1 // SPDX-License-Identifier: GPL-2.0+
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>
10 * DOC: Interesting implementation details of the Maple Tree
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.
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
23 * The following illustrates the layout of a range64 nodes slots and pivots.
26 * Slots -> | 0 | 1 | 2 | ... | 12 | 13 | 14 | 15 |
28 * │ │ │ │ │ │ │ │ └─ Implied maximum
29 * │ │ │ │ │ │ │ └─ Pivot 14
30 * │ │ │ │ │ │ └─ Pivot 13
31 * │ │ │ │ │ └─ Pivot 12
39 * Internal (non-leaf) nodes contain pointers to other nodes.
40 * Leaf nodes contain entries.
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.
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.
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>
62 #define CREATE_TRACE_POINTS
63 #include <trace/events/maple_tree.h>
65 #define MA_ROOT_PARENT 1
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
73 #define MA_STATE_BULK 1
74 #define MA_STATE_REBALANCE 2
75 #define MA_STATE_PREALLOC 4
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;
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,
89 #define mt_node_max(x) mt_max[mte_node_type(x)]
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,
98 #define mt_slot_count(x) mt_slots[mte_node_type(x)]
100 static const unsigned char mt_pivots[] = {
102 [maple_leaf_64] = MAPLE_RANGE64_SLOTS - 1,
103 [maple_range_64] = MAPLE_RANGE64_SLOTS - 1,
104 [maple_arange_64] = MAPLE_ARANGE64_SLOTS - 1,
106 #define mt_pivot_count(x) mt_pivots[mte_node_type(x)]
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,
114 #define mt_min_slot_count(x) mt_min_slots[mte_node_type(x)]
116 #define MAPLE_BIG_NODE_SLOTS (MAPLE_RANGE64_SLOTS * 2 + 2)
117 #define MAPLE_BIG_NODE_GAPS (MAPLE_ARANGE64_SLOTS * 2 + 1)
119 struct maple_big_node {
120 struct maple_pnode *parent;
121 unsigned long pivot[MAPLE_BIG_NODE_SLOTS - 1];
123 struct maple_enode *slot[MAPLE_BIG_NODE_SLOTS];
125 unsigned long padding[MAPLE_BIG_NODE_GAPS];
126 unsigned long gap[MAPLE_BIG_NODE_GAPS];
130 enum maple_type type;
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.
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;
149 #ifdef CONFIG_KASAN_STACK
150 /* Prevent mas_wr_bnode() from exceeding the stack frame limit */
151 #define noinline_for_kasan noinline_for_stack
153 #define noinline_for_kasan inline
157 static inline struct maple_node *mt_alloc_one(gfp_t gfp)
159 return kmem_cache_alloc(maple_node_cache, gfp);
162 static inline int mt_alloc_bulk(gfp_t gfp, size_t size, void **nodes)
164 return kmem_cache_alloc_bulk(maple_node_cache, gfp, size, nodes);
167 static inline void mt_free_bulk(size_t size, void __rcu **nodes)
169 kmem_cache_free_bulk(maple_node_cache, size, (void **)nodes);
172 static void mt_free_rcu(struct rcu_head *head)
174 struct maple_node *node = container_of(head, struct maple_node, rcu);
176 kmem_cache_free(maple_node_cache, node);
180 * ma_free_rcu() - Use rcu callback to free a maple node
181 * @node: The node to free
183 * The maple tree uses the parent pointer to indicate this node is no longer in
184 * use and will be freed.
186 static void ma_free_rcu(struct maple_node *node)
188 WARN_ON(node->parent != ma_parent_ptr(node));
189 call_rcu(&node->rcu, mt_free_rcu);
192 static void mas_set_height(struct ma_state *mas)
194 unsigned int new_flags = mas->tree->ma_flags;
196 new_flags &= ~MT_FLAGS_HEIGHT_MASK;
197 BUG_ON(mas->depth > MAPLE_HEIGHT_MAX);
198 new_flags |= mas->depth << MT_FLAGS_HEIGHT_OFFSET;
199 mas->tree->ma_flags = new_flags;
202 static unsigned int mas_mt_height(struct ma_state *mas)
204 return mt_height(mas->tree);
207 static inline enum maple_type mte_node_type(const struct maple_enode *entry)
209 return ((unsigned long)entry >> MAPLE_NODE_TYPE_SHIFT) &
210 MAPLE_NODE_TYPE_MASK;
213 static inline bool ma_is_dense(const enum maple_type type)
215 return type < maple_leaf_64;
218 static inline bool ma_is_leaf(const enum maple_type type)
220 return type < maple_range_64;
223 static inline bool mte_is_leaf(const struct maple_enode *entry)
225 return ma_is_leaf(mte_node_type(entry));
229 * We also reserve values with the bottom two bits set to '10' which are
232 static inline bool mt_is_reserved(const void *entry)
234 return ((unsigned long)entry < MAPLE_RESERVED_RANGE) &&
235 xa_is_internal(entry);
238 static inline void mas_set_err(struct ma_state *mas, long err)
240 mas->node = MA_ERROR(err);
243 static inline bool mas_is_ptr(struct ma_state *mas)
245 return mas->node == MAS_ROOT;
248 static inline bool mas_is_start(struct ma_state *mas)
250 return mas->node == MAS_START;
253 bool mas_is_err(struct ma_state *mas)
255 return xa_is_err(mas->node);
258 static inline bool mas_searchable(struct ma_state *mas)
260 if (mas_is_none(mas))
269 static inline struct maple_node *mte_to_node(const struct maple_enode *entry)
271 return (struct maple_node *)((unsigned long)entry & ~MAPLE_NODE_MASK);
275 * mte_to_mat() - Convert a maple encoded node to a maple topiary node.
276 * @entry: The maple encoded node
278 * Return: a maple topiary pointer
280 static inline struct maple_topiary *mte_to_mat(const struct maple_enode *entry)
282 return (struct maple_topiary *)
283 ((unsigned long)entry & ~MAPLE_NODE_MASK);
287 * mas_mn() - Get the maple state node.
288 * @mas: The maple state
290 * Return: the maple node (not encoded - bare pointer).
292 static inline struct maple_node *mas_mn(const struct ma_state *mas)
294 return mte_to_node(mas->node);
298 * mte_set_node_dead() - Set a maple encoded node as dead.
299 * @mn: The maple encoded node.
301 static inline void mte_set_node_dead(struct maple_enode *mn)
303 mte_to_node(mn)->parent = ma_parent_ptr(mte_to_node(mn));
304 smp_wmb(); /* Needed for RCU */
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
314 static inline struct maple_enode *mt_mk_node(const struct maple_node *node,
315 enum maple_type type)
317 return (void *)((unsigned long)node |
318 (type << MAPLE_ENODE_TYPE_SHIFT) | MAPLE_ENODE_NULL);
321 static inline void *mte_mk_root(const struct maple_enode *node)
323 return (void *)((unsigned long)node | MAPLE_ROOT_NODE);
326 static inline void *mte_safe_root(const struct maple_enode *node)
328 return (void *)((unsigned long)node & ~MAPLE_ROOT_NODE);
331 static inline void *mte_set_full(const struct maple_enode *node)
333 return (void *)((unsigned long)node & ~MAPLE_ENODE_NULL);
336 static inline void *mte_clear_full(const struct maple_enode *node)
338 return (void *)((unsigned long)node | MAPLE_ENODE_NULL);
341 static inline bool mte_has_null(const struct maple_enode *node)
343 return (unsigned long)node & MAPLE_ENODE_NULL;
346 static inline bool ma_is_root(struct maple_node *node)
348 return ((unsigned long)node->parent & MA_ROOT_PARENT);
351 static inline bool mte_is_root(const struct maple_enode *node)
353 return ma_is_root(mte_to_node(node));
356 static inline bool mas_is_root_limits(const struct ma_state *mas)
358 return !mas->min && mas->max == ULONG_MAX;
361 static inline bool mt_is_alloc(struct maple_tree *mt)
363 return (mt->ma_flags & MT_FLAGS_ALLOC_RANGE);
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.
376 * 0x?00 = 16 bit nodes
377 * 0x010 = 32 bit nodes
378 * 0x110 = 64 bit nodes
380 * Slot size and alignment
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
387 #define MAPLE_PARENT_ROOT 0x01
389 #define MAPLE_PARENT_SLOT_SHIFT 0x03
390 #define MAPLE_PARENT_SLOT_MASK 0xF8
392 #define MAPLE_PARENT_16B_SLOT_SHIFT 0x02
393 #define MAPLE_PARENT_16B_SLOT_MASK 0xFC
395 #define MAPLE_PARENT_RANGE64 0x06
396 #define MAPLE_PARENT_RANGE32 0x04
397 #define MAPLE_PARENT_NOT_RANGE16 0x02
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
404 static inline unsigned long mte_parent_shift(unsigned long parent)
406 /* Note bit 1 == 0 means 16B */
407 if (likely(parent & MAPLE_PARENT_NOT_RANGE16))
408 return MAPLE_PARENT_SLOT_SHIFT;
410 return MAPLE_PARENT_16B_SLOT_SHIFT;
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.
418 static inline unsigned long mte_parent_slot_mask(unsigned long parent)
420 /* Note bit 1 == 0 means 16B */
421 if (likely(parent & MAPLE_PARENT_NOT_RANGE16))
422 return MAPLE_PARENT_SLOT_MASK;
424 return MAPLE_PARENT_16B_SLOT_MASK;
428 * mas_parent_enum() - Return the maple_type of the parent from the stored
430 * @mas: The maple state
431 * @node: The maple_enode to extract the parent's enum
432 * Return: The node->parent maple_type
435 enum maple_type mte_parent_enum(struct maple_enode *p_enode,
436 struct maple_tree *mt)
438 unsigned long p_type;
440 p_type = (unsigned long)p_enode;
441 if (p_type & MAPLE_PARENT_ROOT)
442 return 0; /* Validated in the caller. */
444 p_type &= MAPLE_NODE_MASK;
445 p_type = p_type & ~(MAPLE_PARENT_ROOT | mte_parent_slot_mask(p_type));
448 case MAPLE_PARENT_RANGE64: /* or MAPLE_PARENT_ARANGE64 */
450 return maple_arange_64;
451 return maple_range_64;
458 enum maple_type mas_parent_enum(struct ma_state *mas, struct maple_enode *enode)
460 return mte_parent_enum(ma_enode_ptr(mte_to_node(enode)->parent), mas->tree);
464 * mte_set_parent() - Set the parent node and encode the slot
465 * @enode: The encoded maple node.
466 * @parent: The encoded maple node that is the parent of @enode.
467 * @slot: The slot that @enode resides in @parent.
469 * Slot number is encoded in the enode->parent bit 3-6 or 2-6, depending on the
473 void mte_set_parent(struct maple_enode *enode, const struct maple_enode *parent,
476 unsigned long val = (unsigned long)parent;
479 enum maple_type p_type = mte_node_type(parent);
481 BUG_ON(p_type == maple_dense);
482 BUG_ON(p_type == maple_leaf_64);
486 case maple_arange_64:
487 shift = MAPLE_PARENT_SLOT_SHIFT;
488 type = MAPLE_PARENT_RANGE64;
497 val &= ~MAPLE_NODE_MASK; /* Clear all node metadata in parent */
498 val |= (slot << shift) | type;
499 mte_to_node(enode)->parent = ma_parent_ptr(val);
503 * mte_parent_slot() - get the parent slot of @enode.
504 * @enode: The encoded maple node.
506 * Return: The slot in the parent node where @enode resides.
508 static inline unsigned int mte_parent_slot(const struct maple_enode *enode)
510 unsigned long val = (unsigned long)mte_to_node(enode)->parent;
512 if (val & MA_ROOT_PARENT)
516 * Okay to use MAPLE_PARENT_16B_SLOT_MASK as the last bit will be lost
517 * by shift if the parent shift is MAPLE_PARENT_SLOT_SHIFT
519 return (val & MAPLE_PARENT_16B_SLOT_MASK) >> mte_parent_shift(val);
523 * mte_parent() - Get the parent of @node.
524 * @node: The encoded maple node.
526 * Return: The parent maple node.
528 static inline struct maple_node *mte_parent(const struct maple_enode *enode)
530 return (void *)((unsigned long)
531 (mte_to_node(enode)->parent) & ~MAPLE_NODE_MASK);
535 * ma_dead_node() - check if the @enode is dead.
536 * @enode: The encoded maple node
538 * Return: true if dead, false otherwise.
540 static inline bool ma_dead_node(const struct maple_node *node)
542 struct maple_node *parent;
544 /* Do not reorder reads from the node prior to the parent check */
546 parent = (void *)((unsigned long) node->parent & ~MAPLE_NODE_MASK);
547 return (parent == node);
551 * mte_dead_node() - check if the @enode is dead.
552 * @enode: The encoded maple node
554 * Return: true if dead, false otherwise.
556 static inline bool mte_dead_node(const struct maple_enode *enode)
558 struct maple_node *parent, *node;
560 node = mte_to_node(enode);
561 /* Do not reorder reads from the node prior to the parent check */
563 parent = mte_parent(enode);
564 return (parent == node);
568 * mas_allocated() - Get the number of nodes allocated in a maple state.
569 * @mas: The maple state
571 * The ma_state alloc member is overloaded to hold a pointer to the first
572 * allocated node or to the number of requested nodes to allocate. If bit 0 is
573 * set, then the alloc contains the number of requested nodes. If there is an
574 * allocated node, then the total allocated nodes is in that node.
576 * Return: The total number of nodes allocated
578 static inline unsigned long mas_allocated(const struct ma_state *mas)
580 if (!mas->alloc || ((unsigned long)mas->alloc & 0x1))
583 return mas->alloc->total;
587 * mas_set_alloc_req() - Set the requested number of allocations.
588 * @mas: the maple state
589 * @count: the number of allocations.
591 * The requested number of allocations is either in the first allocated node,
592 * located in @mas->alloc->request_count, or directly in @mas->alloc if there is
593 * no allocated node. Set the request either in the node or do the necessary
594 * encoding to store in @mas->alloc directly.
596 static inline void mas_set_alloc_req(struct ma_state *mas, unsigned long count)
598 if (!mas->alloc || ((unsigned long)mas->alloc & 0x1)) {
602 mas->alloc = (struct maple_alloc *)(((count) << 1U) | 1U);
606 mas->alloc->request_count = count;
610 * mas_alloc_req() - get the requested number of allocations.
611 * @mas: The maple state
613 * The alloc count is either stored directly in @mas, or in
614 * @mas->alloc->request_count if there is at least one node allocated. Decode
615 * the request count if it's stored directly in @mas->alloc.
617 * Return: The allocation request count.
619 static inline unsigned int mas_alloc_req(const struct ma_state *mas)
621 if ((unsigned long)mas->alloc & 0x1)
622 return (unsigned long)(mas->alloc) >> 1;
624 return mas->alloc->request_count;
629 * ma_pivots() - Get a pointer to the maple node pivots.
630 * @node - the maple node
631 * @type - the node type
633 * In the event of a dead node, this array may be %NULL
635 * Return: A pointer to the maple node pivots
637 static inline unsigned long *ma_pivots(struct maple_node *node,
638 enum maple_type type)
641 case maple_arange_64:
642 return node->ma64.pivot;
645 return node->mr64.pivot;
653 * ma_gaps() - Get a pointer to the maple node gaps.
654 * @node - the maple node
655 * @type - the node type
657 * Return: A pointer to the maple node gaps
659 static inline unsigned long *ma_gaps(struct maple_node *node,
660 enum maple_type type)
663 case maple_arange_64:
664 return node->ma64.gap;
674 * mte_pivot() - Get the pivot at @piv of the maple encoded node.
675 * @mn: The maple encoded node.
678 * Return: the pivot at @piv of @mn.
680 static inline unsigned long mte_pivot(const struct maple_enode *mn,
683 struct maple_node *node = mte_to_node(mn);
684 enum maple_type type = mte_node_type(mn);
686 if (piv >= mt_pivots[type]) {
691 case maple_arange_64:
692 return node->ma64.pivot[piv];
695 return node->mr64.pivot[piv];
703 * mas_safe_pivot() - get the pivot at @piv or mas->max.
704 * @mas: The maple state
705 * @pivots: The pointer to the maple node pivots
706 * @piv: The pivot to fetch
707 * @type: The maple node type
709 * Return: The pivot at @piv within the limit of the @pivots array, @mas->max
712 static inline unsigned long
713 mas_safe_pivot(const struct ma_state *mas, unsigned long *pivots,
714 unsigned char piv, enum maple_type type)
716 if (piv >= mt_pivots[type])
723 * mas_safe_min() - Return the minimum for a given offset.
724 * @mas: The maple state
725 * @pivots: The pointer to the maple node pivots
726 * @offset: The offset into the pivot array
728 * Return: The minimum range value that is contained in @offset.
730 static inline unsigned long
731 mas_safe_min(struct ma_state *mas, unsigned long *pivots, unsigned char offset)
734 return pivots[offset - 1] + 1;
740 * mas_logical_pivot() - Get the logical pivot of a given offset.
741 * @mas: The maple state
742 * @pivots: The pointer to the maple node pivots
743 * @offset: The offset into the pivot array
744 * @type: The maple node type
746 * When there is no value at a pivot (beyond the end of the data), then the
747 * pivot is actually @mas->max.
749 * Return: the logical pivot of a given @offset.
751 static inline unsigned long
752 mas_logical_pivot(struct ma_state *mas, unsigned long *pivots,
753 unsigned char offset, enum maple_type type)
755 unsigned long lpiv = mas_safe_pivot(mas, pivots, offset, type);
767 * mte_set_pivot() - Set a pivot to a value in an encoded maple node.
768 * @mn: The encoded maple node
769 * @piv: The pivot offset
770 * @val: The value of the pivot
772 static inline void mte_set_pivot(struct maple_enode *mn, unsigned char piv,
775 struct maple_node *node = mte_to_node(mn);
776 enum maple_type type = mte_node_type(mn);
778 BUG_ON(piv >= mt_pivots[type]);
783 node->mr64.pivot[piv] = val;
785 case maple_arange_64:
786 node->ma64.pivot[piv] = val;
795 * ma_slots() - Get a pointer to the maple node slots.
796 * @mn: The maple node
797 * @mt: The maple node type
799 * Return: A pointer to the maple node slots
801 static inline void __rcu **ma_slots(struct maple_node *mn, enum maple_type mt)
805 case maple_arange_64:
806 return mn->ma64.slot;
809 return mn->mr64.slot;
815 static inline bool mt_locked(const struct maple_tree *mt)
817 return mt_external_lock(mt) ? mt_lock_is_held(mt) :
818 lockdep_is_held(&mt->ma_lock);
821 static inline void *mt_slot(const struct maple_tree *mt,
822 void __rcu **slots, unsigned char offset)
824 return rcu_dereference_check(slots[offset], mt_locked(mt));
827 static inline void *mt_slot_locked(struct maple_tree *mt, void __rcu **slots,
828 unsigned char offset)
830 return rcu_dereference_protected(slots[offset], mt_locked(mt));
833 * mas_slot_locked() - Get the slot value when holding the maple tree lock.
834 * @mas: The maple state
835 * @slots: The pointer to the slots
836 * @offset: The offset into the slots array to fetch
838 * Return: The entry stored in @slots at the @offset.
840 static inline void *mas_slot_locked(struct ma_state *mas, void __rcu **slots,
841 unsigned char offset)
843 return mt_slot_locked(mas->tree, slots, offset);
847 * mas_slot() - Get the slot value when not holding the maple tree lock.
848 * @mas: The maple state
849 * @slots: The pointer to the slots
850 * @offset: The offset into the slots array to fetch
852 * Return: The entry stored in @slots at the @offset
854 static inline void *mas_slot(struct ma_state *mas, void __rcu **slots,
855 unsigned char offset)
857 return mt_slot(mas->tree, slots, offset);
861 * mas_root() - Get the maple tree root.
862 * @mas: The maple state.
864 * Return: The pointer to the root of the tree
866 static inline void *mas_root(struct ma_state *mas)
868 return rcu_dereference_check(mas->tree->ma_root, mt_locked(mas->tree));
871 static inline void *mt_root_locked(struct maple_tree *mt)
873 return rcu_dereference_protected(mt->ma_root, mt_locked(mt));
877 * mas_root_locked() - Get the maple tree root when holding the maple tree lock.
878 * @mas: The maple state.
880 * Return: The pointer to the root of the tree
882 static inline void *mas_root_locked(struct ma_state *mas)
884 return mt_root_locked(mas->tree);
887 static inline struct maple_metadata *ma_meta(struct maple_node *mn,
891 case maple_arange_64:
892 return &mn->ma64.meta;
894 return &mn->mr64.meta;
899 * ma_set_meta() - Set the metadata information of a node.
900 * @mn: The maple node
901 * @mt: The maple node type
902 * @offset: The offset of the highest sub-gap in this node.
903 * @end: The end of the data in this node.
905 static inline void ma_set_meta(struct maple_node *mn, enum maple_type mt,
906 unsigned char offset, unsigned char end)
908 struct maple_metadata *meta = ma_meta(mn, mt);
915 * mt_clear_meta() - clear the metadata information of a node, if it exists
916 * @mt: The maple tree
917 * @mn: The maple node
918 * @type: The maple node type
919 * @offset: The offset of the highest sub-gap in this node.
920 * @end: The end of the data in this node.
922 static inline void mt_clear_meta(struct maple_tree *mt, struct maple_node *mn,
923 enum maple_type type)
925 struct maple_metadata *meta;
926 unsigned long *pivots;
932 pivots = mn->mr64.pivot;
933 if (unlikely(pivots[MAPLE_RANGE64_SLOTS - 2])) {
934 slots = mn->mr64.slot;
935 next = mt_slot_locked(mt, slots,
936 MAPLE_RANGE64_SLOTS - 1);
937 if (unlikely((mte_to_node(next) &&
938 mte_node_type(next))))
939 return; /* no metadata, could be node */
942 case maple_arange_64:
943 meta = ma_meta(mn, type);
954 * ma_meta_end() - Get the data end of a node from the metadata
955 * @mn: The maple node
956 * @mt: The maple node type
958 static inline unsigned char ma_meta_end(struct maple_node *mn,
961 struct maple_metadata *meta = ma_meta(mn, mt);
967 * ma_meta_gap() - Get the largest gap location of a node from the metadata
968 * @mn: The maple node
969 * @mt: The maple node type
971 static inline unsigned char ma_meta_gap(struct maple_node *mn,
974 BUG_ON(mt != maple_arange_64);
976 return mn->ma64.meta.gap;
980 * ma_set_meta_gap() - Set the largest gap location in a nodes metadata
981 * @mn: The maple node
982 * @mn: The maple node type
983 * @offset: The location of the largest gap.
985 static inline void ma_set_meta_gap(struct maple_node *mn, enum maple_type mt,
986 unsigned char offset)
989 struct maple_metadata *meta = ma_meta(mn, mt);
995 * mat_add() - Add a @dead_enode to the ma_topiary of a list of dead nodes.
996 * @mat - the ma_topiary, a linked list of dead nodes.
997 * @dead_enode - the node to be marked as dead and added to the tail of the list
999 * Add the @dead_enode to the linked list in @mat.
1001 static inline void mat_add(struct ma_topiary *mat,
1002 struct maple_enode *dead_enode)
1004 mte_set_node_dead(dead_enode);
1005 mte_to_mat(dead_enode)->next = NULL;
1007 mat->tail = mat->head = dead_enode;
1011 mte_to_mat(mat->tail)->next = dead_enode;
1012 mat->tail = dead_enode;
1015 static void mte_destroy_walk(struct maple_enode *, struct maple_tree *);
1016 static inline void mas_free(struct ma_state *mas, struct maple_enode *used);
1019 * mas_mat_free() - Free all nodes in a dead list.
1020 * @mas - the maple state
1021 * @mat - the ma_topiary linked list of dead nodes to free.
1023 * Free walk a dead list.
1025 static void mas_mat_free(struct ma_state *mas, struct ma_topiary *mat)
1027 struct maple_enode *next;
1030 next = mte_to_mat(mat->head)->next;
1031 mas_free(mas, mat->head);
1037 * mas_mat_destroy() - Free all nodes and subtrees in a dead list.
1038 * @mas - the maple state
1039 * @mat - the ma_topiary linked list of dead nodes to free.
1041 * Destroy walk a dead list.
1043 static void mas_mat_destroy(struct ma_state *mas, struct ma_topiary *mat)
1045 struct maple_enode *next;
1048 next = mte_to_mat(mat->head)->next;
1049 mte_destroy_walk(mat->head, mat->mtree);
1054 * mas_descend() - Descend into the slot stored in the ma_state.
1055 * @mas - the maple state.
1057 * Note: Not RCU safe, only use in write side or debug code.
1059 static inline void mas_descend(struct ma_state *mas)
1061 enum maple_type type;
1062 unsigned long *pivots;
1063 struct maple_node *node;
1067 type = mte_node_type(mas->node);
1068 pivots = ma_pivots(node, type);
1069 slots = ma_slots(node, type);
1072 mas->min = pivots[mas->offset - 1] + 1;
1073 mas->max = mas_safe_pivot(mas, pivots, mas->offset, type);
1074 mas->node = mas_slot(mas, slots, mas->offset);
1078 * mte_set_gap() - Set a maple node gap.
1079 * @mn: The encoded maple node
1080 * @gap: The offset of the gap to set
1081 * @val: The gap value
1083 static inline void mte_set_gap(const struct maple_enode *mn,
1084 unsigned char gap, unsigned long val)
1086 switch (mte_node_type(mn)) {
1089 case maple_arange_64:
1090 mte_to_node(mn)->ma64.gap[gap] = val;
1096 * mas_ascend() - Walk up a level of the tree.
1097 * @mas: The maple state
1099 * Sets the @mas->max and @mas->min to the correct values when walking up. This
1100 * may cause several levels of walking up to find the correct min and max.
1101 * May find a dead node which will cause a premature return.
1102 * Return: 1 on dead node, 0 otherwise
1104 static int mas_ascend(struct ma_state *mas)
1106 struct maple_enode *p_enode; /* parent enode. */
1107 struct maple_enode *a_enode; /* ancestor enode. */
1108 struct maple_node *a_node; /* ancestor node. */
1109 struct maple_node *p_node; /* parent node. */
1110 unsigned char a_slot;
1111 enum maple_type a_type;
1112 unsigned long min, max;
1113 unsigned long *pivots;
1114 unsigned char offset;
1115 bool set_max = false, set_min = false;
1117 a_node = mas_mn(mas);
1118 if (ma_is_root(a_node)) {
1123 p_node = mte_parent(mas->node);
1124 if (unlikely(a_node == p_node))
1126 a_type = mas_parent_enum(mas, mas->node);
1127 offset = mte_parent_slot(mas->node);
1128 a_enode = mt_mk_node(p_node, a_type);
1130 /* Check to make sure all parent information is still accurate */
1131 if (p_node != mte_parent(mas->node))
1134 mas->node = a_enode;
1135 mas->offset = offset;
1137 if (mte_is_root(a_enode)) {
1138 mas->max = ULONG_MAX;
1147 a_type = mas_parent_enum(mas, p_enode);
1148 a_node = mte_parent(p_enode);
1149 a_slot = mte_parent_slot(p_enode);
1150 a_enode = mt_mk_node(a_node, a_type);
1151 pivots = ma_pivots(a_node, a_type);
1153 if (unlikely(ma_dead_node(a_node)))
1156 if (!set_min && a_slot) {
1158 min = pivots[a_slot - 1] + 1;
1161 if (!set_max && a_slot < mt_pivots[a_type]) {
1163 max = pivots[a_slot];
1166 if (unlikely(ma_dead_node(a_node)))
1169 if (unlikely(ma_is_root(a_node)))
1172 } while (!set_min || !set_max);
1180 * mas_pop_node() - Get a previously allocated maple node from the maple state.
1181 * @mas: The maple state
1183 * Return: A pointer to a maple node.
1185 static inline struct maple_node *mas_pop_node(struct ma_state *mas)
1187 struct maple_alloc *ret, *node = mas->alloc;
1188 unsigned long total = mas_allocated(mas);
1189 unsigned int req = mas_alloc_req(mas);
1191 /* nothing or a request pending. */
1192 if (WARN_ON(!total))
1196 /* single allocation in this ma_state */
1202 if (node->node_count == 1) {
1203 /* Single allocation in this node. */
1204 mas->alloc = node->slot[0];
1205 mas->alloc->total = node->total - 1;
1210 ret = node->slot[--node->node_count];
1211 node->slot[node->node_count] = NULL;
1217 mas_set_alloc_req(mas, req);
1220 memset(ret, 0, sizeof(*ret));
1221 return (struct maple_node *)ret;
1225 * mas_push_node() - Push a node back on the maple state allocation.
1226 * @mas: The maple state
1227 * @used: The used maple node
1229 * Stores the maple node back into @mas->alloc for reuse. Updates allocated and
1230 * requested node count as necessary.
1232 static inline void mas_push_node(struct ma_state *mas, struct maple_node *used)
1234 struct maple_alloc *reuse = (struct maple_alloc *)used;
1235 struct maple_alloc *head = mas->alloc;
1236 unsigned long count;
1237 unsigned int requested = mas_alloc_req(mas);
1239 count = mas_allocated(mas);
1241 reuse->request_count = 0;
1242 reuse->node_count = 0;
1243 if (count && (head->node_count < MAPLE_ALLOC_SLOTS)) {
1244 head->slot[head->node_count++] = reuse;
1250 if ((head) && !((unsigned long)head & 0x1)) {
1251 reuse->slot[0] = head;
1252 reuse->node_count = 1;
1253 reuse->total += head->total;
1259 mas_set_alloc_req(mas, requested - 1);
1263 * mas_alloc_nodes() - Allocate nodes into a maple state
1264 * @mas: The maple state
1265 * @gfp: The GFP Flags
1267 static inline void mas_alloc_nodes(struct ma_state *mas, gfp_t gfp)
1269 struct maple_alloc *node;
1270 unsigned long allocated = mas_allocated(mas);
1271 unsigned int requested = mas_alloc_req(mas);
1273 void **slots = NULL;
1274 unsigned int max_req = 0;
1279 mas_set_alloc_req(mas, 0);
1280 if (mas->mas_flags & MA_STATE_PREALLOC) {
1283 WARN_ON(!allocated);
1286 if (!allocated || mas->alloc->node_count == MAPLE_ALLOC_SLOTS) {
1287 node = (struct maple_alloc *)mt_alloc_one(gfp);
1292 node->slot[0] = mas->alloc;
1293 node->node_count = 1;
1295 node->node_count = 0;
1299 node->total = ++allocated;
1304 node->request_count = 0;
1306 max_req = MAPLE_ALLOC_SLOTS;
1307 if (node->node_count) {
1308 unsigned int offset = node->node_count;
1310 slots = (void **)&node->slot[offset];
1313 slots = (void **)&node->slot;
1316 max_req = min(requested, max_req);
1317 count = mt_alloc_bulk(gfp, max_req, slots);
1321 node->node_count += count;
1323 node = node->slot[0];
1324 node->node_count = 0;
1325 node->request_count = 0;
1328 mas->alloc->total = allocated;
1332 /* Clean up potential freed allocations on bulk failure */
1333 memset(slots, 0, max_req * sizeof(unsigned long));
1335 mas_set_alloc_req(mas, requested);
1336 if (mas->alloc && !(((unsigned long)mas->alloc & 0x1)))
1337 mas->alloc->total = allocated;
1338 mas_set_err(mas, -ENOMEM);
1342 * mas_free() - Free an encoded maple node
1343 * @mas: The maple state
1344 * @used: The encoded maple node to free.
1346 * Uses rcu free if necessary, pushes @used back on the maple state allocations
1349 static inline void mas_free(struct ma_state *mas, struct maple_enode *used)
1351 struct maple_node *tmp = mte_to_node(used);
1353 if (mt_in_rcu(mas->tree))
1356 mas_push_node(mas, tmp);
1360 * mas_node_count() - Check if enough nodes are allocated and request more if
1361 * there is not enough nodes.
1362 * @mas: The maple state
1363 * @count: The number of nodes needed
1364 * @gfp: the gfp flags
1366 static void mas_node_count_gfp(struct ma_state *mas, int count, gfp_t gfp)
1368 unsigned long allocated = mas_allocated(mas);
1370 if (allocated < count) {
1371 mas_set_alloc_req(mas, count - allocated);
1372 mas_alloc_nodes(mas, gfp);
1377 * mas_node_count() - Check if enough nodes are allocated and request more if
1378 * there is not enough nodes.
1379 * @mas: The maple state
1380 * @count: The number of nodes needed
1382 * Note: Uses GFP_NOWAIT | __GFP_NOWARN for gfp flags.
1384 static void mas_node_count(struct ma_state *mas, int count)
1386 return mas_node_count_gfp(mas, count, GFP_NOWAIT | __GFP_NOWARN);
1390 * mas_start() - Sets up maple state for operations.
1391 * @mas: The maple state.
1393 * If mas->node == MAS_START, then set the min, max and depth to
1397 * - If mas->node is an error or not MAS_START, return NULL.
1398 * - If it's an empty tree: NULL & mas->node == MAS_NONE
1399 * - If it's a single entry: The entry & mas->node == MAS_ROOT
1400 * - If it's a tree: NULL & mas->node == safe root node.
1402 static inline struct maple_enode *mas_start(struct ma_state *mas)
1404 if (likely(mas_is_start(mas))) {
1405 struct maple_enode *root;
1408 mas->max = ULONG_MAX;
1412 root = mas_root(mas);
1413 /* Tree with nodes */
1414 if (likely(xa_is_node(root))) {
1416 mas->node = mte_safe_root(root);
1418 if (mte_dead_node(mas->node))
1425 if (unlikely(!root)) {
1426 mas->node = MAS_NONE;
1427 mas->offset = MAPLE_NODE_SLOTS;
1431 /* Single entry tree */
1432 mas->node = MAS_ROOT;
1433 mas->offset = MAPLE_NODE_SLOTS;
1435 /* Single entry tree. */
1446 * ma_data_end() - Find the end of the data in a node.
1447 * @node: The maple node
1448 * @type: The maple node type
1449 * @pivots: The array of pivots in the node
1450 * @max: The maximum value in the node
1452 * Uses metadata to find the end of the data when possible.
1453 * Return: The zero indexed last slot with data (may be null).
1455 static inline unsigned char ma_data_end(struct maple_node *node,
1456 enum maple_type type,
1457 unsigned long *pivots,
1460 unsigned char offset;
1465 if (type == maple_arange_64)
1466 return ma_meta_end(node, type);
1468 offset = mt_pivots[type] - 1;
1469 if (likely(!pivots[offset]))
1470 return ma_meta_end(node, type);
1472 if (likely(pivots[offset] == max))
1475 return mt_pivots[type];
1479 * mas_data_end() - Find the end of the data (slot).
1480 * @mas: the maple state
1482 * This method is optimized to check the metadata of a node if the node type
1483 * supports data end metadata.
1485 * Return: The zero indexed last slot with data (may be null).
1487 static inline unsigned char mas_data_end(struct ma_state *mas)
1489 enum maple_type type;
1490 struct maple_node *node;
1491 unsigned char offset;
1492 unsigned long *pivots;
1494 type = mte_node_type(mas->node);
1496 if (type == maple_arange_64)
1497 return ma_meta_end(node, type);
1499 pivots = ma_pivots(node, type);
1500 if (unlikely(ma_dead_node(node)))
1503 offset = mt_pivots[type] - 1;
1504 if (likely(!pivots[offset]))
1505 return ma_meta_end(node, type);
1507 if (likely(pivots[offset] == mas->max))
1510 return mt_pivots[type];
1514 * mas_leaf_max_gap() - Returns the largest gap in a leaf node
1515 * @mas - the maple state
1517 * Return: The maximum gap in the leaf.
1519 static unsigned long mas_leaf_max_gap(struct ma_state *mas)
1522 unsigned long pstart, gap, max_gap;
1523 struct maple_node *mn;
1524 unsigned long *pivots;
1527 unsigned char max_piv;
1529 mt = mte_node_type(mas->node);
1531 slots = ma_slots(mn, mt);
1533 if (unlikely(ma_is_dense(mt))) {
1535 for (i = 0; i < mt_slots[mt]; i++) {
1550 * Check the first implied pivot optimizes the loop below and slot 1 may
1551 * be skipped if there is a gap in slot 0.
1553 pivots = ma_pivots(mn, mt);
1554 if (likely(!slots[0])) {
1555 max_gap = pivots[0] - mas->min + 1;
1561 /* reduce max_piv as the special case is checked before the loop */
1562 max_piv = ma_data_end(mn, mt, pivots, mas->max) - 1;
1564 * Check end implied pivot which can only be a gap on the right most
1567 if (unlikely(mas->max == ULONG_MAX) && !slots[max_piv + 1]) {
1568 gap = ULONG_MAX - pivots[max_piv];
1573 for (; i <= max_piv; i++) {
1574 /* data == no gap. */
1575 if (likely(slots[i]))
1578 pstart = pivots[i - 1];
1579 gap = pivots[i] - pstart;
1583 /* There cannot be two gaps in a row. */
1590 * ma_max_gap() - Get the maximum gap in a maple node (non-leaf)
1591 * @node: The maple node
1592 * @gaps: The pointer to the gaps
1593 * @mt: The maple node type
1594 * @*off: Pointer to store the offset location of the gap.
1596 * Uses the metadata data end to scan backwards across set gaps.
1598 * Return: The maximum gap value
1600 static inline unsigned long
1601 ma_max_gap(struct maple_node *node, unsigned long *gaps, enum maple_type mt,
1604 unsigned char offset, i;
1605 unsigned long max_gap = 0;
1607 i = offset = ma_meta_end(node, mt);
1609 if (gaps[i] > max_gap) {
1620 * mas_max_gap() - find the largest gap in a non-leaf node and set the slot.
1621 * @mas: The maple state.
1623 * If the metadata gap is set to MAPLE_ARANGE64_META_MAX, there is no gap.
1625 * Return: The gap value.
1627 static inline unsigned long mas_max_gap(struct ma_state *mas)
1629 unsigned long *gaps;
1630 unsigned char offset;
1632 struct maple_node *node;
1634 mt = mte_node_type(mas->node);
1636 return mas_leaf_max_gap(mas);
1639 offset = ma_meta_gap(node, mt);
1640 if (offset == MAPLE_ARANGE64_META_MAX)
1643 gaps = ma_gaps(node, mt);
1644 return gaps[offset];
1648 * mas_parent_gap() - Set the parent gap and any gaps above, as needed
1649 * @mas: The maple state
1650 * @offset: The gap offset in the parent to set
1651 * @new: The new gap value.
1653 * Set the parent gap then continue to set the gap upwards, using the metadata
1654 * of the parent to see if it is necessary to check the node above.
1656 static inline void mas_parent_gap(struct ma_state *mas, unsigned char offset,
1659 unsigned long meta_gap = 0;
1660 struct maple_node *pnode;
1661 struct maple_enode *penode;
1662 unsigned long *pgaps;
1663 unsigned char meta_offset;
1664 enum maple_type pmt;
1666 pnode = mte_parent(mas->node);
1667 pmt = mas_parent_enum(mas, mas->node);
1668 penode = mt_mk_node(pnode, pmt);
1669 pgaps = ma_gaps(pnode, pmt);
1672 meta_offset = ma_meta_gap(pnode, pmt);
1673 if (meta_offset == MAPLE_ARANGE64_META_MAX)
1676 meta_gap = pgaps[meta_offset];
1678 pgaps[offset] = new;
1680 if (meta_gap == new)
1683 if (offset != meta_offset) {
1687 ma_set_meta_gap(pnode, pmt, offset);
1688 } else if (new < meta_gap) {
1690 new = ma_max_gap(pnode, pgaps, pmt, &meta_offset);
1691 ma_set_meta_gap(pnode, pmt, meta_offset);
1694 if (ma_is_root(pnode))
1697 /* Go to the parent node. */
1698 pnode = mte_parent(penode);
1699 pmt = mas_parent_enum(mas, penode);
1700 pgaps = ma_gaps(pnode, pmt);
1701 offset = mte_parent_slot(penode);
1702 penode = mt_mk_node(pnode, pmt);
1707 * mas_update_gap() - Update a nodes gaps and propagate up if necessary.
1708 * @mas - the maple state.
1710 static inline void mas_update_gap(struct ma_state *mas)
1712 unsigned char pslot;
1713 unsigned long p_gap;
1714 unsigned long max_gap;
1716 if (!mt_is_alloc(mas->tree))
1719 if (mte_is_root(mas->node))
1722 max_gap = mas_max_gap(mas);
1724 pslot = mte_parent_slot(mas->node);
1725 p_gap = ma_gaps(mte_parent(mas->node),
1726 mas_parent_enum(mas, mas->node))[pslot];
1728 if (p_gap != max_gap)
1729 mas_parent_gap(mas, pslot, max_gap);
1733 * mas_adopt_children() - Set the parent pointer of all nodes in @parent to
1734 * @parent with the slot encoded.
1735 * @mas - the maple state (for the tree)
1736 * @parent - the maple encoded node containing the children.
1738 static inline void mas_adopt_children(struct ma_state *mas,
1739 struct maple_enode *parent)
1741 enum maple_type type = mte_node_type(parent);
1742 struct maple_node *node = mas_mn(mas);
1743 void __rcu **slots = ma_slots(node, type);
1744 unsigned long *pivots = ma_pivots(node, type);
1745 struct maple_enode *child;
1746 unsigned char offset;
1748 offset = ma_data_end(node, type, pivots, mas->max);
1750 child = mas_slot_locked(mas, slots, offset);
1751 mte_set_parent(child, parent, offset);
1756 * mas_replace() - Replace a maple node in the tree with mas->node. Uses the
1757 * parent encoding to locate the maple node in the tree.
1758 * @mas - the ma_state to use for operations.
1759 * @advanced - boolean to adopt the child nodes and free the old node (false) or
1760 * leave the node (true) and handle the adoption and free elsewhere.
1762 static inline void mas_replace(struct ma_state *mas, bool advanced)
1763 __must_hold(mas->tree->lock)
1765 struct maple_node *mn = mas_mn(mas);
1766 struct maple_enode *old_enode;
1767 unsigned char offset = 0;
1768 void __rcu **slots = NULL;
1770 if (ma_is_root(mn)) {
1771 old_enode = mas_root_locked(mas);
1773 offset = mte_parent_slot(mas->node);
1774 slots = ma_slots(mte_parent(mas->node),
1775 mas_parent_enum(mas, mas->node));
1776 old_enode = mas_slot_locked(mas, slots, offset);
1779 if (!advanced && !mte_is_leaf(mas->node))
1780 mas_adopt_children(mas, mas->node);
1782 if (mte_is_root(mas->node)) {
1783 mn->parent = ma_parent_ptr(
1784 ((unsigned long)mas->tree | MA_ROOT_PARENT));
1785 rcu_assign_pointer(mas->tree->ma_root, mte_mk_root(mas->node));
1786 mas_set_height(mas);
1788 rcu_assign_pointer(slots[offset], mas->node);
1792 mte_set_node_dead(old_enode);
1793 mas_free(mas, old_enode);
1798 * mas_new_child() - Find the new child of a node.
1799 * @mas: the maple state
1800 * @child: the maple state to store the child.
1802 static inline bool mas_new_child(struct ma_state *mas, struct ma_state *child)
1803 __must_hold(mas->tree->lock)
1806 unsigned char offset;
1808 unsigned long *pivots;
1809 struct maple_enode *entry;
1810 struct maple_node *node;
1813 mt = mte_node_type(mas->node);
1815 slots = ma_slots(node, mt);
1816 pivots = ma_pivots(node, mt);
1817 end = ma_data_end(node, mt, pivots, mas->max);
1818 for (offset = mas->offset; offset <= end; offset++) {
1819 entry = mas_slot_locked(mas, slots, offset);
1820 if (mte_parent(entry) == node) {
1822 mas->offset = offset + 1;
1823 child->offset = offset;
1833 * mab_shift_right() - Shift the data in mab right. Note, does not clean out the
1834 * old data or set b_node->b_end.
1835 * @b_node: the maple_big_node
1836 * @shift: the shift count
1838 static inline void mab_shift_right(struct maple_big_node *b_node,
1839 unsigned char shift)
1841 unsigned long size = b_node->b_end * sizeof(unsigned long);
1843 memmove(b_node->pivot + shift, b_node->pivot, size);
1844 memmove(b_node->slot + shift, b_node->slot, size);
1845 if (b_node->type == maple_arange_64)
1846 memmove(b_node->gap + shift, b_node->gap, size);
1850 * mab_middle_node() - Check if a middle node is needed (unlikely)
1851 * @b_node: the maple_big_node that contains the data.
1852 * @size: the amount of data in the b_node
1853 * @split: the potential split location
1854 * @slot_count: the size that can be stored in a single node being considered.
1856 * Return: true if a middle node is required.
1858 static inline bool mab_middle_node(struct maple_big_node *b_node, int split,
1859 unsigned char slot_count)
1861 unsigned char size = b_node->b_end;
1863 if (size >= 2 * slot_count)
1866 if (!b_node->slot[split] && (size >= 2 * slot_count - 1))
1873 * mab_no_null_split() - ensure the split doesn't fall on a NULL
1874 * @b_node: the maple_big_node with the data
1875 * @split: the suggested split location
1876 * @slot_count: the number of slots in the node being considered.
1878 * Return: the split location.
1880 static inline int mab_no_null_split(struct maple_big_node *b_node,
1881 unsigned char split, unsigned char slot_count)
1883 if (!b_node->slot[split]) {
1885 * If the split is less than the max slot && the right side will
1886 * still be sufficient, then increment the split on NULL.
1888 if ((split < slot_count - 1) &&
1889 (b_node->b_end - split) > (mt_min_slots[b_node->type]))
1898 * mab_calc_split() - Calculate the split location and if there needs to be two
1900 * @bn: The maple_big_node with the data
1901 * @mid_split: The second split, if required. 0 otherwise.
1903 * Return: The first split location. The middle split is set in @mid_split.
1905 static inline int mab_calc_split(struct ma_state *mas,
1906 struct maple_big_node *bn, unsigned char *mid_split, unsigned long min)
1908 unsigned char b_end = bn->b_end;
1909 int split = b_end / 2; /* Assume equal split. */
1910 unsigned char slot_min, slot_count = mt_slots[bn->type];
1913 * To support gap tracking, all NULL entries are kept together and a node cannot
1914 * end on a NULL entry, with the exception of the left-most leaf. The
1915 * limitation means that the split of a node must be checked for this condition
1916 * and be able to put more data in one direction or the other.
1918 if (unlikely((mas->mas_flags & MA_STATE_BULK))) {
1920 split = b_end - mt_min_slots[bn->type];
1922 if (!ma_is_leaf(bn->type))
1925 mas->mas_flags |= MA_STATE_REBALANCE;
1926 if (!bn->slot[split])
1932 * Although extremely rare, it is possible to enter what is known as the 3-way
1933 * split scenario. The 3-way split comes about by means of a store of a range
1934 * that overwrites the end and beginning of two full nodes. The result is a set
1935 * of entries that cannot be stored in 2 nodes. Sometimes, these two nodes can
1936 * also be located in different parent nodes which are also full. This can
1937 * carry upwards all the way to the root in the worst case.
1939 if (unlikely(mab_middle_node(bn, split, slot_count))) {
1941 *mid_split = split * 2;
1943 slot_min = mt_min_slots[bn->type];
1947 * Avoid having a range less than the slot count unless it
1948 * causes one node to be deficient.
1949 * NOTE: mt_min_slots is 1 based, b_end and split are zero.
1951 while (((bn->pivot[split] - min) < slot_count - 1) &&
1952 (split < slot_count - 1) && (b_end - split > slot_min))
1956 /* Avoid ending a node on a NULL entry */
1957 split = mab_no_null_split(bn, split, slot_count);
1959 if (unlikely(*mid_split))
1960 *mid_split = mab_no_null_split(bn, *mid_split, slot_count);
1966 * mas_mab_cp() - Copy data from a maple state inclusively to a maple_big_node
1967 * and set @b_node->b_end to the next free slot.
1968 * @mas: The maple state
1969 * @mas_start: The starting slot to copy
1970 * @mas_end: The end slot to copy (inclusively)
1971 * @b_node: The maple_big_node to place the data
1972 * @mab_start: The starting location in maple_big_node to store the data.
1974 static inline void mas_mab_cp(struct ma_state *mas, unsigned char mas_start,
1975 unsigned char mas_end, struct maple_big_node *b_node,
1976 unsigned char mab_start)
1979 struct maple_node *node;
1981 unsigned long *pivots, *gaps;
1982 int i = mas_start, j = mab_start;
1983 unsigned char piv_end;
1986 mt = mte_node_type(mas->node);
1987 pivots = ma_pivots(node, mt);
1989 b_node->pivot[j] = pivots[i++];
1990 if (unlikely(i > mas_end))
1995 piv_end = min(mas_end, mt_pivots[mt]);
1996 for (; i < piv_end; i++, j++) {
1997 b_node->pivot[j] = pivots[i];
1998 if (unlikely(!b_node->pivot[j]))
2001 if (unlikely(mas->max == b_node->pivot[j]))
2005 if (likely(i <= mas_end))
2006 b_node->pivot[j] = mas_safe_pivot(mas, pivots, i, mt);
2009 b_node->b_end = ++j;
2011 slots = ma_slots(node, mt);
2012 memcpy(b_node->slot + mab_start, slots + mas_start, sizeof(void *) * j);
2013 if (!ma_is_leaf(mt) && mt_is_alloc(mas->tree)) {
2014 gaps = ma_gaps(node, mt);
2015 memcpy(b_node->gap + mab_start, gaps + mas_start,
2016 sizeof(unsigned long) * j);
2021 * mas_leaf_set_meta() - Set the metadata of a leaf if possible.
2022 * @mas: The maple state
2023 * @node: The maple node
2024 * @pivots: pointer to the maple node pivots
2025 * @mt: The maple type
2026 * @end: The assumed end
2028 * Note, end may be incremented within this function but not modified at the
2029 * source. This is fine since the metadata is the last thing to be stored in a
2030 * node during a write.
2032 static inline void mas_leaf_set_meta(struct ma_state *mas,
2033 struct maple_node *node, unsigned long *pivots,
2034 enum maple_type mt, unsigned char end)
2036 /* There is no room for metadata already */
2037 if (mt_pivots[mt] <= end)
2040 if (pivots[end] && pivots[end] < mas->max)
2043 if (end < mt_slots[mt] - 1)
2044 ma_set_meta(node, mt, 0, end);
2048 * mab_mas_cp() - Copy data from maple_big_node to a maple encoded node.
2049 * @b_node: the maple_big_node that has the data
2050 * @mab_start: the start location in @b_node.
2051 * @mab_end: The end location in @b_node (inclusively)
2052 * @mas: The maple state with the maple encoded node.
2054 static inline void mab_mas_cp(struct maple_big_node *b_node,
2055 unsigned char mab_start, unsigned char mab_end,
2056 struct ma_state *mas, bool new_max)
2059 enum maple_type mt = mte_node_type(mas->node);
2060 struct maple_node *node = mte_to_node(mas->node);
2061 void __rcu **slots = ma_slots(node, mt);
2062 unsigned long *pivots = ma_pivots(node, mt);
2063 unsigned long *gaps = NULL;
2066 if (mab_end - mab_start > mt_pivots[mt])
2069 if (!pivots[mt_pivots[mt] - 1])
2070 slots[mt_pivots[mt]] = NULL;
2074 pivots[j++] = b_node->pivot[i++];
2075 } while (i <= mab_end && likely(b_node->pivot[i]));
2077 memcpy(slots, b_node->slot + mab_start,
2078 sizeof(void *) * (i - mab_start));
2081 mas->max = b_node->pivot[i - 1];
2084 if (likely(!ma_is_leaf(mt) && mt_is_alloc(mas->tree))) {
2085 unsigned long max_gap = 0;
2086 unsigned char offset = 15;
2088 gaps = ma_gaps(node, mt);
2090 gaps[--j] = b_node->gap[--i];
2091 if (gaps[j] > max_gap) {
2097 ma_set_meta(node, mt, offset, end);
2099 mas_leaf_set_meta(mas, node, pivots, mt, end);
2104 * mas_descend_adopt() - Descend through a sub-tree and adopt children.
2105 * @mas: the maple state with the maple encoded node of the sub-tree.
2107 * Descend through a sub-tree and adopt children who do not have the correct
2108 * parents set. Follow the parents which have the correct parents as they are
2109 * the new entries which need to be followed to find other incorrectly set
2112 static inline void mas_descend_adopt(struct ma_state *mas)
2114 struct ma_state list[3], next[3];
2118 * At each level there may be up to 3 correct parent pointers which indicates
2119 * the new nodes which need to be walked to find any new nodes at a lower level.
2122 for (i = 0; i < 3; i++) {
2129 while (!mte_is_leaf(list[0].node)) {
2131 for (i = 0; i < 3; i++) {
2132 if (mas_is_none(&list[i]))
2135 if (i && list[i-1].node == list[i].node)
2138 while ((n < 3) && (mas_new_child(&list[i], &next[n])))
2141 mas_adopt_children(&list[i], list[i].node);
2145 next[n++].node = MAS_NONE;
2147 /* descend by setting the list to the children */
2148 for (i = 0; i < 3; i++)
2154 * mas_bulk_rebalance() - Rebalance the end of a tree after a bulk insert.
2155 * @mas: The maple state
2156 * @end: The maple node end
2157 * @mt: The maple node type
2159 static inline void mas_bulk_rebalance(struct ma_state *mas, unsigned char end,
2162 if (!(mas->mas_flags & MA_STATE_BULK))
2165 if (mte_is_root(mas->node))
2168 if (end > mt_min_slots[mt]) {
2169 mas->mas_flags &= ~MA_STATE_REBALANCE;
2175 * mas_store_b_node() - Store an @entry into the b_node while also copying the
2176 * data from a maple encoded node.
2177 * @wr_mas: the maple write state
2178 * @b_node: the maple_big_node to fill with data
2179 * @offset_end: the offset to end copying
2181 * Return: The actual end of the data stored in @b_node
2183 static noinline_for_kasan void mas_store_b_node(struct ma_wr_state *wr_mas,
2184 struct maple_big_node *b_node, unsigned char offset_end)
2187 unsigned char b_end;
2188 /* Possible underflow of piv will wrap back to 0 before use. */
2190 struct ma_state *mas = wr_mas->mas;
2192 b_node->type = wr_mas->type;
2196 /* Copy start data up to insert. */
2197 mas_mab_cp(mas, 0, slot - 1, b_node, 0);
2198 b_end = b_node->b_end;
2199 piv = b_node->pivot[b_end - 1];
2203 if (piv + 1 < mas->index) {
2204 /* Handle range starting after old range */
2205 b_node->slot[b_end] = wr_mas->content;
2206 if (!wr_mas->content)
2207 b_node->gap[b_end] = mas->index - 1 - piv;
2208 b_node->pivot[b_end++] = mas->index - 1;
2211 /* Store the new entry. */
2212 mas->offset = b_end;
2213 b_node->slot[b_end] = wr_mas->entry;
2214 b_node->pivot[b_end] = mas->last;
2217 if (mas->last >= mas->max)
2220 /* Handle new range ending before old range ends */
2221 piv = mas_logical_pivot(mas, wr_mas->pivots, offset_end, wr_mas->type);
2222 if (piv > mas->last) {
2223 if (piv == ULONG_MAX)
2224 mas_bulk_rebalance(mas, b_node->b_end, wr_mas->type);
2226 if (offset_end != slot)
2227 wr_mas->content = mas_slot_locked(mas, wr_mas->slots,
2230 b_node->slot[++b_end] = wr_mas->content;
2231 if (!wr_mas->content)
2232 b_node->gap[b_end] = piv - mas->last + 1;
2233 b_node->pivot[b_end] = piv;
2236 slot = offset_end + 1;
2237 if (slot > wr_mas->node_end)
2240 /* Copy end data to the end of the node. */
2241 mas_mab_cp(mas, slot, wr_mas->node_end + 1, b_node, ++b_end);
2246 b_node->b_end = b_end;
2250 * mas_prev_sibling() - Find the previous node with the same parent.
2251 * @mas: the maple state
2253 * Return: True if there is a previous sibling, false otherwise.
2255 static inline bool mas_prev_sibling(struct ma_state *mas)
2257 unsigned int p_slot = mte_parent_slot(mas->node);
2259 if (mte_is_root(mas->node))
2266 mas->offset = p_slot - 1;
2272 * mas_next_sibling() - Find the next node with the same parent.
2273 * @mas: the maple state
2275 * Return: true if there is a next sibling, false otherwise.
2277 static inline bool mas_next_sibling(struct ma_state *mas)
2279 MA_STATE(parent, mas->tree, mas->index, mas->last);
2281 if (mte_is_root(mas->node))
2285 mas_ascend(&parent);
2286 parent.offset = mte_parent_slot(mas->node) + 1;
2287 if (parent.offset > mas_data_end(&parent))
2296 * mte_node_or_node() - Return the encoded node or MAS_NONE.
2297 * @enode: The encoded maple node.
2299 * Shorthand to avoid setting %NULLs in the tree or maple_subtree_state.
2301 * Return: @enode or MAS_NONE
2303 static inline struct maple_enode *mte_node_or_none(struct maple_enode *enode)
2308 return ma_enode_ptr(MAS_NONE);
2312 * mas_wr_node_walk() - Find the correct offset for the index in the @mas.
2313 * @wr_mas: The maple write state
2315 * Uses mas_slot_locked() and does not need to worry about dead nodes.
2317 static inline void mas_wr_node_walk(struct ma_wr_state *wr_mas)
2319 struct ma_state *mas = wr_mas->mas;
2320 unsigned char count;
2321 unsigned char offset;
2322 unsigned long index, min, max;
2324 if (unlikely(ma_is_dense(wr_mas->type))) {
2325 wr_mas->r_max = wr_mas->r_min = mas->index;
2326 mas->offset = mas->index = mas->min;
2330 wr_mas->node = mas_mn(wr_mas->mas);
2331 wr_mas->pivots = ma_pivots(wr_mas->node, wr_mas->type);
2332 count = wr_mas->node_end = ma_data_end(wr_mas->node, wr_mas->type,
2333 wr_mas->pivots, mas->max);
2334 offset = mas->offset;
2335 min = mas_safe_min(mas, wr_mas->pivots, offset);
2336 if (unlikely(offset == count))
2339 max = wr_mas->pivots[offset];
2341 if (unlikely(index <= max))
2344 if (unlikely(!max && offset))
2348 while (++offset < count) {
2349 max = wr_mas->pivots[offset];
2352 else if (unlikely(!max))
2361 wr_mas->r_max = max;
2362 wr_mas->r_min = min;
2363 wr_mas->offset_end = mas->offset = offset;
2367 * mas_topiary_range() - Add a range of slots to the topiary.
2368 * @mas: The maple state
2369 * @destroy: The topiary to add the slots (usually destroy)
2370 * @start: The starting slot inclusively
2371 * @end: The end slot inclusively
2373 static inline void mas_topiary_range(struct ma_state *mas,
2374 struct ma_topiary *destroy, unsigned char start, unsigned char end)
2377 unsigned char offset;
2379 MT_BUG_ON(mas->tree, mte_is_leaf(mas->node));
2380 slots = ma_slots(mas_mn(mas), mte_node_type(mas->node));
2381 for (offset = start; offset <= end; offset++) {
2382 struct maple_enode *enode = mas_slot_locked(mas, slots, offset);
2384 if (mte_dead_node(enode))
2387 mat_add(destroy, enode);
2392 * mast_topiary() - Add the portions of the tree to the removal list; either to
2393 * be freed or discarded (destroy walk).
2394 * @mast: The maple_subtree_state.
2396 static inline void mast_topiary(struct maple_subtree_state *mast)
2398 MA_WR_STATE(wr_mas, mast->orig_l, NULL);
2399 unsigned char r_start, r_end;
2400 unsigned char l_start, l_end;
2401 void __rcu **l_slots, **r_slots;
2403 wr_mas.type = mte_node_type(mast->orig_l->node);
2404 mast->orig_l->index = mast->orig_l->last;
2405 mas_wr_node_walk(&wr_mas);
2406 l_start = mast->orig_l->offset + 1;
2407 l_end = mas_data_end(mast->orig_l);
2409 r_end = mast->orig_r->offset;
2414 l_slots = ma_slots(mas_mn(mast->orig_l),
2415 mte_node_type(mast->orig_l->node));
2417 r_slots = ma_slots(mas_mn(mast->orig_r),
2418 mte_node_type(mast->orig_r->node));
2420 if ((l_start < l_end) &&
2421 mte_dead_node(mas_slot_locked(mast->orig_l, l_slots, l_start))) {
2425 if (mte_dead_node(mas_slot_locked(mast->orig_r, r_slots, r_end))) {
2430 if ((l_start > r_end) && (mast->orig_l->node == mast->orig_r->node))
2433 /* At the node where left and right sides meet, add the parts between */
2434 if (mast->orig_l->node == mast->orig_r->node) {
2435 return mas_topiary_range(mast->orig_l, mast->destroy,
2439 /* mast->orig_r is different and consumed. */
2440 if (mte_is_leaf(mast->orig_r->node))
2443 if (mte_dead_node(mas_slot_locked(mast->orig_l, l_slots, l_end)))
2447 if (l_start <= l_end)
2448 mas_topiary_range(mast->orig_l, mast->destroy, l_start, l_end);
2450 if (mte_dead_node(mas_slot_locked(mast->orig_r, r_slots, r_start)))
2453 if (r_start <= r_end)
2454 mas_topiary_range(mast->orig_r, mast->destroy, 0, r_end);
2458 * mast_rebalance_next() - Rebalance against the next node
2459 * @mast: The maple subtree state
2460 * @old_r: The encoded maple node to the right (next node).
2462 static inline void mast_rebalance_next(struct maple_subtree_state *mast)
2464 unsigned char b_end = mast->bn->b_end;
2466 mas_mab_cp(mast->orig_r, 0, mt_slot_count(mast->orig_r->node),
2468 mast->orig_r->last = mast->orig_r->max;
2472 * mast_rebalance_prev() - Rebalance against the previous node
2473 * @mast: The maple subtree state
2474 * @old_l: The encoded maple node to the left (previous node)
2476 static inline void mast_rebalance_prev(struct maple_subtree_state *mast)
2478 unsigned char end = mas_data_end(mast->orig_l) + 1;
2479 unsigned char b_end = mast->bn->b_end;
2481 mab_shift_right(mast->bn, end);
2482 mas_mab_cp(mast->orig_l, 0, end - 1, mast->bn, 0);
2483 mast->l->min = mast->orig_l->min;
2484 mast->orig_l->index = mast->orig_l->min;
2485 mast->bn->b_end = end + b_end;
2486 mast->l->offset += end;
2490 * mast_spanning_rebalance() - Rebalance nodes with nearest neighbour favouring
2491 * the node to the right. Checking the nodes to the right then the left at each
2492 * level upwards until root is reached. Free and destroy as needed.
2493 * Data is copied into the @mast->bn.
2494 * @mast: The maple_subtree_state.
2497 bool mast_spanning_rebalance(struct maple_subtree_state *mast)
2499 struct ma_state r_tmp = *mast->orig_r;
2500 struct ma_state l_tmp = *mast->orig_l;
2501 struct maple_enode *ancestor = NULL;
2502 unsigned char start, end;
2503 unsigned char depth = 0;
2505 r_tmp = *mast->orig_r;
2506 l_tmp = *mast->orig_l;
2508 mas_ascend(mast->orig_r);
2509 mas_ascend(mast->orig_l);
2512 (mast->orig_r->node == mast->orig_l->node)) {
2513 ancestor = mast->orig_r->node;
2514 end = mast->orig_r->offset - 1;
2515 start = mast->orig_l->offset + 1;
2518 if (mast->orig_r->offset < mas_data_end(mast->orig_r)) {
2520 ancestor = mast->orig_r->node;
2524 mast->orig_r->offset++;
2526 mas_descend(mast->orig_r);
2527 mast->orig_r->offset = 0;
2531 mast_rebalance_next(mast);
2533 unsigned char l_off = 0;
2534 struct maple_enode *child = r_tmp.node;
2537 if (ancestor == r_tmp.node)
2543 if (l_off < r_tmp.offset)
2544 mas_topiary_range(&r_tmp, mast->destroy,
2545 l_off, r_tmp.offset);
2547 if (l_tmp.node != child)
2548 mat_add(mast->free, child);
2550 } while (r_tmp.node != ancestor);
2552 *mast->orig_l = l_tmp;
2555 } else if (mast->orig_l->offset != 0) {
2557 ancestor = mast->orig_l->node;
2558 end = mas_data_end(mast->orig_l);
2561 mast->orig_l->offset--;
2563 mas_descend(mast->orig_l);
2564 mast->orig_l->offset =
2565 mas_data_end(mast->orig_l);
2569 mast_rebalance_prev(mast);
2571 unsigned char r_off;
2572 struct maple_enode *child = l_tmp.node;
2575 if (ancestor == l_tmp.node)
2578 r_off = mas_data_end(&l_tmp);
2580 if (l_tmp.offset < r_off)
2583 if (l_tmp.offset < r_off)
2584 mas_topiary_range(&l_tmp, mast->destroy,
2585 l_tmp.offset, r_off);
2587 if (r_tmp.node != child)
2588 mat_add(mast->free, child);
2590 } while (l_tmp.node != ancestor);
2592 *mast->orig_r = r_tmp;
2595 } while (!mte_is_root(mast->orig_r->node));
2597 *mast->orig_r = r_tmp;
2598 *mast->orig_l = l_tmp;
2603 * mast_ascend_free() - Add current original maple state nodes to the free list
2605 * @mast: the maple subtree state.
2607 * Ascend the original left and right sides and add the previous nodes to the
2608 * free list. Set the slots to point to the correct location in the new nodes.
2611 mast_ascend_free(struct maple_subtree_state *mast)
2613 MA_WR_STATE(wr_mas, mast->orig_r, NULL);
2614 struct maple_enode *left = mast->orig_l->node;
2615 struct maple_enode *right = mast->orig_r->node;
2617 mas_ascend(mast->orig_l);
2618 mas_ascend(mast->orig_r);
2619 mat_add(mast->free, left);
2622 mat_add(mast->free, right);
2624 mast->orig_r->offset = 0;
2625 mast->orig_r->index = mast->r->max;
2626 /* last should be larger than or equal to index */
2627 if (mast->orig_r->last < mast->orig_r->index)
2628 mast->orig_r->last = mast->orig_r->index;
2630 * The node may not contain the value so set slot to ensure all
2631 * of the nodes contents are freed or destroyed.
2633 wr_mas.type = mte_node_type(mast->orig_r->node);
2634 mas_wr_node_walk(&wr_mas);
2635 /* Set up the left side of things */
2636 mast->orig_l->offset = 0;
2637 mast->orig_l->index = mast->l->min;
2638 wr_mas.mas = mast->orig_l;
2639 wr_mas.type = mte_node_type(mast->orig_l->node);
2640 mas_wr_node_walk(&wr_mas);
2642 mast->bn->type = wr_mas.type;
2646 * mas_new_ma_node() - Create and return a new maple node. Helper function.
2647 * @mas: the maple state with the allocations.
2648 * @b_node: the maple_big_node with the type encoding.
2650 * Use the node type from the maple_big_node to allocate a new node from the
2651 * ma_state. This function exists mainly for code readability.
2653 * Return: A new maple encoded node
2655 static inline struct maple_enode
2656 *mas_new_ma_node(struct ma_state *mas, struct maple_big_node *b_node)
2658 return mt_mk_node(ma_mnode_ptr(mas_pop_node(mas)), b_node->type);
2662 * mas_mab_to_node() - Set up right and middle nodes
2664 * @mas: the maple state that contains the allocations.
2665 * @b_node: the node which contains the data.
2666 * @left: The pointer which will have the left node
2667 * @right: The pointer which may have the right node
2668 * @middle: the pointer which may have the middle node (rare)
2669 * @mid_split: the split location for the middle node
2671 * Return: the split of left.
2673 static inline unsigned char mas_mab_to_node(struct ma_state *mas,
2674 struct maple_big_node *b_node, struct maple_enode **left,
2675 struct maple_enode **right, struct maple_enode **middle,
2676 unsigned char *mid_split, unsigned long min)
2678 unsigned char split = 0;
2679 unsigned char slot_count = mt_slots[b_node->type];
2681 *left = mas_new_ma_node(mas, b_node);
2686 if (b_node->b_end < slot_count) {
2687 split = b_node->b_end;
2689 split = mab_calc_split(mas, b_node, mid_split, min);
2690 *right = mas_new_ma_node(mas, b_node);
2694 *middle = mas_new_ma_node(mas, b_node);
2701 * mab_set_b_end() - Add entry to b_node at b_node->b_end and increment the end
2703 * @b_node - the big node to add the entry
2704 * @mas - the maple state to get the pivot (mas->max)
2705 * @entry - the entry to add, if NULL nothing happens.
2707 static inline void mab_set_b_end(struct maple_big_node *b_node,
2708 struct ma_state *mas,
2714 b_node->slot[b_node->b_end] = entry;
2715 if (mt_is_alloc(mas->tree))
2716 b_node->gap[b_node->b_end] = mas_max_gap(mas);
2717 b_node->pivot[b_node->b_end++] = mas->max;
2721 * mas_set_split_parent() - combine_then_separate helper function. Sets the parent
2722 * of @mas->node to either @left or @right, depending on @slot and @split
2724 * @mas - the maple state with the node that needs a parent
2725 * @left - possible parent 1
2726 * @right - possible parent 2
2727 * @slot - the slot the mas->node was placed
2728 * @split - the split location between @left and @right
2730 static inline void mas_set_split_parent(struct ma_state *mas,
2731 struct maple_enode *left,
2732 struct maple_enode *right,
2733 unsigned char *slot, unsigned char split)
2735 if (mas_is_none(mas))
2738 if ((*slot) <= split)
2739 mte_set_parent(mas->node, left, *slot);
2741 mte_set_parent(mas->node, right, (*slot) - split - 1);
2747 * mte_mid_split_check() - Check if the next node passes the mid-split
2748 * @**l: Pointer to left encoded maple node.
2749 * @**m: Pointer to middle encoded maple node.
2750 * @**r: Pointer to right encoded maple node.
2752 * @*split: The split location.
2753 * @mid_split: The middle split.
2755 static inline void mte_mid_split_check(struct maple_enode **l,
2756 struct maple_enode **r,
2757 struct maple_enode *right,
2759 unsigned char *split,
2760 unsigned char mid_split)
2765 if (slot < mid_split)
2774 * mast_set_split_parents() - Helper function to set three nodes parents. Slot
2775 * is taken from @mast->l.
2776 * @mast - the maple subtree state
2777 * @left - the left node
2778 * @right - the right node
2779 * @split - the split location.
2781 static inline void mast_set_split_parents(struct maple_subtree_state *mast,
2782 struct maple_enode *left,
2783 struct maple_enode *middle,
2784 struct maple_enode *right,
2785 unsigned char split,
2786 unsigned char mid_split)
2789 struct maple_enode *l = left;
2790 struct maple_enode *r = right;
2792 if (mas_is_none(mast->l))
2798 slot = mast->l->offset;
2800 mte_mid_split_check(&l, &r, right, slot, &split, mid_split);
2801 mas_set_split_parent(mast->l, l, r, &slot, split);
2803 mte_mid_split_check(&l, &r, right, slot, &split, mid_split);
2804 mas_set_split_parent(mast->m, l, r, &slot, split);
2806 mte_mid_split_check(&l, &r, right, slot, &split, mid_split);
2807 mas_set_split_parent(mast->r, l, r, &slot, split);
2811 * mas_wmb_replace() - Write memory barrier and replace
2812 * @mas: The maple state
2813 * @free: the maple topiary list of nodes to free
2814 * @destroy: The maple topiary list of nodes to destroy (walk and free)
2816 * Updates gap as necessary.
2818 static inline void mas_wmb_replace(struct ma_state *mas,
2819 struct ma_topiary *free,
2820 struct ma_topiary *destroy)
2822 /* All nodes must see old data as dead prior to replacing that data */
2823 smp_wmb(); /* Needed for RCU */
2825 /* Insert the new data in the tree */
2826 mas_replace(mas, true);
2828 if (!mte_is_leaf(mas->node))
2829 mas_descend_adopt(mas);
2831 mas_mat_free(mas, free);
2834 mas_mat_destroy(mas, destroy);
2836 if (mte_is_leaf(mas->node))
2839 mas_update_gap(mas);
2843 * mast_new_root() - Set a new tree root during subtree creation
2844 * @mast: The maple subtree state
2845 * @mas: The maple state
2847 static inline void mast_new_root(struct maple_subtree_state *mast,
2848 struct ma_state *mas)
2850 mas_mn(mast->l)->parent =
2851 ma_parent_ptr(((unsigned long)mas->tree | MA_ROOT_PARENT));
2852 if (!mte_dead_node(mast->orig_l->node) &&
2853 !mte_is_root(mast->orig_l->node)) {
2855 mast_ascend_free(mast);
2857 } while (!mte_is_root(mast->orig_l->node));
2859 if ((mast->orig_l->node != mas->node) &&
2860 (mast->l->depth > mas_mt_height(mas))) {
2861 mat_add(mast->free, mas->node);
2866 * mast_cp_to_nodes() - Copy data out to nodes.
2867 * @mast: The maple subtree state
2868 * @left: The left encoded maple node
2869 * @middle: The middle encoded maple node
2870 * @right: The right encoded maple node
2871 * @split: The location to split between left and (middle ? middle : right)
2872 * @mid_split: The location to split between middle and right.
2874 static inline void mast_cp_to_nodes(struct maple_subtree_state *mast,
2875 struct maple_enode *left, struct maple_enode *middle,
2876 struct maple_enode *right, unsigned char split, unsigned char mid_split)
2878 bool new_lmax = true;
2880 mast->l->node = mte_node_or_none(left);
2881 mast->m->node = mte_node_or_none(middle);
2882 mast->r->node = mte_node_or_none(right);
2884 mast->l->min = mast->orig_l->min;
2885 if (split == mast->bn->b_end) {
2886 mast->l->max = mast->orig_r->max;
2890 mab_mas_cp(mast->bn, 0, split, mast->l, new_lmax);
2893 mab_mas_cp(mast->bn, 1 + split, mid_split, mast->m, true);
2894 mast->m->min = mast->bn->pivot[split] + 1;
2898 mast->r->max = mast->orig_r->max;
2900 mab_mas_cp(mast->bn, 1 + split, mast->bn->b_end, mast->r, false);
2901 mast->r->min = mast->bn->pivot[split] + 1;
2906 * mast_combine_cp_left - Copy in the original left side of the tree into the
2907 * combined data set in the maple subtree state big node.
2908 * @mast: The maple subtree state
2910 static inline void mast_combine_cp_left(struct maple_subtree_state *mast)
2912 unsigned char l_slot = mast->orig_l->offset;
2917 mas_mab_cp(mast->orig_l, 0, l_slot - 1, mast->bn, 0);
2921 * mast_combine_cp_right: Copy in the original right side of the tree into the
2922 * combined data set in the maple subtree state big node.
2923 * @mast: The maple subtree state
2925 static inline void mast_combine_cp_right(struct maple_subtree_state *mast)
2927 if (mast->bn->pivot[mast->bn->b_end - 1] >= mast->orig_r->max)
2930 mas_mab_cp(mast->orig_r, mast->orig_r->offset + 1,
2931 mt_slot_count(mast->orig_r->node), mast->bn,
2933 mast->orig_r->last = mast->orig_r->max;
2937 * mast_sufficient: Check if the maple subtree state has enough data in the big
2938 * node to create at least one sufficient node
2939 * @mast: the maple subtree state
2941 static inline bool mast_sufficient(struct maple_subtree_state *mast)
2943 if (mast->bn->b_end > mt_min_slot_count(mast->orig_l->node))
2950 * mast_overflow: Check if there is too much data in the subtree state for a
2952 * @mast: The maple subtree state
2954 static inline bool mast_overflow(struct maple_subtree_state *mast)
2956 if (mast->bn->b_end >= mt_slot_count(mast->orig_l->node))
2962 static inline void *mtree_range_walk(struct ma_state *mas)
2964 unsigned long *pivots;
2965 unsigned char offset;
2966 struct maple_node *node;
2967 struct maple_enode *next, *last;
2968 enum maple_type type;
2971 unsigned long max, min;
2972 unsigned long prev_max, prev_min;
2980 node = mte_to_node(next);
2981 type = mte_node_type(next);
2982 pivots = ma_pivots(node, type);
2983 end = ma_data_end(node, type, pivots, max);
2984 if (unlikely(ma_dead_node(node)))
2987 if (pivots[offset] >= mas->index) {
2990 max = pivots[offset];
2996 } while ((offset < end) && (pivots[offset] < mas->index));
2999 min = pivots[offset - 1] + 1;
3001 if (likely(offset < end && pivots[offset]))
3002 max = pivots[offset];
3005 slots = ma_slots(node, type);
3006 next = mt_slot(mas->tree, slots, offset);
3007 if (unlikely(ma_dead_node(node)))
3009 } while (!ma_is_leaf(type));
3011 mas->offset = offset;
3014 mas->min = prev_min;
3015 mas->max = prev_max;
3017 return (void *)next;
3025 * mas_spanning_rebalance() - Rebalance across two nodes which may not be peers.
3026 * @mas: The starting maple state
3027 * @mast: The maple_subtree_state, keeps track of 4 maple states.
3028 * @count: The estimated count of iterations needed.
3030 * Follow the tree upwards from @l_mas and @r_mas for @count, or until the root
3031 * is hit. First @b_node is split into two entries which are inserted into the
3032 * next iteration of the loop. @b_node is returned populated with the final
3033 * iteration. @mas is used to obtain allocations. orig_l_mas keeps track of the
3034 * nodes that will remain active by using orig_l_mas->index and orig_l_mas->last
3035 * to account of what has been copied into the new sub-tree. The update of
3036 * orig_l_mas->last is used in mas_consume to find the slots that will need to
3037 * be either freed or destroyed. orig_l_mas->depth keeps track of the height of
3038 * the new sub-tree in case the sub-tree becomes the full tree.
3040 * Return: the number of elements in b_node during the last loop.
3042 static int mas_spanning_rebalance(struct ma_state *mas,
3043 struct maple_subtree_state *mast, unsigned char count)
3045 unsigned char split, mid_split;
3046 unsigned char slot = 0;
3047 struct maple_enode *left = NULL, *middle = NULL, *right = NULL;
3049 MA_STATE(l_mas, mas->tree, mas->index, mas->index);
3050 MA_STATE(r_mas, mas->tree, mas->index, mas->last);
3051 MA_STATE(m_mas, mas->tree, mas->index, mas->index);
3052 MA_TOPIARY(free, mas->tree);
3053 MA_TOPIARY(destroy, mas->tree);
3056 * The tree needs to be rebalanced and leaves need to be kept at the same level.
3057 * Rebalancing is done by use of the ``struct maple_topiary``.
3063 mast->destroy = &destroy;
3064 l_mas.node = r_mas.node = m_mas.node = MAS_NONE;
3066 /* Check if this is not root and has sufficient data. */
3067 if (((mast->orig_l->min != 0) || (mast->orig_r->max != ULONG_MAX)) &&
3068 unlikely(mast->bn->b_end <= mt_min_slots[mast->bn->type]))
3069 mast_spanning_rebalance(mast);
3071 mast->orig_l->depth = 0;
3074 * Each level of the tree is examined and balanced, pushing data to the left or
3075 * right, or rebalancing against left or right nodes is employed to avoid
3076 * rippling up the tree to limit the amount of churn. Once a new sub-section of
3077 * the tree is created, there may be a mix of new and old nodes. The old nodes
3078 * will have the incorrect parent pointers and currently be in two trees: the
3079 * original tree and the partially new tree. To remedy the parent pointers in
3080 * the old tree, the new data is swapped into the active tree and a walk down
3081 * the tree is performed and the parent pointers are updated.
3082 * See mas_descend_adopt() for more information..
3086 mast->bn->type = mte_node_type(mast->orig_l->node);
3087 split = mas_mab_to_node(mas, mast->bn, &left, &right, &middle,
3088 &mid_split, mast->orig_l->min);
3089 mast_set_split_parents(mast, left, middle, right, split,
3091 mast_cp_to_nodes(mast, left, middle, right, split, mid_split);
3094 * Copy data from next level in the tree to mast->bn from next
3097 memset(mast->bn, 0, sizeof(struct maple_big_node));
3098 mast->bn->type = mte_node_type(left);
3099 mast->orig_l->depth++;
3101 /* Root already stored in l->node. */
3102 if (mas_is_root_limits(mast->l))
3105 mast_ascend_free(mast);
3106 mast_combine_cp_left(mast);
3107 l_mas.offset = mast->bn->b_end;
3108 mab_set_b_end(mast->bn, &l_mas, left);
3109 mab_set_b_end(mast->bn, &m_mas, middle);
3110 mab_set_b_end(mast->bn, &r_mas, right);
3112 /* Copy anything necessary out of the right node. */
3113 mast_combine_cp_right(mast);
3115 mast->orig_l->last = mast->orig_l->max;
3117 if (mast_sufficient(mast))
3120 if (mast_overflow(mast))
3123 /* May be a new root stored in mast->bn */
3124 if (mas_is_root_limits(mast->orig_l))
3127 mast_spanning_rebalance(mast);
3129 /* rebalancing from other nodes may require another loop. */
3134 l_mas.node = mt_mk_node(ma_mnode_ptr(mas_pop_node(mas)),
3135 mte_node_type(mast->orig_l->node));
3136 mast->orig_l->depth++;
3137 mab_mas_cp(mast->bn, 0, mt_slots[mast->bn->type] - 1, &l_mas, true);
3138 mte_set_parent(left, l_mas.node, slot);
3140 mte_set_parent(middle, l_mas.node, ++slot);
3143 mte_set_parent(right, l_mas.node, ++slot);
3145 if (mas_is_root_limits(mast->l)) {
3147 mast_new_root(mast, mas);
3149 mas_mn(&l_mas)->parent = mas_mn(mast->orig_l)->parent;
3152 if (!mte_dead_node(mast->orig_l->node))
3153 mat_add(&free, mast->orig_l->node);
3155 mas->depth = mast->orig_l->depth;
3156 *mast->orig_l = l_mas;
3157 mte_set_node_dead(mas->node);
3159 /* Set up mas for insertion. */
3160 mast->orig_l->depth = mas->depth;
3161 mast->orig_l->alloc = mas->alloc;
3162 *mas = *mast->orig_l;
3163 mas_wmb_replace(mas, &free, &destroy);
3164 mtree_range_walk(mas);
3165 return mast->bn->b_end;
3169 * mas_rebalance() - Rebalance a given node.
3170 * @mas: The maple state
3171 * @b_node: The big maple node.
3173 * Rebalance two nodes into a single node or two new nodes that are sufficient.
3174 * Continue upwards until tree is sufficient.
3176 * Return: the number of elements in b_node during the last loop.
3178 static inline int mas_rebalance(struct ma_state *mas,
3179 struct maple_big_node *b_node)
3181 char empty_count = mas_mt_height(mas);
3182 struct maple_subtree_state mast;
3183 unsigned char shift, b_end = ++b_node->b_end;
3185 MA_STATE(l_mas, mas->tree, mas->index, mas->last);
3186 MA_STATE(r_mas, mas->tree, mas->index, mas->last);
3188 trace_ma_op(__func__, mas);
3191 * Rebalancing occurs if a node is insufficient. Data is rebalanced
3192 * against the node to the right if it exists, otherwise the node to the
3193 * left of this node is rebalanced against this node. If rebalancing
3194 * causes just one node to be produced instead of two, then the parent
3195 * is also examined and rebalanced if it is insufficient. Every level
3196 * tries to combine the data in the same way. If one node contains the
3197 * entire range of the tree, then that node is used as a new root node.
3199 mas_node_count(mas, 1 + empty_count * 3);
3200 if (mas_is_err(mas))
3203 mast.orig_l = &l_mas;
3204 mast.orig_r = &r_mas;
3206 mast.bn->type = mte_node_type(mas->node);
3208 l_mas = r_mas = *mas;
3210 if (mas_next_sibling(&r_mas)) {
3211 mas_mab_cp(&r_mas, 0, mt_slot_count(r_mas.node), b_node, b_end);
3212 r_mas.last = r_mas.index = r_mas.max;
3214 mas_prev_sibling(&l_mas);
3215 shift = mas_data_end(&l_mas) + 1;
3216 mab_shift_right(b_node, shift);
3217 mas->offset += shift;
3218 mas_mab_cp(&l_mas, 0, shift - 1, b_node, 0);
3219 b_node->b_end = shift + b_end;
3220 l_mas.index = l_mas.last = l_mas.min;
3223 return mas_spanning_rebalance(mas, &mast, empty_count);
3227 * mas_destroy_rebalance() - Rebalance left-most node while destroying the maple
3229 * @mas: The maple state
3230 * @end: The end of the left-most node.
3232 * During a mass-insert event (such as forking), it may be necessary to
3233 * rebalance the left-most node when it is not sufficient.
3235 static inline void mas_destroy_rebalance(struct ma_state *mas, unsigned char end)
3237 enum maple_type mt = mte_node_type(mas->node);
3238 struct maple_node reuse, *newnode, *parent, *new_left, *left, *node;
3239 struct maple_enode *eparent;
3240 unsigned char offset, tmp, split = mt_slots[mt] / 2;
3241 void __rcu **l_slots, **slots;
3242 unsigned long *l_pivs, *pivs, gap;
3243 bool in_rcu = mt_in_rcu(mas->tree);
3245 MA_STATE(l_mas, mas->tree, mas->index, mas->last);
3248 mas_prev_sibling(&l_mas);
3252 /* Allocate for both left and right as well as parent. */
3253 mas_node_count(mas, 3);
3254 if (mas_is_err(mas))
3257 newnode = mas_pop_node(mas);
3263 newnode->parent = node->parent;
3264 slots = ma_slots(newnode, mt);
3265 pivs = ma_pivots(newnode, mt);
3266 left = mas_mn(&l_mas);
3267 l_slots = ma_slots(left, mt);
3268 l_pivs = ma_pivots(left, mt);
3269 if (!l_slots[split])
3271 tmp = mas_data_end(&l_mas) - split;
3273 memcpy(slots, l_slots + split + 1, sizeof(void *) * tmp);
3274 memcpy(pivs, l_pivs + split + 1, sizeof(unsigned long) * tmp);
3275 pivs[tmp] = l_mas.max;
3276 memcpy(slots + tmp, ma_slots(node, mt), sizeof(void *) * end);
3277 memcpy(pivs + tmp, ma_pivots(node, mt), sizeof(unsigned long) * end);
3279 l_mas.max = l_pivs[split];
3280 mas->min = l_mas.max + 1;
3281 eparent = mt_mk_node(mte_parent(l_mas.node),
3282 mas_parent_enum(&l_mas, l_mas.node));
3285 unsigned char max_p = mt_pivots[mt];
3286 unsigned char max_s = mt_slots[mt];
3289 memset(pivs + tmp, 0,
3290 sizeof(unsigned long *) * (max_p - tmp));
3292 if (tmp < mt_slots[mt])
3293 memset(slots + tmp, 0, sizeof(void *) * (max_s - tmp));
3295 memcpy(node, newnode, sizeof(struct maple_node));
3296 ma_set_meta(node, mt, 0, tmp - 1);
3297 mte_set_pivot(eparent, mte_parent_slot(l_mas.node),
3300 /* Remove data from l_pivs. */
3302 memset(l_pivs + tmp, 0, sizeof(unsigned long) * (max_p - tmp));
3303 memset(l_slots + tmp, 0, sizeof(void *) * (max_s - tmp));
3304 ma_set_meta(left, mt, 0, split);
3309 /* RCU requires replacing both l_mas, mas, and parent. */
3310 mas->node = mt_mk_node(newnode, mt);
3311 ma_set_meta(newnode, mt, 0, tmp);
3313 new_left = mas_pop_node(mas);
3314 new_left->parent = left->parent;
3315 mt = mte_node_type(l_mas.node);
3316 slots = ma_slots(new_left, mt);
3317 pivs = ma_pivots(new_left, mt);
3318 memcpy(slots, l_slots, sizeof(void *) * split);
3319 memcpy(pivs, l_pivs, sizeof(unsigned long) * split);
3320 ma_set_meta(new_left, mt, 0, split);
3321 l_mas.node = mt_mk_node(new_left, mt);
3323 /* replace parent. */
3324 offset = mte_parent_slot(mas->node);
3325 mt = mas_parent_enum(&l_mas, l_mas.node);
3326 parent = mas_pop_node(mas);
3327 slots = ma_slots(parent, mt);
3328 pivs = ma_pivots(parent, mt);
3329 memcpy(parent, mte_to_node(eparent), sizeof(struct maple_node));
3330 rcu_assign_pointer(slots[offset], mas->node);
3331 rcu_assign_pointer(slots[offset - 1], l_mas.node);
3332 pivs[offset - 1] = l_mas.max;
3333 eparent = mt_mk_node(parent, mt);
3335 gap = mas_leaf_max_gap(mas);
3336 mte_set_gap(eparent, mte_parent_slot(mas->node), gap);
3337 gap = mas_leaf_max_gap(&l_mas);
3338 mte_set_gap(eparent, mte_parent_slot(l_mas.node), gap);
3342 mas_replace(mas, false);
3344 mas_update_gap(mas);
3348 * mas_split_final_node() - Split the final node in a subtree operation.
3349 * @mast: the maple subtree state
3350 * @mas: The maple state
3351 * @height: The height of the tree in case it's a new root.
3353 static inline bool mas_split_final_node(struct maple_subtree_state *mast,
3354 struct ma_state *mas, int height)
3356 struct maple_enode *ancestor;
3358 if (mte_is_root(mas->node)) {
3359 if (mt_is_alloc(mas->tree))
3360 mast->bn->type = maple_arange_64;
3362 mast->bn->type = maple_range_64;
3363 mas->depth = height;
3366 * Only a single node is used here, could be root.
3367 * The Big_node data should just fit in a single node.
3369 ancestor = mas_new_ma_node(mas, mast->bn);
3370 mte_set_parent(mast->l->node, ancestor, mast->l->offset);
3371 mte_set_parent(mast->r->node, ancestor, mast->r->offset);
3372 mte_to_node(ancestor)->parent = mas_mn(mas)->parent;
3374 mast->l->node = ancestor;
3375 mab_mas_cp(mast->bn, 0, mt_slots[mast->bn->type] - 1, mast->l, true);
3376 mas->offset = mast->bn->b_end - 1;
3381 * mast_fill_bnode() - Copy data into the big node in the subtree state
3382 * @mast: The maple subtree state
3383 * @mas: the maple state
3384 * @skip: The number of entries to skip for new nodes insertion.
3386 static inline void mast_fill_bnode(struct maple_subtree_state *mast,
3387 struct ma_state *mas,
3391 struct maple_enode *old = mas->node;
3392 unsigned char split;
3394 memset(mast->bn->gap, 0, sizeof(unsigned long) * ARRAY_SIZE(mast->bn->gap));
3395 memset(mast->bn->slot, 0, sizeof(unsigned long) * ARRAY_SIZE(mast->bn->slot));
3396 memset(mast->bn->pivot, 0, sizeof(unsigned long) * ARRAY_SIZE(mast->bn->pivot));
3397 mast->bn->b_end = 0;
3399 if (mte_is_root(mas->node)) {
3403 mat_add(mast->free, old);
3404 mas->offset = mte_parent_slot(mas->node);
3407 if (cp && mast->l->offset)
3408 mas_mab_cp(mas, 0, mast->l->offset - 1, mast->bn, 0);
3410 split = mast->bn->b_end;
3411 mab_set_b_end(mast->bn, mast->l, mast->l->node);
3412 mast->r->offset = mast->bn->b_end;
3413 mab_set_b_end(mast->bn, mast->r, mast->r->node);
3414 if (mast->bn->pivot[mast->bn->b_end - 1] == mas->max)
3418 mas_mab_cp(mas, split + skip, mt_slot_count(mas->node) - 1,
3419 mast->bn, mast->bn->b_end);
3422 mast->bn->type = mte_node_type(mas->node);
3426 * mast_split_data() - Split the data in the subtree state big node into regular
3428 * @mast: The maple subtree state
3429 * @mas: The maple state
3430 * @split: The location to split the big node
3432 static inline void mast_split_data(struct maple_subtree_state *mast,
3433 struct ma_state *mas, unsigned char split)
3435 unsigned char p_slot;
3437 mab_mas_cp(mast->bn, 0, split, mast->l, true);
3438 mte_set_pivot(mast->r->node, 0, mast->r->max);
3439 mab_mas_cp(mast->bn, split + 1, mast->bn->b_end, mast->r, false);
3440 mast->l->offset = mte_parent_slot(mas->node);
3441 mast->l->max = mast->bn->pivot[split];
3442 mast->r->min = mast->l->max + 1;
3443 if (mte_is_leaf(mas->node))
3446 p_slot = mast->orig_l->offset;
3447 mas_set_split_parent(mast->orig_l, mast->l->node, mast->r->node,
3449 mas_set_split_parent(mast->orig_r, mast->l->node, mast->r->node,
3454 * mas_push_data() - Instead of splitting a node, it is beneficial to push the
3455 * data to the right or left node if there is room.
3456 * @mas: The maple state
3457 * @height: The current height of the maple state
3458 * @mast: The maple subtree state
3459 * @left: Push left or not.
3461 * Keeping the height of the tree low means faster lookups.
3463 * Return: True if pushed, false otherwise.
3465 static inline bool mas_push_data(struct ma_state *mas, int height,
3466 struct maple_subtree_state *mast, bool left)
3468 unsigned char slot_total = mast->bn->b_end;
3469 unsigned char end, space, split;
3471 MA_STATE(tmp_mas, mas->tree, mas->index, mas->last);
3473 tmp_mas.depth = mast->l->depth;
3475 if (left && !mas_prev_sibling(&tmp_mas))
3477 else if (!left && !mas_next_sibling(&tmp_mas))
3480 end = mas_data_end(&tmp_mas);
3482 space = 2 * mt_slot_count(mas->node) - 2;
3483 /* -2 instead of -1 to ensure there isn't a triple split */
3484 if (ma_is_leaf(mast->bn->type))
3487 if (mas->max == ULONG_MAX)
3490 if (slot_total >= space)
3493 /* Get the data; Fill mast->bn */
3496 mab_shift_right(mast->bn, end + 1);
3497 mas_mab_cp(&tmp_mas, 0, end, mast->bn, 0);
3498 mast->bn->b_end = slot_total + 1;
3500 mas_mab_cp(&tmp_mas, 0, end, mast->bn, mast->bn->b_end);
3503 /* Configure mast for splitting of mast->bn */
3504 split = mt_slots[mast->bn->type] - 2;
3506 /* Switch mas to prev node */
3507 mat_add(mast->free, mas->node);
3509 /* Start using mast->l for the left side. */
3510 tmp_mas.node = mast->l->node;
3513 mat_add(mast->free, tmp_mas.node);
3514 tmp_mas.node = mast->r->node;
3516 split = slot_total - split;
3518 split = mab_no_null_split(mast->bn, split, mt_slots[mast->bn->type]);
3519 /* Update parent slot for split calculation. */
3521 mast->orig_l->offset += end + 1;
3523 mast_split_data(mast, mas, split);
3524 mast_fill_bnode(mast, mas, 2);
3525 mas_split_final_node(mast, mas, height + 1);
3530 * mas_split() - Split data that is too big for one node into two.
3531 * @mas: The maple state
3532 * @b_node: The maple big node
3533 * Return: 1 on success, 0 on failure.
3535 static int mas_split(struct ma_state *mas, struct maple_big_node *b_node)
3537 struct maple_subtree_state mast;
3539 unsigned char mid_split, split = 0;
3542 * Splitting is handled differently from any other B-tree; the Maple
3543 * Tree splits upwards. Splitting up means that the split operation
3544 * occurs when the walk of the tree hits the leaves and not on the way
3545 * down. The reason for splitting up is that it is impossible to know
3546 * how much space will be needed until the leaf is (or leaves are)
3547 * reached. Since overwriting data is allowed and a range could
3548 * overwrite more than one range or result in changing one entry into 3
3549 * entries, it is impossible to know if a split is required until the
3552 * Splitting is a balancing act between keeping allocations to a minimum
3553 * and avoiding a 'jitter' event where a tree is expanded to make room
3554 * for an entry followed by a contraction when the entry is removed. To
3555 * accomplish the balance, there are empty slots remaining in both left
3556 * and right nodes after a split.
3558 MA_STATE(l_mas, mas->tree, mas->index, mas->last);
3559 MA_STATE(r_mas, mas->tree, mas->index, mas->last);
3560 MA_STATE(prev_l_mas, mas->tree, mas->index, mas->last);
3561 MA_STATE(prev_r_mas, mas->tree, mas->index, mas->last);
3562 MA_TOPIARY(mat, mas->tree);
3564 trace_ma_op(__func__, mas);
3565 mas->depth = mas_mt_height(mas);
3566 /* Allocation failures will happen early. */
3567 mas_node_count(mas, 1 + mas->depth * 2);
3568 if (mas_is_err(mas))
3573 mast.orig_l = &prev_l_mas;
3574 mast.orig_r = &prev_r_mas;
3578 while (height++ <= mas->depth) {
3579 if (mt_slots[b_node->type] > b_node->b_end) {
3580 mas_split_final_node(&mast, mas, height);
3584 l_mas = r_mas = *mas;
3585 l_mas.node = mas_new_ma_node(mas, b_node);
3586 r_mas.node = mas_new_ma_node(mas, b_node);
3588 * Another way that 'jitter' is avoided is to terminate a split up early if the
3589 * left or right node has space to spare. This is referred to as "pushing left"
3590 * or "pushing right" and is similar to the B* tree, except the nodes left or
3591 * right can rarely be reused due to RCU, but the ripple upwards is halted which
3592 * is a significant savings.
3594 /* Try to push left. */
3595 if (mas_push_data(mas, height, &mast, true))
3598 /* Try to push right. */
3599 if (mas_push_data(mas, height, &mast, false))
3602 split = mab_calc_split(mas, b_node, &mid_split, prev_l_mas.min);
3603 mast_split_data(&mast, mas, split);
3605 * Usually correct, mab_mas_cp in the above call overwrites
3608 mast.r->max = mas->max;
3609 mast_fill_bnode(&mast, mas, 1);
3610 prev_l_mas = *mast.l;
3611 prev_r_mas = *mast.r;
3614 /* Set the original node as dead */
3615 mat_add(mast.free, mas->node);
3616 mas->node = l_mas.node;
3617 mas_wmb_replace(mas, mast.free, NULL);
3618 mtree_range_walk(mas);
3623 * mas_reuse_node() - Reuse the node to store the data.
3624 * @wr_mas: The maple write state
3625 * @bn: The maple big node
3626 * @end: The end of the data.
3628 * Will always return false in RCU mode.
3630 * Return: True if node was reused, false otherwise.
3632 static inline bool mas_reuse_node(struct ma_wr_state *wr_mas,
3633 struct maple_big_node *bn, unsigned char end)
3635 /* Need to be rcu safe. */
3636 if (mt_in_rcu(wr_mas->mas->tree))
3639 if (end > bn->b_end) {
3640 int clear = mt_slots[wr_mas->type] - bn->b_end;
3642 memset(wr_mas->slots + bn->b_end, 0, sizeof(void *) * clear--);
3643 memset(wr_mas->pivots + bn->b_end, 0, sizeof(void *) * clear);
3645 mab_mas_cp(bn, 0, bn->b_end, wr_mas->mas, false);
3650 * mas_commit_b_node() - Commit the big node into the tree.
3651 * @wr_mas: The maple write state
3652 * @b_node: The maple big node
3653 * @end: The end of the data.
3655 static noinline_for_kasan int mas_commit_b_node(struct ma_wr_state *wr_mas,
3656 struct maple_big_node *b_node, unsigned char end)
3658 struct maple_node *node;
3659 unsigned char b_end = b_node->b_end;
3660 enum maple_type b_type = b_node->type;
3662 if ((b_end < mt_min_slots[b_type]) &&
3663 (!mte_is_root(wr_mas->mas->node)) &&
3664 (mas_mt_height(wr_mas->mas) > 1))
3665 return mas_rebalance(wr_mas->mas, b_node);
3667 if (b_end >= mt_slots[b_type])
3668 return mas_split(wr_mas->mas, b_node);
3670 if (mas_reuse_node(wr_mas, b_node, end))
3673 mas_node_count(wr_mas->mas, 1);
3674 if (mas_is_err(wr_mas->mas))
3677 node = mas_pop_node(wr_mas->mas);
3678 node->parent = mas_mn(wr_mas->mas)->parent;
3679 wr_mas->mas->node = mt_mk_node(node, b_type);
3680 mab_mas_cp(b_node, 0, b_end, wr_mas->mas, false);
3681 mas_replace(wr_mas->mas, false);
3683 mas_update_gap(wr_mas->mas);
3688 * mas_root_expand() - Expand a root to a node
3689 * @mas: The maple state
3690 * @entry: The entry to store into the tree
3692 static inline int mas_root_expand(struct ma_state *mas, void *entry)
3694 void *contents = mas_root_locked(mas);
3695 enum maple_type type = maple_leaf_64;
3696 struct maple_node *node;
3698 unsigned long *pivots;
3701 mas_node_count(mas, 1);
3702 if (unlikely(mas_is_err(mas)))
3705 node = mas_pop_node(mas);
3706 pivots = ma_pivots(node, type);
3707 slots = ma_slots(node, type);
3708 node->parent = ma_parent_ptr(
3709 ((unsigned long)mas->tree | MA_ROOT_PARENT));
3710 mas->node = mt_mk_node(node, type);
3714 rcu_assign_pointer(slots[slot], contents);
3715 if (likely(mas->index > 1))
3718 pivots[slot++] = mas->index - 1;
3721 rcu_assign_pointer(slots[slot], entry);
3723 pivots[slot] = mas->last;
3724 if (mas->last != ULONG_MAX)
3727 mas_set_height(mas);
3728 ma_set_meta(node, maple_leaf_64, 0, slot);
3729 /* swap the new root into the tree */
3730 rcu_assign_pointer(mas->tree->ma_root, mte_mk_root(mas->node));
3734 static inline void mas_store_root(struct ma_state *mas, void *entry)
3736 if (likely((mas->last != 0) || (mas->index != 0)))
3737 mas_root_expand(mas, entry);
3738 else if (((unsigned long) (entry) & 3) == 2)
3739 mas_root_expand(mas, entry);
3741 rcu_assign_pointer(mas->tree->ma_root, entry);
3742 mas->node = MAS_START;
3747 * mas_is_span_wr() - Check if the write needs to be treated as a write that
3749 * @mas: The maple state
3750 * @piv: The pivot value being written
3751 * @type: The maple node type
3752 * @entry: The data to write
3754 * Spanning writes are writes that start in one node and end in another OR if
3755 * the write of a %NULL will cause the node to end with a %NULL.
3757 * Return: True if this is a spanning write, false otherwise.
3759 static bool mas_is_span_wr(struct ma_wr_state *wr_mas)
3762 unsigned long last = wr_mas->mas->last;
3763 unsigned long piv = wr_mas->r_max;
3764 enum maple_type type = wr_mas->type;
3765 void *entry = wr_mas->entry;
3767 /* Contained in this pivot */
3771 max = wr_mas->mas->max;
3772 if (unlikely(ma_is_leaf(type))) {
3773 /* Fits in the node, but may span slots. */
3777 /* Writes to the end of the node but not null. */
3778 if ((last == max) && entry)
3782 * Writing ULONG_MAX is not a spanning write regardless of the
3783 * value being written as long as the range fits in the node.
3785 if ((last == ULONG_MAX) && (last == max))
3787 } else if (piv == last) {
3791 /* Detect spanning store wr walk */
3792 if (last == ULONG_MAX)
3796 trace_ma_write(__func__, wr_mas->mas, piv, entry);
3801 static inline void mas_wr_walk_descend(struct ma_wr_state *wr_mas)
3803 wr_mas->type = mte_node_type(wr_mas->mas->node);
3804 mas_wr_node_walk(wr_mas);
3805 wr_mas->slots = ma_slots(wr_mas->node, wr_mas->type);
3808 static inline void mas_wr_walk_traverse(struct ma_wr_state *wr_mas)
3810 wr_mas->mas->max = wr_mas->r_max;
3811 wr_mas->mas->min = wr_mas->r_min;
3812 wr_mas->mas->node = wr_mas->content;
3813 wr_mas->mas->offset = 0;
3814 wr_mas->mas->depth++;
3817 * mas_wr_walk() - Walk the tree for a write.
3818 * @wr_mas: The maple write state
3820 * Uses mas_slot_locked() and does not need to worry about dead nodes.
3822 * Return: True if it's contained in a node, false on spanning write.
3824 static bool mas_wr_walk(struct ma_wr_state *wr_mas)
3826 struct ma_state *mas = wr_mas->mas;
3829 mas_wr_walk_descend(wr_mas);
3830 if (unlikely(mas_is_span_wr(wr_mas)))
3833 wr_mas->content = mas_slot_locked(mas, wr_mas->slots,
3835 if (ma_is_leaf(wr_mas->type))
3838 mas_wr_walk_traverse(wr_mas);
3844 static bool mas_wr_walk_index(struct ma_wr_state *wr_mas)
3846 struct ma_state *mas = wr_mas->mas;
3849 mas_wr_walk_descend(wr_mas);
3850 wr_mas->content = mas_slot_locked(mas, wr_mas->slots,
3852 if (ma_is_leaf(wr_mas->type))
3854 mas_wr_walk_traverse(wr_mas);
3860 * mas_extend_spanning_null() - Extend a store of a %NULL to include surrounding %NULLs.
3861 * @l_wr_mas: The left maple write state
3862 * @r_wr_mas: The right maple write state
3864 static inline void mas_extend_spanning_null(struct ma_wr_state *l_wr_mas,
3865 struct ma_wr_state *r_wr_mas)
3867 struct ma_state *r_mas = r_wr_mas->mas;
3868 struct ma_state *l_mas = l_wr_mas->mas;
3869 unsigned char l_slot;
3871 l_slot = l_mas->offset;
3872 if (!l_wr_mas->content)
3873 l_mas->index = l_wr_mas->r_min;
3875 if ((l_mas->index == l_wr_mas->r_min) &&
3877 !mas_slot_locked(l_mas, l_wr_mas->slots, l_slot - 1))) {
3879 l_mas->index = l_wr_mas->pivots[l_slot - 2] + 1;
3881 l_mas->index = l_mas->min;
3883 l_mas->offset = l_slot - 1;
3886 if (!r_wr_mas->content) {
3887 if (r_mas->last < r_wr_mas->r_max)
3888 r_mas->last = r_wr_mas->r_max;
3890 } else if ((r_mas->last == r_wr_mas->r_max) &&
3891 (r_mas->last < r_mas->max) &&
3892 !mas_slot_locked(r_mas, r_wr_mas->slots, r_mas->offset + 1)) {
3893 r_mas->last = mas_safe_pivot(r_mas, r_wr_mas->pivots,
3894 r_wr_mas->type, r_mas->offset + 1);
3899 static inline void *mas_state_walk(struct ma_state *mas)
3903 entry = mas_start(mas);
3904 if (mas_is_none(mas))
3907 if (mas_is_ptr(mas))
3910 return mtree_range_walk(mas);
3914 * mtree_lookup_walk() - Internal quick lookup that does not keep maple state up
3917 * @mas: The maple state.
3919 * Note: Leaves mas in undesirable state.
3920 * Return: The entry for @mas->index or %NULL on dead node.
3922 static inline void *mtree_lookup_walk(struct ma_state *mas)
3924 unsigned long *pivots;
3925 unsigned char offset;
3926 struct maple_node *node;
3927 struct maple_enode *next;
3928 enum maple_type type;
3937 node = mte_to_node(next);
3938 type = mte_node_type(next);
3939 pivots = ma_pivots(node, type);
3940 end = ma_data_end(node, type, pivots, max);
3941 if (unlikely(ma_dead_node(node)))
3944 if (pivots[offset] >= mas->index) {
3945 max = pivots[offset];
3948 } while (++offset < end);
3950 slots = ma_slots(node, type);
3951 next = mt_slot(mas->tree, slots, offset);
3952 if (unlikely(ma_dead_node(node)))
3954 } while (!ma_is_leaf(type));
3956 return (void *)next;
3964 * mas_new_root() - Create a new root node that only contains the entry passed
3966 * @mas: The maple state
3967 * @entry: The entry to store.
3969 * Only valid when the index == 0 and the last == ULONG_MAX
3971 * Return 0 on error, 1 on success.
3973 static inline int mas_new_root(struct ma_state *mas, void *entry)
3975 struct maple_enode *root = mas_root_locked(mas);
3976 enum maple_type type = maple_leaf_64;
3977 struct maple_node *node;
3979 unsigned long *pivots;
3981 if (!entry && !mas->index && mas->last == ULONG_MAX) {
3983 mas_set_height(mas);
3984 rcu_assign_pointer(mas->tree->ma_root, entry);
3985 mas->node = MAS_START;
3989 mas_node_count(mas, 1);
3990 if (mas_is_err(mas))
3993 node = mas_pop_node(mas);
3994 pivots = ma_pivots(node, type);
3995 slots = ma_slots(node, type);
3996 node->parent = ma_parent_ptr(
3997 ((unsigned long)mas->tree | MA_ROOT_PARENT));
3998 mas->node = mt_mk_node(node, type);
3999 rcu_assign_pointer(slots[0], entry);
4000 pivots[0] = mas->last;
4002 mas_set_height(mas);
4003 rcu_assign_pointer(mas->tree->ma_root, mte_mk_root(mas->node));
4006 if (xa_is_node(root))
4007 mte_destroy_walk(root, mas->tree);
4012 * mas_wr_spanning_store() - Create a subtree with the store operation completed
4013 * and new nodes where necessary, then place the sub-tree in the actual tree.
4014 * Note that mas is expected to point to the node which caused the store to
4016 * @wr_mas: The maple write state
4018 * Return: 0 on error, positive on success.
4020 static inline int mas_wr_spanning_store(struct ma_wr_state *wr_mas)
4022 struct maple_subtree_state mast;
4023 struct maple_big_node b_node;
4024 struct ma_state *mas;
4025 unsigned char height;
4027 /* Left and Right side of spanning store */
4028 MA_STATE(l_mas, NULL, 0, 0);
4029 MA_STATE(r_mas, NULL, 0, 0);
4031 MA_WR_STATE(r_wr_mas, &r_mas, wr_mas->entry);
4032 MA_WR_STATE(l_wr_mas, &l_mas, wr_mas->entry);
4035 * A store operation that spans multiple nodes is called a spanning
4036 * store and is handled early in the store call stack by the function
4037 * mas_is_span_wr(). When a spanning store is identified, the maple
4038 * state is duplicated. The first maple state walks the left tree path
4039 * to ``index``, the duplicate walks the right tree path to ``last``.
4040 * The data in the two nodes are combined into a single node, two nodes,
4041 * or possibly three nodes (see the 3-way split above). A ``NULL``
4042 * written to the last entry of a node is considered a spanning store as
4043 * a rebalance is required for the operation to complete and an overflow
4044 * of data may happen.
4047 trace_ma_op(__func__, mas);
4049 if (unlikely(!mas->index && mas->last == ULONG_MAX))
4050 return mas_new_root(mas, wr_mas->entry);
4052 * Node rebalancing may occur due to this store, so there may be three new
4053 * entries per level plus a new root.
4055 height = mas_mt_height(mas);
4056 mas_node_count(mas, 1 + height * 3);
4057 if (mas_is_err(mas))
4061 * Set up right side. Need to get to the next offset after the spanning
4062 * store to ensure it's not NULL and to combine both the next node and
4063 * the node with the start together.
4066 /* Avoid overflow, walk to next slot in the tree. */
4070 r_mas.index = r_mas.last;
4071 mas_wr_walk_index(&r_wr_mas);
4072 r_mas.last = r_mas.index = mas->last;
4074 /* Set up left side. */
4076 mas_wr_walk_index(&l_wr_mas);
4078 if (!wr_mas->entry) {
4079 mas_extend_spanning_null(&l_wr_mas, &r_wr_mas);
4080 mas->offset = l_mas.offset;
4081 mas->index = l_mas.index;
4082 mas->last = l_mas.last = r_mas.last;
4085 /* expanding NULLs may make this cover the entire range */
4086 if (!l_mas.index && r_mas.last == ULONG_MAX) {
4087 mas_set_range(mas, 0, ULONG_MAX);
4088 return mas_new_root(mas, wr_mas->entry);
4091 memset(&b_node, 0, sizeof(struct maple_big_node));
4092 /* Copy l_mas and store the value in b_node. */
4093 mas_store_b_node(&l_wr_mas, &b_node, l_wr_mas.node_end);
4094 /* Copy r_mas into b_node. */
4095 if (r_mas.offset <= r_wr_mas.node_end)
4096 mas_mab_cp(&r_mas, r_mas.offset, r_wr_mas.node_end,
4097 &b_node, b_node.b_end + 1);
4101 /* Stop spanning searches by searching for just index. */
4102 l_mas.index = l_mas.last = mas->index;
4105 mast.orig_l = &l_mas;
4106 mast.orig_r = &r_mas;
4107 /* Combine l_mas and r_mas and split them up evenly again. */
4108 return mas_spanning_rebalance(mas, &mast, height + 1);
4112 * mas_wr_node_store() - Attempt to store the value in a node
4113 * @wr_mas: The maple write state
4115 * Attempts to reuse the node, but may allocate.
4117 * Return: True if stored, false otherwise
4119 static inline bool mas_wr_node_store(struct ma_wr_state *wr_mas)
4121 struct ma_state *mas = wr_mas->mas;
4122 void __rcu **dst_slots;
4123 unsigned long *dst_pivots;
4124 unsigned char dst_offset;
4125 unsigned char new_end = wr_mas->node_end;
4126 unsigned char offset;
4127 unsigned char node_slots = mt_slots[wr_mas->type];
4128 struct maple_node reuse, *newnode;
4129 unsigned char copy_size, max_piv = mt_pivots[wr_mas->type];
4130 bool in_rcu = mt_in_rcu(mas->tree);
4132 offset = mas->offset;
4133 if (mas->last == wr_mas->r_max) {
4134 /* runs right to the end of the node */
4135 if (mas->last == mas->max)
4137 /* don't copy this offset */
4138 wr_mas->offset_end++;
4139 } else if (mas->last < wr_mas->r_max) {
4140 /* new range ends in this range */
4141 if (unlikely(wr_mas->r_max == ULONG_MAX))
4142 mas_bulk_rebalance(mas, wr_mas->node_end, wr_mas->type);
4146 if (wr_mas->end_piv == mas->last)
4147 wr_mas->offset_end++;
4149 new_end -= wr_mas->offset_end - offset - 1;
4152 /* new range starts within a range */
4153 if (wr_mas->r_min < mas->index)
4156 /* Not enough room */
4157 if (new_end >= node_slots)
4160 /* Not enough data. */
4161 if (!mte_is_root(mas->node) && (new_end <= mt_min_slots[wr_mas->type]) &&
4162 !(mas->mas_flags & MA_STATE_BULK))
4167 mas_node_count(mas, 1);
4168 if (mas_is_err(mas))
4171 newnode = mas_pop_node(mas);
4173 memset(&reuse, 0, sizeof(struct maple_node));
4177 newnode->parent = mas_mn(mas)->parent;
4178 dst_pivots = ma_pivots(newnode, wr_mas->type);
4179 dst_slots = ma_slots(newnode, wr_mas->type);
4180 /* Copy from start to insert point */
4181 memcpy(dst_pivots, wr_mas->pivots, sizeof(unsigned long) * (offset + 1));
4182 memcpy(dst_slots, wr_mas->slots, sizeof(void *) * (offset + 1));
4183 dst_offset = offset;
4185 /* Handle insert of new range starting after old range */
4186 if (wr_mas->r_min < mas->index) {
4188 rcu_assign_pointer(dst_slots[dst_offset], wr_mas->content);
4189 dst_pivots[dst_offset++] = mas->index - 1;
4192 /* Store the new entry and range end. */
4193 if (dst_offset < max_piv)
4194 dst_pivots[dst_offset] = mas->last;
4195 mas->offset = dst_offset;
4196 rcu_assign_pointer(dst_slots[dst_offset], wr_mas->entry);
4199 * this range wrote to the end of the node or it overwrote the rest of
4202 if (wr_mas->offset_end > wr_mas->node_end || mas->last >= mas->max) {
4203 new_end = dst_offset;
4208 /* Copy to the end of node if necessary. */
4209 copy_size = wr_mas->node_end - wr_mas->offset_end + 1;
4210 memcpy(dst_slots + dst_offset, wr_mas->slots + wr_mas->offset_end,
4211 sizeof(void *) * copy_size);
4212 if (dst_offset < max_piv) {
4213 if (copy_size > max_piv - dst_offset)
4214 copy_size = max_piv - dst_offset;
4216 memcpy(dst_pivots + dst_offset,
4217 wr_mas->pivots + wr_mas->offset_end,
4218 sizeof(unsigned long) * copy_size);
4221 if ((wr_mas->node_end == node_slots - 1) && (new_end < node_slots - 1))
4222 dst_pivots[new_end] = mas->max;
4225 mas_leaf_set_meta(mas, newnode, dst_pivots, maple_leaf_64, new_end);
4227 mte_set_node_dead(mas->node);
4228 mas->node = mt_mk_node(newnode, wr_mas->type);
4229 mas_replace(mas, false);
4231 memcpy(wr_mas->node, newnode, sizeof(struct maple_node));
4233 trace_ma_write(__func__, mas, 0, wr_mas->entry);
4234 mas_update_gap(mas);
4239 * mas_wr_slot_store: Attempt to store a value in a slot.
4240 * @wr_mas: the maple write state
4242 * Return: True if stored, false otherwise
4244 static inline bool mas_wr_slot_store(struct ma_wr_state *wr_mas)
4246 struct ma_state *mas = wr_mas->mas;
4247 unsigned long lmax; /* Logical max. */
4248 unsigned char offset = mas->offset;
4250 if ((wr_mas->r_max > mas->last) && ((wr_mas->r_min != mas->index) ||
4251 (offset != wr_mas->node_end)))
4254 if (offset == wr_mas->node_end - 1)
4257 lmax = wr_mas->pivots[offset + 1];
4259 /* going to overwrite too many slots. */
4260 if (lmax < mas->last)
4263 if (wr_mas->r_min == mas->index) {
4264 /* overwriting two or more ranges with one. */
4265 if (lmax == mas->last)
4268 /* Overwriting all of offset and a portion of offset + 1. */
4269 rcu_assign_pointer(wr_mas->slots[offset], wr_mas->entry);
4270 wr_mas->pivots[offset] = mas->last;
4274 /* Doesn't end on the next range end. */
4275 if (lmax != mas->last)
4278 /* Overwriting a portion of offset and all of offset + 1 */
4279 if ((offset + 1 < mt_pivots[wr_mas->type]) &&
4280 (wr_mas->entry || wr_mas->pivots[offset + 1]))
4281 wr_mas->pivots[offset + 1] = mas->last;
4283 rcu_assign_pointer(wr_mas->slots[offset + 1], wr_mas->entry);
4284 wr_mas->pivots[offset] = mas->index - 1;
4285 mas->offset++; /* Keep mas accurate. */
4288 trace_ma_write(__func__, mas, 0, wr_mas->entry);
4289 mas_update_gap(mas);
4293 static inline void mas_wr_end_piv(struct ma_wr_state *wr_mas)
4295 while ((wr_mas->mas->last > wr_mas->end_piv) &&
4296 (wr_mas->offset_end < wr_mas->node_end))
4297 wr_mas->end_piv = wr_mas->pivots[++wr_mas->offset_end];
4299 if (wr_mas->mas->last > wr_mas->end_piv)
4300 wr_mas->end_piv = wr_mas->mas->max;
4303 static inline void mas_wr_extend_null(struct ma_wr_state *wr_mas)
4305 struct ma_state *mas = wr_mas->mas;
4307 if (mas->last < wr_mas->end_piv && !wr_mas->slots[wr_mas->offset_end])
4308 mas->last = wr_mas->end_piv;
4310 /* Check next slot(s) if we are overwriting the end */
4311 if ((mas->last == wr_mas->end_piv) &&
4312 (wr_mas->node_end != wr_mas->offset_end) &&
4313 !wr_mas->slots[wr_mas->offset_end + 1]) {
4314 wr_mas->offset_end++;
4315 if (wr_mas->offset_end == wr_mas->node_end)
4316 mas->last = mas->max;
4318 mas->last = wr_mas->pivots[wr_mas->offset_end];
4319 wr_mas->end_piv = mas->last;
4322 if (!wr_mas->content) {
4323 /* If this one is null, the next and prev are not */
4324 mas->index = wr_mas->r_min;
4326 /* Check prev slot if we are overwriting the start */
4327 if (mas->index == wr_mas->r_min && mas->offset &&
4328 !wr_mas->slots[mas->offset - 1]) {
4330 wr_mas->r_min = mas->index =
4331 mas_safe_min(mas, wr_mas->pivots, mas->offset);
4332 wr_mas->r_max = wr_mas->pivots[mas->offset];
4337 static inline bool mas_wr_append(struct ma_wr_state *wr_mas)
4339 unsigned char end = wr_mas->node_end;
4340 unsigned char new_end = end + 1;
4341 struct ma_state *mas = wr_mas->mas;
4342 unsigned char node_pivots = mt_pivots[wr_mas->type];
4344 if ((mas->index != wr_mas->r_min) && (mas->last == wr_mas->r_max)) {
4345 if (new_end < node_pivots)
4346 wr_mas->pivots[new_end] = wr_mas->pivots[end];
4348 if (new_end < node_pivots)
4349 ma_set_meta(wr_mas->node, maple_leaf_64, 0, new_end);
4351 rcu_assign_pointer(wr_mas->slots[new_end], wr_mas->entry);
4352 mas->offset = new_end;
4353 wr_mas->pivots[end] = mas->index - 1;
4358 if ((mas->index == wr_mas->r_min) && (mas->last < wr_mas->r_max)) {
4359 if (new_end < node_pivots)
4360 wr_mas->pivots[new_end] = wr_mas->pivots[end];
4362 rcu_assign_pointer(wr_mas->slots[new_end], wr_mas->content);
4363 if (new_end < node_pivots)
4364 ma_set_meta(wr_mas->node, maple_leaf_64, 0, new_end);
4366 wr_mas->pivots[end] = mas->last;
4367 rcu_assign_pointer(wr_mas->slots[end], wr_mas->entry);
4375 * mas_wr_bnode() - Slow path for a modification.
4376 * @wr_mas: The write maple state
4378 * This is where split, rebalance end up.
4380 static void mas_wr_bnode(struct ma_wr_state *wr_mas)
4382 struct maple_big_node b_node;
4384 trace_ma_write(__func__, wr_mas->mas, 0, wr_mas->entry);
4385 memset(&b_node, 0, sizeof(struct maple_big_node));
4386 mas_store_b_node(wr_mas, &b_node, wr_mas->offset_end);
4387 mas_commit_b_node(wr_mas, &b_node, wr_mas->node_end);
4390 static inline void mas_wr_modify(struct ma_wr_state *wr_mas)
4392 unsigned char node_slots;
4393 unsigned char node_size;
4394 struct ma_state *mas = wr_mas->mas;
4396 /* Direct replacement */
4397 if (wr_mas->r_min == mas->index && wr_mas->r_max == mas->last) {
4398 rcu_assign_pointer(wr_mas->slots[mas->offset], wr_mas->entry);
4399 if (!!wr_mas->entry ^ !!wr_mas->content)
4400 mas_update_gap(mas);
4404 /* Attempt to append */
4405 node_slots = mt_slots[wr_mas->type];
4406 node_size = wr_mas->node_end - wr_mas->offset_end + mas->offset + 2;
4407 if (mas->max == ULONG_MAX)
4410 /* slot and node store will not fit, go to the slow path */
4411 if (unlikely(node_size >= node_slots))
4414 if (wr_mas->entry && (wr_mas->node_end < node_slots - 1) &&
4415 (mas->offset == wr_mas->node_end) && mas_wr_append(wr_mas)) {
4416 if (!wr_mas->content || !wr_mas->entry)
4417 mas_update_gap(mas);
4421 if ((wr_mas->offset_end - mas->offset <= 1) && mas_wr_slot_store(wr_mas))
4423 else if (mas_wr_node_store(wr_mas))
4426 if (mas_is_err(mas))
4430 mas_wr_bnode(wr_mas);
4434 * mas_wr_store_entry() - Internal call to store a value
4435 * @mas: The maple state
4436 * @entry: The entry to store.
4438 * Return: The contents that was stored at the index.
4440 static inline void *mas_wr_store_entry(struct ma_wr_state *wr_mas)
4442 struct ma_state *mas = wr_mas->mas;
4444 wr_mas->content = mas_start(mas);
4445 if (mas_is_none(mas) || mas_is_ptr(mas)) {
4446 mas_store_root(mas, wr_mas->entry);
4447 return wr_mas->content;
4450 if (unlikely(!mas_wr_walk(wr_mas))) {
4451 mas_wr_spanning_store(wr_mas);
4452 return wr_mas->content;
4455 /* At this point, we are at the leaf node that needs to be altered. */
4456 wr_mas->end_piv = wr_mas->r_max;
4457 mas_wr_end_piv(wr_mas);
4460 mas_wr_extend_null(wr_mas);
4462 /* New root for a single pointer */
4463 if (unlikely(!mas->index && mas->last == ULONG_MAX)) {
4464 mas_new_root(mas, wr_mas->entry);
4465 return wr_mas->content;
4468 mas_wr_modify(wr_mas);
4469 return wr_mas->content;
4473 * mas_insert() - Internal call to insert a value
4474 * @mas: The maple state
4475 * @entry: The entry to store
4477 * Return: %NULL or the contents that already exists at the requested index
4478 * otherwise. The maple state needs to be checked for error conditions.
4480 static inline void *mas_insert(struct ma_state *mas, void *entry)
4482 MA_WR_STATE(wr_mas, mas, entry);
4485 * Inserting a new range inserts either 0, 1, or 2 pivots within the
4486 * tree. If the insert fits exactly into an existing gap with a value
4487 * of NULL, then the slot only needs to be written with the new value.
4488 * If the range being inserted is adjacent to another range, then only a
4489 * single pivot needs to be inserted (as well as writing the entry). If
4490 * the new range is within a gap but does not touch any other ranges,
4491 * then two pivots need to be inserted: the start - 1, and the end. As
4492 * usual, the entry must be written. Most operations require a new node
4493 * to be allocated and replace an existing node to ensure RCU safety,
4494 * when in RCU mode. The exception to requiring a newly allocated node
4495 * is when inserting at the end of a node (appending). When done
4496 * carefully, appending can reuse the node in place.
4498 wr_mas.content = mas_start(mas);
4502 if (mas_is_none(mas) || mas_is_ptr(mas)) {
4503 mas_store_root(mas, entry);
4507 /* spanning writes always overwrite something */
4508 if (!mas_wr_walk(&wr_mas))
4511 /* At this point, we are at the leaf node that needs to be altered. */
4512 wr_mas.offset_end = mas->offset;
4513 wr_mas.end_piv = wr_mas.r_max;
4515 if (wr_mas.content || (mas->last > wr_mas.r_max))
4521 mas_wr_modify(&wr_mas);
4522 return wr_mas.content;
4525 mas_set_err(mas, -EEXIST);
4526 return wr_mas.content;
4531 * mas_prev_node() - Find the prev non-null entry at the same level in the
4532 * tree. The prev value will be mas->node[mas->offset] or MAS_NONE.
4533 * @mas: The maple state
4534 * @min: The lower limit to search
4536 * The prev node value will be mas->node[mas->offset] or MAS_NONE.
4537 * Return: 1 if the node is dead, 0 otherwise.
4539 static inline int mas_prev_node(struct ma_state *mas, unsigned long min)
4544 struct maple_node *node;
4545 struct maple_enode *enode;
4546 unsigned long *pivots;
4548 if (mas_is_none(mas))
4554 if (ma_is_root(node))
4558 if (unlikely(mas_ascend(mas)))
4560 offset = mas->offset;
4565 mt = mte_node_type(mas->node);
4567 slots = ma_slots(node, mt);
4568 pivots = ma_pivots(node, mt);
4569 if (unlikely(ma_dead_node(node)))
4572 mas->max = pivots[offset];
4574 mas->min = pivots[offset - 1] + 1;
4575 if (unlikely(ma_dead_node(node)))
4583 enode = mas_slot(mas, slots, offset);
4584 if (unlikely(ma_dead_node(node)))
4588 mt = mte_node_type(mas->node);
4590 slots = ma_slots(node, mt);
4591 pivots = ma_pivots(node, mt);
4592 offset = ma_data_end(node, mt, pivots, mas->max);
4593 if (unlikely(ma_dead_node(node)))
4597 mas->min = pivots[offset - 1] + 1;
4599 if (offset < mt_pivots[mt])
4600 mas->max = pivots[offset];
4606 mas->node = mas_slot(mas, slots, offset);
4607 if (unlikely(ma_dead_node(node)))
4610 mas->offset = mas_data_end(mas);
4611 if (unlikely(mte_dead_node(mas->node)))
4617 mas->offset = offset;
4619 mas->min = pivots[offset - 1] + 1;
4621 if (unlikely(ma_dead_node(node)))
4624 mas->node = MAS_NONE;
4629 * mas_next_node() - Get the next node at the same level in the tree.
4630 * @mas: The maple state
4631 * @max: The maximum pivot value to check.
4633 * The next value will be mas->node[mas->offset] or MAS_NONE.
4634 * Return: 1 on dead node, 0 otherwise.
4636 static inline int mas_next_node(struct ma_state *mas, struct maple_node *node,
4639 unsigned long min, pivot;
4640 unsigned long *pivots;
4641 struct maple_enode *enode;
4643 unsigned char offset;
4644 unsigned char node_end;
4648 if (mas->max >= max)
4653 if (ma_is_root(node))
4660 if (unlikely(mas_ascend(mas)))
4663 offset = mas->offset;
4666 mt = mte_node_type(mas->node);
4667 pivots = ma_pivots(node, mt);
4668 node_end = ma_data_end(node, mt, pivots, mas->max);
4669 if (unlikely(ma_dead_node(node)))
4672 } while (unlikely(offset == node_end));
4674 slots = ma_slots(node, mt);
4675 pivot = mas_safe_pivot(mas, pivots, ++offset, mt);
4676 while (unlikely(level > 1)) {
4677 /* Descend, if necessary */
4678 enode = mas_slot(mas, slots, offset);
4679 if (unlikely(ma_dead_node(node)))
4685 mt = mte_node_type(mas->node);
4686 slots = ma_slots(node, mt);
4687 pivots = ma_pivots(node, mt);
4688 if (unlikely(ma_dead_node(node)))
4695 enode = mas_slot(mas, slots, offset);
4696 if (unlikely(ma_dead_node(node)))
4705 if (unlikely(ma_dead_node(node)))
4708 mas->node = MAS_NONE;
4713 * mas_next_nentry() - Get the next node entry
4714 * @mas: The maple state
4715 * @max: The maximum value to check
4716 * @*range_start: Pointer to store the start of the range.
4718 * Sets @mas->offset to the offset of the next node entry, @mas->last to the
4719 * pivot of the entry.
4721 * Return: The next entry, %NULL otherwise
4723 static inline void *mas_next_nentry(struct ma_state *mas,
4724 struct maple_node *node, unsigned long max, enum maple_type type)
4726 unsigned char count;
4727 unsigned long pivot;
4728 unsigned long *pivots;
4732 if (mas->last == mas->max) {
4733 mas->index = mas->max;
4737 slots = ma_slots(node, type);
4738 pivots = ma_pivots(node, type);
4739 count = ma_data_end(node, type, pivots, mas->max);
4740 if (unlikely(ma_dead_node(node)))
4743 mas->index = mas_safe_min(mas, pivots, mas->offset);
4744 if (unlikely(ma_dead_node(node)))
4747 if (mas->index > max)
4750 if (mas->offset > count)
4753 while (mas->offset < count) {
4754 pivot = pivots[mas->offset];
4755 entry = mas_slot(mas, slots, mas->offset);
4756 if (ma_dead_node(node))
4765 mas->index = pivot + 1;
4769 if (mas->index > mas->max) {
4770 mas->index = mas->last;
4774 pivot = mas_safe_pivot(mas, pivots, mas->offset, type);
4775 entry = mas_slot(mas, slots, mas->offset);
4776 if (ma_dead_node(node))
4790 static inline void mas_rewalk(struct ma_state *mas, unsigned long index)
4793 mas_set(mas, index);
4794 mas_state_walk(mas);
4795 if (mas_is_start(mas))
4800 * mas_next_entry() - Internal function to get the next entry.
4801 * @mas: The maple state
4802 * @limit: The maximum range start.
4804 * Set the @mas->node to the next entry and the range_start to
4805 * the beginning value for the entry. Does not check beyond @limit.
4806 * Sets @mas->index and @mas->last to the limit if it is hit.
4807 * Restarts on dead nodes.
4809 * Return: the next entry or %NULL.
4811 static inline void *mas_next_entry(struct ma_state *mas, unsigned long limit)
4814 struct maple_enode *prev_node;
4815 struct maple_node *node;
4816 unsigned char offset;
4820 if (mas->index > limit) {
4821 mas->index = mas->last = limit;
4827 offset = mas->offset;
4828 prev_node = mas->node;
4830 mt = mte_node_type(mas->node);
4832 if (unlikely(mas->offset >= mt_slots[mt])) {
4833 mas->offset = mt_slots[mt] - 1;
4837 while (!mas_is_none(mas)) {
4838 entry = mas_next_nentry(mas, node, limit, mt);
4839 if (unlikely(ma_dead_node(node))) {
4840 mas_rewalk(mas, last);
4847 if (unlikely((mas->index > limit)))
4851 prev_node = mas->node;
4852 offset = mas->offset;
4853 if (unlikely(mas_next_node(mas, node, limit))) {
4854 mas_rewalk(mas, last);
4859 mt = mte_node_type(mas->node);
4862 mas->index = mas->last = limit;
4863 mas->offset = offset;
4864 mas->node = prev_node;
4869 * mas_prev_nentry() - Get the previous node entry.
4870 * @mas: The maple state.
4871 * @limit: The lower limit to check for a value.
4873 * Return: the entry, %NULL otherwise.
4875 static inline void *mas_prev_nentry(struct ma_state *mas, unsigned long limit,
4876 unsigned long index)
4878 unsigned long pivot, min;
4879 unsigned char offset;
4880 struct maple_node *mn;
4882 unsigned long *pivots;
4891 mt = mte_node_type(mas->node);
4892 offset = mas->offset - 1;
4893 if (offset >= mt_slots[mt])
4894 offset = mt_slots[mt] - 1;
4896 slots = ma_slots(mn, mt);
4897 pivots = ma_pivots(mn, mt);
4898 if (unlikely(ma_dead_node(mn))) {
4899 mas_rewalk(mas, index);
4903 if (offset == mt_pivots[mt])
4906 pivot = pivots[offset];
4908 if (unlikely(ma_dead_node(mn))) {
4909 mas_rewalk(mas, index);
4913 while (offset && ((!mas_slot(mas, slots, offset) && pivot >= limit) ||
4915 pivot = pivots[--offset];
4917 min = mas_safe_min(mas, pivots, offset);
4918 entry = mas_slot(mas, slots, offset);
4919 if (unlikely(ma_dead_node(mn))) {
4920 mas_rewalk(mas, index);
4924 if (likely(entry)) {
4925 mas->offset = offset;
4932 static inline void *mas_prev_entry(struct ma_state *mas, unsigned long min)
4936 if (mas->index < min) {
4937 mas->index = mas->last = min;
4938 mas->node = MAS_NONE;
4942 while (likely(!mas_is_none(mas))) {
4943 entry = mas_prev_nentry(mas, min, mas->index);
4944 if (unlikely(mas->last < min))
4950 if (unlikely(mas_prev_node(mas, min))) {
4951 mas_rewalk(mas, mas->index);
4960 mas->index = mas->last = min;
4965 * mas_rev_awalk() - Internal function. Reverse allocation walk. Find the
4966 * highest gap address of a given size in a given node and descend.
4967 * @mas: The maple state
4968 * @size: The needed size.
4970 * Return: True if found in a leaf, false otherwise.
4973 static bool mas_rev_awalk(struct ma_state *mas, unsigned long size)
4975 enum maple_type type = mte_node_type(mas->node);
4976 struct maple_node *node = mas_mn(mas);
4977 unsigned long *pivots, *gaps;
4979 unsigned long gap = 0;
4980 unsigned long max, min;
4981 unsigned char offset;
4983 if (unlikely(mas_is_err(mas)))
4986 if (ma_is_dense(type)) {
4988 mas->offset = (unsigned char)(mas->index - mas->min);
4992 pivots = ma_pivots(node, type);
4993 slots = ma_slots(node, type);
4994 gaps = ma_gaps(node, type);
4995 offset = mas->offset;
4996 min = mas_safe_min(mas, pivots, offset);
4997 /* Skip out of bounds. */
4998 while (mas->last < min)
4999 min = mas_safe_min(mas, pivots, --offset);
5001 max = mas_safe_pivot(mas, pivots, offset, type);
5002 while (mas->index <= max) {
5006 else if (!mas_slot(mas, slots, offset))
5007 gap = max - min + 1;
5010 if ((size <= gap) && (size <= mas->last - min + 1))
5014 /* Skip the next slot, it cannot be a gap. */
5019 max = pivots[offset];
5020 min = mas_safe_min(mas, pivots, offset);
5030 min = mas_safe_min(mas, pivots, offset);
5033 if (unlikely((mas->index > max) || (size - 1 > max - mas->index)))
5036 if (unlikely(ma_is_leaf(type))) {
5037 mas->offset = offset;
5039 mas->max = min + gap - 1;
5043 /* descend, only happens under lock. */
5044 mas->node = mas_slot(mas, slots, offset);
5047 mas->offset = mas_data_end(mas);
5051 if (!mte_is_root(mas->node))
5055 mas_set_err(mas, -EBUSY);
5059 static inline bool mas_anode_descend(struct ma_state *mas, unsigned long size)
5061 enum maple_type type = mte_node_type(mas->node);
5062 unsigned long pivot, min, gap = 0;
5063 unsigned char offset;
5064 unsigned long *gaps;
5065 unsigned long *pivots = ma_pivots(mas_mn(mas), type);
5066 void __rcu **slots = ma_slots(mas_mn(mas), type);
5069 if (ma_is_dense(type)) {
5070 mas->offset = (unsigned char)(mas->index - mas->min);
5074 gaps = ma_gaps(mte_to_node(mas->node), type);
5075 offset = mas->offset;
5076 min = mas_safe_min(mas, pivots, offset);
5077 for (; offset < mt_slots[type]; offset++) {
5078 pivot = mas_safe_pivot(mas, pivots, offset, type);
5079 if (offset && !pivot)
5082 /* Not within lower bounds */
5083 if (mas->index > pivot)
5088 else if (!mas_slot(mas, slots, offset))
5089 gap = min(pivot, mas->last) - max(mas->index, min) + 1;
5094 if (ma_is_leaf(type)) {
5098 if (mas->index <= pivot) {
5099 mas->node = mas_slot(mas, slots, offset);
5108 if (mas->last <= pivot) {
5109 mas_set_err(mas, -EBUSY);
5114 if (mte_is_root(mas->node))
5117 mas->offset = offset;
5122 * mas_walk() - Search for @mas->index in the tree.
5123 * @mas: The maple state.
5125 * mas->index and mas->last will be set to the range if there is a value. If
5126 * mas->node is MAS_NONE, reset to MAS_START.
5128 * Return: the entry at the location or %NULL.
5130 void *mas_walk(struct ma_state *mas)
5135 entry = mas_state_walk(mas);
5136 if (mas_is_start(mas))
5139 if (mas_is_ptr(mas)) {
5144 mas->last = ULONG_MAX;
5149 if (mas_is_none(mas)) {
5151 mas->last = ULONG_MAX;
5156 EXPORT_SYMBOL_GPL(mas_walk);
5158 static inline bool mas_rewind_node(struct ma_state *mas)
5163 if (mte_is_root(mas->node)) {
5173 mas->offset = --slot;
5178 * mas_skip_node() - Internal function. Skip over a node.
5179 * @mas: The maple state.
5181 * Return: true if there is another node, false otherwise.
5183 static inline bool mas_skip_node(struct ma_state *mas)
5185 if (mas_is_err(mas))
5189 if (mte_is_root(mas->node)) {
5190 if (mas->offset >= mas_data_end(mas)) {
5191 mas_set_err(mas, -EBUSY);
5197 } while (mas->offset >= mas_data_end(mas));
5204 * mas_awalk() - Allocation walk. Search from low address to high, for a gap of
5206 * @mas: The maple state
5207 * @size: The size of the gap required
5209 * Search between @mas->index and @mas->last for a gap of @size.
5211 static inline void mas_awalk(struct ma_state *mas, unsigned long size)
5213 struct maple_enode *last = NULL;
5216 * There are 4 options:
5217 * go to child (descend)
5218 * go back to parent (ascend)
5219 * no gap found. (return, slot == MAPLE_NODE_SLOTS)
5220 * found the gap. (return, slot != MAPLE_NODE_SLOTS)
5222 while (!mas_is_err(mas) && !mas_anode_descend(mas, size)) {
5223 if (last == mas->node)
5231 * mas_fill_gap() - Fill a located gap with @entry.
5232 * @mas: The maple state
5233 * @entry: The value to store
5234 * @slot: The offset into the node to store the @entry
5235 * @size: The size of the entry
5236 * @index: The start location
5238 static inline void mas_fill_gap(struct ma_state *mas, void *entry,
5239 unsigned char slot, unsigned long size, unsigned long *index)
5241 MA_WR_STATE(wr_mas, mas, entry);
5242 unsigned char pslot = mte_parent_slot(mas->node);
5243 struct maple_enode *mn = mas->node;
5244 unsigned long *pivots;
5245 enum maple_type ptype;
5247 * mas->index is the start address for the search
5248 * which may no longer be needed.
5249 * mas->last is the end address for the search
5252 *index = mas->index;
5253 mas->last = mas->index + size - 1;
5256 * It is possible that using mas->max and mas->min to correctly
5257 * calculate the index and last will cause an issue in the gap
5258 * calculation, so fix the ma_state here
5261 ptype = mte_node_type(mas->node);
5262 pivots = ma_pivots(mas_mn(mas), ptype);
5263 mas->max = mas_safe_pivot(mas, pivots, pslot, ptype);
5264 mas->min = mas_safe_min(mas, pivots, pslot);
5267 mas_wr_store_entry(&wr_mas);
5271 * mas_sparse_area() - Internal function. Return upper or lower limit when
5272 * searching for a gap in an empty tree.
5273 * @mas: The maple state
5274 * @min: the minimum range
5275 * @max: The maximum range
5276 * @size: The size of the gap
5277 * @fwd: Searching forward or back
5279 static inline void mas_sparse_area(struct ma_state *mas, unsigned long min,
5280 unsigned long max, unsigned long size, bool fwd)
5282 unsigned long start = 0;
5284 if (!unlikely(mas_is_none(mas)))
5293 mas->last = start + size - 1;
5301 * mas_empty_area() - Get the lowest address within the range that is
5302 * sufficient for the size requested.
5303 * @mas: The maple state
5304 * @min: The lowest value of the range
5305 * @max: The highest value of the range
5306 * @size: The size needed
5308 int mas_empty_area(struct ma_state *mas, unsigned long min,
5309 unsigned long max, unsigned long size)
5311 unsigned char offset;
5312 unsigned long *pivots;
5315 if (mas_is_start(mas))
5317 else if (mas->offset >= 2)
5319 else if (!mas_skip_node(mas))
5323 if (mas_is_none(mas) || mas_is_ptr(mas)) {
5324 mas_sparse_area(mas, min, max, size, true);
5328 /* The start of the window can only be within these values */
5331 mas_awalk(mas, size);
5333 if (unlikely(mas_is_err(mas)))
5334 return xa_err(mas->node);
5336 offset = mas->offset;
5337 if (unlikely(offset == MAPLE_NODE_SLOTS))
5340 mt = mte_node_type(mas->node);
5341 pivots = ma_pivots(mas_mn(mas), mt);
5343 mas->min = pivots[offset - 1] + 1;
5345 if (offset < mt_pivots[mt])
5346 mas->max = pivots[offset];
5348 if (mas->index < mas->min)
5349 mas->index = mas->min;
5351 mas->last = mas->index + size - 1;
5354 EXPORT_SYMBOL_GPL(mas_empty_area);
5357 * mas_empty_area_rev() - Get the highest address within the range that is
5358 * sufficient for the size requested.
5359 * @mas: The maple state
5360 * @min: The lowest value of the range
5361 * @max: The highest value of the range
5362 * @size: The size needed
5364 int mas_empty_area_rev(struct ma_state *mas, unsigned long min,
5365 unsigned long max, unsigned long size)
5367 struct maple_enode *last = mas->node;
5369 if (mas_is_start(mas)) {
5371 mas->offset = mas_data_end(mas);
5372 } else if (mas->offset >= 2) {
5374 } else if (!mas_rewind_node(mas)) {
5379 if (mas_is_none(mas) || mas_is_ptr(mas)) {
5380 mas_sparse_area(mas, min, max, size, false);
5384 /* The start of the window can only be within these values. */
5388 while (!mas_rev_awalk(mas, size)) {
5389 if (last == mas->node) {
5390 if (!mas_rewind_node(mas))
5397 if (mas_is_err(mas))
5398 return xa_err(mas->node);
5400 if (unlikely(mas->offset == MAPLE_NODE_SLOTS))
5404 * mas_rev_awalk() has set mas->min and mas->max to the gap values. If
5405 * the maximum is outside the window we are searching, then use the last
5406 * location in the search.
5407 * mas->max and mas->min is the range of the gap.
5408 * mas->index and mas->last are currently set to the search range.
5411 /* Trim the upper limit to the max. */
5412 if (mas->max <= mas->last)
5413 mas->last = mas->max;
5415 mas->index = mas->last - size + 1;
5418 EXPORT_SYMBOL_GPL(mas_empty_area_rev);
5420 static inline int mas_alloc(struct ma_state *mas, void *entry,
5421 unsigned long size, unsigned long *index)
5426 if (mas_is_none(mas) || mas_is_ptr(mas)) {
5427 mas_root_expand(mas, entry);
5428 if (mas_is_err(mas))
5429 return xa_err(mas->node);
5432 return mte_pivot(mas->node, 0);
5433 return mte_pivot(mas->node, 1);
5436 /* Must be walking a tree. */
5437 mas_awalk(mas, size);
5438 if (mas_is_err(mas))
5439 return xa_err(mas->node);
5441 if (mas->offset == MAPLE_NODE_SLOTS)
5445 * At this point, mas->node points to the right node and we have an
5446 * offset that has a sufficient gap.
5450 min = mte_pivot(mas->node, mas->offset - 1) + 1;
5452 if (mas->index < min)
5455 mas_fill_gap(mas, entry, mas->offset, size, index);
5462 static inline int mas_rev_alloc(struct ma_state *mas, unsigned long min,
5463 unsigned long max, void *entry,
5464 unsigned long size, unsigned long *index)
5468 ret = mas_empty_area_rev(mas, min, max, size);
5472 if (mas_is_err(mas))
5473 return xa_err(mas->node);
5475 if (mas->offset == MAPLE_NODE_SLOTS)
5478 mas_fill_gap(mas, entry, mas->offset, size, index);
5486 * mte_dead_leaves() - Mark all leaves of a node as dead.
5487 * @mas: The maple state
5488 * @slots: Pointer to the slot array
5489 * @type: The maple node type
5491 * Must hold the write lock.
5493 * Return: The number of leaves marked as dead.
5496 unsigned char mte_dead_leaves(struct maple_enode *enode, struct maple_tree *mt,
5499 struct maple_node *node;
5500 enum maple_type type;
5504 for (offset = 0; offset < mt_slot_count(enode); offset++) {
5505 entry = mt_slot(mt, slots, offset);
5506 type = mte_node_type(entry);
5507 node = mte_to_node(entry);
5508 /* Use both node and type to catch LE & BE metadata */
5512 mte_set_node_dead(entry);
5514 rcu_assign_pointer(slots[offset], node);
5521 * mte_dead_walk() - Walk down a dead tree to just before the leaves
5522 * @enode: The maple encoded node
5523 * @offset: The starting offset
5525 * Note: This can only be used from the RCU callback context.
5527 static void __rcu **mte_dead_walk(struct maple_enode **enode, unsigned char offset)
5529 struct maple_node *node, *next;
5530 void __rcu **slots = NULL;
5532 next = mte_to_node(*enode);
5534 *enode = ma_enode_ptr(next);
5535 node = mte_to_node(*enode);
5536 slots = ma_slots(node, node->type);
5537 next = rcu_dereference_protected(slots[offset],
5538 lock_is_held(&rcu_callback_map));
5540 } while (!ma_is_leaf(next->type));
5546 * mt_free_walk() - Walk & free a tree in the RCU callback context
5547 * @head: The RCU head that's within the node.
5549 * Note: This can only be used from the RCU callback context.
5551 static void mt_free_walk(struct rcu_head *head)
5554 struct maple_node *node, *start;
5555 struct maple_enode *enode;
5556 unsigned char offset;
5557 enum maple_type type;
5559 node = container_of(head, struct maple_node, rcu);
5561 if (ma_is_leaf(node->type))
5565 enode = mt_mk_node(node, node->type);
5566 slots = mte_dead_walk(&enode, 0);
5567 node = mte_to_node(enode);
5569 mt_free_bulk(node->slot_len, slots);
5570 offset = node->parent_slot + 1;
5571 enode = node->piv_parent;
5572 if (mte_to_node(enode) == node)
5575 type = mte_node_type(enode);
5576 slots = ma_slots(mte_to_node(enode), type);
5577 if ((offset < mt_slots[type]) &&
5578 rcu_dereference_protected(slots[offset],
5579 lock_is_held(&rcu_callback_map)))
5580 slots = mte_dead_walk(&enode, offset);
5581 node = mte_to_node(enode);
5582 } while ((node != start) || (node->slot_len < offset));
5584 slots = ma_slots(node, node->type);
5585 mt_free_bulk(node->slot_len, slots);
5588 mt_free_rcu(&node->rcu);
5591 static inline void __rcu **mte_destroy_descend(struct maple_enode **enode,
5592 struct maple_tree *mt, struct maple_enode *prev, unsigned char offset)
5594 struct maple_node *node;
5595 struct maple_enode *next = *enode;
5596 void __rcu **slots = NULL;
5597 enum maple_type type;
5598 unsigned char next_offset = 0;
5602 node = mte_to_node(*enode);
5603 type = mte_node_type(*enode);
5604 slots = ma_slots(node, type);
5605 next = mt_slot_locked(mt, slots, next_offset);
5606 if ((mte_dead_node(next)))
5607 next = mt_slot_locked(mt, slots, ++next_offset);
5609 mte_set_node_dead(*enode);
5611 node->piv_parent = prev;
5612 node->parent_slot = offset;
5613 offset = next_offset;
5616 } while (!mte_is_leaf(next));
5621 static void mt_destroy_walk(struct maple_enode *enode, struct maple_tree *mt,
5625 struct maple_node *node = mte_to_node(enode);
5626 struct maple_enode *start;
5628 if (mte_is_leaf(enode)) {
5629 node->type = mte_node_type(enode);
5634 slots = mte_destroy_descend(&enode, mt, start, 0);
5635 node = mte_to_node(enode); // Updated in the above call.
5637 enum maple_type type;
5638 unsigned char offset;
5639 struct maple_enode *parent, *tmp;
5641 node->slot_len = mte_dead_leaves(enode, mt, slots);
5643 mt_free_bulk(node->slot_len, slots);
5644 offset = node->parent_slot + 1;
5645 enode = node->piv_parent;
5646 if (mte_to_node(enode) == node)
5649 type = mte_node_type(enode);
5650 slots = ma_slots(mte_to_node(enode), type);
5651 if (offset >= mt_slots[type])
5654 tmp = mt_slot_locked(mt, slots, offset);
5655 if (mte_node_type(tmp) && mte_to_node(tmp)) {
5658 slots = mte_destroy_descend(&enode, mt, parent, offset);
5661 node = mte_to_node(enode);
5662 } while (start != enode);
5664 node = mte_to_node(enode);
5665 node->slot_len = mte_dead_leaves(enode, mt, slots);
5667 mt_free_bulk(node->slot_len, slots);
5671 mt_free_rcu(&node->rcu);
5673 mt_clear_meta(mt, node, node->type);
5677 * mte_destroy_walk() - Free a tree or sub-tree.
5678 * @enode: the encoded maple node (maple_enode) to start
5679 * @mt: the tree to free - needed for node types.
5681 * Must hold the write lock.
5683 static inline void mte_destroy_walk(struct maple_enode *enode,
5684 struct maple_tree *mt)
5686 struct maple_node *node = mte_to_node(enode);
5688 if (mt_in_rcu(mt)) {
5689 mt_destroy_walk(enode, mt, false);
5690 call_rcu(&node->rcu, mt_free_walk);
5692 mt_destroy_walk(enode, mt, true);
5696 static void mas_wr_store_setup(struct ma_wr_state *wr_mas)
5698 if (unlikely(mas_is_paused(wr_mas->mas)))
5699 mas_reset(wr_mas->mas);
5701 if (!mas_is_start(wr_mas->mas)) {
5702 if (mas_is_none(wr_mas->mas)) {
5703 mas_reset(wr_mas->mas);
5705 wr_mas->r_max = wr_mas->mas->max;
5706 wr_mas->type = mte_node_type(wr_mas->mas->node);
5707 if (mas_is_span_wr(wr_mas))
5708 mas_reset(wr_mas->mas);
5716 * mas_store() - Store an @entry.
5717 * @mas: The maple state.
5718 * @entry: The entry to store.
5720 * The @mas->index and @mas->last is used to set the range for the @entry.
5721 * Note: The @mas should have pre-allocated entries to ensure there is memory to
5722 * store the entry. Please see mas_expected_entries()/mas_destroy() for more details.
5724 * Return: the first entry between mas->index and mas->last or %NULL.
5726 void *mas_store(struct ma_state *mas, void *entry)
5728 MA_WR_STATE(wr_mas, mas, entry);
5730 trace_ma_write(__func__, mas, 0, entry);
5731 #ifdef CONFIG_DEBUG_MAPLE_TREE
5732 if (mas->index > mas->last)
5733 pr_err("Error %lu > %lu %p\n", mas->index, mas->last, entry);
5734 MT_BUG_ON(mas->tree, mas->index > mas->last);
5735 if (mas->index > mas->last) {
5736 mas_set_err(mas, -EINVAL);
5743 * Storing is the same operation as insert with the added caveat that it
5744 * can overwrite entries. Although this seems simple enough, one may
5745 * want to examine what happens if a single store operation was to
5746 * overwrite multiple entries within a self-balancing B-Tree.
5748 mas_wr_store_setup(&wr_mas);
5749 mas_wr_store_entry(&wr_mas);
5750 return wr_mas.content;
5752 EXPORT_SYMBOL_GPL(mas_store);
5755 * mas_store_gfp() - Store a value into the tree.
5756 * @mas: The maple state
5757 * @entry: The entry to store
5758 * @gfp: The GFP_FLAGS to use for allocations if necessary.
5760 * Return: 0 on success, -EINVAL on invalid request, -ENOMEM if memory could not
5763 int mas_store_gfp(struct ma_state *mas, void *entry, gfp_t gfp)
5765 MA_WR_STATE(wr_mas, mas, entry);
5767 mas_wr_store_setup(&wr_mas);
5768 trace_ma_write(__func__, mas, 0, entry);
5770 mas_wr_store_entry(&wr_mas);
5771 if (unlikely(mas_nomem(mas, gfp)))
5774 if (unlikely(mas_is_err(mas)))
5775 return xa_err(mas->node);
5779 EXPORT_SYMBOL_GPL(mas_store_gfp);
5782 * mas_store_prealloc() - Store a value into the tree using memory
5783 * preallocated in the maple state.
5784 * @mas: The maple state
5785 * @entry: The entry to store.
5787 void mas_store_prealloc(struct ma_state *mas, void *entry)
5789 MA_WR_STATE(wr_mas, mas, entry);
5791 mas_wr_store_setup(&wr_mas);
5792 trace_ma_write(__func__, mas, 0, entry);
5793 mas_wr_store_entry(&wr_mas);
5794 BUG_ON(mas_is_err(mas));
5797 EXPORT_SYMBOL_GPL(mas_store_prealloc);
5800 * mas_preallocate() - Preallocate enough nodes for a store operation
5801 * @mas: The maple state
5802 * @gfp: The GFP_FLAGS to use for allocations.
5804 * Return: 0 on success, -ENOMEM if memory could not be allocated.
5806 int mas_preallocate(struct ma_state *mas, gfp_t gfp)
5810 mas_node_count_gfp(mas, 1 + mas_mt_height(mas) * 3, gfp);
5811 mas->mas_flags |= MA_STATE_PREALLOC;
5812 if (likely(!mas_is_err(mas)))
5815 mas_set_alloc_req(mas, 0);
5816 ret = xa_err(mas->node);
5824 * mas_destroy() - destroy a maple state.
5825 * @mas: The maple state
5827 * Upon completion, check the left-most node and rebalance against the node to
5828 * the right if necessary. Frees any allocated nodes associated with this maple
5831 void mas_destroy(struct ma_state *mas)
5833 struct maple_alloc *node;
5834 unsigned long total;
5837 * When using mas_for_each() to insert an expected number of elements,
5838 * it is possible that the number inserted is less than the expected
5839 * number. To fix an invalid final node, a check is performed here to
5840 * rebalance the previous node with the final node.
5842 if (mas->mas_flags & MA_STATE_REBALANCE) {
5845 if (mas_is_start(mas))
5848 mtree_range_walk(mas);
5849 end = mas_data_end(mas) + 1;
5850 if (end < mt_min_slot_count(mas->node) - 1)
5851 mas_destroy_rebalance(mas, end);
5853 mas->mas_flags &= ~MA_STATE_REBALANCE;
5855 mas->mas_flags &= ~(MA_STATE_BULK|MA_STATE_PREALLOC);
5857 total = mas_allocated(mas);
5860 mas->alloc = node->slot[0];
5861 if (node->node_count > 1) {
5862 size_t count = node->node_count - 1;
5864 mt_free_bulk(count, (void __rcu **)&node->slot[1]);
5867 kmem_cache_free(maple_node_cache, node);
5873 EXPORT_SYMBOL_GPL(mas_destroy);
5876 * mas_expected_entries() - Set the expected number of entries that will be inserted.
5877 * @mas: The maple state
5878 * @nr_entries: The number of expected entries.
5880 * This will attempt to pre-allocate enough nodes to store the expected number
5881 * of entries. The allocations will occur using the bulk allocator interface
5882 * for speed. Please call mas_destroy() on the @mas after inserting the entries
5883 * to ensure any unused nodes are freed.
5885 * Return: 0 on success, -ENOMEM if memory could not be allocated.
5887 int mas_expected_entries(struct ma_state *mas, unsigned long nr_entries)
5889 int nonleaf_cap = MAPLE_ARANGE64_SLOTS - 2;
5890 struct maple_enode *enode = mas->node;
5895 * Sometimes it is necessary to duplicate a tree to a new tree, such as
5896 * forking a process and duplicating the VMAs from one tree to a new
5897 * tree. When such a situation arises, it is known that the new tree is
5898 * not going to be used until the entire tree is populated. For
5899 * performance reasons, it is best to use a bulk load with RCU disabled.
5900 * This allows for optimistic splitting that favours the left and reuse
5901 * of nodes during the operation.
5904 /* Optimize splitting for bulk insert in-order */
5905 mas->mas_flags |= MA_STATE_BULK;
5908 * Avoid overflow, assume a gap between each entry and a trailing null.
5909 * If this is wrong, it just means allocation can happen during
5910 * insertion of entries.
5912 nr_nodes = max(nr_entries, nr_entries * 2 + 1);
5913 if (!mt_is_alloc(mas->tree))
5914 nonleaf_cap = MAPLE_RANGE64_SLOTS - 2;
5916 /* Leaves; reduce slots to keep space for expansion */
5917 nr_nodes = DIV_ROUND_UP(nr_nodes, MAPLE_RANGE64_SLOTS - 2);
5918 /* Internal nodes */
5919 nr_nodes += DIV_ROUND_UP(nr_nodes, nonleaf_cap);
5920 /* Add working room for split (2 nodes) + new parents */
5921 mas_node_count(mas, nr_nodes + 3);
5923 /* Detect if allocations run out */
5924 mas->mas_flags |= MA_STATE_PREALLOC;
5926 if (!mas_is_err(mas))
5929 ret = xa_err(mas->node);
5935 EXPORT_SYMBOL_GPL(mas_expected_entries);
5938 * mas_next() - Get the next entry.
5939 * @mas: The maple state
5940 * @max: The maximum index to check.
5942 * Returns the next entry after @mas->index.
5943 * Must hold rcu_read_lock or the write lock.
5944 * Can return the zero entry.
5946 * Return: The next entry or %NULL
5948 void *mas_next(struct ma_state *mas, unsigned long max)
5950 if (mas_is_none(mas) || mas_is_paused(mas))
5951 mas->node = MAS_START;
5953 if (mas_is_start(mas))
5954 mas_walk(mas); /* Retries on dead nodes handled by mas_walk */
5956 if (mas_is_ptr(mas)) {
5959 mas->last = ULONG_MAX;
5964 if (mas->last == ULONG_MAX)
5967 /* Retries on dead nodes handled by mas_next_entry */
5968 return mas_next_entry(mas, max);
5970 EXPORT_SYMBOL_GPL(mas_next);
5973 * mt_next() - get the next value in the maple tree
5974 * @mt: The maple tree
5975 * @index: The start index
5976 * @max: The maximum index to check
5978 * Return: The entry at @index or higher, or %NULL if nothing is found.
5980 void *mt_next(struct maple_tree *mt, unsigned long index, unsigned long max)
5983 MA_STATE(mas, mt, index, index);
5986 entry = mas_next(&mas, max);
5990 EXPORT_SYMBOL_GPL(mt_next);
5993 * mas_prev() - Get the previous entry
5994 * @mas: The maple state
5995 * @min: The minimum value to check.
5997 * Must hold rcu_read_lock or the write lock.
5998 * Will reset mas to MAS_START if the node is MAS_NONE. Will stop on not
6001 * Return: the previous value or %NULL.
6003 void *mas_prev(struct ma_state *mas, unsigned long min)
6006 /* Nothing comes before 0 */
6008 mas->node = MAS_NONE;
6012 if (unlikely(mas_is_ptr(mas)))
6015 if (mas_is_none(mas) || mas_is_paused(mas))
6016 mas->node = MAS_START;
6018 if (mas_is_start(mas)) {
6024 if (mas_is_ptr(mas)) {
6030 mas->index = mas->last = 0;
6031 return mas_root_locked(mas);
6033 return mas_prev_entry(mas, min);
6035 EXPORT_SYMBOL_GPL(mas_prev);
6038 * mt_prev() - get the previous value in the maple tree
6039 * @mt: The maple tree
6040 * @index: The start index
6041 * @min: The minimum index to check
6043 * Return: The entry at @index or lower, or %NULL if nothing is found.
6045 void *mt_prev(struct maple_tree *mt, unsigned long index, unsigned long min)
6048 MA_STATE(mas, mt, index, index);
6051 entry = mas_prev(&mas, min);
6055 EXPORT_SYMBOL_GPL(mt_prev);
6058 * mas_pause() - Pause a mas_find/mas_for_each to drop the lock.
6059 * @mas: The maple state to pause
6061 * Some users need to pause a walk and drop the lock they're holding in
6062 * order to yield to a higher priority thread or carry out an operation
6063 * on an entry. Those users should call this function before they drop
6064 * the lock. It resets the @mas to be suitable for the next iteration
6065 * of the loop after the user has reacquired the lock. If most entries
6066 * found during a walk require you to call mas_pause(), the mt_for_each()
6067 * iterator may be more appropriate.
6070 void mas_pause(struct ma_state *mas)
6072 mas->node = MAS_PAUSE;
6074 EXPORT_SYMBOL_GPL(mas_pause);
6077 * mas_find() - On the first call, find the entry at or after mas->index up to
6078 * %max. Otherwise, find the entry after mas->index.
6079 * @mas: The maple state
6080 * @max: The maximum value to check.
6082 * Must hold rcu_read_lock or the write lock.
6083 * If an entry exists, last and index are updated accordingly.
6084 * May set @mas->node to MAS_NONE.
6086 * Return: The entry or %NULL.
6088 void *mas_find(struct ma_state *mas, unsigned long max)
6090 if (unlikely(mas_is_paused(mas))) {
6091 if (unlikely(mas->last == ULONG_MAX)) {
6092 mas->node = MAS_NONE;
6095 mas->node = MAS_START;
6096 mas->index = ++mas->last;
6099 if (unlikely(mas_is_none(mas)))
6100 mas->node = MAS_START;
6102 if (unlikely(mas_is_start(mas))) {
6103 /* First run or continue */
6106 if (mas->index > max)
6109 entry = mas_walk(mas);
6114 if (unlikely(!mas_searchable(mas)))
6117 /* Retries on dead nodes handled by mas_next_entry */
6118 return mas_next_entry(mas, max);
6120 EXPORT_SYMBOL_GPL(mas_find);
6123 * mas_find_rev: On the first call, find the first non-null entry at or below
6124 * mas->index down to %min. Otherwise find the first non-null entry below
6125 * mas->index down to %min.
6126 * @mas: The maple state
6127 * @min: The minimum value to check.
6129 * Must hold rcu_read_lock or the write lock.
6130 * If an entry exists, last and index are updated accordingly.
6131 * May set @mas->node to MAS_NONE.
6133 * Return: The entry or %NULL.
6135 void *mas_find_rev(struct ma_state *mas, unsigned long min)
6137 if (unlikely(mas_is_paused(mas))) {
6138 if (unlikely(mas->last == ULONG_MAX)) {
6139 mas->node = MAS_NONE;
6142 mas->node = MAS_START;
6143 mas->last = --mas->index;
6146 if (unlikely(mas_is_start(mas))) {
6147 /* First run or continue */
6150 if (mas->index < min)
6153 entry = mas_walk(mas);
6158 if (unlikely(!mas_searchable(mas)))
6161 if (mas->index < min)
6164 /* Retries on dead nodes handled by mas_prev_entry */
6165 return mas_prev_entry(mas, min);
6167 EXPORT_SYMBOL_GPL(mas_find_rev);
6170 * mas_erase() - Find the range in which index resides and erase the entire
6172 * @mas: The maple state
6174 * Must hold the write lock.
6175 * Searches for @mas->index, sets @mas->index and @mas->last to the range and
6176 * erases that range.
6178 * Return: the entry that was erased or %NULL, @mas->index and @mas->last are updated.
6180 void *mas_erase(struct ma_state *mas)
6183 MA_WR_STATE(wr_mas, mas, NULL);
6185 if (mas_is_none(mas) || mas_is_paused(mas))
6186 mas->node = MAS_START;
6188 /* Retry unnecessary when holding the write lock. */
6189 entry = mas_state_walk(mas);
6194 /* Must reset to ensure spanning writes of last slot are detected */
6196 mas_wr_store_setup(&wr_mas);
6197 mas_wr_store_entry(&wr_mas);
6198 if (mas_nomem(mas, GFP_KERNEL))
6203 EXPORT_SYMBOL_GPL(mas_erase);
6206 * mas_nomem() - Check if there was an error allocating and do the allocation
6207 * if necessary If there are allocations, then free them.
6208 * @mas: The maple state
6209 * @gfp: The GFP_FLAGS to use for allocations
6210 * Return: true on allocation, false otherwise.
6212 bool mas_nomem(struct ma_state *mas, gfp_t gfp)
6213 __must_hold(mas->tree->lock)
6215 if (likely(mas->node != MA_ERROR(-ENOMEM))) {
6220 if (gfpflags_allow_blocking(gfp) && !mt_external_lock(mas->tree)) {
6221 mtree_unlock(mas->tree);
6222 mas_alloc_nodes(mas, gfp);
6223 mtree_lock(mas->tree);
6225 mas_alloc_nodes(mas, gfp);
6228 if (!mas_allocated(mas))
6231 mas->node = MAS_START;
6235 void __init maple_tree_init(void)
6237 maple_node_cache = kmem_cache_create("maple_node",
6238 sizeof(struct maple_node), sizeof(struct maple_node),
6243 * mtree_load() - Load a value stored in a maple tree
6244 * @mt: The maple tree
6245 * @index: The index to load
6247 * Return: the entry or %NULL
6249 void *mtree_load(struct maple_tree *mt, unsigned long index)
6251 MA_STATE(mas, mt, index, index);
6254 trace_ma_read(__func__, &mas);
6257 entry = mas_start(&mas);
6258 if (unlikely(mas_is_none(&mas)))
6261 if (unlikely(mas_is_ptr(&mas))) {
6268 entry = mtree_lookup_walk(&mas);
6269 if (!entry && unlikely(mas_is_start(&mas)))
6273 if (xa_is_zero(entry))
6278 EXPORT_SYMBOL(mtree_load);
6281 * mtree_store_range() - Store an entry at a given range.
6282 * @mt: The maple tree
6283 * @index: The start of the range
6284 * @last: The end of the range
6285 * @entry: The entry to store
6286 * @gfp: The GFP_FLAGS to use for allocations
6288 * Return: 0 on success, -EINVAL on invalid request, -ENOMEM if memory could not
6291 int mtree_store_range(struct maple_tree *mt, unsigned long index,
6292 unsigned long last, void *entry, gfp_t gfp)
6294 MA_STATE(mas, mt, index, last);
6295 MA_WR_STATE(wr_mas, &mas, entry);
6297 trace_ma_write(__func__, &mas, 0, entry);
6298 if (WARN_ON_ONCE(xa_is_advanced(entry)))
6306 mas_wr_store_entry(&wr_mas);
6307 if (mas_nomem(&mas, gfp))
6311 if (mas_is_err(&mas))
6312 return xa_err(mas.node);
6316 EXPORT_SYMBOL(mtree_store_range);
6319 * mtree_store() - Store an entry at a given index.
6320 * @mt: The maple tree
6321 * @index: The index to store the value
6322 * @entry: The entry to store
6323 * @gfp: The GFP_FLAGS to use for allocations
6325 * Return: 0 on success, -EINVAL on invalid request, -ENOMEM if memory could not
6328 int mtree_store(struct maple_tree *mt, unsigned long index, void *entry,
6331 return mtree_store_range(mt, index, index, entry, gfp);
6333 EXPORT_SYMBOL(mtree_store);
6336 * mtree_insert_range() - Insert an entry at a give range if there is no value.
6337 * @mt: The maple tree
6338 * @first: The start of the range
6339 * @last: The end of the range
6340 * @entry: The entry to store
6341 * @gfp: The GFP_FLAGS to use for allocations.
6343 * Return: 0 on success, -EEXISTS if the range is occupied, -EINVAL on invalid
6344 * request, -ENOMEM if memory could not be allocated.
6346 int mtree_insert_range(struct maple_tree *mt, unsigned long first,
6347 unsigned long last, void *entry, gfp_t gfp)
6349 MA_STATE(ms, mt, first, last);
6351 if (WARN_ON_ONCE(xa_is_advanced(entry)))
6359 mas_insert(&ms, entry);
6360 if (mas_nomem(&ms, gfp))
6364 if (mas_is_err(&ms))
6365 return xa_err(ms.node);
6369 EXPORT_SYMBOL(mtree_insert_range);
6372 * mtree_insert() - Insert an entry at a give index if there is no value.
6373 * @mt: The maple tree
6374 * @index : The index to store the value
6375 * @entry: The entry to store
6376 * @gfp: The FGP_FLAGS to use for allocations.
6378 * Return: 0 on success, -EEXISTS if the range is occupied, -EINVAL on invalid
6379 * request, -ENOMEM if memory could not be allocated.
6381 int mtree_insert(struct maple_tree *mt, unsigned long index, void *entry,
6384 return mtree_insert_range(mt, index, index, entry, gfp);
6386 EXPORT_SYMBOL(mtree_insert);
6388 int mtree_alloc_range(struct maple_tree *mt, unsigned long *startp,
6389 void *entry, unsigned long size, unsigned long min,
6390 unsigned long max, gfp_t gfp)
6394 MA_STATE(mas, mt, min, max - size);
6395 if (!mt_is_alloc(mt))
6398 if (WARN_ON_ONCE(mt_is_reserved(entry)))
6414 mas.last = max - size;
6415 ret = mas_alloc(&mas, entry, size, startp);
6416 if (mas_nomem(&mas, gfp))
6422 EXPORT_SYMBOL(mtree_alloc_range);
6424 int mtree_alloc_rrange(struct maple_tree *mt, unsigned long *startp,
6425 void *entry, unsigned long size, unsigned long min,
6426 unsigned long max, gfp_t gfp)
6430 MA_STATE(mas, mt, min, max - size);
6431 if (!mt_is_alloc(mt))
6434 if (WARN_ON_ONCE(mt_is_reserved(entry)))
6448 ret = mas_rev_alloc(&mas, min, max, entry, size, startp);
6449 if (mas_nomem(&mas, gfp))
6455 EXPORT_SYMBOL(mtree_alloc_rrange);
6458 * mtree_erase() - Find an index and erase the entire range.
6459 * @mt: The maple tree
6460 * @index: The index to erase
6462 * Erasing is the same as a walk to an entry then a store of a NULL to that
6463 * ENTIRE range. In fact, it is implemented as such using the advanced API.
6465 * Return: The entry stored at the @index or %NULL
6467 void *mtree_erase(struct maple_tree *mt, unsigned long index)
6471 MA_STATE(mas, mt, index, index);
6472 trace_ma_op(__func__, &mas);
6475 entry = mas_erase(&mas);
6480 EXPORT_SYMBOL(mtree_erase);
6483 * __mt_destroy() - Walk and free all nodes of a locked maple tree.
6484 * @mt: The maple tree
6486 * Note: Does not handle locking.
6488 void __mt_destroy(struct maple_tree *mt)
6490 void *root = mt_root_locked(mt);
6492 rcu_assign_pointer(mt->ma_root, NULL);
6493 if (xa_is_node(root))
6494 mte_destroy_walk(root, mt);
6498 EXPORT_SYMBOL_GPL(__mt_destroy);
6501 * mtree_destroy() - Destroy a maple tree
6502 * @mt: The maple tree
6504 * Frees all resources used by the tree. Handles locking.
6506 void mtree_destroy(struct maple_tree *mt)
6512 EXPORT_SYMBOL(mtree_destroy);
6515 * mt_find() - Search from the start up until an entry is found.
6516 * @mt: The maple tree
6517 * @index: Pointer which contains the start location of the search
6518 * @max: The maximum value to check
6520 * Handles locking. @index will be incremented to one beyond the range.
6522 * Return: The entry at or after the @index or %NULL
6524 void *mt_find(struct maple_tree *mt, unsigned long *index, unsigned long max)
6526 MA_STATE(mas, mt, *index, *index);
6528 #ifdef CONFIG_DEBUG_MAPLE_TREE
6529 unsigned long copy = *index;
6532 trace_ma_read(__func__, &mas);
6539 entry = mas_state_walk(&mas);
6540 if (mas_is_start(&mas))
6543 if (unlikely(xa_is_zero(entry)))
6549 while (mas_searchable(&mas) && (mas.index < max)) {
6550 entry = mas_next_entry(&mas, max);
6551 if (likely(entry && !xa_is_zero(entry)))
6555 if (unlikely(xa_is_zero(entry)))
6559 if (likely(entry)) {
6560 *index = mas.last + 1;
6561 #ifdef CONFIG_DEBUG_MAPLE_TREE
6562 if ((*index) && (*index) <= copy)
6563 pr_err("index not increased! %lx <= %lx\n",
6565 MT_BUG_ON(mt, (*index) && ((*index) <= copy));
6571 EXPORT_SYMBOL(mt_find);
6574 * mt_find_after() - Search from the start up until an entry is found.
6575 * @mt: The maple tree
6576 * @index: Pointer which contains the start location of the search
6577 * @max: The maximum value to check
6579 * Handles locking, detects wrapping on index == 0
6581 * Return: The entry at or after the @index or %NULL
6583 void *mt_find_after(struct maple_tree *mt, unsigned long *index,
6589 return mt_find(mt, index, max);
6591 EXPORT_SYMBOL(mt_find_after);
6593 #ifdef CONFIG_DEBUG_MAPLE_TREE
6594 atomic_t maple_tree_tests_run;
6595 EXPORT_SYMBOL_GPL(maple_tree_tests_run);
6596 atomic_t maple_tree_tests_passed;
6597 EXPORT_SYMBOL_GPL(maple_tree_tests_passed);
6600 extern void kmem_cache_set_non_kernel(struct kmem_cache *, unsigned int);
6601 void mt_set_non_kernel(unsigned int val)
6603 kmem_cache_set_non_kernel(maple_node_cache, val);
6606 extern unsigned long kmem_cache_get_alloc(struct kmem_cache *);
6607 unsigned long mt_get_alloc_size(void)
6609 return kmem_cache_get_alloc(maple_node_cache);
6612 extern void kmem_cache_zero_nr_tallocated(struct kmem_cache *);
6613 void mt_zero_nr_tallocated(void)
6615 kmem_cache_zero_nr_tallocated(maple_node_cache);
6618 extern unsigned int kmem_cache_nr_tallocated(struct kmem_cache *);
6619 unsigned int mt_nr_tallocated(void)
6621 return kmem_cache_nr_tallocated(maple_node_cache);
6624 extern unsigned int kmem_cache_nr_allocated(struct kmem_cache *);
6625 unsigned int mt_nr_allocated(void)
6627 return kmem_cache_nr_allocated(maple_node_cache);
6631 * mas_dead_node() - Check if the maple state is pointing to a dead node.
6632 * @mas: The maple state
6633 * @index: The index to restore in @mas.
6635 * Used in test code.
6636 * Return: 1 if @mas has been reset to MAS_START, 0 otherwise.
6638 static inline int mas_dead_node(struct ma_state *mas, unsigned long index)
6640 if (unlikely(!mas_searchable(mas) || mas_is_start(mas)))
6643 if (likely(!mte_dead_node(mas->node)))
6646 mas_rewalk(mas, index);
6650 void mt_cache_shrink(void)
6655 * mt_cache_shrink() - For testing, don't use this.
6657 * Certain testcases can trigger an OOM when combined with other memory
6658 * debugging configuration options. This function is used to reduce the
6659 * possibility of an out of memory even due to kmem_cache objects remaining
6660 * around for longer than usual.
6662 void mt_cache_shrink(void)
6664 kmem_cache_shrink(maple_node_cache);
6667 EXPORT_SYMBOL_GPL(mt_cache_shrink);
6669 #endif /* not defined __KERNEL__ */
6671 * mas_get_slot() - Get the entry in the maple state node stored at @offset.
6672 * @mas: The maple state
6673 * @offset: The offset into the slot array to fetch.
6675 * Return: The entry stored at @offset.
6677 static inline struct maple_enode *mas_get_slot(struct ma_state *mas,
6678 unsigned char offset)
6680 return mas_slot(mas, ma_slots(mas_mn(mas), mte_node_type(mas->node)),
6686 * mas_first_entry() - Go the first leaf and find the first entry.
6687 * @mas: the maple state.
6688 * @limit: the maximum index to check.
6689 * @*r_start: Pointer to set to the range start.
6691 * Sets mas->offset to the offset of the entry, r_start to the range minimum.
6693 * Return: The first entry or MAS_NONE.
6695 static inline void *mas_first_entry(struct ma_state *mas, struct maple_node *mn,
6696 unsigned long limit, enum maple_type mt)
6700 unsigned long *pivots;
6704 mas->index = mas->min;
6705 if (mas->index > limit)
6710 while (likely(!ma_is_leaf(mt))) {
6711 MT_BUG_ON(mas->tree, mte_dead_node(mas->node));
6712 slots = ma_slots(mn, mt);
6713 entry = mas_slot(mas, slots, 0);
6714 pivots = ma_pivots(mn, mt);
6715 if (unlikely(ma_dead_node(mn)))
6720 mt = mte_node_type(mas->node);
6722 MT_BUG_ON(mas->tree, mte_dead_node(mas->node));
6725 slots = ma_slots(mn, mt);
6726 entry = mas_slot(mas, slots, 0);
6727 if (unlikely(ma_dead_node(mn)))
6730 /* Slot 0 or 1 must be set */
6731 if (mas->index > limit)
6738 entry = mas_slot(mas, slots, 1);
6739 pivots = ma_pivots(mn, mt);
6740 if (unlikely(ma_dead_node(mn)))
6743 mas->index = pivots[0] + 1;
6744 if (mas->index > limit)
6751 if (likely(!ma_dead_node(mn)))
6752 mas->node = MAS_NONE;
6756 /* Depth first search, post-order */
6757 static void mas_dfs_postorder(struct ma_state *mas, unsigned long max)
6760 struct maple_enode *p = MAS_NONE, *mn = mas->node;
6761 unsigned long p_min, p_max;
6763 mas_next_node(mas, mas_mn(mas), max);
6764 if (!mas_is_none(mas))
6767 if (mte_is_root(mn))
6772 while (mas->node != MAS_NONE) {
6776 mas_prev_node(mas, 0);
6787 /* Tree validations */
6788 static void mt_dump_node(const struct maple_tree *mt, void *entry,
6789 unsigned long min, unsigned long max, unsigned int depth);
6790 static void mt_dump_range(unsigned long min, unsigned long max,
6793 static const char spaces[] = " ";
6796 pr_info("%.*s%lu: ", depth * 2, spaces, min);
6798 pr_info("%.*s%lu-%lu: ", depth * 2, spaces, min, max);
6801 static void mt_dump_entry(void *entry, unsigned long min, unsigned long max,
6804 mt_dump_range(min, max, depth);
6806 if (xa_is_value(entry))
6807 pr_cont("value %ld (0x%lx) [%p]\n", xa_to_value(entry),
6808 xa_to_value(entry), entry);
6809 else if (xa_is_zero(entry))
6810 pr_cont("zero (%ld)\n", xa_to_internal(entry));
6811 else if (mt_is_reserved(entry))
6812 pr_cont("UNKNOWN ENTRY (%p)\n", entry);
6814 pr_cont("%p\n", entry);
6817 static void mt_dump_range64(const struct maple_tree *mt, void *entry,
6818 unsigned long min, unsigned long max, unsigned int depth)
6820 struct maple_range_64 *node = &mte_to_node(entry)->mr64;
6821 bool leaf = mte_is_leaf(entry);
6822 unsigned long first = min;
6825 pr_cont(" contents: ");
6826 for (i = 0; i < MAPLE_RANGE64_SLOTS - 1; i++)
6827 pr_cont("%p %lu ", node->slot[i], node->pivot[i]);
6828 pr_cont("%p\n", node->slot[i]);
6829 for (i = 0; i < MAPLE_RANGE64_SLOTS; i++) {
6830 unsigned long last = max;
6832 if (i < (MAPLE_RANGE64_SLOTS - 1))
6833 last = node->pivot[i];
6834 else if (!node->slot[i] && max != mt_node_max(entry))
6836 if (last == 0 && i > 0)
6839 mt_dump_entry(mt_slot(mt, node->slot, i),
6840 first, last, depth + 1);
6841 else if (node->slot[i])
6842 mt_dump_node(mt, mt_slot(mt, node->slot, i),
6843 first, last, depth + 1);
6848 pr_err("node %p last (%lu) > max (%lu) at pivot %d!\n",
6849 node, last, max, i);
6856 static void mt_dump_arange64(const struct maple_tree *mt, void *entry,
6857 unsigned long min, unsigned long max, unsigned int depth)
6859 struct maple_arange_64 *node = &mte_to_node(entry)->ma64;
6860 bool leaf = mte_is_leaf(entry);
6861 unsigned long first = min;
6864 pr_cont(" contents: ");
6865 for (i = 0; i < MAPLE_ARANGE64_SLOTS; i++)
6866 pr_cont("%lu ", node->gap[i]);
6867 pr_cont("| %02X %02X| ", node->meta.end, node->meta.gap);
6868 for (i = 0; i < MAPLE_ARANGE64_SLOTS - 1; i++)
6869 pr_cont("%p %lu ", node->slot[i], node->pivot[i]);
6870 pr_cont("%p\n", node->slot[i]);
6871 for (i = 0; i < MAPLE_ARANGE64_SLOTS; i++) {
6872 unsigned long last = max;
6874 if (i < (MAPLE_ARANGE64_SLOTS - 1))
6875 last = node->pivot[i];
6876 else if (!node->slot[i])
6878 if (last == 0 && i > 0)
6881 mt_dump_entry(mt_slot(mt, node->slot, i),
6882 first, last, depth + 1);
6883 else if (node->slot[i])
6884 mt_dump_node(mt, mt_slot(mt, node->slot, i),
6885 first, last, depth + 1);
6890 pr_err("node %p last (%lu) > max (%lu) at pivot %d!\n",
6891 node, last, max, i);
6898 static void mt_dump_node(const struct maple_tree *mt, void *entry,
6899 unsigned long min, unsigned long max, unsigned int depth)
6901 struct maple_node *node = mte_to_node(entry);
6902 unsigned int type = mte_node_type(entry);
6905 mt_dump_range(min, max, depth);
6907 pr_cont("node %p depth %d type %d parent %p", node, depth, type,
6908 node ? node->parent : NULL);
6912 for (i = 0; i < MAPLE_NODE_SLOTS; i++) {
6914 pr_cont("OUT OF RANGE: ");
6915 mt_dump_entry(mt_slot(mt, node->slot, i),
6916 min + i, min + i, depth);
6920 case maple_range_64:
6921 mt_dump_range64(mt, entry, min, max, depth);
6923 case maple_arange_64:
6924 mt_dump_arange64(mt, entry, min, max, depth);
6928 pr_cont(" UNKNOWN TYPE\n");
6932 void mt_dump(const struct maple_tree *mt)
6934 void *entry = rcu_dereference_check(mt->ma_root, mt_locked(mt));
6936 pr_info("maple_tree(%p) flags %X, height %u root %p\n",
6937 mt, mt->ma_flags, mt_height(mt), entry);
6938 if (!xa_is_node(entry))
6939 mt_dump_entry(entry, 0, 0, 0);
6941 mt_dump_node(mt, entry, 0, mt_node_max(entry), 0);
6943 EXPORT_SYMBOL_GPL(mt_dump);
6946 * Calculate the maximum gap in a node and check if that's what is reported in
6947 * the parent (unless root).
6949 static void mas_validate_gaps(struct ma_state *mas)
6951 struct maple_enode *mte = mas->node;
6952 struct maple_node *p_mn;
6953 unsigned long gap = 0, max_gap = 0;
6954 unsigned long p_end, p_start = mas->min;
6955 unsigned char p_slot;
6956 unsigned long *gaps = NULL;
6957 unsigned long *pivots = ma_pivots(mte_to_node(mte), mte_node_type(mte));
6960 if (ma_is_dense(mte_node_type(mte))) {
6961 for (i = 0; i < mt_slot_count(mte); i++) {
6962 if (mas_get_slot(mas, i)) {
6973 gaps = ma_gaps(mte_to_node(mte), mte_node_type(mte));
6974 for (i = 0; i < mt_slot_count(mte); i++) {
6975 p_end = mas_logical_pivot(mas, pivots, i, mte_node_type(mte));
6978 if (mas_get_slot(mas, i)) {
6983 gap += p_end - p_start + 1;
6985 void *entry = mas_get_slot(mas, i);
6989 if (gap != p_end - p_start + 1) {
6990 pr_err("%p[%u] -> %p %lu != %lu - %lu + 1\n",
6992 mas_get_slot(mas, i), gap,
6996 MT_BUG_ON(mas->tree,
6997 gap != p_end - p_start + 1);
7000 if (gap > p_end - p_start + 1) {
7001 pr_err("%p[%u] %lu >= %lu - %lu + 1 (%lu)\n",
7002 mas_mn(mas), i, gap, p_end, p_start,
7003 p_end - p_start + 1);
7004 MT_BUG_ON(mas->tree,
7005 gap > p_end - p_start + 1);
7013 p_start = p_end + 1;
7014 if (p_end >= mas->max)
7019 if (mte_is_root(mte))
7022 p_slot = mte_parent_slot(mas->node);
7023 p_mn = mte_parent(mte);
7024 MT_BUG_ON(mas->tree, max_gap > mas->max);
7025 if (ma_gaps(p_mn, mas_parent_enum(mas, mte))[p_slot] != max_gap) {
7026 pr_err("gap %p[%u] != %lu\n", p_mn, p_slot, max_gap);
7030 MT_BUG_ON(mas->tree,
7031 ma_gaps(p_mn, mas_parent_enum(mas, mte))[p_slot] != max_gap);
7034 static void mas_validate_parent_slot(struct ma_state *mas)
7036 struct maple_node *parent;
7037 struct maple_enode *node;
7038 enum maple_type p_type = mas_parent_enum(mas, mas->node);
7039 unsigned char p_slot = mte_parent_slot(mas->node);
7043 if (mte_is_root(mas->node))
7046 parent = mte_parent(mas->node);
7047 slots = ma_slots(parent, p_type);
7048 MT_BUG_ON(mas->tree, mas_mn(mas) == parent);
7050 /* Check prev/next parent slot for duplicate node entry */
7052 for (i = 0; i < mt_slots[p_type]; i++) {
7053 node = mas_slot(mas, slots, i);
7055 if (node != mas->node)
7056 pr_err("parent %p[%u] does not have %p\n",
7057 parent, i, mas_mn(mas));
7058 MT_BUG_ON(mas->tree, node != mas->node);
7059 } else if (node == mas->node) {
7060 pr_err("Invalid child %p at parent %p[%u] p_slot %u\n",
7061 mas_mn(mas), parent, i, p_slot);
7062 MT_BUG_ON(mas->tree, node == mas->node);
7067 static void mas_validate_child_slot(struct ma_state *mas)
7069 enum maple_type type = mte_node_type(mas->node);
7070 void __rcu **slots = ma_slots(mte_to_node(mas->node), type);
7071 unsigned long *pivots = ma_pivots(mte_to_node(mas->node), type);
7072 struct maple_enode *child;
7075 if (mte_is_leaf(mas->node))
7078 for (i = 0; i < mt_slots[type]; i++) {
7079 child = mas_slot(mas, slots, i);
7080 if (!pivots[i] || pivots[i] == mas->max)
7086 if (mte_parent_slot(child) != i) {
7087 pr_err("Slot error at %p[%u]: child %p has pslot %u\n",
7088 mas_mn(mas), i, mte_to_node(child),
7089 mte_parent_slot(child));
7090 MT_BUG_ON(mas->tree, 1);
7093 if (mte_parent(child) != mte_to_node(mas->node)) {
7094 pr_err("child %p has parent %p not %p\n",
7095 mte_to_node(child), mte_parent(child),
7096 mte_to_node(mas->node));
7097 MT_BUG_ON(mas->tree, 1);
7103 * Validate all pivots are within mas->min and mas->max.
7105 static void mas_validate_limits(struct ma_state *mas)
7108 unsigned long prev_piv = 0;
7109 enum maple_type type = mte_node_type(mas->node);
7110 void __rcu **slots = ma_slots(mte_to_node(mas->node), type);
7111 unsigned long *pivots = ma_pivots(mas_mn(mas), type);
7113 /* all limits are fine here. */
7114 if (mte_is_root(mas->node))
7117 for (i = 0; i < mt_slots[type]; i++) {
7120 piv = mas_safe_pivot(mas, pivots, i, type);
7122 if (!piv && (i != 0))
7125 if (!mte_is_leaf(mas->node)) {
7126 void *entry = mas_slot(mas, slots, i);
7129 pr_err("%p[%u] cannot be null\n",
7132 MT_BUG_ON(mas->tree, !entry);
7135 if (prev_piv > piv) {
7136 pr_err("%p[%u] piv %lu < prev_piv %lu\n",
7137 mas_mn(mas), i, piv, prev_piv);
7138 MT_BUG_ON(mas->tree, piv < prev_piv);
7141 if (piv < mas->min) {
7142 pr_err("%p[%u] %lu < %lu\n", mas_mn(mas), i,
7144 MT_BUG_ON(mas->tree, piv < mas->min);
7146 if (piv > mas->max) {
7147 pr_err("%p[%u] %lu > %lu\n", mas_mn(mas), i,
7149 MT_BUG_ON(mas->tree, piv > mas->max);
7152 if (piv == mas->max)
7155 for (i += 1; i < mt_slots[type]; i++) {
7156 void *entry = mas_slot(mas, slots, i);
7158 if (entry && (i != mt_slots[type] - 1)) {
7159 pr_err("%p[%u] should not have entry %p\n", mas_mn(mas),
7161 MT_BUG_ON(mas->tree, entry != NULL);
7164 if (i < mt_pivots[type]) {
7165 unsigned long piv = pivots[i];
7170 pr_err("%p[%u] should not have piv %lu\n",
7171 mas_mn(mas), i, piv);
7172 MT_BUG_ON(mas->tree, i < mt_pivots[type] - 1);
7177 static void mt_validate_nulls(struct maple_tree *mt)
7179 void *entry, *last = (void *)1;
7180 unsigned char offset = 0;
7182 MA_STATE(mas, mt, 0, 0);
7185 if (mas_is_none(&mas) || (mas.node == MAS_ROOT))
7188 while (!mte_is_leaf(mas.node))
7191 slots = ma_slots(mte_to_node(mas.node), mte_node_type(mas.node));
7193 entry = mas_slot(&mas, slots, offset);
7194 if (!last && !entry) {
7195 pr_err("Sequential nulls end at %p[%u]\n",
7196 mas_mn(&mas), offset);
7198 MT_BUG_ON(mt, !last && !entry);
7200 if (offset == mas_data_end(&mas)) {
7201 mas_next_node(&mas, mas_mn(&mas), ULONG_MAX);
7202 if (mas_is_none(&mas))
7205 slots = ma_slots(mte_to_node(mas.node),
7206 mte_node_type(mas.node));
7211 } while (!mas_is_none(&mas));
7215 * validate a maple tree by checking:
7216 * 1. The limits (pivots are within mas->min to mas->max)
7217 * 2. The gap is correctly set in the parents
7219 void mt_validate(struct maple_tree *mt)
7223 MA_STATE(mas, mt, 0, 0);
7226 if (!mas_searchable(&mas))
7229 mas_first_entry(&mas, mas_mn(&mas), ULONG_MAX, mte_node_type(mas.node));
7230 while (!mas_is_none(&mas)) {
7231 MT_BUG_ON(mas.tree, mte_dead_node(mas.node));
7232 if (!mte_is_root(mas.node)) {
7233 end = mas_data_end(&mas);
7234 if ((end < mt_min_slot_count(mas.node)) &&
7235 (mas.max != ULONG_MAX)) {
7236 pr_err("Invalid size %u of %p\n", end,
7238 MT_BUG_ON(mas.tree, 1);
7242 mas_validate_parent_slot(&mas);
7243 mas_validate_child_slot(&mas);
7244 mas_validate_limits(&mas);
7245 if (mt_is_alloc(mt))
7246 mas_validate_gaps(&mas);
7247 mas_dfs_postorder(&mas, ULONG_MAX);
7249 mt_validate_nulls(mt);
7254 EXPORT_SYMBOL_GPL(mt_validate);
7256 #endif /* CONFIG_DEBUG_MAPLE_TREE */