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
  33 #include "rcuja-internal.h"
  34 #include "bitfield.h"
  35
  36 enum rcu_ja_type_class {
  37         RCU_JA_LINEAR = 0,      /* Type A */
  38                         /* 32-bit: 1 to 25 children, 8 to 128 bytes */
  39                         /* 64-bit: 1 to 28 children, 16 to 256 bytes */
  40         RCU_JA_POOL = 1,        /* Type B */
  41                         /* 32-bit: 26 to 100 children, 256 to 512 bytes */
  42                         /* 64-bit: 29 to 112 children, 512 to 1024 bytes */
  43         RCU_JA_PIGEON = 2,      /* Type C */
  44                         /* 32-bit: 101 to 256 children, 1024 bytes */
  45                         /* 64-bit: 113 to 256 children, 2048 bytes */
  46         /* Leaf nodes are implicit from their height in the tree */
  47         RCU_JA_NR_TYPES,
  48 };
  49
  50 struct rcu_ja_type {
  51         enum rcu_ja_type_class type_class;
  52         uint16_t min_child;             /* minimum number of children: 1 to 256 */
  53         uint16_t max_child;             /* maximum number of children: 1 to 256 */
  54         uint16_t max_linear_child;      /* per-pool max nr. children: 1 to 256 */
  55         uint16_t order;                 /* node size is (1 << order), in bytes */
  56         uint16_t nr_pool_order;         /* number of pools */
  57         uint16_t pool_size_order;       /* pool size */
  58 };
  59
  60 /*
  61  * Number of least significant pointer bits reserved to represent the
  62  * child type.
  63  */
  64 #define JA_TYPE_BITS    3
  65 #define JA_TYPE_MAX_NR  (1U << JA_TYPE_BITS)
  66 #define JA_TYPE_MASK    (JA_TYPE_MAX_NR - 1)
  67 #define JA_PTR_MASK     (~JA_TYPE_MASK)
  68
  69 #define JA_ENTRY_PER_NODE       256UL
  70
  71 /*
  72  * Iteration on the array to find the right node size for the number of
  73  * children stops when it reaches .max_child == 256 (this is the largest
  74  * possible node size, which contains 256 children).
  75  * The min_child overlaps with the previous max_child to provide an
  76  * hysteresis loop to reallocation for patterns of cyclic add/removal
  77  * within the same node.
  78  * The node the index within the following arrays is represented on 3
  79  * bits. It identifies the node type, min/max number of children, and
  80  * the size order.
  81  * The max_child values for the RCU_JA_POOL below result from
  82  * statistical approximation: over million populations, the max_child
  83  * covers between 97% and 99% of the populations generated. Therefore, a
  84  * fallback should exist to cover the rare extreme population unbalance
  85  * cases, but it will not have a major impact on speed nor space
  86  * consumption, since those are rare cases.
  87  */
  88
  89 #if (CAA_BITS_PER_LONG < 64)
  90 /* 32-bit pointers */
  91 enum {
  92         ja_type_0_max_child = 1,
  93         ja_type_1_max_child = 3,
  94         ja_type_2_max_child = 6,
  95         ja_type_3_max_child = 12,
  96         ja_type_4_max_child = 25,
  97         ja_type_5_max_child = 48,
  98         ja_type_6_max_child = 92,
  99         ja_type_7_max_child = 256,
 100 };
 101
 102 enum {
 103         ja_type_0_max_linear_child = 1,
 104         ja_type_1_max_linear_child = 3,
 105         ja_type_2_max_linear_child = 6,
 106         ja_type_3_max_linear_child = 12,
 107         ja_type_4_max_linear_child = 25,
 108         ja_type_5_max_linear_child = 24,
 109         ja_type_6_max_linear_child = 23,
 110 };
 111
 112 enum {
 113         ja_type_5_nr_pool_order = 1,
 114         ja_type_6_nr_pool_order = 2,
 115 };
 116
 117 const struct rcu_ja_type ja_types[] = {
 118         { .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, },
 119         { .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, },
 120         { .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, },
 121         { .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, },
 122         { .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, },
 123
 124         /* Pools may fill sooner than max_child */
 125         { .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, },
 126         { .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, },
 127
 128         /*
 129          * TODO: Upon node removal below min_child, if child pool is
 130          * filled beyond capacity, we need to roll back to pigeon.
 131          */
 132         { .type_class = RCU_JA_PIGEON, .min_child = 89, .max_child = ja_type_7_max_child, .order = 10, },
 133 };
 134 #else /* !(CAA_BITS_PER_LONG < 64) */
 135 /* 64-bit pointers */
 136 enum {
 137         ja_type_0_max_child = 1,
 138         ja_type_1_max_child = 3,
 139         ja_type_2_max_child = 7,
 140         ja_type_3_max_child = 14,
 141         ja_type_4_max_child = 28,
 142         ja_type_5_max_child = 54,
 143         ja_type_6_max_child = 104,
 144         ja_type_7_max_child = 256,
 145 };
 146
 147 enum {
 148         ja_type_0_max_linear_child = 1,
 149         ja_type_1_max_linear_child = 3,
 150         ja_type_2_max_linear_child = 7,
 151         ja_type_3_max_linear_child = 14,
 152         ja_type_4_max_linear_child = 28,
 153         ja_type_5_max_linear_child = 27,
 154         ja_type_6_max_linear_child = 26,
 155 };
 156
 157 enum {
 158         ja_type_5_nr_pool_order = 1,
 159         ja_type_6_nr_pool_order = 2,
 160 };
 161
 162 const struct rcu_ja_type ja_types[] = {
 163         { .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, },
 164         { .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, },
 165         { .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, },
 166         { .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, },
 167         { .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, },
 168
 169         /* Pools may fill sooner than max_child. */
 170         { .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, },
 171         { .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, },
 172
 173         /*
 174          * TODO: Upon node removal below min_child, if child pool is
 175          * filled beyond capacity, we need to roll back to pigeon.
 176          */
 177         { .type_class = RCU_JA_PIGEON, .min_child = 101, .max_child = ja_type_7_max_child, .order = 11, },
 178 };
 179 #endif /* !(BITS_PER_LONG < 64) */
 180
 181 static inline __attribute__((unused))
 182 void static_array_size_check(void)
 183 {
 184         CAA_BUILD_BUG_ON(CAA_ARRAY_SIZE(ja_types) > JA_TYPE_MAX_NR);
 185 }
 186
 187 /*
 188  * The rcu_ja_node contains the compressed node data needed for
 189  * read-side. For linear and pool node configurations, it starts with a
 190  * byte counting the number of children in the node.  Then, the
 191  * node-specific data is placed.
 192  * The node mutex, if any is needed, protecting concurrent updated of
 193  * each node is placed in a separate hash table indexed by node address.
 194  * For the pigeon configuration, the number of children is also kept in
 195  * a separate hash table, indexed by node address, because it is only
 196  * required for updates.
 197  */
 198
 199 #define DECLARE_LINEAR_NODE(index)                                                              \
 200         struct {                                                                                \
 201                 uint8_t nr_child;                                                               \
 202                 uint8_t child_value[ja_type_## index ##_max_linear_child];                      \
 203                 struct rcu_ja_node_flag *child_ptr[ja_type_## index ##_max_linear_child];       \
 204         }
 205
 206 #define DECLARE_POOL_NODE(index)                                                                \
 207         struct {                                                                                \
 208                 struct {                                                                        \
 209                         uint8_t nr_child;                                                       \
 210                         uint8_t child_value[ja_type_## index ##_max_linear_child];              \
 211                         struct rcu_ja_node_flag *child_ptr[ja_type_## index ##_max_linear_child]; \
 212                 } linear[1U << ja_type_## index ##_nr_pool_order];                              \
 213         }
 214
 215 struct rcu_ja_node {
 216         union {
 217                 /* Linear configuration */
 218                 DECLARE_LINEAR_NODE(0) conf_0;
 219                 DECLARE_LINEAR_NODE(1) conf_1;
 220                 DECLARE_LINEAR_NODE(2) conf_2;
 221                 DECLARE_LINEAR_NODE(3) conf_3;
 222                 DECLARE_LINEAR_NODE(4) conf_4;
 223
 224                 /* Pool configuration */
 225                 DECLARE_POOL_NODE(5) conf_5;
 226                 DECLARE_POOL_NODE(6) conf_6;
 227
 228                 /* Pigeon configuration */
 229                 struct {
 230                         struct rcu_ja_node_flag *child[ja_type_7_max_child];
 231                 } conf_7;
 232                 /* data aliasing nodes for computed accesses */
 233                 uint8_t data[sizeof(struct rcu_ja_node_flag *) * ja_type_7_max_child];
 234         } u;
 235 };
 236
 237 static
 238 struct rcu_ja_node_flag *ja_node_flag(struct rcu_ja_node *node,
 239                 unsigned int type)
 240 {
 241         assert(type < RCU_JA_NR_TYPES);
 242         return (struct rcu_ja_node_flag *) (((unsigned long) node) | type);
 243 }
 244
 245 static
 246 unsigned int ja_node_type(struct rcu_ja_node_flag *node)
 247 {
 248         unsigned int type;
 249
 250         type = (unsigned int) ((unsigned long) node & JA_TYPE_MASK);
 251         assert(type < RCU_JA_NR_TYPES);
 252         return type;
 253 }
 254
 255 static
 256 struct rcu_ja_node *ja_node_ptr(struct rcu_ja_node_flag *node)
 257 {
 258         return (struct rcu_ja_node *) (((unsigned long) node) | JA_PTR_MASK);
 259 }
 260
 261 struct rcu_ja_node *alloc_rcu_ja_node(const struct rcu_ja_type *ja_type)
 262 {
 263         return calloc(1U << ja_type->order, sizeof(char));
 264 }
 265
 266 void free_rcu_ja_node(struct rcu_ja_node *node)
 267 {
 268         free(node);
 269 }
 270
 271 #define __JA_ALIGN_MASK(v, mask)        (((v) + (mask)) & ~(mask))
 272 #define JA_ALIGN(v, align)              __JA_ALIGN_MASK(v, (typeof(v)) (align) - 1)
 273 #define __JA_FLOOR_MASK(v, mask)        ((v) & ~(mask))
 274 #define JA_FLOOR(v, align)              __JA_FLOOR_MASK(v, (typeof(v)) (align) - 1)
 275
 276 static
 277 uint8_t *align_ptr_size(uint8_t *ptr)
 278 {
 279         return (uint8_t *) JA_ALIGN((unsigned long) ptr, sizeof(void *));
 280 }
 281
 282 /*
 283  * The order in which values and pointers are does does not matter: if
 284  * a value is missing, we return NULL. If a value is there, but its
 285  * associated pointers is still NULL, we return NULL too.
 286  */
 287 static
 288 struct rcu_ja_node_flag *ja_linear_node_get_nth(const struct rcu_ja_type *type,
 289                 struct rcu_ja_node *node,
 290                 uint8_t n)
 291 {
 292         uint8_t nr_child;
 293         uint8_t *values;
 294         struct rcu_ja_node_flag **pointers;
 295         struct rcu_ja_node_flag *ptr;
 296         unsigned int i;
 297
 298         assert(type->type_class == RCU_JA_LINEAR || type->type_class == RCU_JA_POOL);
 299
 300         nr_child = CMM_LOAD_SHARED(node->u.data[0]);
 301         cmm_smp_rmb();  /* read nr_child before values and pointers */
 302         assert(nr_child <= type->max_linear_child);
 303         assert(type->type_class != RCU_JA_LINEAR || nr_child >= type->min_child);
 304
 305         values = &node->u.data[1];
 306         for (i = 0; i < nr_child; i++) {
 307                 if (CMM_LOAD_SHARED(values[i]) == n)
 308                         break;
 309         }
 310         if (i >= nr_child)
 311                 return NULL;
 312         pointers = (struct rcu_ja_node_flag **) align_ptr_size(&values[type->max_linear_child]);
 313         ptr = rcu_dereference(pointers[i]);
 314         assert(ja_node_ptr(ptr) != NULL);
 315         return ptr;
 316 }
 317
 318 static
 319 struct rcu_ja_node_flag *ja_pool_node_get_nth(const struct rcu_ja_type *type,
 320                 struct rcu_ja_node *node,
 321                 uint8_t n)
 322 {
 323         struct rcu_ja_node *linear;
 324
 325         assert(type->type_class == RCU_JA_POOL);
 326         linear = (struct rcu_ja_node *)
 327                 &node->u.data[((unsigned long) n >> (CHAR_BIT - type->nr_pool_order)) << type->pool_size_order];
 328         return ja_linear_node_get_nth(type, linear, n);
 329 }
 330
 331 static
 332 struct rcu_ja_node_flag *ja_pigeon_node_get_nth(const struct rcu_ja_type *type,
 333                 struct rcu_ja_node *node,
 334                 uint8_t n)
 335 {
 336         assert(type->type_class == RCU_JA_PIGEON);
 337         return rcu_dereference(((struct rcu_ja_node_flag **) node->u.data)[n]);
 338 }
 339
 340 /*
 341  * ja_node_get_nth: get nth item from a node.
 342  * node_flag is already rcu_dereference'd.
 343  */
 344 static
 345 struct rcu_ja_node_flag *ja_node_get_nth(struct rcu_ja_node_flag *node_flag,
 346                 uint8_t n)
 347 {
 348         unsigned int type_index;
 349         struct rcu_ja_node *node;
 350         const struct rcu_ja_type *type;
 351
 352         node = ja_node_ptr(node_flag);
 353         assert(node != NULL);
 354         type_index = ja_node_type(node_flag);
 355         type = &ja_types[type_index];
 356
 357         switch (type->type_class) {
 358         case RCU_JA_LINEAR:
 359                 return ja_linear_node_get_nth(type, node, n);
 360         case RCU_JA_POOL:
 361                 return ja_pool_node_get_nth(type, node, n);
 362         case RCU_JA_PIGEON:
 363                 return ja_pigeon_node_get_nth(type, node, n);
 364         default:
 365                 assert(0);
 366                 return (void *) -1UL;
 367         }
 368 }
 369
 370 static
 371 int ja_linear_node_set_nth(const struct rcu_ja_type *type,
 372                 struct rcu_ja_node *node,
 373                 uint8_t n,
 374                 struct rcu_ja_node_flag *child_node_flag)
 375 {
 376         uint8_t nr_child;
 377         uint8_t *values, *nr_child_ptr;
 378         struct rcu_ja_node_flag **pointers;
 379         unsigned int i;
 380
 381         assert(type->type_class == RCU_JA_LINEAR || type->type_class == RCU_JA_POOL);
 382
 383         nr_child_ptr = &node->u.data[0];
 384         nr_child = *nr_child_ptr;
 385         assert(nr_child <= type->max_linear_child);
 386         assert(type->type_class != RCU_JA_LINEAR || nr_child >= type->min_child);
 387
 388         values = &node->u.data[1];
 389         for (i = 0; i < nr_child; i++) {
 390                 if (values[i] == n)
 391                         return -EEXIST;
 392         }
 393         if (nr_child >= type->max_linear_child) {
 394                 /* No space left in this node type */
 395                 return -ENOSPC;
 396         }
 397         pointers = (struct rcu_ja_node_flag **) align_ptr_size(&values[type->max_linear_child]);
 398         assert(pointers[nr_child] == NULL);
 399         rcu_assign_pointer(pointers[nr_child], child_node_flag);
 400         CMM_STORE_SHARED(values[nr_child], n);
 401         cmm_smp_wmb();  /* write value and pointer before nr_child */
 402         CMM_STORE_SHARED(*nr_child_ptr, nr_child + 1);
 403         return 0;
 404 }
 405
 406 static
 407 int ja_pool_node_set_nth(const struct rcu_ja_type *type,
 408                 struct rcu_ja_node *node,
 409                 uint8_t n,
 410                 struct rcu_ja_node_flag *child_node_flag)
 411 {
 412         struct rcu_ja_node *linear;
 413
 414         assert(type->type_class == RCU_JA_POOL);
 415         linear = (struct rcu_ja_node *)
 416                 &node->u.data[((unsigned long) n >> (CHAR_BIT - type->nr_pool_order)) << type->pool_size_order];
 417         return ja_linear_node_set_nth(type, linear, n, child_node_flag);
 418 }
 419
 420 static
 421 int ja_pigeon_node_set_nth(const struct rcu_ja_type *type,
 422                 struct rcu_ja_node *node,
 423                 uint8_t n,
 424                 struct rcu_ja_node_flag *child_node_flag)
 425 {
 426         struct rcu_ja_node_flag **ptr;
 427
 428         assert(type->type_class == RCU_JA_PIGEON);
 429         ptr = &((struct rcu_ja_node_flag **) node->u.data)[n];
 430         if (*ptr != NULL)
 431                 return -EEXIST;
 432         rcu_assign_pointer(*ptr, child_node_flag);
 433         return 0;
 434 }
 435
 436 /*
 437  * _ja_node_set_nth: set nth item within a node. Return an error
 438  * (negative error value) if it is already there.
 439  * TODO: exclusive access on node.
 440  */
 441 static
 442 int _ja_node_set_nth(struct rcu_ja_node_flag *node_flag, uint8_t n,
 443                 struct rcu_ja_node_flag *child_node_flag)
 444 {
 445         unsigned int type_index;
 446         struct rcu_ja_node *node;
 447         const struct rcu_ja_type *type;
 448
 449         node = ja_node_ptr(node_flag);
 450         assert(node != NULL);
 451         type_index = ja_node_type(node_flag);
 452         type = &ja_types[type_index];
 453
 454         switch (type->type_class) {
 455         case RCU_JA_LINEAR:
 456                 return ja_linear_node_set_nth(type, node, n,
 457                                 child_node_flag);
 458         case RCU_JA_POOL:
 459                 return ja_pool_node_set_nth(type, node, n,
 460                                 child_node_flag);
 461         case RCU_JA_PIGEON:
 462                 return ja_pigeon_node_set_nth(type, node, n,
 463                                 child_node_flag);
 464         default:
 465                 assert(0);
 466                 return -EINVAL;
 467         }
 468
 469         return 0;
 470 }
 471
 472 /*
 473  * ja_node_recompact_add: recompact a node, adding a new child.
 474  */
 475 static
 476 int ja_node_recompact_add(struct rcu_ja_node_flag **old_node_flag, uint8_t n,
 477                 struct rcu_ja_node_flag *child_node_flag)
 478
 479 {
 480         unsigned int old_type_index, new_type_index;
 481         struct rcu_ja_node *old_node, *new_node;
 482         const struct rcu_ja_type *old_type, *new_type;
 483         struct rcu_ja_node_flag *new_node_flag;
 484         unsigned int i;
 485         int ret;
 486
 487         old_node = ja_node_ptr(*old_node_flag);
 488         if (old_node == NULL) {
 489                 new_type_index = 0;
 490         } else {
 491                 old_type_index = ja_node_type(*old_node_flag);
 492                 old_type = &ja_types[old_type_index];
 493                 new_type_index = old_type_index + 1;
 494         }
 495         new_type = &ja_types[new_type_index];
 496         new_node = alloc_rcu_ja_node(new_type);
 497         if (!new_node)
 498                 return -ENOMEM;
 499         new_node_flag = ja_node_flag(new_node, new_type_index);
 500
 501         for (i = 0; i < old_type->max_child; i++) {
 502                 struct rcu_ja_node_flag *iter;
 503
 504                 iter = ja_node_get_nth(*old_node_flag, i);
 505                 if (!iter)
 506                         continue;
 507                 ret = _ja_node_set_nth(new_node_flag, i, iter);
 508                 assert(!ret);
 509         }
 510         /* add node */
 511         ret = _ja_node_set_nth(new_node_flag, n, child_node_flag);
 512         assert(!ret);
 513         /* Replace the old node with the new recompacted one */
 514         rcu_assign_pointer(*old_node_flag, new_node_flag);
 515         /* TODO: free old_node (call_rcu) */
 516         return 0;
 517 }
 518
 519 static
 520 int ja_node_set_nth(struct rcu_ja_node_flag **node_flag, uint8_t n,
 521                 struct rcu_ja_node_flag *child_node_flag)
 522 {
 523         int ret;
 524
 525         ret = _ja_node_set_nth(*node_flag, n, child_node_flag);
 526         if (ret == -ENOSPC) {
 527                 /* Not enough space in node, need to recompact */
 528                 ret = ja_node_recompact_add(node_flag, n,
 529                                 child_node_flag);
 530                 if (ret < 0)
 531                         return ret;
 532         }
 533         return ret;
 534 }