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 #include <stdint.h>
  24 #include <errno.h>
  25 #include <limits.h>
  26 #include <urcu/rcuja.h>
  27 #include <urcu/compiler.h>
  28 #include <urcu/arch.h>
  29 #include <assert.h>
  30 #include <urcu-pointer.h>
  31
  32 #include "rcuja-internal.h"
  33 #include "bitfield.h"
  34
  35 enum rcu_ja_type_class {
  36         RCU_JA_LINEAR = 0,      /* Type A */
  37                         /* 32-bit: 1 to 25 children, 8 to 128 bytes */
  38                         /* 64-bit: 1 to 28 children, 16 to 256 bytes */
  39         RCU_JA_POOL = 1,        /* Type B */
  40                         /* 32-bit: 26 to 100 children, 256 to 512 bytes */
  41                         /* 64-bit: 29 to 112 children, 512 to 1024 bytes */
  42         RCU_JA_PIGEON = 2,      /* Type C */
  43                         /* 32-bit: 101 to 256 children, 1024 bytes */
  44                         /* 64-bit: 113 to 256 children, 2048 bytes */
  45         /* Leaf nodes are implicit from their height in the tree */
  46         RCU_JA_NR_TYPES,
  47 };
  48
  49 struct rcu_ja_type {
  50         enum rcu_ja_type_class type_class;
  51         uint16_t min_child;             /* minimum number of children: 1 to 256 */
  52         uint16_t max_child;             /* maximum number of children: 1 to 256 */
  53         uint16_t max_linear_child;      /* per-pool max nr. children: 1 to 256 */
  54         uint16_t order;                 /* node size is (1 << order), in bytes */
  55         uint16_t nr_pool_order;         /* number of pools */
  56         uint16_t pool_size_order;       /* pool size */
  57 };
  58
  59 /*
  60  * Number of least significant pointer bits reserved to represent the
  61  * child type.
  62  */
  63 #define JA_TYPE_BITS    3
  64 #define JA_TYPE_MAX_NR  (1U << JA_TYPE_BITS)
  65 #define JA_TYPE_MASK    (JA_TYPE_MAX_NR - 1)
  66 #define JA_PTR_MASK     (~JA_TYPE_MASK)
  67
  68 #define JA_ENTRY_PER_NODE       256UL
  69
  70 /*
  71  * Iteration on the array to find the right node size for the number of
  72  * children stops when it reaches .max_child == 256 (this is the largest
  73  * possible node size, which contains 256 children).
  74  * The min_child overlaps with the previous max_child to provide an
  75  * hysteresis loop to reallocation for patterns of cyclic add/removal
  76  * within the same node.
  77  * The node the index within the following arrays is represented on 3
  78  * bits. It identifies the node type, min/max number of children, and
  79  * the size order.
  80  * The max_child values for the RCU_JA_POOL below result from
  81  * statistical approximation: over million populations, the max_child
  82  * covers between 97% and 99% of the populations generated. Therefore, a
  83  * fallback should exist to cover the rare extreme population unbalance
  84  * cases, but it will not have a major impact on speed nor space
  85  * consumption, since those are rare cases.
  86  */
  87
  88 #if (CAA_BITS_PER_LONG < 64)
  89 /* 32-bit pointers */
  90 enum {
  91         ja_type_0_max_child = 1,
  92         ja_type_1_max_child = 3,
  93         ja_type_2_max_child = 6,
  94         ja_type_3_max_child = 12,
  95         ja_type_4_max_child = 25,
  96         ja_type_5_max_child = 48,
  97         ja_type_6_max_child = 92,
  98         ja_type_7_max_child = 256,
  99 };
 100
 101 enum {
 102         ja_type_0_max_linear_child = 1,
 103         ja_type_1_max_linear_child = 3,
 104         ja_type_2_max_linear_child = 6,
 105         ja_type_3_max_linear_child = 12,
 106         ja_type_4_max_linear_child = 25,
 107         ja_type_5_max_linear_child = 24,
 108         ja_type_6_max_linear_child = 23,
 109 };
 110
 111 enum {
 112         ja_type_5_nr_pool_order = 1,
 113         ja_type_6_nr_pool_order = 2,
 114 };
 115
 116 const struct rcu_ja_type ja_types[] = {
 117         { .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, },
 118         { .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, },
 119         { .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, },
 120         { .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, },
 121         { .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, },
 122
 123         /* Pools may fill sooner than max_child */
 124         { .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, },
 125         { .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, },
 126
 127         /*
 128          * TODO: Upon node removal below min_child, if child pool is
 129          * filled beyond capacity, we need to roll back to pigeon.
 130          */
 131         { .type_class = RCU_JA_PIGEON, .min_child = 89, .max_child = ja_type_7_max_child, .order = 10, },
 132 };
 133 #else /* !(CAA_BITS_PER_LONG < 64) */
 134 /* 64-bit pointers */
 135 enum {
 136         ja_type_0_max_child = 1,
 137         ja_type_1_max_child = 3,
 138         ja_type_2_max_child = 7,
 139         ja_type_3_max_child = 14,
 140         ja_type_4_max_child = 28,
 141         ja_type_5_max_child = 54,
 142         ja_type_6_max_child = 104,
 143         ja_type_7_max_child = 256,
 144 };
 145
 146 enum {
 147         ja_type_0_max_linear_child = 1,
 148         ja_type_1_max_linear_child = 3,
 149         ja_type_2_max_linear_child = 7,
 150         ja_type_3_max_linear_child = 14,
 151         ja_type_4_max_linear_child = 28,
 152         ja_type_5_max_linear_child = 27,
 153         ja_type_6_max_linear_child = 26,
 154 };
 155
 156 enum {
 157         ja_type_5_nr_pool_order = 1,
 158         ja_type_6_nr_pool_order = 2,
 159 };
 160
 161 const struct rcu_ja_type ja_types[] = {
 162         { .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, },
 163         { .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, },
 164         { .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, },
 165         { .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, },
 166         { .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, },
 167
 168         /* Pools may fill sooner than max_child. */
 169         { .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, },
 170         { .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, },
 171
 172         /*
 173          * TODO: Upon node removal below min_child, if child pool is
 174          * filled beyond capacity, we need to roll back to pigeon.
 175          */
 176         { .type_class = RCU_JA_PIGEON, .min_child = 101, .max_child = ja_type_7_max_child, .order = 11, },
 177 };
 178 #endif /* !(BITS_PER_LONG < 64) */
 179
 180 static inline __attribute__((unused))
 181 void static_array_size_check(void)
 182 {
 183         CAA_BUILD_BUG_ON(CAA_ARRAY_SIZE(ja_types) > JA_TYPE_MAX_NR);
 184 }
 185
 186 /*
 187  * The rcu_ja_node contains the compressed node data needed for
 188  * read-side. For linear and pool node configurations, it starts with a
 189  * byte counting the number of children in the node.  Then, the
 190  * node-specific data is placed.
 191  * The node mutex, if any is needed, protecting concurrent updated of
 192  * each node is placed in a separate hash table indexed by node address.
 193  * For the pigeon configuration, the number of children is also kept in
 194  * a separate hash table, indexed by node address, because it is only
 195  * required for updates.
 196  */
 197
 198 #define DECLARE_LINEAR_NODE(index)                                                              \
 199         struct {                                                                                \
 200                 uint8_t nr_child;                                                               \
 201                 uint8_t child_value[ja_type_## index ##_max_linear_child];                      \
 202                 struct rcu_ja_node_flag *child_ptr[ja_type_## index ##_max_linear_child];       \
 203         }
 204
 205 #define DECLARE_POOL_NODE(index)                                                                \
 206         struct {                                                                                \
 207                 struct {                                                                        \
 208                         uint8_t nr_child;                                                       \
 209                         uint8_t child_value[ja_type_## index ##_max_linear_child];              \
 210                         struct rcu_ja_node_flag *child_ptr[ja_type_## index ##_max_linear_child]; \
 211                 } linear[1U << ja_type_## index ##_nr_pool_order];                              \
 212         }
 213
 214 struct rcu_ja_node {
 215         union {
 216                 /* Linear configuration */
 217                 DECLARE_LINEAR_NODE(0) conf_0;
 218                 DECLARE_LINEAR_NODE(1) conf_1;
 219                 DECLARE_LINEAR_NODE(2) conf_2;
 220                 DECLARE_LINEAR_NODE(3) conf_3;
 221                 DECLARE_LINEAR_NODE(4) conf_4;
 222
 223                 /* Pool configuration */
 224                 DECLARE_POOL_NODE(5) conf_5;
 225                 DECLARE_POOL_NODE(6) conf_6;
 226
 227                 /* Pigeon configuration */
 228                 struct {
 229                         struct rcu_ja_node_flag *child[ja_type_7_max_child];
 230                 } conf_7;
 231                 /* data aliasing nodes for computed accesses */
 232                 uint8_t data[sizeof(struct rcu_ja_node_flag *) * ja_type_7_max_child];
 233         } u;
 234 };
 235
 236 static
 237 struct rcu_ja_node_flag *ja_node_flag(struct rcu_ja_node *node,
 238                 unsigned int type)
 239 {
 240         assert(type < RCU_JA_NR_TYPES);
 241         return (struct rcu_ja_node_flag *) (((unsigned long) node) | type);
 242 }
 243
 244 static
 245 unsigned int ja_node_type(struct rcu_ja_node_flag *node)
 246 {
 247         unsigned int type;
 248
 249         type = (unsigned int) ((unsigned long) node & JA_TYPE_MASK);
 250         assert(type < RCU_JA_NR_TYPES);
 251         return type;
 252 }
 253
 254 static
 255 struct rcu_ja_node *ja_node_ptr(struct rcu_ja_node_flag *node)
 256 {
 257         return (struct rcu_ja_node *) (((unsigned long) node) | JA_PTR_MASK);
 258 }
 259
 260 struct rcu_ja_node *alloc_rcu_ja_node(const struct rcu_ja_type *ja_type)
 261 {
 262         return calloc(1U << ja_type->order, sizeof(char));
 263 }
 264
 265 void free_rcu_ja_node(struct rcu_ja_node *node)
 266 {
 267         free(node);
 268 }
 269
 270 #define __JA_ALIGN_MASK(v, mask)        (((v) + (mask)) & ~(mask))
 271 #define JA_ALIGN(v, align)              __JA_ALIGN_MASK(v, (typeof(v)) (align) - 1)
 272 #define __JA_FLOOR_MASK(v, mask)        ((v) & ~(mask))
 273 #define JA_FLOOR(v, align)              __JA_FLOOR_MASK(v, (typeof(v)) (align) - 1)
 274
 275 static
 276 uint8_t *align_ptr_size(uint8_t *ptr)
 277 {
 278         return (uint8_t *) JA_ALIGN((unsigned long) ptr, sizeof(void *));
 279 }
 280
 281 /*
 282  * The order in which values and pointers are does does not matter: if
 283  * a value is missing, we return NULL. If a value is there, but its
 284  * associated pointers is still NULL, we return NULL too.
 285  */
 286 static
 287 struct rcu_ja_node_flag *ja_linear_node_get_nth(const struct rcu_ja_type *type,
 288                 struct rcu_ja_node *node,
 289                 uint8_t n)
 290 {
 291         uint8_t nr_child;
 292         uint8_t *values;
 293         struct rcu_ja_node_flag **pointers;
 294         struct rcu_ja_node_flag *ptr;
 295         unsigned int i;
 296
 297         assert(type->type_class == RCU_JA_LINEAR || type->type_class == RCU_JA_POOL);
 298
 299         nr_child = CMM_LOAD_SHARED(node->u.data[0]);
 300         cmm_smp_rmb();  /* read nr_child before values and pointers */
 301         assert(nr_child <= type->max_linear_child);
 302         assert(type->type_class != RCU_JA_LINEAR || nr_child >= type->min_child);
 303
 304         values = &node->u.data[1];
 305         for (i = 0; i < nr_child; i++) {
 306                 if (CMM_LOAD_SHARED(values[i]) == n)
 307                         break;
 308         }
 309         if (i >= nr_child)
 310                 return NULL;
 311         pointers = (struct rcu_ja_node_flag **) align_ptr_size(&values[type->max_linear_child]);
 312         ptr = rcu_dereference(pointers[i]);
 313         assert(ja_node_ptr(ptr) != NULL);
 314         return ptr;
 315 }
 316
 317 static
 318 struct rcu_ja_node_flag *ja_pool_node_get_nth(const struct rcu_ja_type *type,
 319                 struct rcu_ja_node *node,
 320                 uint8_t n)
 321 {
 322         struct rcu_ja_node *linear;
 323
 324         assert(type->type_class == RCU_JA_POOL);
 325         linear = (struct rcu_ja_node *)
 326                 &node->u.data[((unsigned long) n >> (CHAR_BIT - type->nr_pool_order)) << type->pool_size_order];
 327         return ja_linear_node_get_nth(type, linear, n);
 328 }
 329
 330 static
 331 struct rcu_ja_node_flag *ja_pigeon_node_get_nth(const struct rcu_ja_type *type,
 332                 struct rcu_ja_node *node,
 333                 uint8_t n)
 334 {
 335         assert(type->type_class == RCU_JA_PIGEON);
 336         return rcu_dereference(((struct rcu_ja_node_flag **) node->u.data)[n]);
 337 }
 338
 339 /*
 340  * ja_node_get_nth: get nth item from a node.
 341  * node_flag is already rcu_dereference'd.
 342  */
 343 static
 344 struct rcu_ja_node_flag *ja_node_get_nth(struct rcu_ja_node_flag *node_flag,
 345                 uint8_t n)
 346 {
 347         unsigned int type_index;
 348         struct rcu_ja_node *node;
 349         const struct rcu_ja_type *type;
 350
 351         node = ja_node_ptr(node_flag);
 352         assert(node != NULL);
 353         type_index = ja_node_type(node_flag);
 354         type = &ja_types[type_index];
 355
 356         switch (type->type_class) {
 357         case RCU_JA_LINEAR:
 358                 return ja_linear_node_get_nth(type, node, n);
 359         case RCU_JA_POOL:
 360                 return ja_pool_node_get_nth(type, node, n);
 361         case RCU_JA_PIGEON:
 362                 return ja_pigeon_node_get_nth(type, node, n);
 363         default:
 364                 assert(0);
 365                 return (void *) -1UL;
 366         }
 367 }
 368
 369 static
 370 int ja_linear_node_set_nth(const struct rcu_ja_type *type,
 371                 struct rcu_ja_node *node,
 372                 uint8_t n,
 373                 struct rcu_ja_node_flag *child_node_flag)
 374 {
 375         uint8_t nr_child;
 376         uint8_t *values, *nr_child_ptr;
 377         struct rcu_ja_node_flag **pointers;
 378         unsigned int i;
 379
 380         assert(type->type_class == RCU_JA_LINEAR || type->type_class == RCU_JA_POOL);
 381
 382         nr_child_ptr = &node->u.data[0];
 383         nr_child = *nr_child_ptr;
 384         assert(nr_child <= type->max_linear_child);
 385         assert(type->type_class != RCU_JA_LINEAR || nr_child >= type->min_child);
 386
 387         values = &node->u.data[1];
 388         for (i = 0; i < nr_child; i++) {
 389                 if (values[i] == n)
 390                         return -EEXIST;
 391         }
 392         if (nr_child >= type->max_linear_child) {
 393                 /* No space left in this node type */
 394                 return -ENOSPC;
 395         }
 396         pointers = (struct rcu_ja_node_flag **) align_ptr_size(&values[type->max_linear_child]);
 397         assert(pointers[nr_child] == NULL);
 398         rcu_assign_pointer(pointers[nr_child], child_node_flag);
 399         CMM_STORE_SHARED(values[nr_child], n);
 400         cmm_smp_wmb();  /* write value and pointer before nr_child */
 401         CMM_STORE_SHARED(*nr_child_ptr, nr_child + 1);
 402         return 0;
 403 }
 404
 405 static
 406 int ja_pool_node_set_nth(const struct rcu_ja_type *type,
 407                 struct rcu_ja_node *node,
 408                 uint8_t n,
 409                 struct rcu_ja_node_flag *child_node_flag)
 410 {
 411         struct rcu_ja_node *linear;
 412
 413         assert(type->type_class == RCU_JA_POOL);
 414         linear = (struct rcu_ja_node *)
 415                 &node->u.data[((unsigned long) n >> (CHAR_BIT - type->nr_pool_order)) << type->pool_size_order];
 416         return ja_linear_node_set_nth(type, linear, n, child_node_flag);
 417 }
 418
 419 static
 420 int ja_pigeon_node_set_nth(const struct rcu_ja_type *type,
 421                 struct rcu_ja_node *node,
 422                 uint8_t n,
 423                 struct rcu_ja_node_flag *child_node_flag)
 424 {
 425         struct rcu_ja_node_flag **ptr;
 426
 427         assert(type->type_class == RCU_JA_PIGEON);
 428         ptr = &((struct rcu_ja_node_flag **) node->u.data)[n];
 429         if (*ptr != NULL)
 430                 return -EEXIST;
 431         rcu_assign_pointer(*ptr, child_node_flag);
 432         return 0;
 433 }
 434
 435 /*
 436  * _ja_node_set_nth: set nth item within a node. Return an error
 437  * (negative error value) if it is already there.
 438  * TODO: exclusive access on node.
 439  */
 440 static
 441 int _ja_node_set_nth(struct rcu_ja_node_flag *node_flag, uint8_t n,
 442                 struct rcu_ja_node_flag *child_node_flag)
 443 {
 444         unsigned int type_index;
 445         struct rcu_ja_node *node;
 446         const struct rcu_ja_type *type;
 447
 448         node = ja_node_ptr(node_flag);
 449         assert(node != NULL);
 450         type_index = ja_node_type(node_flag);
 451         type = &ja_types[type_index];
 452
 453         switch (type->type_class) {
 454         case RCU_JA_LINEAR:
 455                 return ja_linear_node_set_nth(type, node, n,
 456                                 child_node_flag);
 457         case RCU_JA_POOL:
 458                 return ja_pool_node_set_nth(type, node, n,
 459                                 child_node_flag);
 460         case RCU_JA_PIGEON:
 461                 return ja_pigeon_node_set_nth(type, node, n,
 462                                 child_node_flag);
 463         default:
 464                 assert(0);
 465                 return -EINVAL;
 466         }
 467
 468         return 0;
 469 }
 470
 471 /*
 472  * ja_node_recompact_add: recompact a node, adding a new child.
 473  */
 474 static
 475 int ja_node_recompact_add(struct rcu_ja_node_flag **old_node_flag, uint8_t n,
 476                 struct rcu_ja_node_flag *child_node_flag)
 477
 478 {
 479         unsigned int old_type_index, new_type_index;
 480         struct rcu_ja_node *old_node, *new_node;
 481         const struct rcu_ja_type *old_type, *new_type;
 482         struct rcu_ja_node_flag *new_node_flag;
 483         unsigned int i;
 484         int ret;
 485
 486         old_node = ja_node_ptr(*old_node_flag);
 487         if (old_node == NULL) {
 488                 new_type_index = 0;
 489         } else {
 490                 old_type_index = ja_node_type(*old_node_flag);
 491                 old_type = &ja_types[old_type_index];
 492                 new_type_index = old_type_index + 1;
 493         }
 494         new_type = &ja_types[new_type_index];
 495         new_node = alloc_rcu_ja_node(new_type);
 496         if (!new_node)
 497                 return -ENOMEM;
 498         new_node_flag = ja_node_flag(new_node, new_type_index);
 499
 500         for (i = 0; i < old_type->max_child; i++) {
 501                 struct rcu_ja_node_flag *iter;
 502
 503                 iter = ja_node_get_nth(*old_node_flag, i);
 504                 if (!iter)
 505                         continue;
 506                 ret = _ja_node_set_nth(new_node_flag, i, iter);
 507                 assert(!ret);
 508         }
 509         /* add node */
 510         ret = _ja_node_set_nth(new_node_flag, n, child_node_flag);
 511         assert(!ret);
 512         /* Replace the old node with the new recompacted one */
 513         rcu_assign_pointer(*old_node_flag, new_node_flag);
 514         /* TODO: free old_node (call_rcu) */
 515         return 0;
 516 }
 517
 518 static
 519 int ja_node_set_nth(struct rcu_ja_node_flag **node_flag, uint8_t n,
 520                 struct rcu_ja_node_flag *child_node_flag)
 521 {
 522         int ret;
 523
 524         ret = _ja_node_set_nth(*node_flag, n, child_node_flag);
 525         if (ret == -ENOSPC) {
 526                 /* Not enough space in node, need to recompact */
 527                 ret = ja_node_recompact_add(node_flag, n,
 528                                 child_node_flag);
 529                 if (ret < 0)
 530                         return ret;
 531         }
 532         return ret;
 533 }