slots, mas->offset); | |
4757 | if (unlikely(mas_rewalk_if_dead(mas, node, save_point))) | |
4758 | goto retry; | |
54a611b6 | 4759 | |
fff4a58c LH |
4760 | if (entry) |
4761 | return entry; | |
54a611b6 | 4762 | |
fff4a58c LH |
4763 | if (!empty) { |
4764 | if (!mas->offset) | |
4765 | data_end = 2; | |
4766 | goto again; | |
4767 | } | |
4768 | ||
4769 | return entry; | |
54a611b6 LH |
4770 | } |
4771 | ||
4772 | /* | |
4773 | * mas_next_entry() - Internal function to get the next entry. | |
4774 | * @mas: The maple state | |
4775 | * @limit: The maximum range start. | |
4776 | * | |
4777 | * Set the @mas->node to the next entry and the range_start to | |
4778 | * the beginning value for the entry. Does not check beyond @limit. | |
4779 | * Sets @mas->index and @mas->last to the limit if it is hit. | |
4780 | * Restarts on dead nodes. | |
4781 | * | |
4782 | * Return: the next entry or %NULL. | |
4783 | */ | |
4784 | static inline void *mas_next_entry(struct ma_state *mas, unsigned long limit) | |
4785 | { | |
ca80f610 | 4786 | if (mas->last >= limit) |
50e81c82 | 4787 | return NULL; |
ca80f610 | 4788 | |
6169b553 | 4789 | return mas_next_slot(mas, limit, false); |
54a611b6 LH |
4790 | } |
4791 | ||
54a611b6 LH |
4792 | /* |
4793 | * mas_rev_awalk() - Internal function. Reverse allocation walk. Find the | |
4794 | * highest gap address of a given size in a given node and descend. | |
4795 | * @mas: The maple state | |
4796 | * @size: The needed size. | |
4797 | * | |
4798 | * Return: True if found in a leaf, false otherwise. | |
4799 | * | |
4800 | */ | |
fad8e429 LH |
4801 | static bool mas_rev_awalk(struct ma_state *mas, unsigned long size, |
4802 | unsigned long *gap_min, unsigned long *gap_max) | |
54a611b6 LH |
4803 | { |
4804 | enum maple_type type = mte_node_type(mas->node); | |
4805 | struct maple_node *node = mas_mn(mas); | |
4806 | unsigned long *pivots, *gaps; | |
4807 | void __rcu **slots; | |
4808 | unsigned long gap = 0; | |
7327e811 | 4809 | unsigned long max, min; |
54a611b6 LH |
4810 | unsigned char offset; |
4811 | ||
4812 | if (unlikely(mas_is_err(mas))) | |
4813 | return true; | |
4814 | ||
4815 | if (ma_is_dense(type)) { | |
4816 | /* dense nodes. */ | |
4817 | mas->offset = (unsigned char)(mas->index - mas->min); | |
4818 | return true; | |
4819 | } | |
4820 | ||
4821 | pivots = ma_pivots(node, type); | |
4822 | slots = ma_slots(node, type); | |
4823 | gaps = ma_gaps(node, type); | |
4824 | offset = mas->offset; | |
4825 | min = mas_safe_min(mas, pivots, offset); | |
4826 | /* Skip out of bounds. */ | |
4827 | while (mas->last < min) | |
4828 | min = mas_safe_min(mas, pivots, --offset); | |
4829 | ||
4830 | max = mas_safe_pivot(mas, pivots, offset, type); | |
7327e811 | 4831 | while (mas->index <= max) { |
54a611b6 LH |
4832 | gap = 0; |
4833 | if (gaps) | |
4834 | gap = gaps[offset]; | |
4835 | else if (!mas_slot(mas, slots, offset)) | |
4836 | gap = max - min + 1; | |
4837 | ||
4838 | if (gap) { | |
4839 | if ((size <= gap) && (size <= mas->last - min + 1)) | |
4840 | break; | |
4841 | ||
4842 | if (!gaps) { | |
4843 | /* Skip the next slot, it cannot be a gap. */ | |
4844 | if (offset < 2) | |
4845 | goto ascend; | |
4846 | ||
4847 | offset -= 2; | |
4848 | max = pivots[offset]; | |
4849 | min = mas_safe_min(mas, pivots, offset); | |
4850 | continue; | |
4851 | } | |
4852 | } | |
4853 | ||
4854 | if (!offset) | |
4855 | goto ascend; | |
4856 | ||
4857 | offset--; | |
4858 | max = min - 1; | |
4859 | min = mas_safe_min(mas, pivots, offset); | |
4860 | } | |
4861 | ||
7327e811 LH |
4862 | if (unlikely((mas->index > max) || (size - 1 > max - mas->index))) |
4863 | goto no_space; | |
54a611b6 LH |
4864 | |
4865 | if (unlikely(ma_is_leaf(type))) { | |
4866 | mas->offset = offset; | |
fad8e429 LH |
4867 | *gap_min = min; |
4868 | *gap_max = min + gap - 1; | |
54a611b6 LH |
4869 | return true; |
4870 | } | |
4871 | ||
4872 | /* descend, only happens under lock. */ | |
4873 | mas->node = mas_slot(mas, slots, offset); | |
4874 | mas->min = min; | |
4875 | mas->max = max; | |
4876 | mas->offset = mas_data_end(mas); | |
4877 | return false; | |
4878 | ||
4879 | ascend: | |
7327e811 LH |
4880 | if (!mte_is_root(mas->node)) |
4881 | return false; | |
54a611b6 | 4882 | |
7327e811 LH |
4883 | no_space: |
4884 | mas_set_err(mas, -EBUSY); | |
54a611b6 LH |
4885 | return false; |
4886 | } | |
4887 | ||
4888 | static inline bool mas_anode_descend(struct ma_state *mas, unsigned long size) | |
4889 | { | |
4890 | enum maple_type type = mte_node_type(mas->node); | |
4891 | unsigned long pivot, min, gap = 0; | |
06e8fd99 LH |
4892 | unsigned char offset, data_end; |
4893 | unsigned long *gaps, *pivots; | |
4894 | void __rcu **slots; | |
4895 | struct maple_node *node; | |
54a611b6 LH |
4896 | bool found = false; |
4897 | ||
4898 | if (ma_is_dense(type)) { | |
4899 | mas->offset = (unsigned char)(mas->index - mas->min); | |
4900 | return true; | |
4901 | } | |
4902 | ||
06e8fd99 LH |
4903 | node = mas_mn(mas); |
4904 | pivots = ma_pivots(node, type); | |
4905 | slots = ma_slots(node, type); | |
4906 | gaps = ma_gaps(node, type); | |
54a611b6 | 4907 | offset = mas->offset; |
54a611b6 | 4908 | min = mas_safe_min(mas, pivots, offset); |
06e8fd99 LH |
4909 | data_end = ma_data_end(node, type, pivots, mas->max); |
4910 | for (; offset <= data_end; offset++) { | |
29b2681f | 4911 | pivot = mas_safe_pivot(mas, pivots, offset, type); |
54a611b6 LH |
4912 | |
4913 | /* Not within lower bounds */ | |
4914 | if (mas->index > pivot) | |
4915 | goto next_slot; | |
4916 | ||
4917 | if (gaps) | |
4918 | gap = gaps[offset]; | |
4919 | else if (!mas_slot(mas, slots, offset)) | |
4920 | gap = min(pivot, mas->last) - max(mas->index, min) + 1; | |
4921 | else | |
4922 | goto next_slot; | |
4923 | ||
4924 | if (gap >= size) { | |
4925 | if (ma_is_leaf(type)) { | |
4926 | found = true; | |
4927 | goto done; | |
4928 | } | |
4929 | if (mas->index <= pivot) { | |
4930 | mas->node = mas_slot(mas, slots, offset); | |
4931 | mas->min = min; | |
4932 | mas->max = pivot; | |
4933 | offset = 0; | |
54a611b6 LH |
4934 | break; |
4935 | } | |
4936 | } | |
4937 | next_slot: | |
4938 | min = pivot + 1; | |
4939 | if (mas->last <= pivot) { | |
4940 | mas_set_err(mas, -EBUSY); | |
4941 | return true; | |
4942 | } | |
4943 | } | |
4944 | ||
4945 | if (mte_is_root(mas->node)) | |
4946 | found = true; | |
4947 | done: | |
4948 | mas->offset = offset; | |
4949 | return found; | |
4950 | } | |
4951 | ||
4952 | /** | |
4953 | * mas_walk() - Search for @mas->index in the tree. | |
4954 | * @mas: The maple state. | |
4955 | * | |
4956 | * mas->index and mas->last will be set to the range if there is a value. If | |
4957 | * mas->node is MAS_NONE, reset to MAS_START. | |
4958 | * | |
4959 | * Return: the entry at the location or %NULL. | |
4960 | */ | |
4961 | void *mas_walk(struct ma_state *mas) | |
4962 | { | |
4963 | void *entry; | |
4964 | ||
6b23a290 LH |
4965 | if (mas_is_none(mas) || mas_is_paused(mas) || mas_is_ptr(mas)) |
4966 | mas->node = MAS_START; | |
54a611b6 LH |
4967 | retry: |
4968 | entry = mas_state_walk(mas); | |
6b23a290 | 4969 | if (mas_is_start(mas)) { |
54a611b6 | 4970 | goto retry; |
6b23a290 LH |
4971 | } else if (mas_is_none(mas)) { |
4972 | mas->index = 0; | |
4973 | mas->last = ULONG_MAX; | |
4974 | } else if (mas_is_ptr(mas)) { | |
54a611b6 LH |
4975 | if (!mas->index) { |
4976 | mas->last = 0; | |
6b23a290 | 4977 | return entry; |
54a611b6 | 4978 | } |
54a611b6 | 4979 | |
6b23a290 | 4980 | mas->index = 1; |
54a611b6 | 4981 | mas->last = ULONG_MAX; |
6b23a290 LH |
4982 | mas->node = MAS_NONE; |
4983 | return NULL; | |
54a611b6 LH |
4984 | } |
4985 | ||
4986 | return entry; | |
4987 | } | |
120b1162 | 4988 | EXPORT_SYMBOL_GPL(mas_walk); |
54a611b6 LH |
4989 | |
4990 | static inline bool mas_rewind_node(struct ma_state *mas) | |
4991 | { | |
4992 | unsigned char slot; | |
4993 | ||
4994 | do { | |
4995 | if (mte_is_root(mas->node)) { | |
4996 | slot = mas->offset; | |
4997 | if (!slot) | |
4998 | return false; | |
4999 | } else { | |
5000 | mas_ascend(mas); | |
5001 | slot = mas->offset; | |
5002 | } | |
5003 | } while (!slot); | |
5004 | ||
5005 | mas->offset = --slot; | |
5006 | return true; | |
5007 | } | |
5008 | ||
5009 | /* | |
5010 | * mas_skip_node() - Internal function. Skip over a node. | |
5011 | * @mas: The maple state. | |
5012 | * | |
5013 | * Return: true if there is another node, false otherwise. | |
5014 | */ | |
5015 | static inline bool mas_skip_node(struct ma_state *mas) | |
5016 | { | |
0fa99fdf LH |
5017 | if (mas_is_err(mas)) |
5018 | return false; | |
54a611b6 | 5019 | |
54a611b6 LH |
5020 | do { |
5021 | if (mte_is_root(mas->node)) { | |
0fa99fdf | 5022 | if (mas->offset >= mas_data_end(mas)) { |
54a611b6 LH |
5023 | mas_set_err(mas, -EBUSY); |
5024 | return false; | |
5025 | } | |
5026 | } else { | |
5027 | mas_ascend(mas); | |
54a611b6 | 5028 | } |
0fa99fdf | 5029 | } while (mas->offset >= mas_data_end(mas)); |
54a611b6 | 5030 | |
0fa99fdf | 5031 | mas->offset++; |
54a611b6 LH |
5032 | return true; |
5033 | } | |
5034 | ||
5035 | /* | |
5036 | * mas_awalk() - Allocation walk. Search from low address to high, for a gap of | |
5037 | * @size | |
5038 | * @mas: The maple state | |
5039 | * @size: The size of the gap required | |
5040 | * | |
5041 | * Search between @mas->index and @mas->last for a gap of @size. | |
5042 | */ | |
5043 | static inline void mas_awalk(struct ma_state *mas, unsigned long size) | |
5044 | { | |
5045 | struct maple_enode *last = NULL; | |
5046 | ||
5047 | /* | |
5048 | * There are 4 options: | |
5049 | * go to child (descend) | |
5050 | * go back to parent (ascend) | |
5051 | * no gap found. (return, slot == MAPLE_NODE_SLOTS) | |
5052 | * found the gap. (return, slot != MAPLE_NODE_SLOTS) | |
5053 | */ | |
5054 | while (!mas_is_err(mas) && !mas_anode_descend(mas, size)) { | |
5055 | if (last == mas->node) | |
5056 | mas_skip_node(mas); | |
5057 | else | |
5058 | last = mas->node; | |
5059 | } | |
5060 | } | |
5061 | ||
54a611b6 LH |
5062 | /* |
5063 | * mas_sparse_area() - Internal function. Return upper or lower limit when | |
5064 | * searching for a gap in an empty tree. | |
5065 | * @mas: The maple state | |
5066 | * @min: the minimum range | |
5067 | * @max: The maximum range | |
5068 | * @size: The size of the gap | |
5069 | * @fwd: Searching forward or back | |
5070 | */ | |
29ad6bb3 | 5071 | static inline int mas_sparse_area(struct ma_state *mas, unsigned long min, |
54a611b6 LH |
5072 | unsigned long max, unsigned long size, bool fwd) |
5073 | { | |
29ad6bb3 PZ |
5074 | if (!unlikely(mas_is_none(mas)) && min == 0) { |
5075 | min++; | |
5076 | /* | |
5077 | * At this time, min is increased, we need to recheck whether | |
5078 | * the size is satisfied. | |
5079 | */ | |
5080 | if (min > max || max - min + 1 < size) | |
5081 | return -EBUSY; | |
5082 | } | |
54a611b6 LH |
5083 | /* mas_is_ptr */ |
5084 | ||
54a611b6 | 5085 | if (fwd) { |
29ad6bb3 PZ |
5086 | mas->index = min; |
5087 | mas->last = min + size - 1; | |
5088 | } else { | |
5089 | mas->last = max; | |
5090 | mas->index = max - size + 1; | |
54a611b6 | 5091 | } |
29ad6bb3 | 5092 | return 0; |
54a611b6 LH |
5093 | } |
5094 | ||
5095 | /* | |
5096 | * mas_empty_area() - Get the lowest address within the range that is | |
5097 | * sufficient for the size requested. | |
5098 | * @mas: The maple state | |
5099 | * @min: The lowest value of the range | |
5100 | * @max: The highest value of the range | |
5101 | * @size: The size needed | |
5102 | */ | |
5103 | int mas_empty_area(struct ma_state *mas, unsigned long min, | |
5104 | unsigned long max, unsigned long size) | |
5105 | { | |
5106 | unsigned char offset; | |
5107 | unsigned long *pivots; | |
5108 | enum maple_type mt; | |
5109 | ||
ba997212 LH |
5110 | if (min > max) |
5111 | return -EINVAL; | |
5112 | ||
5113 | if (size == 0 || max - min < size - 1) | |
fad8e429 LH |
5114 | return -EINVAL; |
5115 | ||
54a611b6 LH |
5116 | if (mas_is_start(mas)) |
5117 | mas_start(mas); | |
5118 | else if (mas->offset >= 2) | |
5119 | mas->offset -= 2; | |
5120 | else if (!mas_skip_node(mas)) | |
5121 | return -EBUSY; | |
5122 | ||
5123 | /* Empty set */ | |
29ad6bb3 PZ |
5124 | if (mas_is_none(mas) || mas_is_ptr(mas)) |
5125 | return mas_sparse_area(mas, min, max, size, true); | |
54a611b6 LH |
5126 | |
5127 | /* The start of the window can only be within these values */ | |
5128 | mas->index = min; | |
5129 | mas->last = max; | |
5130 | mas_awalk(mas, size); | |
5131 | ||
5132 | if (unlikely(mas_is_err(mas))) | |
5133 | return xa_err(mas->node); | |
5134 | ||
5135 | offset = mas->offset; | |
5136 | if (unlikely(offset == MAPLE_NODE_SLOTS)) | |
5137 | return -EBUSY; | |
5138 | ||
5139 | mt = mte_node_type(mas->node); | |
5140 | pivots = ma_pivots(mas_mn(mas), mt); | |
0257d990 PZ |
5141 | min = mas_safe_min(mas, pivots, offset); |
5142 | if (mas->index < min) | |
5143 | mas->index = min; | |
54a611b6 LH |
5144 | mas->last = mas->index + size - 1; |
5145 | return 0; | |
5146 | } | |
120b1162 | 5147 | EXPORT_SYMBOL_GPL(mas_empty_area); |
54a611b6 LH |
5148 | |
5149 | /* | |
5150 | * mas_empty_area_rev() - Get the highest address within the range that is | |
5151 | * sufficient for the size requested. | |
5152 | * @mas: The maple state | |
5153 | * @min: The lowest value of the range | |
5154 | * @max: The highest value of the range | |
5155 | * @size: The size needed | |
5156 | */ | |
5157 | int mas_empty_area_rev(struct ma_state *mas, unsigned long min, | |
5158 | unsigned long max, unsigned long size) | |
5159 | { | |
5160 | struct maple_enode *last = mas->node; | |
5161 | ||
ba997212 LH |
5162 | if (min > max) |
5163 | return -EINVAL; | |
5164 | ||
5165 | if (size == 0 || max - min < size - 1) | |
fad8e429 LH |
5166 | return -EINVAL; |
5167 | ||
54a611b6 LH |
5168 | if (mas_is_start(mas)) { |
5169 | mas_start(mas); | |
5170 | mas->offset = mas_data_end(mas); | |
5171 | } else if (mas->offset >= 2) { | |
5172 | mas->offset -= 2; | |
5173 | } else if (!mas_rewind_node(mas)) { | |
5174 | return -EBUSY; | |
5175 | } | |
5176 | ||
5177 | /* Empty set. */ | |
29ad6bb3 PZ |
5178 | if (mas_is_none(mas) || mas_is_ptr(mas)) |
5179 | return mas_sparse_area(mas, min, max, size, false); | |
54a611b6 LH |
5180 | |
5181 | /* The start of the window can only be within these values. */ | |
5182 | mas->index = min; | |
5183 | mas->last = max; | |
5184 | ||
fad8e429 | 5185 | while (!mas_rev_awalk(mas, size, &min, &max)) { |
54a611b6 LH |
5186 | if (last == mas->node) { |
5187 | if (!mas_rewind_node(mas)) | |
5188 | return -EBUSY; | |
5189 | } else { | |
5190 | last = mas->node; | |
5191 | } | |
5192 | } | |
5193 | ||
5194 | if (mas_is_err(mas)) | |
5195 | return xa_err(mas->node); | |
5196 | ||
5197 | if (unlikely(mas->offset == MAPLE_NODE_SLOTS)) | |
5198 | return -EBUSY; | |
5199 | ||
54a611b6 | 5200 | /* Trim the upper limit to the max. */ |
ba997212 | 5201 | if (max < mas->last) |
fad8e429 | 5202 | mas->last = max; |
54a611b6 LH |
5203 | |
5204 | mas->index = mas->last - size + 1; | |
5205 | return 0; | |
5206 | } | |
120b1162 | 5207 | EXPORT_SYMBOL_GPL(mas_empty_area_rev); |
54a611b6 | 5208 | |
54a611b6 | 5209 | /* |
790e1fa8 | 5210 | * mte_dead_leaves() - Mark all leaves of a node as dead. |
54a611b6 LH |
5211 | * @mas: The maple state |
5212 | * @slots: Pointer to the slot array | |
2e5b4921 | 5213 | * @type: The maple node type |
54a611b6 LH |
5214 | * |
5215 | * Must hold the write lock. | |
5216 | * | |
5217 | * Return: The number of leaves marked as dead. | |
5218 | */ | |
5219 | static inline | |
790e1fa8 LH |
5220 | unsigned char mte_dead_leaves(struct maple_enode *enode, struct maple_tree *mt, |
5221 | void __rcu **slots) | |
54a611b6 LH |
5222 | { |
5223 | struct maple_node *node; | |
5224 | enum maple_type type; | |
5225 | void *entry; | |
5226 | int offset; | |
5227 | ||
790e1fa8 LH |
5228 | for (offset = 0; offset < mt_slot_count(enode); offset++) { |
5229 | entry = mt_slot(mt, slots, offset); | |
54a611b6 LH |
5230 | type = mte_node_type(entry); |
5231 | node = mte_to_node(entry); | |
5232 | /* Use both node and type to catch LE & BE metadata */ | |
5233 | if (!node || !type) | |
5234 | break; | |
5235 | ||
5236 | mte_set_node_dead(entry); | |
54a611b6 LH |
5237 | node->type = type; |
5238 | rcu_assign_pointer(slots[offset], node); | |
5239 | } | |
5240 | ||
5241 | return offset; | |
5242 | } | |
5243 | ||
790e1fa8 LH |
5244 | /** |
5245 | * mte_dead_walk() - Walk down a dead tree to just before the leaves | |
5246 | * @enode: The maple encoded node | |
5247 | * @offset: The starting offset | |
5248 | * | |
5249 | * Note: This can only be used from the RCU callback context. | |
5250 | */ | |
5251 | static void __rcu **mte_dead_walk(struct maple_enode **enode, unsigned char offset) | |
54a611b6 LH |
5252 | { |
5253 | struct maple_node *node, *next; | |
5254 | void __rcu **slots = NULL; | |
5255 | ||
790e1fa8 | 5256 | next = mte_to_node(*enode); |
54a611b6 | 5257 | do { |
790e1fa8 LH |
5258 | *enode = ma_enode_ptr(next); |
5259 | node = mte_to_node(*enode); | |
54a611b6 | 5260 | slots = ma_slots(node, node->type); |
790e1fa8 LH |
5261 | next = rcu_dereference_protected(slots[offset], |
5262 | lock_is_held(&rcu_callback_map)); | |
54a611b6 LH |
5263 | offset = 0; |
5264 | } while (!ma_is_leaf(next->type)); | |
5265 | ||
5266 | return slots; | |
5267 | } | |
5268 | ||
790e1fa8 LH |
5269 | /** |
5270 | * mt_free_walk() - Walk & free a tree in the RCU callback context | |
5271 | * @head: The RCU head that's within the node. | |
5272 | * | |
5273 | * Note: This can only be used from the RCU callback context. | |
5274 | */ | |
54a611b6 LH |
5275 | static void mt_free_walk(struct rcu_head *head) |
5276 | { | |
5277 | void __rcu **slots; | |
5278 | struct maple_node *node, *start; | |
790e1fa8 | 5279 | struct maple_enode *enode; |
54a611b6 LH |
5280 | unsigned char offset; |
5281 | enum maple_type type; | |
54a611b6 LH |
5282 | |
5283 | node = container_of(head, struct maple_node, rcu); | |
5284 | ||
5285 | if (ma_is_leaf(node->type)) | |
5286 | goto free_leaf; | |
5287 | ||
54a611b6 | 5288 | start = node; |
790e1fa8 LH |
5289 | enode = mt_mk_node(node, node->type); |
5290 | slots = mte_dead_walk(&enode, 0); | |
5291 | node = mte_to_node(enode); | |
54a611b6 LH |
5292 | do { |
5293 | mt_free_bulk(node->slot_len, slots); | |
5294 | offset = node->parent_slot + 1; | |
790e1fa8 LH |
5295 | enode = node->piv_parent; |
5296 | if (mte_to_node(enode) == node) | |
5297 | goto free_leaf; | |
5298 | ||
5299 | type = mte_node_type(enode); | |
5300 | slots = ma_slots(mte_to_node(enode), type); | |
5301 | if ((offset < mt_slots[type]) && | |
5302 | rcu_dereference_protected(slots[offset], | |
5303 | lock_is_held(&rcu_callback_map))) | |
5304 | slots = mte_dead_walk(&enode, offset); | |
5305 | node = mte_to_node(enode); | |
54a611b6 LH |
5306 | } while ((node != start) || (node->slot_len < offset)); |
5307 | ||
5308 | slots = ma_slots(node, node->type); | |
5309 | mt_free_bulk(node->slot_len, slots); | |
5310 | ||
54a611b6 LH |
5311 | free_leaf: |
5312 | mt_free_rcu(&node->rcu); | |
5313 | } | |
5314 | ||
790e1fa8 LH |
5315 | static inline void __rcu **mte_destroy_descend(struct maple_enode **enode, |
5316 | struct maple_tree *mt, struct maple_enode *prev, unsigned char offset) | |
54a611b6 LH |
5317 | { |
5318 | struct maple_node *node; | |
790e1fa8 | 5319 | struct maple_enode *next = *enode; |
54a611b6 | 5320 | void __rcu **slots = NULL; |
790e1fa8 LH |
5321 | enum maple_type type; |
5322 | unsigned char next_offset = 0; | |
54a611b6 LH |
5323 | |
5324 | do { | |
790e1fa8 LH |
5325 | *enode = next; |
5326 | node = mte_to_node(*enode); | |
5327 | type = mte_node_type(*enode); | |
5328 | slots = ma_slots(node, type); | |
5329 | next = mt_slot_locked(mt, slots, next_offset); | |
54a611b6 | 5330 | if ((mte_dead_node(next))) |
790e1fa8 | 5331 | next = mt_slot_locked(mt, slots, ++next_offset); |
54a611b6 | 5332 | |
790e1fa8 LH |
5333 | mte_set_node_dead(*enode); |
5334 | node->type = type; | |
54a611b6 LH |
5335 | node->piv_parent = prev; |
5336 | node->parent_slot = offset; | |
790e1fa8 LH |
5337 | offset = next_offset; |
5338 | next_offset = 0; | |
5339 | prev = *enode; | |
54a611b6 LH |
5340 | } while (!mte_is_leaf(next)); |
5341 | ||
5342 | return slots; | |
5343 | } | |
5344 | ||
790e1fa8 | 5345 | static void mt_destroy_walk(struct maple_enode *enode, struct maple_tree *mt, |
54a611b6 LH |
5346 | bool free) |
5347 | { | |
5348 | void __rcu **slots; | |
5349 | struct maple_node *node = mte_to_node(enode); | |
5350 | struct maple_enode *start; | |
54a611b6 | 5351 | |
2e5b4921 LH |
5352 | if (mte_is_leaf(enode)) { |
5353 | node->type = mte_node_type(enode); | |
54a611b6 | 5354 | goto free_leaf; |
2e5b4921 | 5355 | } |
54a611b6 | 5356 | |
2e5b4921 | 5357 | start = enode; |
790e1fa8 LH |
5358 | slots = mte_destroy_descend(&enode, mt, start, 0); |
5359 | node = mte_to_node(enode); // Updated in the above call. | |
54a611b6 LH |
5360 | do { |
5361 | enum maple_type type; | |
5362 | unsigned char offset; | |
5363 | struct maple_enode *parent, *tmp; | |
5364 | ||
790e1fa8 | 5365 | node->slot_len = mte_dead_leaves(enode, mt, slots); |
54a611b6 LH |
5366 | if (free) |
5367 | mt_free_bulk(node->slot_len, slots); | |
5368 | offset = node->parent_slot + 1; | |
790e1fa8 LH |
5369 | enode = node->piv_parent; |
5370 | if (mte_to_node(enode) == node) | |
5371 | goto free_leaf; | |
54a611b6 | 5372 | |
790e1fa8 LH |
5373 | type = mte_node_type(enode); |
5374 | slots = ma_slots(mte_to_node(enode), type); | |
54a611b6 LH |
5375 | if (offset >= mt_slots[type]) |
5376 | goto next; | |
5377 | ||
790e1fa8 | 5378 | tmp = mt_slot_locked(mt, slots, offset); |
54a611b6 | 5379 | if (mte_node_type(tmp) && mte_to_node(tmp)) { |
790e1fa8 LH |
5380 | parent = enode; |
5381 | enode = tmp; | |
5382 | slots = mte_destroy_descend(&enode, mt, parent, offset); | |
54a611b6 LH |
5383 | } |
5384 | next: | |
790e1fa8 LH |
5385 | node = mte_to_node(enode); |
5386 | } while (start != enode); | |
54a611b6 | 5387 | |
790e1fa8 LH |
5388 | node = mte_to_node(enode); |
5389 | node->slot_len = mte_dead_leaves(enode, mt, slots); | |
54a611b6 LH |
5390 | if (free) |
5391 | mt_free_bulk(node->slot_len, slots); | |
5392 | ||
54a611b6 LH |
5393 | free_leaf: |
5394 | if (free) | |
5395 | mt_free_rcu(&node->rcu); | |
2e5b4921 | 5396 | else |
790e1fa8 | 5397 | mt_clear_meta(mt, node, node->type); |
54a611b6 LH |
5398 | } |
5399 | ||
5400 | /* | |
5401 | * mte_destroy_walk() - Free a tree or sub-tree. | |
f942b0f0 VY |
5402 | * @enode: the encoded maple node (maple_enode) to start |
5403 | * @mt: the tree to free - needed for node types. | |
54a611b6 LH |
5404 | * |
5405 | * Must hold the write lock. | |
5406 | */ | |
5407 | static inline void mte_destroy_walk(struct maple_enode *enode, | |
5408 | struct maple_tree *mt) | |
5409 | { | |
5410 | struct maple_node *node = mte_to_node(enode); | |
5411 | ||
5412 | if (mt_in_rcu(mt)) { | |
790e1fa8 | 5413 | mt_destroy_walk(enode, mt, false); |
54a611b6 LH |
5414 | call_rcu(&node->rcu, mt_free_walk); |
5415 | } else { | |
790e1fa8 | 5416 | mt_destroy_walk(enode, mt, true); |
54a611b6 LH |
5417 | } |
5418 | } | |
5419 | ||
5420 | static void mas_wr_store_setup(struct ma_wr_state *wr_mas) | |
5421 | { | |
1202700c LH |
5422 | if (unlikely(mas_is_paused(wr_mas->mas))) |
5423 | mas_reset(wr_mas->mas); | |
5424 | ||
54a611b6 LH |
5425 | if (!mas_is_start(wr_mas->mas)) { |
5426 | if (mas_is_none(wr_mas->mas)) { | |
5427 | mas_reset(wr_mas->mas); | |
5428 | } else { | |
5429 | wr_mas->r_max = wr_mas->mas->max; | |
5430 | wr_mas->type = mte_node_type(wr_mas->mas->node); | |
5431 | if (mas_is_span_wr(wr_mas)) | |
5432 | mas_reset(wr_mas->mas); | |
5433 | } | |
5434 | } | |
54a611b6 LH |
5435 | } |
5436 | ||
5437 | /* Interface */ | |
5438 | ||
5439 | /** | |
5440 | * mas_store() - Store an @entry. | |
5441 | * @mas: The maple state. | |
5442 | * @entry: The entry to store. | |
5443 | * | |
5444 | * The @mas->index and @mas->last is used to set the range for the @entry. | |
5445 | * Note: The @mas should have pre-allocated entries to ensure there is memory to | |
5446 | * store the entry. Please see mas_expected_entries()/mas_destroy() for more details. | |
5447 | * | |
5448 | * Return: the first entry between mas->index and mas->last or %NULL. | |
5449 | */ | |
5450 | void *mas_store(struct ma_state *mas, void *entry) | |
5451 | { | |
5452 | MA_WR_STATE(wr_mas, mas, entry); | |
5453 | ||
5454 | trace_ma_write(__func__, mas, 0, entry); | |
5455 | #ifdef CONFIG_DEBUG_MAPLE_TREE | |
e6d6792a | 5456 | if (MAS_WARN_ON(mas, mas->index > mas->last)) |
89f499f3 | 5457 | pr_err("Error %lX > %lX %p\n", mas->index, mas->last, entry); |
e6d6792a | 5458 | |
54a611b6 LH |
5459 | if (mas->index > mas->last) { |
5460 | mas_set_err(mas, -EINVAL); | |
5461 | return NULL; | |
5462 | } | |
5463 | ||
5464 | #endif | |
5465 | ||
5466 | /* | |
5467 | * Storing is the same operation as insert with the added caveat that it | |
5468 | * can overwrite entries. Although this seems simple enough, one may | |
5469 | * want to examine what happens if a single store operation was to | |
5470 | * overwrite multiple entries within a self-balancing B-Tree. | |
5471 | */ | |
5472 | mas_wr_store_setup(&wr_mas); | |
5473 | mas_wr_store_entry(&wr_mas); | |
5474 | return wr_mas.content; | |
5475 | } | |
120b1162 | 5476 | EXPORT_SYMBOL_GPL(mas_store); |
54a611b6 LH |
5477 | |
5478 | /** | |
5479 | * mas_store_gfp() - Store a value into the tree. | |
5480 | * @mas: The maple state | |
5481 | * @entry: The entry to store | |
5482 | * @gfp: The GFP_FLAGS to use for allocations if necessary. | |
5483 | * | |
5484 | * Return: 0 on success, -EINVAL on invalid request, -ENOMEM if memory could not | |
5485 | * be allocated. | |
5486 | */ | |
5487 | int mas_store_gfp(struct ma_state *mas, void *entry, gfp_t gfp) | |
5488 | { | |
5489 | MA_WR_STATE(wr_mas, mas, entry); | |
5490 | ||
5491 | mas_wr_store_setup(&wr_mas); | |
5492 | trace_ma_write(__func__, mas, 0, entry); | |
5493 | retry: | |
5494 | mas_wr_store_entry(&wr_mas); | |
5495 | if (unlikely(mas_nomem(mas, gfp))) | |
5496 | goto retry; | |
5497 | ||
5498 | if (unlikely(mas_is_err(mas))) | |
5499 | return xa_err(mas->node); | |
5500 | ||
5501 | return 0; | |
5502 | } | |
120b1162 | 5503 | EXPORT_SYMBOL_GPL(mas_store_gfp); |
54a611b6 LH |
5504 | |
5505 | /** | |
5506 | * mas_store_prealloc() - Store a value into the tree using memory | |
5507 | * preallocated in the maple state. | |
5508 | * @mas: The maple state | |
5509 | * @entry: The entry to store. | |
5510 | */ | |
5511 | void mas_store_prealloc(struct ma_state *mas, void *entry) | |
5512 | { | |
5513 | MA_WR_STATE(wr_mas, mas, entry); | |
5514 | ||
5515 | mas_wr_store_setup(&wr_mas); | |
5516 | trace_ma_write(__func__, mas, 0, entry); | |
5517 | mas_wr_store_entry(&wr_mas); | |
1c414c6a | 5518 | MAS_WR_BUG_ON(&wr_mas, mas_is_err(mas)); |
54a611b6 LH |
5519 | mas_destroy(mas); |
5520 | } | |
120b1162 | 5521 | EXPORT_SYMBOL_GPL(mas_store_prealloc); |
54a611b6 LH |
5522 | |
5523 | /** | |
5524 | * mas_preallocate() - Preallocate enough nodes for a store operation | |
5525 | * @mas: The maple state | |
54a611b6 LH |
5526 | * @gfp: The GFP_FLAGS to use for allocations. |
5527 | * | |
5528 | * Return: 0 on success, -ENOMEM if memory could not be allocated. | |
5529 | */ | |
c5d5546e | 5530 | int mas_preallocate(struct ma_state *mas, gfp_t gfp) |
54a611b6 LH |
5531 | { |
5532 | int ret; | |
5533 | ||
5534 | mas_node_count_gfp(mas, 1 + mas_mt_height(mas) * 3, gfp); | |
5535 | mas->mas_flags |= MA_STATE_PREALLOC; | |
5536 | if (likely(!mas_is_err(mas))) | |
5537 | return 0; | |
5538 | ||
5539 | mas_set_alloc_req(mas, 0); | |
5540 | ret = xa_err(mas->node); | |
5541 | mas_reset(mas); | |
5542 | mas_destroy(mas); | |
5543 | mas_reset(mas); | |
5544 | return ret; | |
5545 | } | |
5c63a7c3 | 5546 | EXPORT_SYMBOL_GPL(mas_preallocate); |
54a611b6 LH |
5547 | |
5548 | /* | |
5549 | * mas_destroy() - destroy a maple state. | |
5550 | * @mas: The maple state | |
5551 | * | |
5552 | * Upon completion, check the left-most node and rebalance against the node to | |
5553 | * the right if necessary. Frees any allocated nodes associated with this maple | |
5554 | * state. | |
5555 | */ | |
5556 | void mas_destroy(struct ma_state *mas) | |
5557 | { | |
5558 | struct maple_alloc *node; | |
541e06b7 | 5559 | unsigned long total; |
54a611b6 LH |
5560 | |
5561 | /* | |
5562 | * When using mas_for_each() to insert an expected number of elements, | |
5563 | * it is possible that the number inserted is less than the expected | |
5564 | * number. To fix an invalid final node, a check is performed here to | |
5565 | * rebalance the previous node with the final node. | |
5566 | */ | |
5567 | if (mas->mas_flags & MA_STATE_REBALANCE) { | |
5568 | unsigned char end; | |
5569 | ||
23e734ec | 5570 | mas_start(mas); |
54a611b6 LH |
5571 | mtree_range_walk(mas); |
5572 | end = mas_data_end(mas) + 1; | |
5573 | if (end < mt_min_slot_count(mas->node) - 1) | |
5574 | mas_destroy_rebalance(mas, end); | |
5575 | ||
5576 | mas->mas_flags &= ~MA_STATE_REBALANCE; | |
5577 | } | |
5578 | mas->mas_flags &= ~(MA_STATE_BULK|MA_STATE_PREALLOC); | |
5579 | ||
541e06b7 LH |
5580 | total = mas_allocated(mas); |
5581 | while (total) { | |
54a611b6 LH |
5582 | node = mas->alloc; |
5583 | mas->alloc = node->slot[0]; | |
541e06b7 LH |
5584 | if (node->node_count > 1) { |
5585 | size_t count = node->node_count - 1; | |
5586 | ||
5587 | mt_free_bulk(count, (void __rcu **)&node->slot[1]); | |
5588 | total -= count; | |
5589 | } | |
54a611b6 | 5590 | kmem_cache_free(maple_node_cache, node); |
541e06b7 | 5591 | total--; |
54a611b6 | 5592 | } |
541e06b7 | 5593 | |
54a611b6 LH |
5594 | mas->alloc = NULL; |
5595 | } | |
120b1162 | 5596 | EXPORT_SYMBOL_GPL(mas_destroy); |
54a611b6 LH |
5597 | |
5598 | /* | |
5599 | * mas_expected_entries() - Set the expected number of entries that will be inserted. | |
5600 | * @mas: The maple state | |
5601 | * @nr_entries: The number of expected entries. | |
5602 | * | |
5603 | * This will attempt to pre-allocate enough nodes to store the expected number | |
5604 | * of entries. The allocations will occur using the bulk allocator interface | |
5605 | * for speed. Please call mas_destroy() on the @mas after inserting the entries | |
5606 | * to ensure any unused nodes are freed. | |
5607 | * | |
5608 | * Return: 0 on success, -ENOMEM if memory could not be allocated. | |
5609 | */ | |
5610 | int mas_expected_entries(struct ma_state *mas, unsigned long nr_entries) | |
5611 | { | |
5612 | int nonleaf_cap = MAPLE_ARANGE64_SLOTS - 2; | |
5613 | struct maple_enode *enode = mas->node; | |
5614 | int nr_nodes; | |
5615 | int ret; | |
5616 | ||
5617 | /* | |
5618 | * Sometimes it is necessary to duplicate a tree to a new tree, such as | |
5619 | * forking a process and duplicating the VMAs from one tree to a new | |
5620 | * tree. When such a situation arises, it is known that the new tree is | |
5621 | * not going to be used until the entire tree is populated. For | |
5622 | * performance reasons, it is best to use a bulk load with RCU disabled. | |
5623 | * This allows for optimistic splitting that favours the left and reuse | |
5624 | * of nodes during the operation. | |
5625 | */ | |
5626 | ||
5627 | /* Optimize splitting for bulk insert in-order */ | |
5628 | mas->mas_flags |= MA_STATE_BULK; | |
5629 | ||
5630 | /* | |
5631 | * Avoid overflow, assume a gap between each entry and a trailing null. | |
5632 | * If this is wrong, it just means allocation can happen during | |
5633 | * insertion of entries. | |
5634 | */ | |
5635 | nr_nodes = max(nr_entries, nr_entries * 2 + 1); | |
5636 | if (!mt_is_alloc(mas->tree)) | |
5637 | nonleaf_cap = MAPLE_RANGE64_SLOTS - 2; | |
5638 | ||
5639 | /* Leaves; reduce slots to keep space for expansion */ | |
5640 | nr_nodes = DIV_ROUND_UP(nr_nodes, MAPLE_RANGE64_SLOTS - 2); | |
5641 | /* Internal nodes */ | |
5642 | nr_nodes += DIV_ROUND_UP(nr_nodes, nonleaf_cap); | |
5643 | /* Add working room for split (2 nodes) + new parents */ | |
5644 | mas_node_count(mas, nr_nodes + 3); | |
5645 | ||
5646 | /* Detect if allocations run out */ | |
5647 | mas->mas_flags |= MA_STATE_PREALLOC; | |
5648 | ||
5649 | if (!mas_is_err(mas)) | |
5650 | return 0; | |
5651 | ||
5652 | ret = xa_err(mas->node); | |
5653 | mas->node = enode; | |
5654 | mas_destroy(mas); | |
5655 | return ret; | |
5656 | ||
5657 | } | |
120b1162 | 5658 | EXPORT_SYMBOL_GPL(mas_expected_entries); |
54a611b6 | 5659 | |
6169b553 LH |
5660 | static inline bool mas_next_setup(struct ma_state *mas, unsigned long max, |
5661 | void **entry) | |
54a611b6 | 5662 | { |
ca80f610 LH |
5663 | bool was_none = mas_is_none(mas); |
5664 | ||
54a611b6 LH |
5665 | if (mas_is_none(mas) || mas_is_paused(mas)) |
5666 | mas->node = MAS_START; | |
5667 | ||
5668 | if (mas_is_start(mas)) | |
6169b553 | 5669 | *entry = mas_walk(mas); /* Retries on dead nodes handled by mas_walk */ |
54a611b6 LH |
5670 | |
5671 | if (mas_is_ptr(mas)) { | |
6169b553 | 5672 | *entry = NULL; |
ca80f610 LH |
5673 | if (was_none && mas->index == 0) { |
5674 | mas->index = mas->last = 0; | |
6169b553 | 5675 | return true; |
54a611b6 | 5676 | } |
ca80f610 LH |
5677 | mas->index = 1; |
5678 | mas->last = ULONG_MAX; | |
5679 | mas->node = MAS_NONE; | |
6169b553 | 5680 | return true; |
54a611b6 LH |
5681 | } |
5682 | ||
6169b553 LH |
5683 | if (mas_is_none(mas)) |
5684 | return true; | |
5685 | return false; | |
5686 | } | |
5687 | ||
5688 | /** | |
5689 | * mas_next() - Get the next entry. | |
5690 | * @mas: The maple state | |
5691 | * @max: The maximum index to check. | |
5692 | * | |
5693 | * Returns the next entry after @mas->index. | |
5694 | * Must hold rcu_read_lock or the write lock. | |
5695 | * Can return the zero entry. | |
5696 | * | |
5697 | * Return: The next entry or %NULL | |
5698 | */ | |
5699 | void *mas_next(struct ma_state *mas, unsigned long max) | |
5700 | { | |
5701 | void *entry = NULL; | |
5702 | ||
5703 | if (mas_next_setup(mas, max, &entry)) | |
5704 | return entry; | |
5705 | ||
5706 | /* Retries on dead nodes handled by mas_next_slot */ | |
5707 | return mas_next_slot(mas, max, false); | |
54a611b6 LH |
5708 | } |
5709 | EXPORT_SYMBOL_GPL(mas_next); | |
5710 | ||
6169b553 LH |
5711 | /** |
5712 | * mas_next_range() - Advance the maple state to the next range | |
5713 | * @mas: The maple state | |
5714 | * @max: The maximum index to check. | |
5715 | * | |
5716 | * Sets @mas->index and @mas->last to the range. | |
5717 | * Must hold rcu_read_lock or the write lock. | |
5718 | * Can return the zero entry. | |
5719 | * | |
5720 | * Return: The next entry or %NULL | |
5721 | */ | |
5722 | void *mas_next_range(struct ma_state *mas, unsigned long max) | |
5723 | { | |
5724 | void *entry = NULL; | |
5725 | ||
5726 | if (mas_next_setup(mas, max, &entry)) | |
5727 | return entry; | |
5728 | ||
5729 | /* Retries on dead nodes handled by mas_next_slot */ | |
5730 | return mas_next_slot(mas, max, true); | |
5731 | } | |
5732 | EXPORT_SYMBOL_GPL(mas_next_range); | |
5733 | ||
54a611b6 LH |
5734 | /** |
5735 | * mt_next() - get the next value in the maple tree | |
5736 | * @mt: The maple tree | |
5737 | * @index: The start index | |
5738 | * @max: The maximum index to check | |
5739 | * | |
fad9c80e TG |
5740 | * Takes RCU read lock internally to protect the search, which does not |
5741 | * protect the returned pointer after dropping RCU read lock. | |
5742 | * See also: Documentation/core-api/maple_tree.rst | |
5743 | * | |
5744 | * Return: The entry higher than @index or %NULL if nothing is found. | |
54a611b6 LH |
5745 | */ |
5746 | void *mt_next(struct maple_tree *mt, unsigned long index, unsigned long max) | |
5747 | { | |
5748 | void *entry = NULL; | |
5749 | MA_STATE(mas, mt, index, index); | |
5750 | ||
5751 | rcu_read_lock(); | |
5752 | entry = mas_next(&mas, max); | |
5753 | rcu_read_unlock(); | |
5754 | return entry; | |
5755 | } | |
5756 | EXPORT_SYMBOL_GPL(mt_next); | |
5757 | ||
6b9e93e0 LH |
5758 | static inline bool mas_prev_setup(struct ma_state *mas, unsigned long min, |
5759 | void **entry) | |
54a611b6 | 5760 | { |
39193685 LH |
5761 | if (mas->index <= min) |
5762 | goto none; | |
54a611b6 LH |
5763 | |
5764 | if (mas_is_none(mas) || mas_is_paused(mas)) | |
5765 | mas->node = MAS_START; | |
5766 | ||
5767 | if (mas_is_start(mas)) { | |
5768 | mas_walk(mas); | |
5769 | if (!mas->index) | |
39193685 | 5770 | goto none; |
54a611b6 LH |
5771 | } |
5772 | ||
39193685 LH |
5773 | if (unlikely(mas_is_ptr(mas))) { |
5774 | if (!mas->index) | |
5775 | goto none; | |
54a611b6 | 5776 | mas->index = mas->last = 0; |
6b9e93e0 LH |
5777 | *entry = mas_root(mas); |
5778 | return true; | |
39193685 LH |
5779 | } |
5780 | ||
5781 | if (mas_is_none(mas)) { | |
5782 | if (mas->index) { | |
5783 | /* Walked to out-of-range pointer? */ | |
5784 | mas->index = mas->last = 0; | |
5785 | mas->node = MAS_ROOT; | |
6b9e93e0 LH |
5786 | *entry = mas_root(mas); |
5787 | return true; | |
39193685 | 5788 | } |
6b9e93e0 | 5789 | return true; |
54a611b6 | 5790 | } |
6b9e93e0 LH |
5791 | |
5792 | return false; | |
39193685 LH |
5793 | |
5794 | none: | |
5795 | mas->node = MAS_NONE; | |
6b9e93e0 LH |
5796 | return true; |
5797 | } | |
5798 | ||
5799 | /** | |
5800 | * mas_prev() - Get the previous entry | |
5801 | * @mas: The maple state | |
5802 | * @min: The minimum value to check. | |
5803 | * | |
5804 | * Must hold rcu_read_lock or the write lock. | |
5805 | * Will reset mas to MAS_START if the node is MAS_NONE. Will stop on not | |
5806 | * searchable nodes. | |
5807 | * | |
5808 | * Return: the previous value or %NULL. | |
5809 | */ | |
5810 | void *mas_prev(struct ma_state *mas, unsigned long min) | |
5811 | { | |
5812 | void *entry = NULL; | |
5813 | ||
5814 | if (mas_prev_setup(mas, min, &entry)) | |
5815 | return entry; | |
5816 | ||
5817 | return mas_prev_slot(mas, min, false); | |
54a611b6 LH |
5818 | } |
5819 | EXPORT_SYMBOL_GPL(mas_prev); | |
5820 | ||
6b9e93e0 LH |
5821 | /** |
5822 | * mas_prev_range() - Advance to the previous range | |
5823 | * @mas: The maple state | |
5824 | * @min: The minimum value to check. | |
5825 | * | |
5826 | * Sets @mas->index and @mas->last to the range. | |
5827 | * Must hold rcu_read_lock or the write lock. | |
5828 | * Will reset mas to MAS_START if the node is MAS_NONE. Will stop on not | |
5829 | * searchable nodes. | |
5830 | * | |
5831 | * Return: the previous value or %NULL. | |
5832 | */ | |
5833 | void *mas_prev_range(struct ma_state *mas, unsigned long min) | |
5834 | { | |
5835 | void *entry = NULL; | |
5836 | ||
5837 | if (mas_prev_setup(mas, min, &entry)) | |
5838 | return entry; | |
5839 | ||
5840 | return mas_prev_slot(mas, min, true); | |
5841 | } | |
5842 | EXPORT_SYMBOL_GPL(mas_prev_range); | |
5843 | ||
54a611b6 LH |
5844 | /** |
5845 | * mt_prev() - get the previous value in the maple tree | |
5846 | * @mt: The maple tree | |
5847 | * @index: The start index | |
5848 | * @min: The minimum index to check | |
5849 | * | |
fad9c80e TG |
5850 | * Takes RCU read lock internally to protect the search, which does not |
5851 | * protect the returned pointer after dropping RCU read lock. | |
5852 | * See also: Documentation/core-api/maple_tree.rst | |
5853 | * | |
5854 | * Return: The entry before @index or %NULL if nothing is found. | |
54a611b6 LH |
5855 | */ |
5856 | void *mt_prev(struct maple_tree *mt, unsigned long index, unsigned long min) | |
5857 | { | |
5858 | void *entry = NULL; | |
5859 | MA_STATE(mas, mt, index, index); | |
5860 | ||
5861 | rcu_read_lock(); | |
5862 | entry = mas_prev(&mas, min); | |
5863 | rcu_read_unlock(); | |
5864 | return entry; | |
5865 | } | |
5866 | EXPORT_SYMBOL_GPL(mt_prev); | |
5867 | ||
5868 | /** | |
5869 | * mas_pause() - Pause a mas_find/mas_for_each to drop the lock. | |
5870 | * @mas: The maple state to pause | |
5871 | * | |
5872 | * Some users need to pause a walk and drop the lock they're holding in | |
5873 | * order to yield to a higher priority thread or carry out an operation | |
5874 | * on an entry. Those users should call this function before they drop | |
5875 | * the lock. It resets the @mas to be suitable for the next iteration | |
5876 | * of the loop after the user has reacquired the lock. If most entries | |
5877 | * found during a walk require you to call mas_pause(), the mt_for_each() | |
5878 | * iterator may be more appropriate. | |
5879 | * | |
5880 | */ | |
5881 | void mas_pause(struct ma_state *mas) | |
5882 | { | |
5883 | mas->node = MAS_PAUSE; | |
5884 | } | |
5885 | EXPORT_SYMBOL_GPL(mas_pause); | |
5886 | ||
5887 | /** | |
6169b553 | 5888 | * mas_find_setup() - Internal function to set up mas_find*(). |
54a611b6 | 5889 | * @mas: The maple state |
6169b553 LH |
5890 | * @max: The maximum index |
5891 | * @entry: Pointer to the entry | |
54a611b6 | 5892 | * |
6169b553 | 5893 | * Returns: True if entry is the answer, false otherwise. |
54a611b6 | 5894 | */ |
6169b553 LH |
5895 | static inline bool mas_find_setup(struct ma_state *mas, unsigned long max, |
5896 | void **entry) | |
54a611b6 | 5897 | { |
6169b553 LH |
5898 | *entry = NULL; |
5899 | ||
ca80f610 LH |
5900 | if (unlikely(mas_is_none(mas))) { |
5901 | if (unlikely(mas->last >= max)) | |
6169b553 | 5902 | return true; |
ca80f610 LH |
5903 | |
5904 | mas->index = mas->last; | |
5905 | mas->node = MAS_START; | |
6169b553 | 5906 | } else if (unlikely(mas_is_paused(mas))) { |
ca80f610 | 5907 | if (unlikely(mas->last >= max)) |
6169b553 | 5908 | return true; |
ca80f610 | 5909 | |
54a611b6 LH |
5910 | mas->node = MAS_START; |
5911 | mas->index = ++mas->last; | |
6169b553 | 5912 | } else if (unlikely(mas_is_ptr(mas))) |
ca80f610 | 5913 | goto ptr_out_of_range; |
17dc622c | 5914 | |
54a611b6 LH |
5915 | if (unlikely(mas_is_start(mas))) { |
5916 | /* First run or continue */ | |
54a611b6 | 5917 | if (mas->index > max) |
6169b553 | 5918 | return true; |
54a611b6 | 5919 | |
6169b553 LH |
5920 | *entry = mas_walk(mas); |
5921 | if (*entry) | |
5922 | return true; | |
ca80f610 | 5923 | |
54a611b6 LH |
5924 | } |
5925 | ||
ca80f610 LH |
5926 | if (unlikely(!mas_searchable(mas))) { |
5927 | if (unlikely(mas_is_ptr(mas))) | |
5928 | goto ptr_out_of_range; | |
5929 | ||
6169b553 | 5930 | return true; |
ca80f610 LH |
5931 | } |
5932 | ||
5933 | if (mas->index == max) | |
6169b553 | 5934 | return true; |
54a611b6 | 5935 | |
6169b553 | 5936 | return false; |
ca80f610 LH |
5937 | |
5938 | ptr_out_of_range: | |
5939 | mas->node = MAS_NONE; | |
5940 | mas->index = 1; | |
5941 | mas->last = ULONG_MAX; | |
6169b553 LH |
5942 | return true; |
5943 | } | |
5944 | ||
5945 | /** | |
5946 | * mas_find() - On the first call, find the entry at or after mas->index up to | |
5947 | * %max. Otherwise, find the entry after mas->index. | |
5948 | * @mas: The maple state | |
5949 | * @max: The maximum value to check. | |
5950 | * | |
5951 | * Must hold rcu_read_lock or the write lock. | |
5952 | * If an entry exists, last and index are updated accordingly. | |
5953 | * May set @mas->node to MAS_NONE. | |
5954 | * | |
5955 | * Return: The entry or %NULL. | |
5956 | */ | |
5957 | void *mas_find(struct ma_state *mas, unsigned long max) | |
5958 | { | |
5959 | void *entry = NULL; | |
5960 | ||
5961 | if (mas_find_setup(mas, max, &entry)) | |
5962 | return entry; | |
5963 | ||
5964 | /* Retries on dead nodes handled by mas_next_slot */ | |
5965 | return mas_next_slot(mas, max, false); | |
54a611b6 | 5966 | } |
120b1162 | 5967 | EXPORT_SYMBOL_GPL(mas_find); |
54a611b6 | 5968 | |
6169b553 LH |
5969 | /** |
5970 | * mas_find_range() - On the first call, find the entry at or after | |
5971 | * mas->index up to %max. Otherwise, advance to the next slot mas->index. | |
5972 | * @mas: The maple state | |
5973 | * @max: The maximum value to check. | |
5974 | * | |
5975 | * Must hold rcu_read_lock or the write lock. | |
5976 | * If an entry exists, last and index are updated accordingly. | |
5977 | * May set @mas->node to MAS_NONE. | |
5978 | * | |
5979 | * Return: The entry or %NULL. | |
5980 | */ | |
5981 | void *mas_find_range(struct ma_state *mas, unsigned long max) | |
5982 | { | |
5983 | void *entry; | |
5984 | ||
5985 | if (mas_find_setup(mas, max, &entry)) | |
5986 | return entry; | |
5987 | ||
5988 | /* Retries on dead nodes handled by mas_next_slot */ | |
5989 | return mas_next_slot(mas, max, true); | |
5990 | } | |
5991 | EXPORT_SYMBOL_GPL(mas_find_range); | |
5992 | ||
54a611b6 | 5993 | /** |
6b9e93e0 | 5994 | * mas_find_rev_setup() - Internal function to set up mas_find_*_rev() |
54a611b6 | 5995 | * @mas: The maple state |
6b9e93e0 LH |
5996 | * @min: The minimum index |
5997 | * @entry: Pointer to the entry | |
54a611b6 | 5998 | * |
6b9e93e0 | 5999 | * Returns: True if entry is the answer, false otherwise. |
54a611b6 | 6000 | */ |
6b9e93e0 LH |
6001 | static inline bool mas_find_rev_setup(struct ma_state *mas, unsigned long min, |
6002 | void **entry) | |
54a611b6 | 6003 | { |
6b9e93e0 LH |
6004 | *entry = NULL; |
6005 | ||
39193685 LH |
6006 | if (unlikely(mas_is_none(mas))) { |
6007 | if (mas->index <= min) | |
6008 | goto none; | |
6009 | ||
6010 | mas->last = mas->index; | |
6011 | mas->node = MAS_START; | |
6012 | } | |
6013 | ||
54a611b6 | 6014 | if (unlikely(mas_is_paused(mas))) { |
39193685 | 6015 | if (unlikely(mas->index <= min)) { |
54a611b6 | 6016 | mas->node = MAS_NONE; |
6b9e93e0 | 6017 | return true; |
54a611b6 LH |
6018 | } |
6019 | mas->node = MAS_START; | |
6020 | mas->last = --mas->index; | |
6021 | } | |
6022 | ||
6023 | if (unlikely(mas_is_start(mas))) { | |
6024 | /* First run or continue */ | |
54a611b6 | 6025 | if (mas->index < min) |
6b9e93e0 | 6026 | return true; |
54a611b6 | 6027 | |
6b9e93e0 LH |
6028 | *entry = mas_walk(mas); |
6029 | if (*entry) | |
6030 | return true; | |
54a611b6 LH |
6031 | } |
6032 | ||
39193685 LH |
6033 | if (unlikely(!mas_searchable(mas))) { |
6034 | if (mas_is_ptr(mas)) | |
6035 | goto none; | |
6036 | ||
6037 | if (mas_is_none(mas)) { | |
6038 | /* | |
6039 | * Walked to the location, and there was nothing so the | |
6040 | * previous location is 0. | |
6041 | */ | |
6042 | mas->last = mas->index = 0; | |
6043 | mas->node = MAS_ROOT; | |
6b9e93e0 LH |
6044 | *entry = mas_root(mas); |
6045 | return true; | |
39193685 LH |
6046 | } |
6047 | } | |
54a611b6 LH |
6048 | |
6049 | if (mas->index < min) | |
6b9e93e0 | 6050 | return true; |
54a611b6 | 6051 | |
6b9e93e0 | 6052 | return false; |
39193685 LH |
6053 | |
6054 | none: | |
6055 | mas->node = MAS_NONE; | |
6b9e93e0 LH |
6056 | return true; |
6057 | } | |
6058 | ||
6059 | /** | |
6060 | * mas_find_rev: On the first call, find the first non-null entry at or below | |
6061 | * mas->index down to %min. Otherwise find the first non-null entry below | |
6062 | * mas->index down to %min. | |
6063 | * @mas: The maple state | |
6064 | * @min: The minimum value to check. | |
6065 | * | |
6066 | * Must hold rcu_read_lock or the write lock. | |
6067 | * If an entry exists, last and index are updated accordingly. | |
6068 | * May set @mas->node to MAS_NONE. | |
6069 | * | |
6070 | * Return: The entry or %NULL. | |
6071 | */ | |
6072 | void *mas_find_rev(struct ma_state *mas, unsigned long min) | |
6073 | { | |
6074 | void *entry; | |
6075 | ||
6076 | if (mas_find_rev_setup(mas, min, &entry)) | |
6077 | return entry; | |
6078 | ||
6079 | /* Retries on dead nodes handled by mas_prev_slot */ | |
6080 | return mas_prev_slot(mas, min, false); | |
6081 | ||
54a611b6 | 6082 | } |
120b1162 | 6083 | EXPORT_SYMBOL_GPL(mas_find_rev); |
54a611b6 | 6084 | |
6b9e93e0 LH |
6085 | /** |
6086 | * mas_find_range_rev: On the first call, find the first non-null entry at or | |
6087 | * below mas->index down to %min. Otherwise advance to the previous slot after | |
6088 | * mas->index down to %min. | |
6089 | * @mas: The maple state | |
6090 | * @min: The minimum value to check. | |
6091 | * | |
6092 | * Must hold rcu_read_lock or the write lock. | |
6093 | * If an entry exists, last and index are updated accordingly. | |
6094 | * May set @mas->node to MAS_NONE. | |
6095 | * | |
6096 | * Return: The entry or %NULL. | |
6097 | */ | |
6098 | void *mas_find_range_rev(struct ma_state *mas, unsigned long min) | |
6099 | { | |
6100 | void *entry; | |
6101 | ||
6102 | if (mas_find_rev_setup(mas, min, &entry)) | |
6103 | return entry; | |
6104 | ||
6105 | /* Retries on dead nodes handled by mas_prev_slot */ | |
6106 | return mas_prev_slot(mas, min, true); | |
6107 | } | |
6108 | EXPORT_SYMBOL_GPL(mas_find_range_rev); | |
6109 | ||
54a611b6 LH |
6110 | /** |
6111 | * mas_erase() - Find the range in which index resides and erase the entire | |
6112 | * range. | |
6113 | * @mas: The maple state | |
6114 | * | |
6115 | * Must hold the write lock. | |
6116 | * Searches for @mas->index, sets @mas->index and @mas->last to the range and | |
6117 | * erases that range. | |
6118 | * | |
6119 | * Return: the entry that was erased or %NULL, @mas->index and @mas->last are updated. | |
6120 | */ | |
6121 | void *mas_erase(struct ma_state *mas) | |
6122 | { | |
6123 | void *entry; | |
6124 | MA_WR_STATE(wr_mas, mas, NULL); | |
6125 | ||
6126 | if (mas_is_none(mas) || mas_is_paused(mas)) | |
6127 | mas->node = MAS_START; | |
6128 | ||
6129 | /* Retry unnecessary when holding the write lock. */ | |
6130 | entry = mas_state_walk(mas); | |
6131 | if (!entry) | |
6132 | return NULL; | |
6133 | ||
6134 | write_retry: | |
6135 | /* Must reset to ensure spanning writes of last slot are detected */ | |
6136 | mas_reset(mas); | |
6137 | mas_wr_store_setup(&wr_mas); | |
6138 | mas_wr_store_entry(&wr_mas); | |
6139 | if (mas_nomem(mas, GFP_KERNEL)) | |
6140 | goto write_retry; | |
6141 | ||
6142 | return entry; | |
6143 | } | |
6144 | EXPORT_SYMBOL_GPL(mas_erase); | |
6145 | ||
6146 | /** | |
6147 | * mas_nomem() - Check if there was an error allocating and do the allocation | |
6148 | * if necessary If there are allocations, then free them. | |
6149 | * @mas: The maple state | |
6150 | * @gfp: The GFP_FLAGS to use for allocations | |
6151 | * Return: true on allocation, false otherwise. | |
6152 | */ | |
6153 | bool mas_nomem(struct ma_state *mas, gfp_t gfp) | |
14c4b5ab | 6154 | __must_hold(mas->tree->ma_lock) |
54a611b6 LH |
6155 | { |
6156 | if (likely(mas->node != MA_ERROR(-ENOMEM))) { | |
6157 | mas_destroy(mas); | |
6158 | return false; | |
6159 | } | |
6160 | ||
6161 | if (gfpflags_allow_blocking(gfp) && !mt_external_lock(mas->tree)) { | |
6162 | mtree_unlock(mas->tree); | |
6163 | mas_alloc_nodes(mas, gfp); | |
6164 | mtree_lock(mas->tree); | |
6165 | } else { | |
6166 | mas_alloc_nodes(mas, gfp); | |
6167 | } | |
6168 | ||
6169 | if (!mas_allocated(mas)) | |
6170 | return false; | |
6171 | ||
6172 | mas->node = MAS_START; | |
6173 | return true; | |
6174 | } | |
6175 | ||
6176 | void __init maple_tree_init(void) | |
6177 | { | |
6178 | maple_node_cache = kmem_cache_create("maple_node", | |
6179 | sizeof(struct maple_node), sizeof(struct maple_node), | |
6180 | SLAB_PANIC, NULL); | |
6181 | } | |
6182 | ||
6183 | /** | |
6184 | * mtree_load() - Load a value stored in a maple tree | |
6185 | * @mt: The maple tree | |
6186 | * @index: The index to load | |
6187 | * | |
6188 | * Return: the entry or %NULL | |
6189 | */ | |
6190 | void *mtree_load(struct maple_tree *mt, unsigned long index) | |
6191 | { | |
6192 | MA_STATE(mas, mt, index, index); | |
6193 | void *entry; | |
6194 | ||
6195 | trace_ma_read(__func__, &mas); | |
6196 | rcu_read_lock(); | |
6197 | retry: | |
6198 | entry = mas_start(&mas); | |
6199 | if (unlikely(mas_is_none(&mas))) | |
6200 | goto unlock; | |
6201 | ||
6202 | if (unlikely(mas_is_ptr(&mas))) { | |
6203 | if (index) | |
6204 | entry = NULL; | |
6205 | ||
6206 | goto unlock; | |
6207 | } | |
6208 | ||
6209 | entry = mtree_lookup_walk(&mas); | |
6210 | if (!entry && unlikely(mas_is_start(&mas))) | |
6211 | goto retry; | |
6212 | unlock: | |
6213 | rcu_read_unlock(); | |
6214 | if (xa_is_zero(entry)) | |
6215 | return NULL; | |
6216 | ||
6217 | return entry; | |
6218 | } | |
6219 | EXPORT_SYMBOL(mtree_load); | |
6220 | ||
6221 | /** | |
6222 | * mtree_store_range() - Store an entry at a given range. | |
6223 | * @mt: The maple tree | |
6224 | * @index: The start of the range | |
6225 | * @last: The end of the range | |
6226 | * @entry: The entry to store | |
6227 | * @gfp: The GFP_FLAGS to use for allocations | |
6228 | * | |
6229 | * Return: 0 on success, -EINVAL on invalid request, -ENOMEM if memory could not | |
6230 | * be allocated. | |
6231 | */ | |
6232 | int mtree_store_range(struct maple_tree *mt, unsigned long index, | |
6233 | unsigned long last, void *entry, gfp_t gfp) | |
6234 | { | |
6235 | MA_STATE(mas, mt, index, last); | |
6236 | MA_WR_STATE(wr_mas, &mas, entry); | |
6237 | ||
6238 | trace_ma_write(__func__, &mas, 0, entry); | |
6239 | if (WARN_ON_ONCE(xa_is_advanced(entry))) | |
6240 | return -EINVAL; | |
6241 | ||
6242 | if (index > last) | |
6243 | return -EINVAL; | |
6244 | ||
6245 | mtree_lock(mt); | |
6246 | retry: | |
6247 | mas_wr_store_entry(&wr_mas); | |
6248 | if (mas_nomem(&mas, gfp)) | |
6249 | goto retry; | |
6250 | ||
6251 | mtree_unlock(mt); | |
6252 | if (mas_is_err(&mas)) | |
6253 | return xa_err(mas.node); | |
6254 | ||
6255 | return 0; | |
6256 | } | |
6257 | EXPORT_SYMBOL(mtree_store_range); | |
6258 | ||
6259 | /** | |
6260 | * mtree_store() - Store an entry at a given index. | |
6261 | * @mt: The maple tree | |
6262 | * @index: The index to store the value | |
6263 | * @entry: The entry to store | |
6264 | * @gfp: The GFP_FLAGS to use for allocations | |
6265 | * | |
6266 | * Return: 0 on success, -EINVAL on invalid request, -ENOMEM if memory could not | |
6267 | * be allocated. | |
6268 | */ | |
6269 | int mtree_store(struct maple_tree *mt, unsigned long index, void *entry, | |
6270 | gfp_t gfp) | |
6271 | { | |
6272 | return mtree_store_range(mt, index, index, entry, gfp); | |
6273 | } | |
6274 | EXPORT_SYMBOL(mtree_store); | |
6275 | ||
6276 | /** | |
4445e582 | 6277 | * mtree_insert_range() - Insert an entry at a given range if there is no value. |
54a611b6 LH |
6278 | * @mt: The maple tree |
6279 | * @first: The start of the range | |
6280 | * @last: The end of the range | |
6281 | * @entry: The entry to store | |
6282 | * @gfp: The GFP_FLAGS to use for allocations. | |
6283 | * | |
6284 | * Return: 0 on success, -EEXISTS if the range is occupied, -EINVAL on invalid | |
6285 | * request, -ENOMEM if memory could not be allocated. | |
6286 | */ | |
6287 | int mtree_insert_range(struct maple_tree *mt, unsigned long first, | |
6288 | unsigned long last, void *entry, gfp_t gfp) | |
6289 | { | |
6290 | MA_STATE(ms, mt, first, last); | |
6291 | ||
6292 | if (WARN_ON_ONCE(xa_is_advanced(entry))) | |
6293 | return -EINVAL; | |
6294 | ||
6295 | if (first > last) | |
6296 | return -EINVAL; | |
6297 | ||
6298 | mtree_lock(mt); | |
6299 | retry: | |
6300 | mas_insert(&ms, entry); | |
6301 | if (mas_nomem(&ms, gfp)) | |
6302 | goto retry; | |
6303 | ||
6304 | mtree_unlock(mt); | |
6305 | if (mas_is_err(&ms)) | |
6306 | return xa_err(ms.node); | |
6307 | ||
6308 | return 0; | |
6309 | } | |
6310 | EXPORT_SYMBOL(mtree_insert_range); | |
6311 | ||
6312 | /** | |
4445e582 | 6313 | * mtree_insert() - Insert an entry at a given index if there is no value. |
54a611b6 LH |
6314 | * @mt: The maple tree |
6315 | * @index : The index to store the value | |
6316 | * @entry: The entry to store | |
6317 | * @gfp: The FGP_FLAGS to use for allocations. | |
6318 | * | |
6319 | * Return: 0 on success, -EEXISTS if the range is occupied, -EINVAL on invalid | |
6320 | * request, -ENOMEM if memory could not be allocated. | |
6321 | */ | |
6322 | int mtree_insert(struct maple_tree *mt, unsigned long index, void *entry, | |
6323 | gfp_t gfp) | |
6324 | { | |
6325 | return mtree_insert_range(mt, index, index, entry, gfp); | |
6326 | } | |
6327 | EXPORT_SYMBOL(mtree_insert); | |
6328 | ||
6329 | int mtree_alloc_range(struct maple_tree *mt, unsigned long *startp, | |
6330 | void *entry, unsigned long size, unsigned long min, | |
6331 | unsigned long max, gfp_t gfp) | |
6332 | { | |
6333 | int ret = 0; | |
6334 | ||
52371677 | 6335 | MA_STATE(mas, mt, 0, 0); |
54a611b6 LH |
6336 | if (!mt_is_alloc(mt)) |
6337 | return -EINVAL; | |
6338 | ||
6339 | if (WARN_ON_ONCE(mt_is_reserved(entry))) | |
6340 | return -EINVAL; | |
6341 | ||
54a611b6 LH |
6342 | mtree_lock(mt); |
6343 | retry: | |
52371677 PZ |
6344 | ret = mas_empty_area(&mas, min, max, size); |
6345 | if (ret) | |
6346 | goto unlock; | |
6347 | ||
6348 | mas_insert(&mas, entry); | |
6349 | /* | |
6350 | * mas_nomem() may release the lock, causing the allocated area | |
6351 | * to be unavailable, so try to allocate a free area again. | |
6352 | */ | |
54a611b6 LH |
6353 | if (mas_nomem(&mas, gfp)) |
6354 | goto retry; | |
6355 | ||
52371677 PZ |
6356 | if (mas_is_err(&mas)) |
6357 | ret = xa_err(mas.node); | |
6358 | else | |
6359 | *startp = mas.index; | |
6360 | ||
6361 | unlock: | |
54a611b6 LH |
6362 | mtree_unlock(mt); |
6363 | return ret; | |
6364 | } | |
6365 | EXPORT_SYMBOL(mtree_alloc_range); | |
6366 | ||
6367 | int mtree_alloc_rrange(struct maple_tree *mt, unsigned long *startp, | |
6368 | void *entry, unsigned long size, unsigned long min, | |
6369 | unsigned long max, gfp_t gfp) | |
6370 | { | |
6371 | int ret = 0; | |
6372 | ||
52371677 | 6373 | MA_STATE(mas, mt, 0, 0); |
54a611b6 LH |
6374 | if (!mt_is_alloc(mt)) |
6375 | return -EINVAL; | |
6376 | ||
6377 | if (WARN_ON_ONCE(mt_is_reserved(entry))) | |
6378 | return -EINVAL; | |
6379 | ||
54a611b6 LH |
6380 | mtree_lock(mt); |
6381 | retry: | |
52371677 PZ |
6382 | ret = mas_empty_area_rev(&mas, min, max, size); |
6383 | if (ret) | |
6384 | goto unlock; | |
6385 | ||
6386 | mas_insert(&mas, entry); | |
6387 | /* | |
6388 | * mas_nomem() may release the lock, causing the allocated area | |
6389 | * to be unavailable, so try to allocate a free area again. | |
6390 | */ | |
54a611b6 LH |
6391 | if (mas_nomem(&mas, gfp)) |
6392 | goto retry; | |
6393 | ||
52371677 PZ |
6394 | if (mas_is_err(&mas)) |
6395 | ret = xa_err(mas.node); | |
6396 | else | |
6397 | *startp = mas.index; | |
6398 | ||
6399 | unlock: | |
54a611b6 LH |
6400 | mtree_unlock(mt); |
6401 | return ret; | |
6402 | } | |
6403 | EXPORT_SYMBOL(mtree_alloc_rrange); | |
6404 | ||
6405 | /** | |
6406 | * mtree_erase() - Find an index and erase the entire range. | |
6407 | * @mt: The maple tree | |
6408 | * @index: The index to erase | |
6409 | * | |
6410 | * Erasing is the same as a walk to an entry then a store of a NULL to that | |
6411 | * ENTIRE range. In fact, it is implemented as such using the advanced API. | |
6412 | * | |
6413 | * Return: The entry stored at the @index or %NULL | |
6414 | */ | |
6415 | void *mtree_erase(struct maple_tree *mt, unsigned long index) | |
6416 | { | |
6417 | void *entry = NULL; | |
6418 | ||
6419 | MA_STATE(mas, mt, index, index); | |
6420 | trace_ma_op(__func__, &mas); | |
6421 | ||
6422 | mtree_lock(mt); | |
6423 | entry = mas_erase(&mas); | |
6424 | mtree_unlock(mt); | |
6425 | ||
6426 | return entry; | |
6427 | } | |
6428 | EXPORT_SYMBOL(mtree_erase); | |
6429 | ||
6430 | /** | |
6431 | * __mt_destroy() - Walk and free all nodes of a locked maple tree. | |
6432 | * @mt: The maple tree | |
6433 | * | |
6434 | * Note: Does not handle locking. | |
6435 | */ | |
6436 | void __mt_destroy(struct maple_tree *mt) | |
6437 | { | |
6438 | void *root = mt_root_locked(mt); | |
6439 | ||
6440 | rcu_assign_pointer(mt->ma_root, NULL); | |
6441 | if (xa_is_node(root)) | |
6442 | mte_destroy_walk(root, mt); | |
6443 | ||
6444 | mt->ma_flags = 0; | |
6445 | } | |
6446 | EXPORT_SYMBOL_GPL(__mt_destroy); | |
6447 | ||
6448 | /** | |
6449 | * mtree_destroy() - Destroy a maple tree | |
6450 | * @mt: The maple tree | |
6451 | * | |
6452 | * Frees all resources used by the tree. Handles locking. | |
6453 | */ | |
6454 | void mtree_destroy(struct maple_tree *mt) | |
6455 | { | |
6456 | mtree_lock(mt); | |
6457 | __mt_destroy(mt); | |
6458 | mtree_unlock(mt); | |
6459 | } | |
6460 | EXPORT_SYMBOL(mtree_destroy); | |
6461 | ||
6462 | /** | |
6463 | * mt_find() - Search from the start up until an entry is found. | |
6464 | * @mt: The maple tree | |
6465 | * @index: Pointer which contains the start location of the search | |
fad9c80e TG |
6466 | * @max: The maximum value of the search range |
6467 | * | |
6468 | * Takes RCU read lock internally to protect the search, which does not | |
6469 | * protect the returned pointer after dropping RCU read lock. | |
6470 | * See also: Documentation/core-api/maple_tree.rst | |
54a611b6 | 6471 | * |
fad9c80e TG |
6472 | * In case that an entry is found @index is updated to point to the next |
6473 | * possible entry independent whether the found entry is occupying a | |
6474 | * single index or a range if indices. | |
54a611b6 LH |
6475 | * |
6476 | * Return: The entry at or after the @index or %NULL | |
6477 | */ | |
6478 | void *mt_find(struct maple_tree *mt, unsigned long *index, unsigned long max) | |
6479 | { | |
6480 | MA_STATE(mas, mt, *index, *index); | |
6481 | void *entry; | |
6482 | #ifdef CONFIG_DEBUG_MAPLE_TREE | |
6483 | unsigned long copy = *index; | |
6484 | #endif | |
6485 | ||
6486 | trace_ma_read(__func__, &mas); | |
6487 | ||
6488 | if ((*index) > max) | |
6489 | return NULL; | |
6490 | ||
6491 | rcu_read_lock(); | |
6492 | retry: | |
6493 | entry = mas_state_walk(&mas); | |
6494 | if (mas_is_start(&mas)) | |
6495 | goto retry; | |
6496 | ||
6497 | if (unlikely(xa_is_zero(entry))) | |
6498 | entry = NULL; | |
6499 | ||
6500 | if (entry) | |
6501 | goto unlock; | |
6502 | ||
ca80f610 | 6503 | while (mas_searchable(&mas) && (mas.last < max)) { |
54a611b6 LH |
6504 | entry = mas_next_entry(&mas, max); |
6505 | if (likely(entry && !xa_is_zero(entry))) | |
6506 | break; | |
6507 | } | |
6508 | ||
6509 | if (unlikely(xa_is_zero(entry))) | |
6510 | entry = NULL; | |
6511 | unlock: | |
6512 | rcu_read_unlock(); | |
6513 | if (likely(entry)) { | |
6514 | *index = mas.last + 1; | |
6515 | #ifdef CONFIG_DEBUG_MAPLE_TREE | |
e6d6792a | 6516 | if (MT_WARN_ON(mt, (*index) && ((*index) <= copy))) |
54a611b6 LH |
6517 | pr_err("index not increased! %lx <= %lx\n", |
6518 | *index, copy); | |
54a611b6 LH |
6519 | #endif |
6520 | } | |
6521 | ||
6522 | return entry; | |
6523 | } | |
6524 | EXPORT_SYMBOL(mt_find); | |
6525 | ||
6526 | /** | |
6527 | * mt_find_after() - Search from the start up until an entry is found. | |
6528 | * @mt: The maple tree | |
6529 | * @index: Pointer which contains the start location of the search | |
6530 | * @max: The maximum value to check | |
6531 | * | |
fad9c80e TG |
6532 | * Same as mt_find() except that it checks @index for 0 before |
6533 | * searching. If @index == 0, the search is aborted. This covers a wrap | |
6534 | * around of @index to 0 in an iterator loop. | |
54a611b6 LH |
6535 | * |
6536 | * Return: The entry at or after the @index or %NULL | |
6537 | */ | |
6538 | void *mt_find_after(struct maple_tree *mt, unsigned long *index, | |
6539 | unsigned long max) | |
6540 | { | |
6541 | if (!(*index)) | |
6542 | return NULL; | |
6543 | ||
6544 | return mt_find(mt, index, max); | |
6545 | } | |
6546 | EXPORT_SYMBOL(mt_find_after); | |
6547 | ||
6548 | #ifdef CONFIG_DEBUG_MAPLE_TREE | |
6549 | atomic_t maple_tree_tests_run; | |
6550 | EXPORT_SYMBOL_GPL(maple_tree_tests_run); | |
6551 | atomic_t maple_tree_tests_passed; | |
6552 | EXPORT_SYMBOL_GPL(maple_tree_tests_passed); | |
6553 | ||
6554 | #ifndef __KERNEL__ | |
6555 | extern void kmem_cache_set_non_kernel(struct kmem_cache *, unsigned int); | |
6556 | void mt_set_non_kernel(unsigned int val) | |
6557 | { | |
6558 | kmem_cache_set_non_kernel(maple_node_cache, val); | |
6559 | } | |
6560 | ||
6561 | extern unsigned long kmem_cache_get_alloc(struct kmem_cache *); | |
6562 | unsigned long mt_get_alloc_size(void) | |
6563 | { | |
6564 | return kmem_cache_get_alloc(maple_node_cache); | |
6565 | } | |
6566 | ||
6567 | extern void kmem_cache_zero_nr_tallocated(struct kmem_cache *); | |
6568 | void mt_zero_nr_tallocated(void) | |
6569 | { | |
6570 | kmem_cache_zero_nr_tallocated(maple_node_cache); | |
6571 | } | |
6572 | ||
6573 | extern unsigned int kmem_cache_nr_tallocated(struct kmem_cache *); | |
6574 | unsigned int mt_nr_tallocated(void) | |
6575 | { | |
6576 | return kmem_cache_nr_tallocated(maple_node_cache); | |
6577 | } | |
6578 | ||
6579 | extern unsigned int kmem_cache_nr_allocated(struct kmem_cache *); | |
6580 | unsigned int mt_nr_allocated(void) | |
6581 | { | |
6582 | return kmem_cache_nr_allocated(maple_node_cache); | |
6583 | } | |
6584 | ||
6585 | /* | |
6586 | * mas_dead_node() - Check if the maple state is pointing to a dead node. | |
6587 | * @mas: The maple state | |
6588 | * @index: The index to restore in @mas. | |
6589 | * | |
6590 | * Used in test code. | |
6591 | * Return: 1 if @mas has been reset to MAS_START, 0 otherwise. | |
6592 | */ | |
6593 | static inline int mas_dead_node(struct ma_state *mas, unsigned long index) | |
6594 | { | |
6595 | if (unlikely(!mas_searchable(mas) || mas_is_start(mas))) | |
6596 | return 0; | |
6597 | ||
6598 | if (likely(!mte_dead_node(mas->node))) | |
6599 | return 0; | |
6600 | ||
6601 | mas_rewalk(mas, index); | |
6602 | return 1; | |
6603 | } | |
54a611b6 | 6604 | |
120b1162 LH |
6605 | void mt_cache_shrink(void) |
6606 | { | |
6607 | } | |
6608 | #else | |
6609 | /* | |
6610 | * mt_cache_shrink() - For testing, don't use this. | |
6611 | * | |
6612 | * Certain testcases can trigger an OOM when combined with other memory | |
6613 | * debugging configuration options. This function is used to reduce the | |
6614 | * possibility of an out of memory even due to kmem_cache objects remaining | |
6615 | * around for longer than usual. | |
6616 | */ | |
6617 | void mt_cache_shrink(void) | |
6618 | { | |
6619 | kmem_cache_shrink(maple_node_cache); | |
6620 | ||
6621 | } | |
6622 | EXPORT_SYMBOL_GPL(mt_cache_shrink); | |
6623 | ||
6624 | #endif /* not defined __KERNEL__ */ | |
54a611b6 LH |
6625 | /* |
6626 | * mas_get_slot() - Get the entry in the maple state node stored at @offset. | |
6627 | * @mas: The maple state | |
6628 | * @offset: The offset into the slot array to fetch. | |
6629 | * | |
6630 | * Return: The entry stored at @offset. | |
6631 | */ | |
6632 | static inline struct maple_enode *mas_get_slot(struct ma_state *mas, | |
6633 | unsigned char offset) | |
6634 | { | |
6635 | return mas_slot(mas, ma_slots(mas_mn(mas), mte_node_type(mas->node)), | |
6636 | offset); | |
6637 | } | |
6638 | ||
54a611b6 LH |
6639 | /* Depth first search, post-order */ |
6640 | static void mas_dfs_postorder(struct ma_state *mas, unsigned long max) | |
6641 | { | |
6642 | ||
6643 | struct maple_enode *p = MAS_NONE, *mn = mas->node; | |
6644 | unsigned long p_min, p_max; | |
6645 | ||
6646 | mas_next_node(mas, mas_mn(mas), max); | |
6647 | if (!mas_is_none(mas)) | |
6648 | return; | |
6649 | ||
6650 | if (mte_is_root(mn)) | |
6651 | return; | |
6652 | ||
6653 | mas->node = mn; | |
6654 | mas_ascend(mas); | |
c3eb787e | 6655 | do { |
54a611b6 LH |
6656 | p = mas->node; |
6657 | p_min = mas->min; | |
6658 | p_max = mas->max; | |
6659 | mas_prev_node(mas, 0); | |
c3eb787e | 6660 | } while (!mas_is_none(mas)); |
54a611b6 LH |
6661 | |
6662 | mas->node = p; | |
6663 | mas->max = p_max; | |
6664 | mas->min = p_min; | |
6665 | } | |
6666 | ||
6667 | /* Tree validations */ | |
6668 | static void mt_dump_node(const struct maple_tree *mt, void *entry, | |
89f499f3 LH |
6669 | unsigned long min, unsigned long max, unsigned int depth, |
6670 | enum mt_dump_format format); | |
54a611b6 | 6671 | static void mt_dump_range(unsigned long min, unsigned long max, |
89f499f3 | 6672 | unsigned int depth, enum mt_dump_format format) |
54a611b6 LH |
6673 | { |
6674 | static const char spaces[] = " "; | |
6675 | ||
89f499f3 LH |
6676 | switch(format) { |
6677 | case mt_dump_hex: | |
6678 | if (min == max) | |
6679 | pr_info("%.*s%lx: ", depth * 2, spaces, min); | |
6680 | else | |
6681 | pr_info("%.*s%lx-%lx: ", depth * 2, spaces, min, max); | |
6682 | break; | |
6683 | default: | |
6684 | case mt_dump_dec: | |
6685 | if (min == max) | |
6686 | pr_info("%.*s%lu: ", depth * 2, spaces, min); | |
6687 | else | |
6688 | pr_info("%.*s%lu-%lu: ", depth * 2, spaces, min, max); | |
6689 | } | |
54a611b6 LH |
6690 | } |
6691 | ||
6692 | static void mt_dump_entry(void *entry, unsigned long min, unsigned long max, | |
89f499f3 | 6693 | unsigned int depth, enum mt_dump_format format) |
54a611b6 | 6694 | { |
89f499f3 | 6695 | mt_dump_range(min, max, depth, format); |
54a611b6 LH |
6696 | |
6697 | if (xa_is_value(entry)) | |
6698 | pr_cont("value %ld (0x%lx) [%p]\n", xa_to_value(entry), | |
6699 | xa_to_value(entry), entry); | |
6700 | else if (xa_is_zero(entry)) | |
6701 | pr_cont("zero (%ld)\n", xa_to_internal(entry)); | |
6702 | else if (mt_is_reserved(entry)) | |
6703 | pr_cont("UNKNOWN ENTRY (%p)\n", entry); | |
6704 | else | |
6705 | pr_cont("%p\n", entry); | |
6706 | } | |
6707 | ||
6708 | static void mt_dump_range64(const struct maple_tree *mt, void *entry, | |
89f499f3 LH |
6709 | unsigned long min, unsigned long max, unsigned int depth, |
6710 | enum mt_dump_format format) | |
54a611b6 LH |
6711 | { |
6712 | struct maple_range_64 *node = &mte_to_node(entry)->mr64; | |
6713 | bool leaf = mte_is_leaf(entry); | |
6714 | unsigned long first = min; | |
6715 | int i; | |
6716 | ||
6717 | pr_cont(" contents: "); | |
89f499f3 LH |
6718 | for (i = 0; i < MAPLE_RANGE64_SLOTS - 1; i++) { |
6719 | switch(format) { | |
6720 | case mt_dump_hex: | |
6721 | pr_cont("%p %lX ", node->slot[i], node->pivot[i]); | |
6722 | break; | |
6723 | default: | |
6724 | case mt_dump_dec: | |
6725 | pr_cont("%p %lu ", node->slot[i], node->pivot[i]); | |
6726 | } | |
6727 | } | |
54a611b6 LH |
6728 | pr_cont("%p\n", node->slot[i]); |
6729 | for (i = 0; i < MAPLE_RANGE64_SLOTS; i++) { | |
6730 | unsigned long last = max; | |
6731 | ||
6732 | if (i < (MAPLE_RANGE64_SLOTS - 1)) | |
6733 | last = node->pivot[i]; | |
bd592703 | 6734 | else if (!node->slot[i] && max != mt_node_max(entry)) |
54a611b6 LH |
6735 | break; |
6736 | if (last == 0 && i > 0) | |
6737 | break; | |
6738 | if (leaf) | |
6739 | mt_dump_entry(mt_slot(mt, node->slot, i), | |
89f499f3 | 6740 | first, last, depth + 1, format); |
54a611b6 LH |
6741 | else if (node->slot[i]) |
6742 | mt_dump_node(mt, mt_slot(mt, node->slot, i), | |
89f499f3 | 6743 | first, last, depth + 1, format); |
54a611b6 LH |
6744 | |
6745 | if (last == max) | |
6746 | break; | |
6747 | if (last > max) { | |
89f499f3 LH |
6748 | switch(format) { |
6749 | case mt_dump_hex: | |
6750 | pr_err("node %p last (%lx) > max (%lx) at pivot %d!\n", | |
54a611b6 | 6751 | node, last, max, i); |
89f499f3 LH |
6752 | break; |
6753 | default: | |
6754 | case mt_dump_dec: | |
6755 | pr_err("node %p last (%lu) > max (%lu) at pivot %d!\n", | |
6756 | node, last, max, i); | |
6757 | } | |
54a611b6 LH |
6758 | } |
6759 | first = last + 1; | |
6760 | } | |
6761 | } | |
6762 | ||
6763 | static void mt_dump_arange64(const struct maple_tree *mt, void *entry, | |
89f499f3 LH |
6764 | unsigned long min, unsigned long max, unsigned int depth, |
6765 | enum mt_dump_format format) | |
54a611b6 LH |
6766 | { |
6767 | struct maple_arange_64 *node = &mte_to_node(entry)->ma64; | |
6768 | bool leaf = mte_is_leaf(entry); | |
6769 | unsigned long first = min; | |
6770 | int i; | |
6771 | ||
6772 | pr_cont(" contents: "); | |
6773 | for (i = 0; i < MAPLE_ARANGE64_SLOTS; i++) | |
6774 | pr_cont("%lu ", node->gap[i]); | |
6775 | pr_cont("| %02X %02X| ", node->meta.end, node->meta.gap); | |
6776 | for (i = 0; i < MAPLE_ARANGE64_SLOTS - 1; i++) | |
6777 | pr_cont("%p %lu ", node->slot[i], node->pivot[i]); | |
6778 | pr_cont("%p\n", node->slot[i]); | |
6779 | for (i = 0; i < MAPLE_ARANGE64_SLOTS; i++) { | |
6780 | unsigned long last = max; | |
6781 | ||
6782 | if (i < (MAPLE_ARANGE64_SLOTS - 1)) | |
6783 | last = node->pivot[i]; | |
6784 | else if (!node->slot[i]) | |
6785 | break; | |
6786 | if (last == 0 && i > 0) | |
6787 | break; | |
6788 | if (leaf) | |
6789 | mt_dump_entry(mt_slot(mt, node->slot, i), | |
89f499f3 | 6790 | first, last, depth + 1, format); |
54a611b6 LH |
6791 | else if (node->slot[i]) |
6792 | mt_dump_node(mt, mt_slot(mt, node->slot, i), | |
89f499f3 | 6793 | first, last, depth + 1, format); |
54a611b6 LH |
6794 | |
6795 | if (last == max) | |
6796 | break; | |
6797 | if (last > max) { | |
6798 | pr_err("node %p last (%lu) > max (%lu) at pivot %d!\n", | |
6799 | node, last, max, i); | |
6800 | break; | |
6801 | } | |
6802 | first = last + 1; | |
6803 | } | |
6804 | } | |
6805 | ||
6806 | static void mt_dump_node(const struct maple_tree *mt, void *entry, | |
89f499f3 LH |
6807 | unsigned long min, unsigned long max, unsigned int depth, |
6808 | enum mt_dump_format format) | |
54a611b6 LH |
6809 | { |
6810 | struct maple_node *node = mte_to_node(entry); | |
6811 | unsigned int type = mte_node_type(entry); | |
6812 | unsigned int i; | |
6813 | ||
89f499f3 | 6814 | mt_dump_range(min, max, depth, format); |
54a611b6 LH |
6815 | |
6816 | pr_cont("node %p depth %d type %d parent %p", node, depth, type, | |
6817 | node ? node->parent : NULL); | |
6818 | switch (type) { | |
6819 | case maple_dense: | |
6820 | pr_cont("\n"); | |
6821 | for (i = 0; i < MAPLE_NODE_SLOTS; i++) { | |
6822 | if (min + i > max) | |
6823 | pr_cont("OUT OF RANGE: "); | |
6824 | mt_dump_entry(mt_slot(mt, node->slot, i), | |
89f499f3 | 6825 | min + i, min + i, depth, format); |
54a611b6 LH |
6826 | } |
6827 | break; | |
6828 | case maple_leaf_64: | |
6829 | case maple_range_64: | |
89f499f3 | 6830 | mt_dump_range64(mt, entry, min, max, depth, format); |
54a611b6 LH |
6831 | break; |
6832 | case maple_arange_64: | |
89f499f3 | 6833 | mt_dump_arange64(mt, entry, min, max, depth, format); |
54a611b6 LH |
6834 | break; |
6835 | ||
6836 | default: | |
6837 | pr_cont(" UNKNOWN TYPE\n"); | |
6838 | } | |
6839 | } | |
6840 | ||
89f499f3 | 6841 | void mt_dump(const struct maple_tree *mt, enum mt_dump_format format) |
54a611b6 LH |
6842 | { |
6843 | void *entry = rcu_dereference_check(mt->ma_root, mt_locked(mt)); | |
6844 | ||
6845 | pr_info("maple_tree(%p) flags %X, height %u root %p\n", | |
6846 | mt, mt->ma_flags, mt_height(mt), entry); | |
6847 | if (!xa_is_node(entry)) | |
89f499f3 | 6848 | mt_dump_entry(entry, 0, 0, 0, format); |
54a611b6 | 6849 | else if (entry) |
89f499f3 | 6850 | mt_dump_node(mt, entry, 0, mt_node_max(entry), 0, format); |
54a611b6 | 6851 | } |
120b1162 | 6852 | EXPORT_SYMBOL_GPL(mt_dump); |
54a611b6 LH |
6853 | |
6854 | /* | |
6855 | * Calculate the maximum gap in a node and check if that's what is reported in | |
6856 | * the parent (unless root). | |
6857 | */ | |
6858 | static void mas_validate_gaps(struct ma_state *mas) | |
6859 | { | |
6860 | struct maple_enode *mte = mas->node; | |
f8e5eac8 PZ |
6861 | struct maple_node *p_mn, *node = mte_to_node(mte); |
6862 | enum maple_type mt = mte_node_type(mas->node); | |
54a611b6 LH |
6863 | unsigned long gap = 0, max_gap = 0; |
6864 | unsigned long p_end, p_start = mas->min; | |
f8e5eac8 | 6865 | unsigned char p_slot, offset; |
54a611b6 | 6866 | unsigned long *gaps = NULL; |
f8e5eac8 PZ |
6867 | unsigned long *pivots = ma_pivots(node, mt); |
6868 | unsigned int i; | |
54a611b6 | 6869 | |
f8e5eac8 | 6870 | if (ma_is_dense(mt)) { |
54a611b6 LH |
6871 | for (i = 0; i < mt_slot_count(mte); i++) { |
6872 | if (mas_get_slot(mas, i)) { | |
6873 | if (gap > max_gap) | |
6874 | max_gap = gap; | |
6875 | gap = 0; | |
6876 | continue; | |
6877 | } | |
6878 | gap++; | |
6879 | } | |
6880 | goto counted; | |
6881 | } | |
6882 | ||
f8e5eac8 | 6883 | gaps = ma_gaps(node, mt); |
54a611b6 | 6884 | for (i = 0; i < mt_slot_count(mte); i++) { |
29b2681f | 6885 | p_end = mas_safe_pivot(mas, pivots, i, mt); |
54a611b6 LH |
6886 | |
6887 | if (!gaps) { | |
f8e5eac8 PZ |
6888 | if (!mas_get_slot(mas, i)) |
6889 | gap = p_end - p_start + 1; | |
54a611b6 LH |
6890 | } else { |
6891 | void *entry = mas_get_slot(mas, i); | |
6892 | ||
6893 | gap = gaps[i]; | |
f8e5eac8 PZ |
6894 | MT_BUG_ON(mas->tree, !entry); |
6895 | ||
6896 | if (gap > p_end - p_start + 1) { | |
6897 | pr_err("%p[%u] %lu >= %lu - %lu + 1 (%lu)\n", | |
6898 | mas_mn(mas), i, gap, p_end, p_start, | |
6899 | p_end - p_start + 1); | |
6900 | MT_BUG_ON(mas->tree, gap > p_end - p_start + 1); | |
54a611b6 LH |
6901 | } |
6902 | } | |
6903 | ||
6904 | if (gap > max_gap) | |
6905 | max_gap = gap; | |
f8e5eac8 | 6906 | |
54a611b6 LH |
6907 | p_start = p_end + 1; |
6908 | if (p_end >= mas->max) | |
6909 | break; | |
6910 | } | |
6911 | ||
6912 | counted: | |
f8e5eac8 PZ |
6913 | if (mt == maple_arange_64) { |
6914 | offset = ma_meta_gap(node, mt); | |
6915 | if (offset > i) { | |
6916 | pr_err("gap offset %p[%u] is invalid\n", node, offset); | |
6917 | MT_BUG_ON(mas->tree, 1); | |
6918 | } | |
6919 | ||
6920 | if (gaps[offset] != max_gap) { | |
6921 | pr_err("gap %p[%u] is not the largest gap %lu\n", | |
6922 | node, offset, max_gap); | |
6923 | MT_BUG_ON(mas->tree, 1); | |
6924 | } | |
6925 | ||
6926 | MT_BUG_ON(mas->tree, !gaps); | |
6927 | for (i++ ; i < mt_slot_count(mte); i++) { | |
6928 | if (gaps[i] != 0) { | |
6929 | pr_err("gap %p[%u] beyond node limit != 0\n", | |
6930 | node, i); | |
6931 | MT_BUG_ON(mas->tree, 1); | |
6932 | } | |
6933 | } | |
6934 | } | |
6935 | ||
54a611b6 LH |
6936 | if (mte_is_root(mte)) |
6937 | return; | |
6938 | ||
6939 | p_slot = mte_parent_slot(mas->node); | |
6940 | p_mn = mte_parent(mte); | |
6941 | MT_BUG_ON(mas->tree, max_gap > mas->max); | |
afc754c6 | 6942 | if (ma_gaps(p_mn, mas_parent_type(mas, mte))[p_slot] != max_gap) { |
54a611b6 | 6943 | pr_err("gap %p[%u] != %lu\n", p_mn, p_slot, max_gap); |
89f499f3 | 6944 | mt_dump(mas->tree, mt_dump_hex); |
f8e5eac8 | 6945 | MT_BUG_ON(mas->tree, 1); |
54a611b6 | 6946 | } |
54a611b6 LH |
6947 | } |
6948 | ||
6949 | static void mas_validate_parent_slot(struct ma_state *mas) | |
6950 | { | |
6951 | struct maple_node *parent; | |
6952 | struct maple_enode *node; | |
afc754c6 LH |
6953 | enum maple_type p_type; |
6954 | unsigned char p_slot; | |
54a611b6 LH |
6955 | void __rcu **slots; |
6956 | int i; | |
6957 | ||
6958 | if (mte_is_root(mas->node)) | |
6959 | return; | |
6960 | ||
afc754c6 LH |
6961 | p_slot = mte_parent_slot(mas->node); |
6962 | p_type = mas_parent_type(mas, mas->node); | |
54a611b6 LH |
6963 | parent = mte_parent(mas->node); |
6964 | slots = ma_slots(parent, p_type); | |
6965 | MT_BUG_ON(mas->tree, mas_mn(mas) == parent); | |
6966 | ||
6967 | /* Check prev/next parent slot for duplicate node entry */ | |
6968 | ||
6969 | for (i = 0; i < mt_slots[p_type]; i++) { | |
6970 | node = mas_slot(mas, slots, i); | |
6971 | if (i == p_slot) { | |
6972 | if (node != mas->node) | |
6973 | pr_err("parent %p[%u] does not have %p\n", | |
6974 | parent, i, mas_mn(mas)); | |
6975 | MT_BUG_ON(mas->tree, node != mas->node); | |
6976 | } else if (node == mas->node) { | |
6977 | pr_err("Invalid child %p at parent %p[%u] p_slot %u\n", | |
6978 | mas_mn(mas), parent, i, p_slot); | |
6979 | MT_BUG_ON(mas->tree, node == mas->node); | |
6980 | } | |
6981 | } | |
6982 | } | |
6983 | ||
6984 | static void mas_validate_child_slot(struct ma_state *mas) | |
6985 | { | |
6986 | enum maple_type type = mte_node_type(mas->node); | |
6987 | void __rcu **slots = ma_slots(mte_to_node(mas->node), type); | |
6988 | unsigned long *pivots = ma_pivots(mte_to_node(mas->node), type); | |
6989 | struct maple_enode *child; | |
6990 | unsigned char i; | |
6991 | ||
6992 | if (mte_is_leaf(mas->node)) | |
6993 | return; | |
6994 | ||
6995 | for (i = 0; i < mt_slots[type]; i++) { | |
6996 | child = mas_slot(mas, slots, i); | |
54a611b6 | 6997 | |
e93fda5a PZ |
6998 | if (!child) { |
6999 | pr_err("Non-leaf node lacks child at %p[%u]\n", | |
7000 | mas_mn(mas), i); | |
7001 | MT_BUG_ON(mas->tree, 1); | |
7002 | } | |
54a611b6 LH |
7003 | |
7004 | if (mte_parent_slot(child) != i) { | |
7005 | pr_err("Slot error at %p[%u]: child %p has pslot %u\n", | |
7006 | mas_mn(mas), i, mte_to_node(child), | |
7007 | mte_parent_slot(child)); | |
7008 | MT_BUG_ON(mas->tree, 1); | |
7009 | } | |
7010 | ||
7011 | if (mte_parent(child) != mte_to_node(mas->node)) { | |
7012 | pr_err("child %p has parent %p not %p\n", | |
7013 | mte_to_node(child), mte_parent(child), | |
7014 | mte_to_node(mas->node)); | |
7015 | MT_BUG_ON(mas->tree, 1); | |
7016 | } | |
e93fda5a PZ |
7017 | |
7018 | if (i < mt_pivots[type] && pivots[i] == mas->max) | |
7019 | break; | |
54a611b6 LH |
7020 | } |
7021 | } | |
7022 | ||
7023 | /* | |
33af39d0 PZ |
7024 | * Validate all pivots are within mas->min and mas->max, check metadata ends |
7025 | * where the maximum ends and ensure there is no slots or pivots set outside of | |
7026 | * the end of the data. | |
54a611b6 LH |
7027 | */ |
7028 | static void mas_validate_limits(struct ma_state *mas) | |
7029 | { | |
7030 | int i; | |
7031 | unsigned long prev_piv = 0; | |
7032 | enum maple_type type = mte_node_type(mas->node); | |
7033 | void __rcu **slots = ma_slots(mte_to_node(mas->node), type); | |
7034 | unsigned long *pivots = ma_pivots(mas_mn(mas), type); | |
7035 | ||
54a611b6 LH |
7036 | for (i = 0; i < mt_slots[type]; i++) { |
7037 | unsigned long piv; | |
7038 | ||
7039 | piv = mas_safe_pivot(mas, pivots, i, type); | |
7040 | ||
33af39d0 PZ |
7041 | if (!piv && (i != 0)) { |
7042 | pr_err("Missing node limit pivot at %p[%u]", | |
7043 | mas_mn(mas), i); | |
7044 | MAS_WARN_ON(mas, 1); | |
54a611b6 LH |
7045 | } |
7046 | ||
7047 | if (prev_piv > piv) { | |
7048 | pr_err("%p[%u] piv %lu < prev_piv %lu\n", | |
7049 | mas_mn(mas), i, piv, prev_piv); | |
e6d6792a | 7050 | MAS_WARN_ON(mas, piv < prev_piv); |
54a611b6 LH |
7051 | } |
7052 | ||
7053 | if (piv < mas->min) { | |
7054 | pr_err("%p[%u] %lu < %lu\n", mas_mn(mas), i, | |
7055 | piv, mas->min); | |
e6d6792a | 7056 | MAS_WARN_ON(mas, piv < mas->min); |
54a611b6 LH |
7057 | } |
7058 | if (piv > mas->max) { | |
7059 | pr_err("%p[%u] %lu > %lu\n", mas_mn(mas), i, | |
7060 | piv, mas->max); | |
e6d6792a | 7061 | MAS_WARN_ON(mas, piv > mas->max); |
54a611b6 LH |
7062 | } |
7063 | prev_piv = piv; | |
7064 | if (piv == mas->max) | |
7065 | break; | |
7066 | } | |
33af39d0 PZ |
7067 | |
7068 | if (mas_data_end(mas) != i) { | |
7069 | pr_err("node%p: data_end %u != the last slot offset %u\n", | |
7070 | mas_mn(mas), mas_data_end(mas), i); | |
7071 | MT_BUG_ON(mas->tree, 1); | |
7072 | } | |
7073 | ||
54a611b6 LH |
7074 | for (i += 1; i < mt_slots[type]; i++) { |
7075 | void *entry = mas_slot(mas, slots, i); | |
7076 | ||
7077 | if (entry && (i != mt_slots[type] - 1)) { | |
7078 | pr_err("%p[%u] should not have entry %p\n", mas_mn(mas), | |
7079 | i, entry); | |
7080 | MT_BUG_ON(mas->tree, entry != NULL); | |
7081 | } | |
7082 | ||
7083 | if (i < mt_pivots[type]) { | |
7084 | unsigned long piv = pivots[i]; | |
7085 | ||
7086 | if (!piv) | |
7087 | continue; | |
7088 | ||
7089 | pr_err("%p[%u] should not have piv %lu\n", | |
7090 | mas_mn(mas), i, piv); | |
e6d6792a | 7091 | MAS_WARN_ON(mas, i < mt_pivots[type] - 1); |
54a611b6 LH |
7092 | } |
7093 | } | |
7094 | } | |
7095 | ||
7096 | static void mt_validate_nulls(struct maple_tree *mt) | |
7097 | { | |
7098 | void *entry, *last = (void *)1; | |
7099 | unsigned char offset = 0; | |
7100 | void __rcu **slots; | |
7101 | MA_STATE(mas, mt, 0, 0); | |
7102 | ||
7103 | mas_start(&mas); | |
7104 | if (mas_is_none(&mas) || (mas.node == MAS_ROOT)) | |
7105 | return; | |
7106 | ||
7107 | while (!mte_is_leaf(mas.node)) | |
7108 | mas_descend(&mas); | |
7109 | ||
7110 | slots = ma_slots(mte_to_node(mas.node), mte_node_type(mas.node)); | |
7111 | do { | |
7112 | entry = mas_slot(&mas, slots, offset); | |
7113 | if (!last && !entry) { | |
7114 | pr_err("Sequential nulls end at %p[%u]\n", | |
7115 | mas_mn(&mas), offset); | |
7116 | } | |
7117 | MT_BUG_ON(mt, !last && !entry); | |
7118 | last = entry; | |
7119 | if (offset == mas_data_end(&mas)) { | |
7120 | mas_next_node(&mas, mas_mn(&mas), ULONG_MAX); | |
7121 | if (mas_is_none(&mas)) | |
7122 | return; | |
7123 | offset = 0; | |
7124 | slots = ma_slots(mte_to_node(mas.node), | |
7125 | mte_node_type(mas.node)); | |
7126 | } else { | |
7127 | offset++; | |
7128 | } | |
7129 | ||
7130 | } while (!mas_is_none(&mas)); | |
7131 | } | |
7132 | ||
7133 | /* | |
7134 | * validate a maple tree by checking: | |
7135 | * 1. The limits (pivots are within mas->min to mas->max) | |
7136 | * 2. The gap is correctly set in the parents | |
7137 | */ | |
7138 | void mt_validate(struct maple_tree *mt) | |
7139 | { | |
7140 | unsigned char end; | |
7141 | ||
7142 | MA_STATE(mas, mt, 0, 0); | |
7143 | rcu_read_lock(); | |
7144 | mas_start(&mas); | |
7145 | if (!mas_searchable(&mas)) | |
7146 | goto done; | |
7147 | ||
a489539e PZ |
7148 | while (!mte_is_leaf(mas.node)) |
7149 | mas_descend(&mas); | |
7150 | ||
54a611b6 | 7151 | while (!mas_is_none(&mas)) { |
e6d6792a | 7152 | MAS_WARN_ON(&mas, mte_dead_node(mas.node)); |
a489539e PZ |
7153 | end = mas_data_end(&mas); |
7154 | if (MAS_WARN_ON(&mas, (end < mt_min_slot_count(mas.node)) && | |
7155 | (mas.max != ULONG_MAX))) { | |
7156 | pr_err("Invalid size %u of %p\n", end, mas_mn(&mas)); | |
54a611b6 | 7157 | } |
a489539e | 7158 | |
54a611b6 | 7159 | mas_validate_parent_slot(&mas); |
54a611b6 | 7160 | mas_validate_limits(&mas); |
a489539e | 7161 | mas_validate_child_slot(&mas); |
54a611b6 LH |
7162 | if (mt_is_alloc(mt)) |
7163 | mas_validate_gaps(&mas); | |
7164 | mas_dfs_postorder(&mas, ULONG_MAX); | |
7165 | } | |
7166 | mt_validate_nulls(mt); | |
7167 | done: | |
7168 | rcu_read_unlock(); | |
7169 | ||
7170 | } | |
120b1162 | 7171 | EXPORT_SYMBOL_GPL(mt_validate); |
54a611b6 | 7172 | |
f0a1f866 LH |
7173 | void mas_dump(const struct ma_state *mas) |
7174 | { | |
7175 | pr_err("MAS: tree=%p enode=%p ", mas->tree, mas->node); | |
7176 | if (mas_is_none(mas)) | |
7177 | pr_err("(MAS_NONE) "); | |
7178 | else if (mas_is_ptr(mas)) | |
7179 | pr_err("(MAS_ROOT) "); | |
7180 | else if (mas_is_start(mas)) | |
7181 | pr_err("(MAS_START) "); | |
7182 | else if (mas_is_paused(mas)) | |
7183 | pr_err("(MAS_PAUSED) "); | |
7184 | ||
7185 | pr_err("[%u] index=%lx last=%lx\n", mas->offset, mas->index, mas->last); | |
7186 | pr_err(" min=%lx max=%lx alloc=%p, depth=%u, flags=%x\n", | |
7187 | mas->min, mas->max, mas->alloc, mas->depth, mas->mas_flags); | |
7188 | if (mas->index > mas->last) | |
7189 | pr_err("Check index & last\n"); | |
7190 | } | |
7191 | EXPORT_SYMBOL_GPL(mas_dump); | |
7192 | ||
7193 | void mas_wr_dump(const struct ma_wr_state *wr_mas) | |
7194 | { | |
7195 | pr_err("WR_MAS: node=%p r_min=%lx r_max=%lx\n", | |
7196 | wr_mas->node, wr_mas->r_min, wr_mas->r_max); | |
7197 | pr_err(" type=%u off_end=%u, node_end=%u, end_piv=%lx\n", | |
7198 | wr_mas->type, wr_mas->offset_end, wr_mas->node_end, | |
7199 | wr_mas->end_piv); | |
7200 | } | |
7201 | EXPORT_SYMBOL_GPL(mas_wr_dump); | |
7202 | ||
54a611b6 | 7203 | #endif /* CONFIG_DEBUG_MAPLE_TREE */ |