1 /* SPDX-License-Identifier: GPL-2.0+ */
2 #ifndef _LINUX_MAPLE_TREE_H
3 #define _LINUX_MAPLE_TREE_H
5 * Maple Tree - An RCU-safe adaptive tree for storing ranges
6 * Copyright (c) 2018-2022 Oracle
11 #include <linux/kernel.h>
12 #include <linux/rcupdate.h>
13 #include <linux/spinlock.h>
14 /* #define CONFIG_MAPLE_RCU_DISABLED */
15 /* #define CONFIG_DEBUG_MAPLE_TREE_VERBOSE */
18 * Allocated nodes are mutable until they have been inserted into the tree,
19 * at which time they cannot change their type until they have been removed
20 * from the tree and an RCU grace period has passed.
22 * Removed nodes have their ->parent set to point to themselves. RCU readers
23 * check ->parent before relying on the value that they loaded from the
24 * slots array. This lets us reuse the slots array for the RCU head.
26 * Nodes in the tree point to their parent unless bit 0 is set.
28 #if defined(CONFIG_64BIT) || defined(BUILD_VDSO32_64)
30 #define MAPLE_NODE_SLOTS 31 /* 256 bytes including ->parent */
31 #define MAPLE_RANGE64_SLOTS 16 /* 256 bytes */
32 #define MAPLE_ARANGE64_SLOTS 10 /* 240 bytes */
33 #define MAPLE_ARANGE64_META_MAX 15 /* Out of range for metadata */
34 #define MAPLE_ALLOC_SLOTS (MAPLE_NODE_SLOTS - 1)
37 #define MAPLE_NODE_SLOTS 63 /* 256 bytes including ->parent */
38 #define MAPLE_RANGE64_SLOTS 32 /* 256 bytes */
39 #define MAPLE_ARANGE64_SLOTS 21 /* 240 bytes */
40 #define MAPLE_ARANGE64_META_MAX 31 /* Out of range for metadata */
41 #define MAPLE_ALLOC_SLOTS (MAPLE_NODE_SLOTS - 2)
42 #endif /* defined(CONFIG_64BIT) || defined(BUILD_VDSO32_64) */
44 #define MAPLE_NODE_MASK 255UL
47 * The node->parent of the root node has bit 0 set and the rest of the pointer
48 * is a pointer to the tree itself. No more bits are available in this pointer
49 * (on m68k, the data structure may only be 2-byte aligned).
51 * Internal non-root nodes can only have maple_range_* nodes as parents. The
52 * parent pointer is 256B aligned like all other tree nodes. When storing a 32
53 * or 64 bit values, the offset can fit into 4 bits. The 16 bit values need an
54 * extra bit to store the offset. This extra bit comes from a reuse of the last
55 * bit in the node type. This is possible by using bit 1 to indicate if bit 2
56 * is part of the type or the slot.
58 * Once the type is decided, the decision of an allocation range type or a range
59 * type is done by examining the immutable tree flag for the MAPLE_ALLOC_RANGE
64 * 0x?00 = 16 bit nodes
65 * 0x010 = 32 bit nodes
66 * 0x110 = 64 bit nodes
68 * Slot size and location in the parent pointer:
69 * type : slot location
71 * 0x?00 : 16 bit values, type in 0-1, slot in 2-6
72 * 0x010 : 32 bit values, type in 0-2, slot in 3-6
73 * 0x110 : 64 bit values, type in 0-2, slot in 3-6
77 * This metadata is used to optimize the gap updating code and in reverse
78 * searching for gaps or any other code that needs to find the end of the data.
80 struct maple_metadata {
86 * Leaf nodes do not store pointers to nodes, they store user data. Users may
87 * store almost any bit pattern. As noted above, the optimisation of storing an
88 * entry at 0 in the root pointer cannot be done for data which have the bottom
89 * two bits set to '10'. We also reserve values with the bottom two bits set to
90 * '10' which are below 4096 (ie 2, 6, 10 .. 4094) for internal use. Some APIs
91 * return errnos as a negative errno shifted right by two bits and the bottom
92 * two bits set to '10', and while choosing to store these values in the array
93 * is not an error, it may lead to confusion if you're testing for an error with
96 * Non-leaf nodes store the type of the node pointed to (enum maple_type in bits
97 * 3-6), bit 2 is reserved. That leaves bits 0-1 unused for now.
99 * In regular B-Tree terms, pivots are called keys. The term pivot is used to
100 * indicate that the tree is specifying ranges, Pivots may appear in the
101 * subtree with an entry attached to the value whereas keys are unique to a
102 * specific position of a B-tree. Pivot values are inclusive of the slot with
106 struct maple_range_64 {
107 struct maple_pnode *parent;
108 unsigned long pivot[MAPLE_RANGE64_SLOTS - 1];
110 void __rcu *slot[MAPLE_RANGE64_SLOTS];
112 void __rcu *pad[MAPLE_RANGE64_SLOTS - 1];
113 struct maple_metadata meta;
119 * At tree creation time, the user can specify that they're willing to trade off
120 * storing fewer entries in a tree in return for storing more information in
123 * The maple tree supports recording the largest range of NULL entries available
124 * in this node, also called gaps. This optimises the tree for allocating a
127 struct maple_arange_64 {
128 struct maple_pnode *parent;
129 unsigned long pivot[MAPLE_ARANGE64_SLOTS - 1];
130 void __rcu *slot[MAPLE_ARANGE64_SLOTS];
131 unsigned long gap[MAPLE_ARANGE64_SLOTS];
132 struct maple_metadata meta;
137 unsigned char node_count;
138 unsigned int request_count;
139 struct maple_alloc *slot[MAPLE_ALLOC_SLOTS];
142 struct maple_topiary {
143 struct maple_pnode *parent;
144 struct maple_enode *next; /* Overlaps the pivot */
156 * DOC: Maple tree flags
158 * * MT_FLAGS_ALLOC_RANGE - Track gaps in this tree
159 * * MT_FLAGS_USE_RCU - Operate in RCU mode
160 * * MT_FLAGS_HEIGHT_OFFSET - The position of the tree height in the flags
161 * * MT_FLAGS_HEIGHT_MASK - The mask for the maple tree height value
162 * * MT_FLAGS_LOCK_MASK - How the mt_lock is used
163 * * MT_FLAGS_LOCK_IRQ - Acquired irq-safe
164 * * MT_FLAGS_LOCK_BH - Acquired bh-safe
165 * * MT_FLAGS_LOCK_EXTERN - mt_lock is not used
167 * MAPLE_HEIGHT_MAX The largest height that can be stored
169 #define MT_FLAGS_ALLOC_RANGE 0x01
170 #define MT_FLAGS_USE_RCU 0x02
171 #define MT_FLAGS_HEIGHT_OFFSET 0x02
172 #define MT_FLAGS_HEIGHT_MASK 0x7C
173 #define MT_FLAGS_LOCK_MASK 0x300
174 #define MT_FLAGS_LOCK_IRQ 0x100
175 #define MT_FLAGS_LOCK_BH 0x200
176 #define MT_FLAGS_LOCK_EXTERN 0x300
178 #define MAPLE_HEIGHT_MAX 31
181 #define MAPLE_NODE_TYPE_MASK 0x0F
182 #define MAPLE_NODE_TYPE_SHIFT 0x03
184 #define MAPLE_RESERVED_RANGE 4096
186 #ifdef CONFIG_LOCKDEP
187 typedef struct lockdep_map *lockdep_map_p;
188 #define mt_lock_is_held(mt) lock_is_held(mt->ma_external_lock)
189 #define mt_set_external_lock(mt, lock) \
190 (mt)->ma_external_lock = &(lock)->dep_map
192 typedef struct { /* nothing */ } lockdep_map_p;
193 #define mt_lock_is_held(mt) 1
194 #define mt_set_external_lock(mt, lock) do { } while (0)
198 * If the tree contains a single entry at index 0, it is usually stored in
199 * tree->ma_root. To optimise for the page cache, an entry which ends in '00',
200 * '01' or '11' is stored in the root, but an entry which ends in '10' will be
201 * stored in a node. Bits 3-6 are used to store enum maple_type.
203 * The flags are used both to store some immutable information about this tree
204 * (set at tree creation time) and dynamic information set under the spinlock.
206 * Another use of flags are to indicate global states of the tree. This is the
207 * case with the MAPLE_USE_RCU flag, which indicates the tree is currently in
208 * RCU mode. This mode was added to allow the tree to reuse nodes instead of
209 * re-allocating and RCU freeing nodes when there is a single user.
214 lockdep_map_p ma_external_lock;
217 unsigned int ma_flags;
221 * MTREE_INIT() - Initialize a maple tree
222 * @name: The maple tree name
223 * @__flags: The maple tree flags
226 #define MTREE_INIT(name, __flags) { \
227 .ma_lock = __SPIN_LOCK_UNLOCKED((name).ma_lock), \
228 .ma_flags = __flags, \
233 * MTREE_INIT_EXT() - Initialize a maple tree with an external lock.
234 * @name: The tree name
235 * @__flags: The maple tree flags
236 * @__lock: The external lock
238 #ifdef CONFIG_LOCKDEP
239 #define MTREE_INIT_EXT(name, __flags, __lock) { \
240 .ma_external_lock = &(__lock).dep_map, \
241 .ma_flags = (__flags), \
245 #define MTREE_INIT_EXT(name, __flags, __lock) MTREE_INIT(name, __flags)
248 #define DEFINE_MTREE(name) \
249 struct maple_tree name = MTREE_INIT(name, 0)
251 #define mtree_lock(mt) spin_lock((&(mt)->ma_lock))
252 #define mtree_unlock(mt) spin_unlock((&(mt)->ma_lock))
255 * The Maple Tree squeezes various bits in at various points which aren't
256 * necessarily obvious. Usually, this is done by observing that pointers are
257 * N-byte aligned and thus the bottom log_2(N) bits are available for use. We
258 * don't use the high bits of pointers to store additional information because
259 * we don't know what bits are unused on any given architecture.
261 * Nodes are 256 bytes in size and are also aligned to 256 bytes, giving us 8
262 * low bits for our own purposes. Nodes are currently of 4 types:
263 * 1. Single pointer (Range is 0-0)
264 * 2. Non-leaf Allocation Range nodes
265 * 3. Non-leaf Range nodes
266 * 4. Leaf Range nodes All nodes consist of a number of node slots,
267 * pivots, and a parent pointer.
273 struct maple_pnode *parent;
274 void __rcu *slot[MAPLE_NODE_SLOTS];
279 struct maple_enode *piv_parent;
280 unsigned char parent_slot;
281 enum maple_type type;
282 unsigned char slot_len;
283 unsigned int ma_flags;
285 struct maple_range_64 mr64;
286 struct maple_arange_64 ma64;
287 struct maple_alloc alloc;
292 * More complicated stores can cause two nodes to become one or three and
293 * potentially alter the height of the tree. Either half of the tree may need
294 * to be rebalanced against the other. The ma_topiary struct is used to track
295 * which nodes have been 'cut' from the tree so that the change can be done
296 * safely at a later date. This is done to support RCU.
299 struct maple_enode *head;
300 struct maple_enode *tail;
301 struct maple_tree *mtree;
304 void *mtree_load(struct maple_tree *mt, unsigned long index);
306 int mtree_insert(struct maple_tree *mt, unsigned long index,
307 void *entry, gfp_t gfp);
308 int mtree_insert_range(struct maple_tree *mt, unsigned long first,
309 unsigned long last, void *entry, gfp_t gfp);
310 int mtree_alloc_range(struct maple_tree *mt, unsigned long *startp,
311 void *entry, unsigned long size, unsigned long min,
312 unsigned long max, gfp_t gfp);
313 int mtree_alloc_rrange(struct maple_tree *mt, unsigned long *startp,
314 void *entry, unsigned long size, unsigned long min,
315 unsigned long max, gfp_t gfp);
317 int mtree_store_range(struct maple_tree *mt, unsigned long first,
318 unsigned long last, void *entry, gfp_t gfp);
319 int mtree_store(struct maple_tree *mt, unsigned long index,
320 void *entry, gfp_t gfp);
321 void *mtree_erase(struct maple_tree *mt, unsigned long index);
323 void mtree_destroy(struct maple_tree *mt);
324 void __mt_destroy(struct maple_tree *mt);
327 * mtree_empty() - Determine if a tree has any present entries.
330 * Context: Any context.
331 * Return: %true if the tree contains only NULL pointers.
333 static inline bool mtree_empty(const struct maple_tree *mt)
335 return mt->ma_root == NULL;
341 * The maple state is defined in the struct ma_state and is used to keep track
342 * of information during operations, and even between operations when using the
345 * If state->node has bit 0 set then it references a tree location which is not
346 * a node (eg the root). If bit 1 is set, the rest of the bits are a negative
347 * errno. Bit 2 (the 'unallocated slots' bit) is clear. Bits 3-6 indicate the
350 * state->alloc either has a request number of nodes or an allocated node. If
351 * stat->alloc has a requested number of nodes, the first bit will be set (0x1)
352 * and the remaining bits are the value. If state->alloc is a node, then the
353 * node will be of type maple_alloc. maple_alloc has MAPLE_NODE_SLOTS - 1 for
354 * storing more allocated nodes, a total number of nodes allocated, and the
355 * node_count in this node. node_count is the number of allocated nodes in this
356 * node. The scaling beyond MAPLE_NODE_SLOTS - 1 is handled by storing further
357 * nodes into state->alloc->slot[0]'s node. Nodes are taken from state->alloc
358 * by removing a node from the state->alloc node until state->alloc->node_count
359 * is 1, when state->alloc is returned and the state->alloc->slot[0] is promoted
360 * to state->alloc. Nodes are pushed onto state->alloc by putting the current
361 * state->alloc into the pushed node's slot[0].
363 * The state also contains the implied min/max of the state->node, the depth of
364 * this search, and the offset. The implied min/max are either from the parent
365 * node or are 0-oo for the root node. The depth is incremented or decremented
366 * every time a node is walked down or up. The offset is the slot/pivot of
367 * interest in the node - either for reading or writing.
369 * When returning a value the maple state index and last respectively contain
370 * the start and end of the range for the entry. Ranges are inclusive in the
374 struct maple_tree *tree; /* The tree we're operating in */
375 unsigned long index; /* The index we're operating on - range start */
376 unsigned long last; /* The last index we're operating on - range end */
377 struct maple_enode *node; /* The node containing this entry */
378 unsigned long min; /* The minimum index of this node - implied pivot min */
379 unsigned long max; /* The maximum index of this node - implied pivot max */
380 struct maple_alloc *alloc; /* Allocated nodes for this operation */
381 unsigned char depth; /* depth of tree descent during write */
382 unsigned char offset;
383 unsigned char mas_flags;
387 struct ma_state *mas;
388 struct maple_node *node; /* Decoded mas->node */
389 unsigned long r_min; /* range min */
390 unsigned long r_max; /* range max */
391 enum maple_type type; /* mas->node type */
392 unsigned char offset_end; /* The offset where the write ends */
393 unsigned char node_end; /* mas->node end */
394 unsigned long *pivots; /* mas->node->pivots pointer */
395 unsigned long end_piv; /* The pivot at the offset end */
396 void __rcu **slots; /* mas->node->slots pointer */
397 void *entry; /* The entry to write */
398 void *content; /* The existing entry that is being overwritten */
401 #define mas_lock(mas) spin_lock(&((mas)->tree->ma_lock))
402 #define mas_unlock(mas) spin_unlock(&((mas)->tree->ma_lock))
406 * Special values for ma_state.node.
407 * MAS_START means we have not searched the tree.
408 * MAS_ROOT means we have searched the tree and the entry we found lives in
409 * the root of the tree (ie it has index 0, length 1 and is the only entry in
411 * MAS_NONE means we have searched the tree and there is no node in the
412 * tree for this entry. For example, we searched for index 1 in an empty
413 * tree. Or we have a tree which points to a full leaf node and we
414 * searched for an entry which is larger than can be contained in that
416 * MA_ERROR represents an errno. After dropping the lock and attempting
417 * to resolve the error, the walk would have to be restarted from the
418 * top of the tree as the tree may have been modified.
420 #define MAS_START ((struct maple_enode *)1UL)
421 #define MAS_ROOT ((struct maple_enode *)5UL)
422 #define MAS_NONE ((struct maple_enode *)9UL)
423 #define MAS_PAUSE ((struct maple_enode *)17UL)
424 #define MA_ERROR(err) \
425 ((struct maple_enode *)(((unsigned long)err << 2) | 2UL))
427 #define MA_STATE(name, mt, first, end) \
428 struct ma_state name = { \
438 #define MA_WR_STATE(name, ma_state, wr_entry) \
439 struct ma_wr_state name = { \
445 #define MA_TOPIARY(name, tree) \
446 struct ma_topiary name = { \
452 void *mas_walk(struct ma_state *mas);
453 void *mas_store(struct ma_state *mas, void *entry);
454 void *mas_erase(struct ma_state *mas);
455 int mas_store_gfp(struct ma_state *mas, void *entry, gfp_t gfp);
456 void mas_store_prealloc(struct ma_state *mas, void *entry);
457 void *mas_find(struct ma_state *mas, unsigned long max);
458 void *mas_find_rev(struct ma_state *mas, unsigned long min);
459 int mas_preallocate(struct ma_state *mas, void *entry, gfp_t gfp);
460 bool mas_is_err(struct ma_state *mas);
462 bool mas_nomem(struct ma_state *mas, gfp_t gfp);
463 void mas_pause(struct ma_state *mas);
464 void maple_tree_init(void);
465 void mas_destroy(struct ma_state *mas);
466 int mas_expected_entries(struct ma_state *mas, unsigned long nr_entries);
468 void *mas_prev(struct ma_state *mas, unsigned long min);
469 void *mas_next(struct ma_state *mas, unsigned long max);
471 int mas_empty_area(struct ma_state *mas, unsigned long min, unsigned long max,
474 /* Checks if a mas has not found anything */
475 static inline bool mas_is_none(struct ma_state *mas)
477 return mas->node == MAS_NONE;
480 /* Checks if a mas has been paused */
481 static inline bool mas_is_paused(struct ma_state *mas)
483 return mas->node == MAS_PAUSE;
486 void mas_dup_tree(struct ma_state *oldmas, struct ma_state *mas);
487 void mas_dup_store(struct ma_state *mas, void *entry);
490 * This finds an empty area from the highest address to the lowest.
491 * AKA "Topdown" version,
493 int mas_empty_area_rev(struct ma_state *mas, unsigned long min,
494 unsigned long max, unsigned long size);
496 * mas_reset() - Reset a Maple Tree operation state.
497 * @mas: Maple Tree operation state.
499 * Resets the error or walk state of the @mas so future walks of the
500 * array will start from the root. Use this if you have dropped the
501 * lock and want to reuse the ma_state.
503 * Context: Any context.
505 static inline void mas_reset(struct ma_state *mas)
507 mas->node = MAS_START;
511 * mas_for_each() - Iterate over a range of the maple tree.
512 * @__mas: Maple Tree operation state (maple_state)
513 * @__entry: Entry retrieved from the tree
514 * @__max: maximum index to retrieve from the tree
516 * When returned, mas->index and mas->last will hold the entire range for the
519 * Note: may return the zero entry.
522 #define mas_for_each(__mas, __entry, __max) \
523 while (((__entry) = mas_find((__mas), (__max))) != NULL)
527 * mas_set_range() - Set up Maple Tree operation state for a different index.
528 * @mas: Maple Tree operation state.
529 * @start: New start of range in the Maple Tree.
530 * @last: New end of range in the Maple Tree.
532 * Move the operation state to refer to a different range. This will
533 * have the effect of starting a walk from the top; see mas_next()
534 * to move to an adjacent index.
537 void mas_set_range(struct ma_state *mas, unsigned long start, unsigned long last)
541 mas->node = MAS_START;
545 * mas_set() - Set up Maple Tree operation state for a different index.
546 * @mas: Maple Tree operation state.
547 * @index: New index into the Maple Tree.
549 * Move the operation state to refer to a different index. This will
550 * have the effect of starting a walk from the top; see mas_next()
551 * to move to an adjacent index.
553 static inline void mas_set(struct ma_state *mas, unsigned long index)
556 mas_set_range(mas, index, index);
559 static inline bool mt_external_lock(const struct maple_tree *mt)
561 return (mt->ma_flags & MT_FLAGS_LOCK_MASK) == MT_FLAGS_LOCK_EXTERN;
565 * mt_init_flags() - Initialise an empty maple tree with flags.
567 * @flags: maple tree flags.
569 * If you need to initialise a Maple Tree with special flags (eg, an
570 * allocation tree), use this function.
572 * Context: Any context.
574 static inline void mt_init_flags(struct maple_tree *mt, unsigned int flags)
576 mt->ma_flags = flags;
577 if (!mt_external_lock(mt))
578 spin_lock_init(&mt->ma_lock);
579 rcu_assign_pointer(mt->ma_root, NULL);
583 * mt_init() - Initialise an empty maple tree.
586 * An empty Maple Tree.
588 * Context: Any context.
590 static inline void mt_init(struct maple_tree *mt)
592 mt_init_flags(mt, 0);
595 static inline bool mt_in_rcu(struct maple_tree *mt)
597 #ifdef CONFIG_MAPLE_RCU_DISABLED
600 return mt->ma_flags & MT_FLAGS_USE_RCU;
604 * mt_clear_in_rcu() - Switch the tree to non-RCU mode.
605 * @mt: The Maple Tree
607 static inline void mt_clear_in_rcu(struct maple_tree *mt)
612 if (mt_external_lock(mt)) {
613 BUG_ON(!mt_lock_is_held(mt));
614 mt->ma_flags &= ~MT_FLAGS_USE_RCU;
617 mt->ma_flags &= ~MT_FLAGS_USE_RCU;
623 * mt_set_in_rcu() - Switch the tree to RCU safe mode.
624 * @mt: The Maple Tree
626 static inline void mt_set_in_rcu(struct maple_tree *mt)
631 if (mt_external_lock(mt)) {
632 BUG_ON(!mt_lock_is_held(mt));
633 mt->ma_flags |= MT_FLAGS_USE_RCU;
636 mt->ma_flags |= MT_FLAGS_USE_RCU;
641 static inline unsigned int mt_height(const struct maple_tree *mt)
644 return (mt->ma_flags & MT_FLAGS_HEIGHT_MASK) >> MT_FLAGS_HEIGHT_OFFSET;
647 void *mt_find(struct maple_tree *mt, unsigned long *index, unsigned long max);
648 void *mt_find_after(struct maple_tree *mt, unsigned long *index,
650 void *mt_prev(struct maple_tree *mt, unsigned long index, unsigned long min);
651 void *mt_next(struct maple_tree *mt, unsigned long index, unsigned long max);
654 * mt_for_each - Iterate over each entry starting at index until max.
655 * @__tree: The Maple Tree
656 * @__entry: The current entry
657 * @__index: The index to update to track the location in the tree
658 * @__max: The maximum limit for @index
660 * Note: Will not return the zero entry.
662 #define mt_for_each(__tree, __entry, __index, __max) \
663 for (__entry = mt_find(__tree, &(__index), __max); \
664 __entry; __entry = mt_find_after(__tree, &(__index), __max))
667 #ifdef CONFIG_DEBUG_MAPLE_TREE
668 extern atomic_t maple_tree_tests_run;
669 extern atomic_t maple_tree_tests_passed;
671 void mt_dump(const struct maple_tree *mt);
672 void mt_validate(struct maple_tree *mt);
673 void mt_cache_shrink(void);
674 #define MT_BUG_ON(__tree, __x) do { \
675 atomic_inc(&maple_tree_tests_run); \
677 pr_info("BUG at %s:%d (%u)\n", \
678 __func__, __LINE__, __x); \
680 pr_info("Pass: %u Run:%u\n", \
681 atomic_read(&maple_tree_tests_passed), \
682 atomic_read(&maple_tree_tests_run)); \
685 atomic_inc(&maple_tree_tests_passed); \
689 #define MT_BUG_ON(__tree, __x) BUG_ON(__x)
690 #endif /* CONFIG_DEBUG_MAPLE_TREE */
692 #endif /*_LINUX_MAPLE_TREE_H */