1 /*
2  * Copyright © 2019 Intel Corporation
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21  * IN THE SOFTWARE.
22  */
23 
24 #include "sparse_array.h"
25 #include "os_memory.h"
26 
27 /* Aligning our allocations to 64 has two advantages:
28  *
29  *  1. On x86 platforms, it means that they are cache-line aligned so we
30  *     reduce the likelihood that one of our allocations shares a cache line
31  *     with some other allocation.
32  *
33  *  2. It lets us use the bottom 6 bits of the pointer to store the tree level
34  *     of the node so we can avoid some pointer indirections.
35  */
36 #define NODE_ALLOC_ALIGN 64
37 
38 void
util_sparse_array_init(struct util_sparse_array * arr,size_t elem_size,size_t node_size)39 util_sparse_array_init(struct util_sparse_array *arr,
40                        size_t elem_size, size_t node_size)
41 {
42    memset(arr, 0, sizeof(*arr));
43    arr->elem_size = elem_size;
44    arr->node_size_log2 = util_logbase2_64(node_size);
45    assert(node_size >= 2 && node_size == (1ull << arr->node_size_log2));
46 }
47 
48 #define NODE_PTR_MASK (~((uintptr_t)NODE_ALLOC_ALIGN - 1))
49 #define NODE_LEVEL_MASK ((uintptr_t)NODE_ALLOC_ALIGN - 1)
50 #define NULL_NODE 0
51 
52 static inline uintptr_t
_util_sparse_array_node(void * data,unsigned level)53 _util_sparse_array_node(void *data, unsigned level)
54 {
55    assert(data != NULL);
56    assert(((uintptr_t)data & NODE_LEVEL_MASK) == 0);
57    assert((level & NODE_PTR_MASK) == 0);
58    return (uintptr_t)data | level;
59 }
60 
61 static inline void *
_util_sparse_array_node_data(uintptr_t handle)62 _util_sparse_array_node_data(uintptr_t handle)
63 {
64    return (void *)(handle & NODE_PTR_MASK);
65 }
66 
67 static inline unsigned
_util_sparse_array_node_level(uintptr_t handle)68 _util_sparse_array_node_level(uintptr_t handle)
69 {
70    return handle & NODE_LEVEL_MASK;
71 }
72 
73 static inline void
_util_sparse_array_node_finish(struct util_sparse_array * arr,uintptr_t node)74 _util_sparse_array_node_finish(struct util_sparse_array *arr,
75                                uintptr_t node)
76 {
77    if (_util_sparse_array_node_level(node) > 0) {
78       uintptr_t *children = _util_sparse_array_node_data(node);
79       size_t node_size = 1ull << arr->node_size_log2;
80       for (size_t i = 0; i < node_size; i++) {
81          if (children[i])
82             _util_sparse_array_node_finish(arr, children[i]);
83       }
84    }
85 
86    os_free_aligned(_util_sparse_array_node_data(node));
87 }
88 
89 void
util_sparse_array_finish(struct util_sparse_array * arr)90 util_sparse_array_finish(struct util_sparse_array *arr)
91 {
92    if (arr->root)
93       _util_sparse_array_node_finish(arr, arr->root);
94 }
95 
96 static inline uintptr_t
_util_sparse_array_node_alloc(struct util_sparse_array * arr,unsigned level)97 _util_sparse_array_node_alloc(struct util_sparse_array *arr,
98                               unsigned level)
99 {
100    size_t size;
101    if (level == 0) {
102       size = arr->elem_size << arr->node_size_log2;
103    } else {
104       size = sizeof(uintptr_t) << arr->node_size_log2;
105    }
106 
107    void *data = os_malloc_aligned(size, NODE_ALLOC_ALIGN);
108    memset(data, 0, size);
109 
110    return _util_sparse_array_node(data, level);
111 }
112 
113 static inline uintptr_t
_util_sparse_array_set_or_free_node(uintptr_t * node_ptr,uintptr_t cmp_node,uintptr_t node)114 _util_sparse_array_set_or_free_node(uintptr_t *node_ptr,
115                                     uintptr_t cmp_node,
116                                     uintptr_t node)
117 {
118    uintptr_t prev_node = p_atomic_cmpxchg(node_ptr, cmp_node, node);
119 
120    if (prev_node != cmp_node) {
121       /* We lost the race.  Free this one and return the one that was already
122        * allocated.
123        */
124       os_free_aligned(_util_sparse_array_node_data(node));
125       return prev_node;
126    } else {
127       return node;
128    }
129 }
130 
131 void *
util_sparse_array_get(struct util_sparse_array * arr,uint64_t idx)132 util_sparse_array_get(struct util_sparse_array *arr, uint64_t idx)
133 {
134    const unsigned node_size_log2 = arr->node_size_log2;
135    uintptr_t root = p_atomic_read(&arr->root);
136    if (unlikely(!root)) {
137       unsigned root_level = 0;
138       uint64_t idx_iter = idx >> node_size_log2;
139       while (idx_iter) {
140          idx_iter >>= node_size_log2;
141          root_level++;
142       }
143       uintptr_t new_root = _util_sparse_array_node_alloc(arr, root_level);
144       root = _util_sparse_array_set_or_free_node(&arr->root,
145                                                  NULL_NODE, new_root);
146    }
147 
148    while (1) {
149       unsigned root_level = _util_sparse_array_node_level(root);
150       uint64_t root_idx = idx >> (root_level * node_size_log2);
151       if (likely(root_idx < (1ull << node_size_log2)))
152          break;
153 
154       /* In this case, we have a root but its level is low enough that the
155        * requested index is out-of-bounds.
156        */
157       uintptr_t new_root = _util_sparse_array_node_alloc(arr, root_level + 1);
158 
159       uintptr_t *new_root_children = _util_sparse_array_node_data(new_root);
160       new_root_children[0] = root;
161 
162       /* We only add one at a time instead of the whole tree because it's
163        * easier to ensure correctness of both the tree building and the
164        * clean-up path.  Because we're only adding one node we never have to
165        * worry about trying to free multiple things without freeing the old
166        * things.
167        */
168       root = _util_sparse_array_set_or_free_node(&arr->root, root, new_root);
169    }
170 
171    void *node_data = _util_sparse_array_node_data(root);
172    unsigned node_level = _util_sparse_array_node_level(root);
173    while (node_level > 0) {
174       uint64_t child_idx = (idx >> (node_level * node_size_log2)) &
175                            ((1ull << node_size_log2) - 1);
176 
177       uintptr_t *children = node_data;
178       uintptr_t child = p_atomic_read(&children[child_idx]);
179 
180       if (unlikely(!child)) {
181          child = _util_sparse_array_node_alloc(arr, node_level - 1);
182          child = _util_sparse_array_set_or_free_node(&children[child_idx],
183                                                      NULL_NODE, child);
184       }
185 
186       node_data = _util_sparse_array_node_data(child);
187       node_level = _util_sparse_array_node_level(child);
188    }
189 
190    uint64_t elem_idx = idx & ((1ull << node_size_log2) - 1);
191    return (void *)((char *)node_data + (elem_idx * arr->elem_size));
192 }
193 
194 static void
validate_node_level(struct util_sparse_array * arr,uintptr_t node,unsigned level)195 validate_node_level(struct util_sparse_array *arr,
196                     uintptr_t node, unsigned level)
197 {
198    assert(_util_sparse_array_node_level(node) == level);
199 
200    if (_util_sparse_array_node_level(node) > 0) {
201       uintptr_t *children = _util_sparse_array_node_data(node);
202       size_t node_size = 1ull << arr->node_size_log2;
203       for (size_t i = 0; i < node_size; i++) {
204          if (children[i])
205             validate_node_level(arr, children[i], level - 1);
206       }
207    }
208 }
209 
210 void
util_sparse_array_validate(struct util_sparse_array * arr)211 util_sparse_array_validate(struct util_sparse_array *arr)
212 {
213    uintptr_t root = p_atomic_read(&arr->root);
214    validate_node_level(arr, root, _util_sparse_array_node_level(root));
215 }
216 
217 void
util_sparse_array_free_list_init(struct util_sparse_array_free_list * fl,struct util_sparse_array * arr,uint32_t sentinel,uint32_t next_offset)218 util_sparse_array_free_list_init(struct util_sparse_array_free_list *fl,
219                                  struct util_sparse_array *arr,
220                                  uint32_t sentinel,
221                                  uint32_t next_offset)
222 {
223    fl->head = sentinel;
224    fl->arr = arr;
225    fl->sentinel = sentinel;
226    fl->next_offset = next_offset;
227 }
228 
229 static uint64_t
free_list_head(uint64_t old,uint32_t next)230 free_list_head(uint64_t old, uint32_t next)
231 {
232    return ((old & 0xffffffff00000000ull) + 0x100000000ull) | next;
233 }
234 
235 void
util_sparse_array_free_list_push(struct util_sparse_array_free_list * fl,uint32_t * items,unsigned num_items)236 util_sparse_array_free_list_push(struct util_sparse_array_free_list *fl,
237                                  uint32_t *items, unsigned num_items)
238 {
239    assert(num_items > 0);
240    assert(items[0] != fl->sentinel);
241    void *last_elem = util_sparse_array_get(fl->arr, items[0]);
242    uint32_t *last_next = (uint32_t *)((char *)last_elem + fl->next_offset);
243    for (unsigned i = 1; i < num_items; i++) {
244       p_atomic_set(last_next, items[i]);
245       assert(items[i] != fl->sentinel);
246       last_elem = util_sparse_array_get(fl->arr, items[i]);
247       last_next = (uint32_t *)((char *)last_elem + fl->next_offset);
248    }
249 
250    uint64_t current_head, old_head;
251    old_head = p_atomic_read(&fl->head);
252    do {
253       current_head = old_head;
254       p_atomic_set(last_next, (uint32_t)current_head); /* Index is the bottom 32 bits */
255       uint64_t new_head = free_list_head(current_head, items[0]);
256       old_head = p_atomic_cmpxchg(&fl->head, current_head, new_head);
257    } while (old_head != current_head);
258 }
259 
260 uint32_t
util_sparse_array_free_list_pop_idx(struct util_sparse_array_free_list * fl)261 util_sparse_array_free_list_pop_idx(struct util_sparse_array_free_list *fl)
262 {
263    uint64_t current_head;
264 
265    current_head = p_atomic_read(&fl->head);
266    while (1) {
267       if ((uint32_t)current_head == fl->sentinel)
268          return fl->sentinel;
269 
270       uint32_t head_idx = current_head; /* Index is the bottom 32 bits */
271       void *head_elem = util_sparse_array_get(fl->arr, head_idx);
272       uint32_t *head_next = (uint32_t *)((char *)head_elem + fl->next_offset);
273       uint64_t new_head = free_list_head(current_head, p_atomic_read(head_next));
274       uint64_t old_head = p_atomic_cmpxchg(&fl->head, current_head, new_head);
275       if (old_head == current_head)
276          return head_idx;
277       current_head = old_head;
278    }
279 }
280 
281 void *
util_sparse_array_free_list_pop_elem(struct util_sparse_array_free_list * fl)282 util_sparse_array_free_list_pop_elem(struct util_sparse_array_free_list *fl)
283 {
284    uint64_t current_head;
285 
286    current_head = p_atomic_read(&fl->head);
287    while (1) {
288       if ((uint32_t)current_head == fl->sentinel)
289          return NULL;
290 
291       uint32_t head_idx = current_head; /* Index is the bottom 32 bits */
292       void *head_elem = util_sparse_array_get(fl->arr, head_idx);
293       uint32_t *head_next = (uint32_t *)((char *)head_elem + fl->next_offset);
294       uint64_t new_head = free_list_head(current_head, p_atomic_read(head_next));
295       uint64_t old_head = p_atomic_cmpxchg(&fl->head, current_head, new_head);
296       if (old_head == current_head)
297          return head_elem;
298       current_head = old_head;
299    }
300 }
301