src/util/sparse_array.c

   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
  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
  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 *
  62 _util_sparse_array_node_data(uintptr_t handle)
  63 {
  64    return (void *)(handle & NODE_PTR_MASK);
  65 }
  66
  67 static inline unsigned
  68 _util_sparse_array_node_level(uintptr_t handle)
  69 {
  70    return handle & NODE_LEVEL_MASK;
  71 }
  72
  73 static inline void
  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
  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
  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
 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 *
 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
 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
 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
 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
 230 free_list_head(uint64_t old, uint32_t next)
 231 {
 232    return ((old & 0xffffffff00000000ull) + 0x100000000ull) | next;
 233 }
 234
 235 void
 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
 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 *
 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 }