rcuja/rcuja.c

   1 /*
   2  * rcuja/rcuja.c
   3  *
   4  * Userspace RCU library - RCU Judy Array
   5  *
   6  * Copyright 2012 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   7  *
   8  * This library is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * This library is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with this library; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 #define _LGPL_SOURCE
  24 #include <stdint.h>
  25 #include <errno.h>
  26 #include <limits.h>
  27 #include <urcu/rcuja.h>
  28 #include <urcu/compiler.h>
  29 #include <urcu/arch.h>
  30 #include <assert.h>
  31 #include <urcu-pointer.h>
  32 #include <stdint.h>
  33
  34 #include "rcuja-internal.h"
  35 #include "bitfield.h"
  36
  37 enum cds_ja_type_class {
  38         RCU_JA_LINEAR = 0,      /* Type A */
  39                         /* 32-bit: 1 to 25 children, 8 to 128 bytes */
  40                         /* 64-bit: 1 to 28 children, 16 to 256 bytes */
  41         RCU_JA_POOL = 1,        /* Type B */
  42                         /* 32-bit: 26 to 100 children, 256 to 512 bytes */
  43                         /* 64-bit: 29 to 112 children, 512 to 1024 bytes */
  44         RCU_JA_PIGEON = 2,      /* Type C */
  45                         /* 32-bit: 101 to 256 children, 1024 bytes */
  46                         /* 64-bit: 113 to 256 children, 2048 bytes */
  47         /* Leaf nodes are implicit from their height in the tree */
  48         RCU_JA_NR_TYPES,
  49
  50         RCU_JA_NULL,    /* not an encoded type, but keeps code regular */
  51 };
  52
  53 struct cds_ja_type {
  54         enum cds_ja_type_class type_class;
  55         uint16_t min_child;             /* minimum number of children: 1 to 256 */
  56         uint16_t max_child;             /* maximum number of children: 1 to 256 */
  57         uint16_t max_linear_child;      /* per-pool max nr. children: 1 to 256 */
  58         uint16_t order;                 /* node size is (1 << order), in bytes */
  59         uint16_t nr_pool_order;         /* number of pools */
  60         uint16_t pool_size_order;       /* pool size */
  61 };
  62
  63 /*
  64  * Number of least significant pointer bits reserved to represent the
  65  * child type.
  66  */
  67 #define JA_TYPE_BITS    3
  68 #define JA_TYPE_MAX_NR  (1U << JA_TYPE_BITS)
  69 #define JA_TYPE_MASK    (JA_TYPE_MAX_NR - 1)
  70 #define JA_PTR_MASK     (~JA_TYPE_MASK)
  71
  72 #define JA_ENTRY_PER_NODE       256UL
  73 #define JA_BITS_PER_BYTE        3
  74
  75 /*
  76  * Entry for NULL node is at index 8 of the table. It is never encoded
  77  * in flags.
  78  */
  79 #define NODE_INDEX_NULL         8
  80
  81 /*
  82  * Iteration on the array to find the right node size for the number of
  83  * children stops when it reaches .max_child == 256 (this is the largest
  84  * possible node size, which contains 256 children).
  85  * The min_child overlaps with the previous max_child to provide an
  86  * hysteresis loop to reallocation for patterns of cyclic add/removal
  87  * within the same node.
  88  * The node the index within the following arrays is represented on 3
  89  * bits. It identifies the node type, min/max number of children, and
  90  * the size order.
  91  * The max_child values for the RCU_JA_POOL below result from
  92  * statistical approximation: over million populations, the max_child
  93  * covers between 97% and 99% of the populations generated. Therefore, a
  94  * fallback should exist to cover the rare extreme population unbalance
  95  * cases, but it will not have a major impact on speed nor space
  96  * consumption, since those are rare cases.
  97  */
  98
  99 #if (CAA_BITS_PER_LONG < 64)
 100 /* 32-bit pointers */
 101 enum {
 102         ja_type_0_max_child = 1,
 103         ja_type_1_max_child = 3,
 104         ja_type_2_max_child = 6,
 105         ja_type_3_max_child = 12,
 106         ja_type_4_max_child = 25,
 107         ja_type_5_max_child = 48,
 108         ja_type_6_max_child = 92,
 109         ja_type_7_max_child = 256,
 110         ja_type_8_max_child = 0,        /* NULL */
 111 };
 112
 113 enum {
 114         ja_type_0_max_linear_child = 1,
 115         ja_type_1_max_linear_child = 3,
 116         ja_type_2_max_linear_child = 6,
 117         ja_type_3_max_linear_child = 12,
 118         ja_type_4_max_linear_child = 25,
 119         ja_type_5_max_linear_child = 24,
 120         ja_type_6_max_linear_child = 23,
 121 };
 122
 123 enum {
 124         ja_type_5_nr_pool_order = 1,
 125         ja_type_6_nr_pool_order = 2,
 126 };
 127
 128 const struct cds_ja_type ja_types[] = {
 129         { .type_class = RCU_JA_LINEAR, .min_child = 1, .max_child = ja_type_0_max_child, .max_linear_child = ja_type_0_max_linear_child, .order = 3, },
 130         { .type_class = RCU_JA_LINEAR, .min_child = 1, .max_child = ja_type_1_max_child, .max_linear_child = ja_type_1_max_linear_child, .order = 4, },
 131         { .type_class = RCU_JA_LINEAR, .min_child = 3, .max_child = ja_type_2_max_child, .max_linear_child = ja_type_2_max_linear_child, .order = 5, },
 132         { .type_class = RCU_JA_LINEAR, .min_child = 4, .max_child = ja_type_3_max_child, .max_linear_child = ja_type_3_max_linear_child, .order = 6, },
 133         { .type_class = RCU_JA_LINEAR, .min_child = 10, .max_child = ja_type_4_max_child, .max_linear_child = ja_type_4_max_linear_child, .order = 7, },
 134
 135         /* Pools may fill sooner than max_child */
 136         { .type_class = RCU_JA_POOL, .min_child = 20, .max_child = ja_type_5_max_child, .max_linear_child = ja_type_5_max_linear_child, .order = 8, .nr_pool_order = ja_type_5_nr_pool_order, .pool_size_order = 7, },
 137         { .type_class = RCU_JA_POOL, .min_child = 45, .max_child = ja_type_6_max_child, .max_linear_child = ja_type_6_max_linear_child, .order = 9, .nr_pool_order = ja_type_6_nr_pool_order, .pool_size_order = 7, },
 138
 139         /*
 140          * TODO: Upon node removal below min_child, if child pool is
 141          * filled beyond capacity, we need to roll back to pigeon.
 142          */
 143         { .type_class = RCU_JA_PIGEON, .min_child = 89, .max_child = ja_type_7_max_child, .order = 10, },
 144
 145         { .type_class = RCU_JA_NULL, .min_child = 0, .max_child = ja_type_8_max_child, },
 146 };
 147 #else /* !(CAA_BITS_PER_LONG < 64) */
 148 /* 64-bit pointers */
 149 enum {
 150         ja_type_0_max_child = 1,
 151         ja_type_1_max_child = 3,
 152         ja_type_2_max_child = 7,
 153         ja_type_3_max_child = 14,
 154         ja_type_4_max_child = 28,
 155         ja_type_5_max_child = 54,
 156         ja_type_6_max_child = 104,
 157         ja_type_7_max_child = 256,
 158         ja_type_8_max_child = 256,
 159 };
 160
 161 enum {
 162         ja_type_0_max_linear_child = 1,
 163         ja_type_1_max_linear_child = 3,
 164         ja_type_2_max_linear_child = 7,
 165         ja_type_3_max_linear_child = 14,
 166         ja_type_4_max_linear_child = 28,
 167         ja_type_5_max_linear_child = 27,
 168         ja_type_6_max_linear_child = 26,
 169 };
 170
 171 enum {
 172         ja_type_5_nr_pool_order = 1,
 173         ja_type_6_nr_pool_order = 2,
 174 };
 175
 176 const struct cds_ja_type ja_types[] = {
 177         { .type_class = RCU_JA_LINEAR, .min_child = 1, .max_child = ja_type_0_max_child, .max_linear_child = ja_type_0_max_linear_child, .order = 4, },
 178         { .type_class = RCU_JA_LINEAR, .min_child = 1, .max_child = ja_type_1_max_child, .max_linear_child = ja_type_1_max_linear_child, .order = 5, },
 179         { .type_class = RCU_JA_LINEAR, .min_child = 3, .max_child = ja_type_2_max_child, .max_linear_child = ja_type_2_max_linear_child, .order = 6, },
 180         { .type_class = RCU_JA_LINEAR, .min_child = 5, .max_child = ja_type_3_max_child, .max_linear_child = ja_type_3_max_linear_child, .order = 7, },
 181         { .type_class = RCU_JA_LINEAR, .min_child = 10, .max_child = ja_type_4_max_child, .max_linear_child = ja_type_4_max_linear_child, .order = 8, },
 182
 183         /* Pools may fill sooner than max_child. */
 184         { .type_class = RCU_JA_POOL, .min_child = 22, .max_child = ja_type_5_max_child, .max_linear_child = ja_type_5_max_linear_child, .order = 9, .nr_pool_order = ja_type_5_nr_pool_order, .pool_size_order = 8, },
 185         { .type_class = RCU_JA_POOL, .min_child = 51, .max_child = ja_type_6_max_child, .max_linear_child = ja_type_6_max_linear_child, .order = 10, .nr_pool_order = ja_type_6_nr_pool_order, .pool_size_order = 8, },
 186
 187         /*
 188          * TODO: Upon node removal below min_child, if child pool is
 189          * filled beyond capacity, we need to roll back to pigeon.
 190          */
 191         { .type_class = RCU_JA_PIGEON, .min_child = 101, .max_child = ja_type_7_max_child, .order = 11, },
 192
 193         { .type_class = RCU_JA_NULL, .min_child = 0, .max_child = ja_type_8_max_child, },
 194 };
 195 #endif /* !(BITS_PER_LONG < 64) */
 196
 197 static inline __attribute__((unused))
 198 void static_array_size_check(void)
 199 {
 200         CAA_BUILD_BUG_ON(CAA_ARRAY_SIZE(ja_types) < JA_TYPE_MAX_NR);
 201 }
 202
 203 /*
 204  * The cds_ja_node contains the compressed node data needed for
 205  * read-side. For linear and pool node configurations, it starts with a
 206  * byte counting the number of children in the node.  Then, the
 207  * node-specific data is placed.
 208  * The node mutex, if any is needed, protecting concurrent updated of
 209  * each node is placed in a separate hash table indexed by node address.
 210  * For the pigeon configuration, the number of children is also kept in
 211  * a separate hash table, indexed by node address, because it is only
 212  * required for updates.
 213  */
 214
 215 #define DECLARE_LINEAR_NODE(index)                                                              \
 216         struct {                                                                                \
 217                 uint8_t nr_child;                                                               \
 218                 uint8_t child_value[ja_type_## index ##_max_linear_child];                      \
 219                 struct cds_ja_inode_flag *child_ptr[ja_type_## index ##_max_linear_child];      \
 220         }
 221
 222 #define DECLARE_POOL_NODE(index)                                                                \
 223         struct {                                                                                \
 224                 struct {                                                                        \
 225                         uint8_t nr_child;                                                       \
 226                         uint8_t child_value[ja_type_## index ##_max_linear_child];              \
 227                         struct cds_ja_inode_flag *child_ptr[ja_type_## index ##_max_linear_child]; \
 228                 } linear[1U << ja_type_## index ##_nr_pool_order];                              \
 229         }
 230
 231 struct cds_ja_inode {
 232         union {
 233                 /* Linear configuration */
 234                 DECLARE_LINEAR_NODE(0) conf_0;
 235                 DECLARE_LINEAR_NODE(1) conf_1;
 236                 DECLARE_LINEAR_NODE(2) conf_2;
 237                 DECLARE_LINEAR_NODE(3) conf_3;
 238                 DECLARE_LINEAR_NODE(4) conf_4;
 239
 240                 /* Pool configuration */
 241                 DECLARE_POOL_NODE(5) conf_5;
 242                 DECLARE_POOL_NODE(6) conf_6;
 243
 244                 /* Pigeon configuration */
 245                 struct {
 246                         struct cds_ja_inode_flag *child[ja_type_7_max_child];
 247                 } conf_7;
 248                 /* data aliasing nodes for computed accesses */
 249                 uint8_t data[sizeof(struct cds_ja_inode_flag *) * ja_type_7_max_child];
 250         } u;
 251 };
 252
 253 static
 254 struct cds_ja_inode_flag *ja_node_flag(struct cds_ja_inode *node,
 255                 unsigned int type)
 256 {
 257         assert(type < RCU_JA_NR_TYPES);
 258         return (struct cds_ja_inode_flag *) (((unsigned long) node) | type);
 259 }
 260
 261 static
 262 struct cds_ja_inode *ja_node_ptr(struct cds_ja_inode_flag *node)
 263 {
 264         return (struct cds_ja_inode *) (((unsigned long) node) | JA_PTR_MASK);
 265 }
 266
 267 static
 268 unsigned int ja_node_type(struct cds_ja_inode_flag *node)
 269 {
 270         unsigned int type;
 271
 272         if (ja_node_ptr(node) == NULL) {
 273                 return NODE_INDEX_NULL;
 274         }
 275         type = (unsigned int) ((unsigned long) node & JA_TYPE_MASK);
 276         assert(type < RCU_JA_NR_TYPES);
 277         return type;
 278 }
 279
 280 struct cds_ja_inode *alloc_cds_ja_node(const struct cds_ja_type *ja_type)
 281 {
 282         return calloc(1U << ja_type->order, sizeof(char));
 283 }
 284
 285 void free_cds_ja_node(struct cds_ja_inode *node)
 286 {
 287         free(node);
 288 }
 289
 290 #define __JA_ALIGN_MASK(v, mask)        (((v) + (mask)) & ~(mask))
 291 #define JA_ALIGN(v, align)              __JA_ALIGN_MASK(v, (typeof(v)) (align) - 1)
 292 #define __JA_FLOOR_MASK(v, mask)        ((v) & ~(mask))
 293 #define JA_FLOOR(v, align)              __JA_FLOOR_MASK(v, (typeof(v)) (align) - 1)
 294
 295 static
 296 uint8_t *align_ptr_size(uint8_t *ptr)
 297 {
 298         return (uint8_t *) JA_ALIGN((unsigned long) ptr, sizeof(void *));
 299 }
 300
 301 static
 302 uint8_t ja_linear_node_get_nr_child(const struct cds_ja_type *type,
 303                 struct cds_ja_inode *node)
 304 {
 305         assert(type->type_class == RCU_JA_LINEAR || type->type_class == RCU_JA_POOL);
 306         return CMM_LOAD_SHARED(node->u.data[0]);
 307 }
 308
 309 /*
 310  * The order in which values and pointers are does does not matter: if
 311  * a value is missing, we return NULL. If a value is there, but its
 312  * associated pointers is still NULL, we return NULL too.
 313  */
 314 static
 315 struct cds_ja_inode_flag *ja_linear_node_get_nth(const struct cds_ja_type *type,
 316                 struct cds_ja_inode *node,
 317                 uint8_t n)
 318 {
 319         uint8_t nr_child;
 320         uint8_t *values;
 321         struct cds_ja_inode_flag **pointers;
 322         struct cds_ja_inode_flag *ptr;
 323         unsigned int i;
 324
 325         assert(type->type_class == RCU_JA_LINEAR || type->type_class == RCU_JA_POOL);
 326
 327         nr_child = ja_linear_node_get_nr_child(type, node);
 328         cmm_smp_rmb();  /* read nr_child before values and pointers */
 329         assert(nr_child <= type->max_linear_child);
 330         assert(type->type_class != RCU_JA_LINEAR || nr_child >= type->min_child);
 331
 332         values = &node->u.data[1];
 333         for (i = 0; i < nr_child; i++) {
 334                 if (CMM_LOAD_SHARED(values[i]) == n)
 335                         break;
 336         }
 337         if (i >= nr_child)
 338                 return NULL;
 339         pointers = (struct cds_ja_inode_flag **) align_ptr_size(&values[type->max_linear_child]);
 340         ptr = rcu_dereference(pointers[i]);
 341         assert(ja_node_ptr(ptr) != NULL);
 342         return ptr;
 343 }
 344
 345 static
 346 struct cds_ja_inode_flag *ja_linear_node_get_ith_pos(const struct cds_ja_type *type,
 347                 struct cds_ja_inode *node,
 348                 uint8_t i,
 349                 uint8_t *v,
 350                 struct cds_ja_inode_flag **iter)
 351 {
 352         uint8_t *values;
 353         struct cds_ja_inode_flag **pointers;
 354
 355         assert(type->type_class == RCU_JA_LINEAR || type->type_class == RCU_JA_POOL);
 356         assert(i < ja_linear_node_get_nr_child(type, node));
 357
 358         values = &node->u.data[1];
 359         *v = values[i];
 360         pointers = (struct cds_ja_inode_flag **) align_ptr_size(&values[type->max_linear_child]);
 361         *iter = pointers[i];
 362 }
 363
 364 static
 365 struct cds_ja_inode_flag *ja_pool_node_get_nth(const struct cds_ja_type *type,
 366                 struct cds_ja_inode *node,
 367                 uint8_t n)
 368 {
 369         struct cds_ja_inode *linear;
 370
 371         assert(type->type_class == RCU_JA_POOL);
 372         /*
 373          * TODO: currently, we select the pool by highest bits. We
 374          * should support various encodings.
 375          */
 376         linear = (struct cds_ja_inode *)
 377                 &node->u.data[((unsigned long) n >> (CHAR_BIT - type->nr_pool_order)) << type->pool_size_order];
 378         return ja_linear_node_get_nth(type, linear, n);
 379 }
 380
 381 static
 382 struct cds_ja_inode *ja_pool_node_get_ith_pool(const struct cds_ja_type *type,
 383                 struct cds_ja_inode *node,
 384                 uint8_t i)
 385 {
 386         assert(type->type_class == RCU_JA_POOL);
 387         return (struct cds_ja_inode *)
 388                 &node->u.data[(unsigned int) i << type->pool_size_order];
 389 }
 390
 391 static
 392 struct cds_ja_inode_flag *ja_pigeon_node_get_nth(const struct cds_ja_type *type,
 393                 struct cds_ja_inode *node,
 394                 uint8_t n)
 395 {
 396         assert(type->type_class == RCU_JA_PIGEON);
 397         return rcu_dereference(((struct cds_ja_inode_flag **) node->u.data)[n]);
 398 }
 399
 400 /*
 401  * ja_node_get_nth: get nth item from a node.
 402  * node_flag is already rcu_dereference'd.
 403  */
 404 static
 405 struct cds_ja_inode_flag *ja_node_get_nth(struct cds_ja_inode_flag *node_flag,
 406                 uint8_t n)
 407 {
 408         unsigned int type_index;
 409         struct cds_ja_inode *node;
 410         const struct cds_ja_type *type;
 411
 412         node = ja_node_ptr(node_flag);
 413         assert(node != NULL);
 414         type_index = ja_node_type(node_flag);
 415         type = &ja_types[type_index];
 416
 417         switch (type->type_class) {
 418         case RCU_JA_LINEAR:
 419                 return ja_linear_node_get_nth(type, node, n);
 420         case RCU_JA_POOL:
 421                 return ja_pool_node_get_nth(type, node, n);
 422         case RCU_JA_PIGEON:
 423                 return ja_pigeon_node_get_nth(type, node, n);
 424         default:
 425                 assert(0);
 426                 return (void *) -1UL;
 427         }
 428 }
 429
 430 /*
 431  * TODO: use ja_get_nr_child to monitor limits triggering shrink
 432  * recompaction.
 433  * Also use ja_get_nr_child to make the difference between resize and
 434  * pool change of compaction bit(s).
 435  */
 436 static
 437 unsigned int ja_get_nr_child(struct cds_ja_shadow_node *shadow_node)
 438 {
 439         return shadow_node->nr_child;
 440 }
 441
 442 static
 443 int ja_linear_node_set_nth(const struct cds_ja_type *type,
 444                 struct cds_ja_inode *node,
 445                 struct cds_ja_shadow_node *shadow_node,
 446                 uint8_t n,
 447                 struct cds_ja_inode_flag *child_node_flag)
 448 {
 449         uint8_t nr_child;
 450         uint8_t *values, *nr_child_ptr;
 451         struct cds_ja_inode_flag **pointers;
 452         unsigned int i;
 453
 454         assert(type->type_class == RCU_JA_LINEAR || type->type_class == RCU_JA_POOL);
 455
 456         nr_child_ptr = &node->u.data[0];
 457         nr_child = *nr_child_ptr;
 458         assert(nr_child <= type->max_linear_child);
 459         assert(type->type_class != RCU_JA_LINEAR || nr_child >= type->min_child);
 460
 461         values = &node->u.data[1];
 462         for (i = 0; i < nr_child; i++) {
 463                 if (values[i] == n)
 464                         return -EEXIST;
 465         }
 466         if (nr_child >= type->max_linear_child) {
 467                 /* No space left in this node type */
 468                 return -ENOSPC;
 469         }
 470         pointers = (struct cds_ja_inode_flag **) align_ptr_size(&values[type->max_linear_child]);
 471         assert(pointers[nr_child] == NULL);
 472         rcu_assign_pointer(pointers[nr_child], child_node_flag);
 473         CMM_STORE_SHARED(values[nr_child], n);
 474         cmm_smp_wmb();  /* write value and pointer before nr_child */
 475         CMM_STORE_SHARED(*nr_child_ptr, nr_child + 1);
 476         shadow_node->nr_child++;
 477         return 0;
 478 }
 479
 480 static
 481 int ja_pool_node_set_nth(const struct cds_ja_type *type,
 482                 struct cds_ja_inode *node,
 483                 struct cds_ja_shadow_node *shadow_node,
 484                 uint8_t n,
 485                 struct cds_ja_inode_flag *child_node_flag)
 486 {
 487         struct cds_ja_inode *linear;
 488
 489         assert(type->type_class == RCU_JA_POOL);
 490         linear = (struct cds_ja_inode *)
 491                 &node->u.data[((unsigned long) n >> (CHAR_BIT - type->nr_pool_order)) << type->pool_size_order];
 492         return ja_linear_node_set_nth(type, linear, shadow_node,
 493                         n, child_node_flag);
 494 }
 495
 496 static
 497 int ja_pigeon_node_set_nth(const struct cds_ja_type *type,
 498                 struct cds_ja_inode *node,
 499                 struct cds_ja_shadow_node *shadow_node,
 500                 uint8_t n,
 501                 struct cds_ja_inode_flag *child_node_flag)
 502 {
 503         struct cds_ja_inode_flag **ptr;
 504
 505         assert(type->type_class == RCU_JA_PIGEON);
 506         ptr = &((struct cds_ja_inode_flag **) node->u.data)[n];
 507         if (*ptr != NULL)
 508                 return -EEXIST;
 509         rcu_assign_pointer(*ptr, child_node_flag);
 510         shadow_node->nr_child++;
 511         return 0;
 512 }
 513
 514 /*
 515  * _ja_node_set_nth: set nth item within a node. Return an error
 516  * (negative error value) if it is already there.
 517  * TODO: exclusive access on node.
 518  */
 519 static
 520 int _ja_node_set_nth(const struct cds_ja_type *type,
 521                 struct cds_ja_inode *node,
 522                 struct cds_ja_shadow_node *shadow_node,
 523                 uint8_t n,
 524                 struct cds_ja_inode_flag *child_node_flag)
 525 {
 526         switch (type->type_class) {
 527         case RCU_JA_LINEAR:
 528                 return ja_linear_node_set_nth(type, node, shadow_node, n,
 529                                 child_node_flag);
 530         case RCU_JA_POOL:
 531                 return ja_pool_node_set_nth(type, node, shadow_node, n,
 532                                 child_node_flag);
 533         case RCU_JA_PIGEON:
 534                 return ja_pigeon_node_set_nth(type, node, shadow_node, n,
 535                                 child_node_flag);
 536         case RCU_JA_NULL:
 537                 return -ENOSPC;
 538         default:
 539                 assert(0);
 540                 return -EINVAL;
 541         }
 542
 543         return 0;
 544 }
 545
 546 /*
 547  * ja_node_recompact_add: recompact a node, adding a new child.
 548  * TODO: for pool type, take selection bit(s) into account.
 549  */
 550 static
 551 int ja_node_recompact_add(struct cds_ja *ja,
 552                 unsigned int old_type_index,
 553                 const struct cds_ja_type *old_type,
 554                 struct cds_ja_inode *old_node,
 555                 struct cds_ja_shadow_node **shadow_node,
 556                 struct cds_ja_inode_flag **old_node_flag, uint8_t n,
 557                 struct cds_ja_inode_flag *child_node_flag)
 558 {
 559         unsigned int new_type_index;
 560         struct cds_ja_inode *new_node;
 561         const struct cds_ja_type *new_type;
 562         struct cds_ja_inode_flag *new_node_flag;
 563         int ret;
 564
 565         if (*shadow_node == NULL) {
 566                 new_type_index = 0;
 567         } else {
 568                 new_type_index = old_type_index + 1;
 569         }
 570         new_type = &ja_types[new_type_index];
 571         new_node = alloc_cds_ja_node(new_type);
 572         if (!new_node)
 573                 return -ENOMEM;
 574         new_node_flag = ja_node_flag(new_node, new_type_index);
 575
 576         ret = rcuja_shadow_set(ja->ht, new_node, *shadow_node);
 577         if (ret)
 578                 return ret;
 579
 580         if (*shadow_node == NULL) {
 581                 *shadow_node = rcuja_shadow_lookup_lock(ja->ht, new_node);
 582                 assert(*shadow_node);
 583         }
 584
 585         /*
 586          * We need to clear nr_child, because it will be re-incremented
 587          * by _ja_node_set_nth().
 588          */
 589         (*shadow_node)->nr_child = 0;
 590
 591         assert(old_type->type_class != RCU_JA_PIGEON);
 592         switch (old_type->type_class) {
 593         case RCU_JA_LINEAR:
 594         {
 595                 uint8_t nr_child =
 596                         ja_linear_node_get_nr_child(old_type, old_node);
 597                 unsigned int i;
 598
 599                 for (i = 0; i < nr_child; i++) {
 600                         struct cds_ja_inode_flag *iter;
 601                         uint8_t v;
 602
 603                         ja_linear_node_get_ith_pos(old_type, old_node, i, &v, &iter);
 604                         if (!iter)
 605                                 continue;
 606                         ret = _ja_node_set_nth(new_type, new_node, *shadow_node,
 607                                         v, iter);
 608                         assert(!ret);
 609                 }
 610                 break;
 611         }
 612         case RCU_JA_POOL:
 613         {
 614                 unsigned int pool_nr;
 615
 616                 for (pool_nr = 0; pool_nr < (1U << old_type->nr_pool_order); pool_nr++) {
 617                         struct cds_ja_inode *pool =
 618                                 ja_pool_node_get_ith_pool(old_type,
 619                                         old_node, pool_nr);
 620                         uint8_t nr_child =
 621                                 ja_linear_node_get_nr_child(old_type, pool);
 622                         unsigned int j;
 623
 624                         for (j = 0; j < nr_child; j++) {
 625                                 struct cds_ja_inode_flag *iter;
 626                                 uint8_t v;
 627
 628                                 ja_linear_node_get_ith_pos(old_type, pool,
 629                                                 j, &v, &iter);
 630                                 if (!iter)
 631                                         continue;
 632                                 ret = _ja_node_set_nth(new_type, new_node, *shadow_node,
 633                                                 v, iter);
 634                                 assert(!ret);
 635                         }
 636                 }
 637                 break;
 638         }
 639         case RCU_JA_PIGEON:
 640         default:
 641                 assert(0);
 642                 return -EINVAL;
 643         }
 644
 645         /* add node */
 646         ret = _ja_node_set_nth(new_type, new_node, *shadow_node,
 647                         n, child_node_flag);
 648         assert(!ret);
 649         /* Replace the old node with the new recompacted one */
 650         rcu_assign_pointer(*old_node_flag, new_node_flag);
 651         ret = rcuja_shadow_clear(ja->ht, old_node,
 652                         RCUJA_SHADOW_CLEAR_FREE_NODE);
 653         assert(!ret);
 654         return 0;
 655 }
 656
 657 static
 658 int ja_node_set_nth(struct cds_ja *ja,
 659                 struct cds_ja_inode_flag **node_flag, uint8_t n,
 660                 struct cds_ja_inode_flag *child_node_flag)
 661 {
 662         int ret;
 663         unsigned int type_index;
 664         const struct cds_ja_type *type;
 665         struct cds_ja_inode *node;
 666         struct cds_ja_shadow_node *shadow_node = NULL;
 667
 668         node = ja_node_ptr(*node_flag);
 669         type_index = ja_node_type(*node_flag);
 670         type = &ja_types[type_index];
 671         if (node != NULL) {
 672                 shadow_node = rcuja_shadow_lookup_lock(ja->ht, node);
 673                 assert(shadow_node);
 674         }
 675         ret = _ja_node_set_nth(type, node, shadow_node,
 676                         n, child_node_flag);
 677         if (ret == -ENOSPC) {
 678                 /* Not enough space in node, need to recompact. */
 679                 ret = ja_node_recompact_add(ja, type_index, type, node,
 680                                 &shadow_node, node_flag, n, child_node_flag);
 681                 /* recompact always leave shadow_node locked */
 682         }
 683         rcuja_shadow_unlock(shadow_node);
 684         return ret;
 685 }
 686
 687 struct cds_ja_node *cds_ja_lookup(uint64_t key)
 688 {
 689         /* TODO */
 690 }
 691
 692 struct cds_ja *_cds_ja_new(unsigned int key_bits,
 693                 const struct rcu_flavor_struct *flavor)
 694 {
 695         struct cds_ja *ja;
 696
 697
 698         ja = calloc(sizeof(*ja), 1);
 699         if (!ja)
 700                 goto ja_error;
 701
 702         switch (key_bits) {
 703         case 8:
 704                 ja->key_max = UINT8_MAX;
 705                 break;
 706         case 16:
 707                 ja->key_max = UINT16_MAX;
 708                 break;
 709         case 32:
 710                 ja->key_max = UINT32_MAX;
 711                 break;
 712         case 64:
 713                 ja->key_max = UINT64_MAX;
 714                 break;
 715         default:
 716                 goto check_error;
 717         }
 718
 719         /* ja->root is NULL */
 720         ja->tree_depth = key_bits >> JA_BITS_PER_BYTE;
 721         ja->ht = rcuja_create_ht(flavor);
 722         if (!ja->ht)
 723                 goto ht_error;
 724         return ja;
 725
 726 ht_error:
 727 check_error:
 728         free(ja);
 729 ja_error:
 730         return NULL;
 731 }
 732
 733 /*
 734  * There should be no more concurrent add to the judy array while it is
 735  * being destroyed (ensured by the caller).
 736  */
 737 int cds_ja_destroy(struct cds_ja *ja)
 738 {
 739         int ret;
 740
 741         rcuja_shadow_prune(ja->ht,
 742                 RCUJA_SHADOW_CLEAR_FREE_NODE | RCUJA_SHADOW_CLEAR_FREE_LOCK);
 743         ret = rcuja_delete_ht(ja->ht);
 744         if (ret)
 745                 return ret;
 746         free(ja);
 747 }