From: Mathieu Desnoyers Date: Mon, 5 Dec 2011 23:41:49 +0000 (-0500) Subject: Merge branch 'master' into urcu/ht-shrink X-Git-Tag: v0.7.0~43^2~20 X-Git-Url: https://git.liburcu.org/?p=urcu.git;a=commitdiff_plain;h=1644042068ae8de26c9f6b85dcb2a5cd0b0f7608;hp=9056e3b82eab303fcfc1262fd9889b23173ee282 Merge branch 'master' into urcu/ht-shrink --- diff --git a/Makefile.am b/Makefile.am index a7d3316..e1bf095 100644 --- a/Makefile.am +++ b/Makefile.am @@ -8,19 +8,21 @@ AM_CFLAGS=-Wall SUBDIRS = . tests include_HEADERS = urcu.h urcu-bp.h urcu-call-rcu.h urcu-defer.h \ - urcu-pointer.h urcu-qsbr.h + urcu-pointer.h urcu-qsbr.h urcu-flavor.h nobase_dist_include_HEADERS = urcu/compiler.h urcu/hlist.h urcu/list.h \ urcu/rculist.h urcu/rcuhlist.h urcu/system.h urcu/futex.h \ urcu/uatomic/generic.h urcu/arch/generic.h urcu/wfstack.h \ urcu/wfqueue.h urcu/rculfstack.h urcu/rculfqueue.h \ urcu/ref.h urcu/map/*.h urcu/static/*.h urcu/cds.h \ - urcu/urcu_ref.h urcu/urcu-futex.h urcu/uatomic_arch.h + urcu/urcu_ref.h urcu/urcu-futex.h urcu/uatomic_arch.h \ + urcu/rculfhash.h nobase_nodist_include_HEADERS = urcu/arch.h urcu/uatomic.h urcu/config.h EXTRA_DIST = $(top_srcdir)/urcu/arch/*.h $(top_srcdir)/urcu/uatomic/*.h \ gpl-2.0.txt lgpl-2.1.txt lgpl-relicensing.txt \ README LICENSE compat_arch_x86.c \ urcu-call-rcu-impl.h urcu-defer-impl.h \ + rculfhash-internal.h \ ChangeLog API.txt \ $(top_srcdir)/tests/*.sh @@ -34,6 +36,9 @@ if COMPAT_FUTEX COMPAT+=compat_futex.c endif +RCULFHASH = rculfhash.c rculfhash-mm-order.c rculfhash-mm-chunk.c \ + rculfhash-mm-mmap.c + lib_LTLIBRARIES = liburcu-common.la \ liburcu.la liburcu-qsbr.la \ liburcu-mb.la liburcu-signal.la liburcu-bp.la \ @@ -62,7 +67,7 @@ liburcu_signal_la_LIBADD = liburcu-common.la liburcu_bp_la_SOURCES = urcu-bp.c urcu-pointer.c $(COMPAT) liburcu_bp_la_LIBADD = liburcu-common.la -liburcu_cds_la_SOURCES = rculfqueue.c rculfstack.c $(COMPAT) +liburcu_cds_la_SOURCES = rculfqueue.c rculfstack.c $(RCULFHASH) $(COMPAT) liburcu_cds_la_LIBADD = liburcu-common.la pkgconfigdir = $(libdir)/pkgconfig diff --git a/rculfhash-internal.h b/rculfhash-internal.h new file mode 100644 index 0000000..a8a1ce4 --- /dev/null +++ b/rculfhash-internal.h @@ -0,0 +1,177 @@ +#ifndef _URCU_RCULFHASH_INTERNAL_H +#define _URCU_RCULFHASH_INTERNAL_H + +/* + * urcu/rculfhash-internal.h + * + * Internal header for Lock-Free RCU Hash Table + * + * Copyright 2011 - Mathieu Desnoyers + * Copyright 2011 - Lai Jiangshan + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include + +#ifdef DEBUG +#define dbg_printf(fmt, args...) printf("[debug rculfhash] " fmt, ## args) +#else +#define dbg_printf(fmt, args...) +#endif + +#if (CAA_BITS_PER_LONG == 32) +#define MAX_TABLE_ORDER 32 +#else +#define MAX_TABLE_ORDER 64 +#endif + +#define MAX_CHUNK_TABLE (1UL << 10) + +#ifndef min +#define min(a, b) ((a) < (b) ? (a) : (b)) +#endif + +#ifndef max +#define max(a, b) ((a) > (b) ? (a) : (b)) +#endif + +struct ht_items_count; + +/* + * cds_lfht: Top-level data structure representing a lock-free hash + * table. Defined in the implementation file to make it be an opaque + * cookie to users. + * + * The fields used in fast-paths are placed near the end of the + * structure, because we need to have a variable-sized union to contain + * the mm plugin fields, which are used in the fast path. + */ +struct cds_lfht { + /* Initial configuration items */ + unsigned long max_nr_buckets; + const struct cds_lfht_mm_type *mm; /* memory management plugin */ + const struct rcu_flavor_struct *flavor; /* RCU flavor */ + + long count; /* global approximate item count */ + + /* + * We need to put the work threads offline (QSBR) when taking this + * mutex, because we use synchronize_rcu within this mutex critical + * section, which waits on read-side critical sections, and could + * therefore cause grace-period deadlock if we hold off RCU G.P. + * completion. + */ + pthread_mutex_t resize_mutex; /* resize mutex: add/del mutex */ + pthread_attr_t *resize_attr; /* Resize threads attributes */ + unsigned int in_progress_resize, in_progress_destroy; + unsigned long resize_target; + int resize_initiated; + + /* + * Variables needed for add and remove fast-paths. + */ + int flags; + unsigned long min_alloc_buckets_order; + unsigned long min_nr_alloc_buckets; + struct ht_items_count *split_count; /* split item count */ + + /* + * Variables needed for the lookup, add and remove fast-paths. + */ + unsigned long size; /* always a power of 2, shared (RCU) */ + /* + * bucket_at pointer is kept here to skip the extra level of + * dereference needed to get to "mm" (this is a fast-path). + */ + struct cds_lfht_node *(*bucket_at)(struct cds_lfht *ht, + unsigned long index); + /* + * Dynamic length "tbl_chunk" needs to be at the end of + * cds_lfht. + */ + union { + /* + * Contains the per order-index-level bucket node table. + * The size of each bucket node table is half the number + * of hashes contained in this order (except for order 0). + * The minimum allocation buckets size parameter allows + * combining the bucket node arrays of the lowermost + * levels to improve cache locality for small index orders. + */ + struct cds_lfht_node *tbl_order[MAX_TABLE_ORDER]; + + /* + * Contains the bucket node chunks. The size of each + * bucket node chunk is ->min_alloc_size (we avoid to + * allocate chunks with different size). Chunks improve + * cache locality for small index orders, and are more + * friendly with environments where allocation of large + * contiguous memory areas is challenging due to memory + * fragmentation concerns or inability to use virtual + * memory addressing. + */ + struct cds_lfht_node *tbl_chunk[0]; + + /* + * Memory mapping with room for all possible buckets. + * Their memory is allocated when needed. + */ + struct cds_lfht_node *tbl_mmap; + }; + /* + * End of variables needed for the lookup, add and remove + * fast-paths. + */ +}; + +extern unsigned int fls_ulong(unsigned long x); +extern int get_count_order_ulong(unsigned long x); + +#ifdef POISON_FREE +#define poison_free(ptr) \ + do { \ + if (ptr) { \ + memset(ptr, 0x42, sizeof(*(ptr))); \ + free(ptr); \ + } \ + } while (0) +#else +#define poison_free(ptr) free(ptr) +#endif + +static inline +struct cds_lfht *__default_alloc_cds_lfht( + const struct cds_lfht_mm_type *mm, + unsigned long cds_lfht_size, + unsigned long min_nr_alloc_buckets, + unsigned long max_nr_buckets) +{ + struct cds_lfht *ht; + + ht = calloc(1, cds_lfht_size); + assert(ht); + + ht->mm = mm; + ht->bucket_at = mm->bucket_at; + ht->min_nr_alloc_buckets = min_nr_alloc_buckets; + ht->min_alloc_buckets_order = + get_count_order_ulong(min_nr_alloc_buckets); + ht->max_nr_buckets = max_nr_buckets; + + return ht; +} + +#endif /* _URCU_RCULFHASH_INTERNAL_H */ diff --git a/rculfhash-mm-chunk.c b/rculfhash-mm-chunk.c new file mode 100644 index 0000000..a7a9b76 --- /dev/null +++ b/rculfhash-mm-chunk.c @@ -0,0 +1,97 @@ +/* + * rculfhash-mm-chunk.c + * + * Chunk based memory management for Lock-Free RCU Hash Table + * + * Copyright 2011 - Lai Jiangshan + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include +#include + +static +void cds_lfht_alloc_bucket_table(struct cds_lfht *ht, unsigned long order) +{ + if (order == 0) { + ht->tbl_chunk[0] = calloc(ht->min_nr_alloc_buckets, + sizeof(struct cds_lfht_node)); + assert(ht->tbl_chunk[0]); + } else if (order > ht->min_alloc_buckets_order) { + unsigned long i, len = 1UL << (order - 1 - ht->min_alloc_buckets_order); + + for (i = len; i < 2 * len; i++) { + ht->tbl_chunk[i] = calloc(ht->min_nr_alloc_buckets, + sizeof(struct cds_lfht_node)); + assert(ht->tbl_chunk[i]); + } + } + /* Nothing to do for 0 < order && order <= ht->min_alloc_buckets_order */ +} + +/* + * cds_lfht_free_bucket_table() should be called with decreasing order. + * When cds_lfht_free_bucket_table(0) is called, it means the whole + * lfht is destroyed. + */ +static +void cds_lfht_free_bucket_table(struct cds_lfht *ht, unsigned long order) +{ + if (order == 0) + poison_free(ht->tbl_chunk[0]); + else if (order > ht->min_alloc_buckets_order) { + unsigned long i, len = 1UL << (order - 1 - ht->min_alloc_buckets_order); + + for (i = len; i < 2 * len; i++) + poison_free(ht->tbl_chunk[i]); + } + /* Nothing to do for 0 < order && order <= ht->min_alloc_buckets_order */ +} + +static +struct cds_lfht_node *bucket_at(struct cds_lfht *ht, unsigned long index) +{ + unsigned long chunk, offset; + + chunk = index >> ht->min_alloc_buckets_order; + offset = index & (ht->min_nr_alloc_buckets - 1); + return &ht->tbl_chunk[chunk][offset]; +} + +static +struct cds_lfht *alloc_cds_lfht(unsigned long min_nr_alloc_buckets, + unsigned long max_nr_buckets) +{ + unsigned long nr_chunks, cds_lfht_size; + + min_nr_alloc_buckets = max(min_nr_alloc_buckets, + max_nr_buckets / MAX_CHUNK_TABLE); + nr_chunks = max_nr_buckets / min_nr_alloc_buckets; + cds_lfht_size = offsetof(struct cds_lfht, tbl_chunk) + + sizeof(struct cds_lfht_node *) * nr_chunks; + cds_lfht_size = max(cds_lfht_size, sizeof(struct cds_lfht)); + + return __default_alloc_cds_lfht( + &cds_lfht_mm_chunk, cds_lfht_size, + min_nr_alloc_buckets, max_nr_buckets); +} + +const struct cds_lfht_mm_type cds_lfht_mm_chunk = { + .alloc_cds_lfht = alloc_cds_lfht, + .alloc_bucket_table = cds_lfht_alloc_bucket_table, + .free_bucket_table = cds_lfht_free_bucket_table, + .bucket_at = bucket_at, +}; diff --git a/rculfhash-mm-mmap.c b/rculfhash-mm-mmap.c new file mode 100644 index 0000000..4554ed6 --- /dev/null +++ b/rculfhash-mm-mmap.c @@ -0,0 +1,156 @@ +/* + * rculfhash-mm-mmap.c + * + * mmap/reservation based memory management for Lock-Free RCU Hash Table + * + * Copyright 2011 - Lai Jiangshan + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include +#include +#include "rculfhash-internal.h" + +/* reserve inaccessible memory space without allocation any memory */ +static void *memory_map(size_t length) +{ + void *ret = mmap(NULL, length, PROT_NONE, + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); + + assert(ret != MAP_FAILED); + return ret; +} + +static void memory_unmap(void *ptr, size_t length) +{ + int ret __attribute__((unused)); + + ret = munmap(ptr, length); + + assert(ret == 0); +} + +static void memory_populate(void *ptr, size_t length) +{ + void *ret __attribute__((unused)); + + ret = mmap(ptr, length, PROT_READ | PROT_WRITE, + MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); + + assert(ret == ptr); +} + +/* + * Discard garbage memory and avoid system save it when try to swap it out. + * Make it still reserved, inaccessible. + */ +static void memory_discard(void *ptr, size_t length) +{ + void *ret __attribute__((unused)); + + ret = mmap(ptr, length, PROT_NONE, + MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); + + assert(ret == ptr); +} + +static +void cds_lfht_alloc_bucket_table(struct cds_lfht *ht, unsigned long order) +{ + if (order == 0) { + if (ht->min_nr_alloc_buckets == ht->max_nr_buckets) { + /* small table */ + ht->tbl_mmap = calloc(ht->max_nr_buckets, + sizeof(*ht->tbl_mmap)); + assert(ht->tbl_mmap); + return; + } + /* large table */ + ht->tbl_mmap = memory_map(ht->max_nr_buckets + * sizeof(*ht->tbl_mmap)); + memory_populate(ht->tbl_mmap, + ht->min_nr_alloc_buckets * sizeof(*ht->tbl_mmap)); + } else if (order > ht->min_alloc_buckets_order) { + /* large table */ + unsigned long len = 1UL << (order - 1); + + assert(ht->min_nr_alloc_buckets < ht->max_nr_buckets); + memory_populate(ht->tbl_mmap + len, + len * sizeof(*ht->tbl_mmap)); + } + /* Nothing to do for 0 < order && order <= ht->min_alloc_buckets_order */ +} + +/* + * cds_lfht_free_bucket_table() should be called with decreasing order. + * When cds_lfht_free_bucket_table(0) is called, it means the whole + * lfht is destroyed. + */ +static +void cds_lfht_free_bucket_table(struct cds_lfht *ht, unsigned long order) +{ + if (order == 0) { + if (ht->min_nr_alloc_buckets == ht->max_nr_buckets) { + /* small table */ + poison_free(ht->tbl_mmap); + return; + } + /* large table */ + memory_unmap(ht->tbl_mmap, + ht->max_nr_buckets * sizeof(*ht->tbl_mmap)); + } else if (order > ht->min_alloc_buckets_order) { + /* large table */ + unsigned long len = 1UL << (order - 1); + + assert(ht->min_nr_alloc_buckets < ht->max_nr_buckets); + memory_discard(ht->tbl_mmap + len, len * sizeof(*ht->tbl_mmap)); + } + /* Nothing to do for 0 < order && order <= ht->min_alloc_buckets_order */ +} + +static +struct cds_lfht_node *bucket_at(struct cds_lfht *ht, unsigned long index) +{ + return &ht->tbl_mmap[index]; +} + +static +struct cds_lfht *alloc_cds_lfht(unsigned long min_nr_alloc_buckets, + unsigned long max_nr_buckets) +{ + unsigned long page_bucket_size; + + page_bucket_size = getpagesize() / sizeof(struct cds_lfht_node); + if (max_nr_buckets <= page_bucket_size) { + /* small table */ + min_nr_alloc_buckets = max_nr_buckets; + } else { + /* large table */ + min_nr_alloc_buckets = max(min_nr_alloc_buckets, + page_bucket_size); + } + + return __default_alloc_cds_lfht( + &cds_lfht_mm_mmap, sizeof(struct cds_lfht), + min_nr_alloc_buckets, max_nr_buckets); +} + +const struct cds_lfht_mm_type cds_lfht_mm_mmap = { + .alloc_cds_lfht = alloc_cds_lfht, + .alloc_bucket_table = cds_lfht_alloc_bucket_table, + .free_bucket_table = cds_lfht_free_bucket_table, + .bucket_at = bucket_at, +}; diff --git a/rculfhash-mm-order.c b/rculfhash-mm-order.c new file mode 100644 index 0000000..237d4cf --- /dev/null +++ b/rculfhash-mm-order.c @@ -0,0 +1,90 @@ +/* + * rculfhash-mm-order.c + * + * Order based memory management for Lock-Free RCU Hash Table + * + * Copyright 2011 - Mathieu Desnoyers + * Copyright 2011 - Lai Jiangshan + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include + +static +void cds_lfht_alloc_bucket_table(struct cds_lfht *ht, unsigned long order) +{ + if (order == 0) { + ht->tbl_order[0] = calloc(ht->min_nr_alloc_buckets, + sizeof(struct cds_lfht_node)); + assert(ht->tbl_order[0]); + } else if (order > ht->min_alloc_buckets_order) { + ht->tbl_order[order] = calloc(1UL << (order -1), + sizeof(struct cds_lfht_node)); + assert(ht->tbl_order[order]); + } + /* Nothing to do for 0 < order && order <= ht->min_alloc_buckets_order */ +} + +/* + * cds_lfht_free_bucket_table() should be called with decreasing order. + * When cds_lfht_free_bucket_table(0) is called, it means the whole + * lfht is destroyed. + */ +static +void cds_lfht_free_bucket_table(struct cds_lfht *ht, unsigned long order) +{ + if (order == 0) + poison_free(ht->tbl_order[0]); + else if (order > ht->min_alloc_buckets_order) + poison_free(ht->tbl_order[order]); + /* Nothing to do for 0 < order && order <= ht->min_alloc_buckets_order */ +} + +static +struct cds_lfht_node *bucket_at(struct cds_lfht *ht, unsigned long index) +{ + unsigned long order; + + if (index < ht->min_nr_alloc_buckets) { + dbg_printf("bucket index %lu order 0 aridx 0\n", index); + return &ht->tbl_order[0][index]; + } + /* + * equivalent to get_count_order_ulong(index + 1), but optimizes + * away the non-existing 0 special-case for + * get_count_order_ulong. + */ + order = fls_ulong(index); + dbg_printf("bucket index %lu order %lu aridx %lu\n", + index, order, index & ((1UL << (order - 1)) - 1)); + return &ht->tbl_order[order][index & ((1UL << (order - 1)) - 1)]; +} + +static +struct cds_lfht *alloc_cds_lfht(unsigned long min_nr_alloc_buckets, + unsigned long max_nr_buckets) +{ + return __default_alloc_cds_lfht( + &cds_lfht_mm_order, sizeof(struct cds_lfht), + min_nr_alloc_buckets, max_nr_buckets); +} + +const struct cds_lfht_mm_type cds_lfht_mm_order = { + .alloc_cds_lfht = alloc_cds_lfht, + .alloc_bucket_table = cds_lfht_alloc_bucket_table, + .free_bucket_table = cds_lfht_free_bucket_table, + .bucket_at = bucket_at, +}; diff --git a/rculfhash.c b/rculfhash.c new file mode 100644 index 0000000..d04451f --- /dev/null +++ b/rculfhash.c @@ -0,0 +1,1781 @@ +/* + * rculfhash.c + * + * Userspace RCU library - Lock-Free Resizable RCU Hash Table + * + * Copyright 2010-2011 - Mathieu Desnoyers + * Copyright 2011 - Lai Jiangshan + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/* + * Based on the following articles: + * - Ori Shalev and Nir Shavit. Split-ordered lists: Lock-free + * extensible hash tables. J. ACM 53, 3 (May 2006), 379-405. + * - Michael, M. M. High performance dynamic lock-free hash tables + * and list-based sets. In Proceedings of the fourteenth annual ACM + * symposium on Parallel algorithms and architectures, ACM Press, + * (2002), 73-82. + * + * Some specificities of this Lock-Free Resizable RCU Hash Table + * implementation: + * + * - RCU read-side critical section allows readers to perform hash + * table lookups and use the returned objects safely by delaying + * memory reclaim of a grace period. + * - Add and remove operations are lock-free, and do not need to + * allocate memory. They need to be executed within RCU read-side + * critical section to ensure the objects they read are valid and to + * deal with the cmpxchg ABA problem. + * - add and add_unique operations are supported. add_unique checks if + * the node key already exists in the hash table. It ensures no key + * duplicata exists. + * - The resize operation executes concurrently with add/remove/lookup. + * - Hash table nodes are contained within a split-ordered list. This + * list is ordered by incrementing reversed-bits-hash value. + * - An index of bucket nodes is kept. These bucket nodes are the hash + * table "buckets", and they are also chained together in the + * split-ordered list, which allows recursive expansion. + * - The resize operation for small tables only allows expanding the hash table. + * It is triggered automatically by detecting long chains in the add + * operation. + * - The resize operation for larger tables (and available through an + * API) allows both expanding and shrinking the hash table. + * - Split-counters are used to keep track of the number of + * nodes within the hash table for automatic resize triggering. + * - Resize operation initiated by long chain detection is executed by a + * call_rcu thread, which keeps lock-freedom of add and remove. + * - Resize operations are protected by a mutex. + * - The removal operation is split in two parts: first, a "removed" + * flag is set in the next pointer within the node to remove. Then, + * a "garbage collection" is performed in the bucket containing the + * removed node (from the start of the bucket up to the removed node). + * All encountered nodes with "removed" flag set in their next + * pointers are removed from the linked-list. If the cmpxchg used for + * removal fails (due to concurrent garbage-collection or concurrent + * add), we retry from the beginning of the bucket. This ensures that + * the node with "removed" flag set is removed from the hash table + * (not visible to lookups anymore) before the RCU read-side critical + * section held across removal ends. Furthermore, this ensures that + * the node with "removed" flag set is removed from the linked-list + * before its memory is reclaimed. Only the thread which removal + * successfully set the "removed" flag (with a cmpxchg) into a node's + * next pointer is considered to have succeeded its removal (and thus + * owns the node to reclaim). Because we garbage-collect starting from + * an invariant node (the start-of-bucket bucket node) up to the + * "removed" node (or find a reverse-hash that is higher), we are sure + * that a successful traversal of the chain leads to a chain that is + * present in the linked-list (the start node is never removed) and + * that is does not contain the "removed" node anymore, even if + * concurrent delete/add operations are changing the structure of the + * list concurrently. + * - The add operation performs gargage collection of buckets if it + * encounters nodes with removed flag set in the bucket where it wants + * to add its new node. This ensures lock-freedom of add operation by + * helping the remover unlink nodes from the list rather than to wait + * for it do to so. + * - A RCU "order table" indexed by log2(hash index) is copied and + * expanded by the resize operation. This order table allows finding + * the "bucket node" tables. + * - There is one bucket node table per hash index order. The size of + * each bucket node table is half the number of hashes contained in + * this order (except for order 0). + * - synchronzie_rcu is used to garbage-collect the old bucket node table. + * - The per-order bucket node tables contain a compact version of the + * hash table nodes. These tables are invariant after they are + * populated into the hash table. + * + * Bucket node tables: + * + * hash table hash table the last all bucket node tables + * order size bucket node 0 1 2 3 4 5 6(index) + * table size + * 0 1 1 1 + * 1 2 1 1 1 + * 2 4 2 1 1 2 + * 3 8 4 1 1 2 4 + * 4 16 8 1 1 2 4 8 + * 5 32 16 1 1 2 4 8 16 + * 6 64 32 1 1 2 4 8 16 32 + * + * When growing/shrinking, we only focus on the last bucket node table + * which size is (!order ? 1 : (1 << (order -1))). + * + * Example for growing/shrinking: + * grow hash table from order 5 to 6: init the index=6 bucket node table + * shrink hash table from order 6 to 5: fini the index=6 bucket node table + * + * A bit of ascii art explanation: + * + * Order index is the off-by-one compare to the actual power of 2 because + * we use index 0 to deal with the 0 special-case. + * + * This shows the nodes for a small table ordered by reversed bits: + * + * bits reverse + * 0 000 000 + * 4 100 001 + * 2 010 010 + * 6 110 011 + * 1 001 100 + * 5 101 101 + * 3 011 110 + * 7 111 111 + * + * This shows the nodes in order of non-reversed bits, linked by + * reversed-bit order. + * + * order bits reverse + * 0 0 000 000 + * 1 | 1 001 100 <- + * 2 | | 2 010 010 <- | + * | | | 3 011 110 | <- | + * 3 -> | | | 4 100 001 | | + * -> | | 5 101 101 | + * -> | 6 110 011 + * -> 7 111 111 + */ + +#define _LGPL_SOURCE +#include +#include +#include +#include +#include +#include + +#include "config.h" +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +/* + * Split-counters lazily update the global counter each 1024 + * addition/removal. It automatically keeps track of resize required. + * We use the bucket length as indicator for need to expand for small + * tables and machines lacking per-cpu data suppport. + */ +#define COUNT_COMMIT_ORDER 10 +#define DEFAULT_SPLIT_COUNT_MASK 0xFUL +#define CHAIN_LEN_TARGET 1 +#define CHAIN_LEN_RESIZE_THRESHOLD 3 + +/* + * Define the minimum table size. + */ +#define MIN_TABLE_ORDER 0 +#define MIN_TABLE_SIZE (1UL << MIN_TABLE_ORDER) + +/* + * Minimum number of bucket nodes to touch per thread to parallelize grow/shrink. + */ +#define MIN_PARTITION_PER_THREAD_ORDER 12 +#define MIN_PARTITION_PER_THREAD (1UL << MIN_PARTITION_PER_THREAD_ORDER) + +/* + * The removed flag needs to be updated atomically with the pointer. + * It indicates that no node must attach to the node scheduled for + * removal, and that node garbage collection must be performed. + * The bucket flag does not require to be updated atomically with the + * pointer, but it is added as a pointer low bit flag to save space. + */ +#define REMOVED_FLAG (1UL << 0) +#define BUCKET_FLAG (1UL << 1) +#define FLAGS_MASK ((1UL << 2) - 1) + +/* Value of the end pointer. Should not interact with flags. */ +#define END_VALUE NULL + +/* + * ht_items_count: Split-counters counting the number of node addition + * and removal in the table. Only used if the CDS_LFHT_ACCOUNTING flag + * is set at hash table creation. + * + * These are free-running counters, never reset to zero. They count the + * number of add/remove, and trigger every (1 << COUNT_COMMIT_ORDER) + * operations to update the global counter. We choose a power-of-2 value + * for the trigger to deal with 32 or 64-bit overflow of the counter. + */ +struct ht_items_count { + unsigned long add, del; +} __attribute__((aligned(CAA_CACHE_LINE_SIZE))); + +/* + * rcu_resize_work: Contains arguments passed to RCU worker thread + * responsible for performing lazy resize. + */ +struct rcu_resize_work { + struct rcu_head head; + struct cds_lfht *ht; +}; + +/* + * partition_resize_work: Contains arguments passed to worker threads + * executing the hash table resize on partitions of the hash table + * assigned to each processor's worker thread. + */ +struct partition_resize_work { + pthread_t thread_id; + struct cds_lfht *ht; + unsigned long i, start, len; + void (*fct)(struct cds_lfht *ht, unsigned long i, + unsigned long start, unsigned long len); +}; + +/* + * Algorithm to reverse bits in a word by lookup table, extended to + * 64-bit words. + * Source: + * http://graphics.stanford.edu/~seander/bithacks.html#BitReverseTable + * Originally from Public Domain. + */ + +static const uint8_t BitReverseTable256[256] = +{ +#define R2(n) (n), (n) + 2*64, (n) + 1*64, (n) + 3*64 +#define R4(n) R2(n), R2((n) + 2*16), R2((n) + 1*16), R2((n) + 3*16) +#define R6(n) R4(n), R4((n) + 2*4 ), R4((n) + 1*4 ), R4((n) + 3*4 ) + R6(0), R6(2), R6(1), R6(3) +}; +#undef R2 +#undef R4 +#undef R6 + +static +uint8_t bit_reverse_u8(uint8_t v) +{ + return BitReverseTable256[v]; +} + +static __attribute__((unused)) +uint32_t bit_reverse_u32(uint32_t v) +{ + return ((uint32_t) bit_reverse_u8(v) << 24) | + ((uint32_t) bit_reverse_u8(v >> 8) << 16) | + ((uint32_t) bit_reverse_u8(v >> 16) << 8) | + ((uint32_t) bit_reverse_u8(v >> 24)); +} + +static __attribute__((unused)) +uint64_t bit_reverse_u64(uint64_t v) +{ + return ((uint64_t) bit_reverse_u8(v) << 56) | + ((uint64_t) bit_reverse_u8(v >> 8) << 48) | + ((uint64_t) bit_reverse_u8(v >> 16) << 40) | + ((uint64_t) bit_reverse_u8(v >> 24) << 32) | + ((uint64_t) bit_reverse_u8(v >> 32) << 24) | + ((uint64_t) bit_reverse_u8(v >> 40) << 16) | + ((uint64_t) bit_reverse_u8(v >> 48) << 8) | + ((uint64_t) bit_reverse_u8(v >> 56)); +} + +static +unsigned long bit_reverse_ulong(unsigned long v) +{ +#if (CAA_BITS_PER_LONG == 32) + return bit_reverse_u32(v); +#else + return bit_reverse_u64(v); +#endif +} + +/* + * fls: returns the position of the most significant bit. + * Returns 0 if no bit is set, else returns the position of the most + * significant bit (from 1 to 32 on 32-bit, from 1 to 64 on 64-bit). + */ +#if defined(__i386) || defined(__x86_64) +static inline +unsigned int fls_u32(uint32_t x) +{ + int r; + + asm("bsrl %1,%0\n\t" + "jnz 1f\n\t" + "movl $-1,%0\n\t" + "1:\n\t" + : "=r" (r) : "rm" (x)); + return r + 1; +} +#define HAS_FLS_U32 +#endif + +#if defined(__x86_64) +static inline +unsigned int fls_u64(uint64_t x) +{ + long r; + + asm("bsrq %1,%0\n\t" + "jnz 1f\n\t" + "movq $-1,%0\n\t" + "1:\n\t" + : "=r" (r) : "rm" (x)); + return r + 1; +} +#define HAS_FLS_U64 +#endif + +#ifndef HAS_FLS_U64 +static __attribute__((unused)) +unsigned int fls_u64(uint64_t x) +{ + unsigned int r = 64; + + if (!x) + return 0; + + if (!(x & 0xFFFFFFFF00000000ULL)) { + x <<= 32; + r -= 32; + } + if (!(x & 0xFFFF000000000000ULL)) { + x <<= 16; + r -= 16; + } + if (!(x & 0xFF00000000000000ULL)) { + x <<= 8; + r -= 8; + } + if (!(x & 0xF000000000000000ULL)) { + x <<= 4; + r -= 4; + } + if (!(x & 0xC000000000000000ULL)) { + x <<= 2; + r -= 2; + } + if (!(x & 0x8000000000000000ULL)) { + x <<= 1; + r -= 1; + } + return r; +} +#endif + +#ifndef HAS_FLS_U32 +static __attribute__((unused)) +unsigned int fls_u32(uint32_t x) +{ + unsigned int r = 32; + + if (!x) + return 0; + if (!(x & 0xFFFF0000U)) { + x <<= 16; + r -= 16; + } + if (!(x & 0xFF000000U)) { + x <<= 8; + r -= 8; + } + if (!(x & 0xF0000000U)) { + x <<= 4; + r -= 4; + } + if (!(x & 0xC0000000U)) { + x <<= 2; + r -= 2; + } + if (!(x & 0x80000000U)) { + x <<= 1; + r -= 1; + } + return r; +} +#endif + +unsigned int fls_ulong(unsigned long x) +{ +#if (CAA_BITS_PER_LONG == 32) + return fls_u32(x); +#else + return fls_u64(x); +#endif +} + +/* + * Return the minimum order for which x <= (1UL << order). + * Return -1 if x is 0. + */ +int get_count_order_u32(uint32_t x) +{ + if (!x) + return -1; + + return fls_u32(x - 1); +} + +/* + * Return the minimum order for which x <= (1UL << order). + * Return -1 if x is 0. + */ +int get_count_order_ulong(unsigned long x) +{ + if (!x) + return -1; + + return fls_ulong(x - 1); +} + +static +void cds_lfht_resize_lazy_grow(struct cds_lfht *ht, unsigned long size, int growth); + +static +void cds_lfht_resize_lazy_count(struct cds_lfht *ht, unsigned long size, + unsigned long count); + +static long nr_cpus_mask = -1; +static long split_count_mask = -1; + +#if defined(HAVE_SYSCONF) +static void ht_init_nr_cpus_mask(void) +{ + long maxcpus; + + maxcpus = sysconf(_SC_NPROCESSORS_CONF); + if (maxcpus <= 0) { + nr_cpus_mask = -2; + return; + } + /* + * round up number of CPUs to next power of two, so we + * can use & for modulo. + */ + maxcpus = 1UL << get_count_order_ulong(maxcpus); + nr_cpus_mask = maxcpus - 1; +} +#else /* #if defined(HAVE_SYSCONF) */ +static void ht_init_nr_cpus_mask(void) +{ + nr_cpus_mask = -2; +} +#endif /* #else #if defined(HAVE_SYSCONF) */ + +static +void alloc_split_items_count(struct cds_lfht *ht) +{ + struct ht_items_count *count; + + if (nr_cpus_mask == -1) { + ht_init_nr_cpus_mask(); + if (nr_cpus_mask < 0) + split_count_mask = DEFAULT_SPLIT_COUNT_MASK; + else + split_count_mask = nr_cpus_mask; + } + + assert(split_count_mask >= 0); + + if (ht->flags & CDS_LFHT_ACCOUNTING) { + ht->split_count = calloc(split_count_mask + 1, sizeof(*count)); + assert(ht->split_count); + } else { + ht->split_count = NULL; + } +} + +static +void free_split_items_count(struct cds_lfht *ht) +{ + poison_free(ht->split_count); +} + +#if defined(HAVE_SCHED_GETCPU) +static +int ht_get_split_count_index(unsigned long hash) +{ + int cpu; + + assert(split_count_mask >= 0); + cpu = sched_getcpu(); + if (caa_unlikely(cpu < 0)) + return hash & split_count_mask; + else + return cpu & split_count_mask; +} +#else /* #if defined(HAVE_SCHED_GETCPU) */ +static +int ht_get_split_count_index(unsigned long hash) +{ + return hash & split_count_mask; +} +#endif /* #else #if defined(HAVE_SCHED_GETCPU) */ + +static +void ht_count_add(struct cds_lfht *ht, unsigned long size, unsigned long hash) +{ + unsigned long split_count; + int index; + long count; + + if (caa_unlikely(!ht->split_count)) + return; + index = ht_get_split_count_index(hash); + split_count = uatomic_add_return(&ht->split_count[index].add, 1); + if (caa_likely(split_count & ((1UL << COUNT_COMMIT_ORDER) - 1))) + return; + /* Only if number of add multiple of 1UL << COUNT_COMMIT_ORDER */ + + dbg_printf("add split count %lu\n", split_count); + count = uatomic_add_return(&ht->count, + 1UL << COUNT_COMMIT_ORDER); + if (caa_likely(count & (count - 1))) + return; + /* Only if global count is power of 2 */ + + if ((count >> CHAIN_LEN_RESIZE_THRESHOLD) < size) + return; + dbg_printf("add set global %ld\n", count); + cds_lfht_resize_lazy_count(ht, size, + count >> (CHAIN_LEN_TARGET - 1)); +} + +static +void ht_count_del(struct cds_lfht *ht, unsigned long size, unsigned long hash) +{ + unsigned long split_count; + int index; + long count; + + if (caa_unlikely(!ht->split_count)) + return; + index = ht_get_split_count_index(hash); + split_count = uatomic_add_return(&ht->split_count[index].del, 1); + if (caa_likely(split_count & ((1UL << COUNT_COMMIT_ORDER) - 1))) + return; + /* Only if number of deletes multiple of 1UL << COUNT_COMMIT_ORDER */ + + dbg_printf("del split count %lu\n", split_count); + count = uatomic_add_return(&ht->count, + -(1UL << COUNT_COMMIT_ORDER)); + if (caa_likely(count & (count - 1))) + return; + /* Only if global count is power of 2 */ + + if ((count >> CHAIN_LEN_RESIZE_THRESHOLD) >= size) + return; + dbg_printf("del set global %ld\n", count); + /* + * Don't shrink table if the number of nodes is below a + * certain threshold. + */ + if (count < (1UL << COUNT_COMMIT_ORDER) * (split_count_mask + 1)) + return; + cds_lfht_resize_lazy_count(ht, size, + count >> (CHAIN_LEN_TARGET - 1)); +} + +static +void check_resize(struct cds_lfht *ht, unsigned long size, uint32_t chain_len) +{ + unsigned long count; + + if (!(ht->flags & CDS_LFHT_AUTO_RESIZE)) + return; + count = uatomic_read(&ht->count); + /* + * Use bucket-local length for small table expand and for + * environments lacking per-cpu data support. + */ + if (count >= (1UL << COUNT_COMMIT_ORDER)) + return; + if (chain_len > 100) + dbg_printf("WARNING: large chain length: %u.\n", + chain_len); + if (chain_len >= CHAIN_LEN_RESIZE_THRESHOLD) + cds_lfht_resize_lazy_grow(ht, size, + get_count_order_u32(chain_len - (CHAIN_LEN_TARGET - 1))); +} + +static +struct cds_lfht_node *clear_flag(struct cds_lfht_node *node) +{ + return (struct cds_lfht_node *) (((unsigned long) node) & ~FLAGS_MASK); +} + +static +int is_removed(struct cds_lfht_node *node) +{ + return ((unsigned long) node) & REMOVED_FLAG; +} + +static +struct cds_lfht_node *flag_removed(struct cds_lfht_node *node) +{ + return (struct cds_lfht_node *) (((unsigned long) node) | REMOVED_FLAG); +} + +static +int is_bucket(struct cds_lfht_node *node) +{ + return ((unsigned long) node) & BUCKET_FLAG; +} + +static +struct cds_lfht_node *flag_bucket(struct cds_lfht_node *node) +{ + return (struct cds_lfht_node *) (((unsigned long) node) | BUCKET_FLAG); +} + +static +struct cds_lfht_node *get_end(void) +{ + return (struct cds_lfht_node *) END_VALUE; +} + +static +int is_end(struct cds_lfht_node *node) +{ + return clear_flag(node) == (struct cds_lfht_node *) END_VALUE; +} + +static +unsigned long _uatomic_xchg_monotonic_increase(unsigned long *ptr, + unsigned long v) +{ + unsigned long old1, old2; + + old1 = uatomic_read(ptr); + do { + old2 = old1; + if (old2 >= v) + return old2; + } while ((old1 = uatomic_cmpxchg(ptr, old2, v)) != old2); + return old2; +} + +static +void cds_lfht_alloc_bucket_table(struct cds_lfht *ht, unsigned long order) +{ + return ht->mm->alloc_bucket_table(ht, order); +} + +/* + * cds_lfht_free_bucket_table() should be called with decreasing order. + * When cds_lfht_free_bucket_table(0) is called, it means the whole + * lfht is destroyed. + */ +static +void cds_lfht_free_bucket_table(struct cds_lfht *ht, unsigned long order) +{ + return ht->mm->free_bucket_table(ht, order); +} + +static inline +struct cds_lfht_node *bucket_at(struct cds_lfht *ht, unsigned long index) +{ + return ht->bucket_at(ht, index); +} + +static inline +struct cds_lfht_node *lookup_bucket(struct cds_lfht *ht, unsigned long size, + unsigned long hash) +{ + assert(size > 0); + return bucket_at(ht, hash & (size - 1)); +} + +/* + * Remove all logically deleted nodes from a bucket up to a certain node key. + */ +static +void _cds_lfht_gc_bucket(struct cds_lfht_node *bucket, struct cds_lfht_node *node) +{ + struct cds_lfht_node *iter_prev, *iter, *next, *new_next; + + assert(!is_bucket(bucket)); + assert(!is_removed(bucket)); + assert(!is_bucket(node)); + assert(!is_removed(node)); + for (;;) { + iter_prev = bucket; + /* We can always skip the bucket node initially */ + iter = rcu_dereference(iter_prev->next); + assert(!is_removed(iter)); + assert(iter_prev->reverse_hash <= node->reverse_hash); + /* + * We should never be called with bucket (start of chain) + * and logically removed node (end of path compression + * marker) being the actual same node. This would be a + * bug in the algorithm implementation. + */ + assert(bucket != node); + for (;;) { + if (caa_unlikely(is_end(iter))) + return; + if (caa_likely(clear_flag(iter)->reverse_hash > node->reverse_hash)) + return; + next = rcu_dereference(clear_flag(iter)->next); + if (caa_likely(is_removed(next))) + break; + iter_prev = clear_flag(iter); + iter = next; + } + assert(!is_removed(iter)); + if (is_bucket(iter)) + new_next = flag_bucket(clear_flag(next)); + else + new_next = clear_flag(next); + (void) uatomic_cmpxchg(&iter_prev->next, iter, new_next); + } + return; +} + +static +int _cds_lfht_replace(struct cds_lfht *ht, unsigned long size, + struct cds_lfht_node *old_node, + struct cds_lfht_node *old_next, + struct cds_lfht_node *new_node) +{ + struct cds_lfht_node *bucket, *ret_next; + + if (!old_node) /* Return -ENOENT if asked to replace NULL node */ + return -ENOENT; + + assert(!is_removed(old_node)); + assert(!is_bucket(old_node)); + assert(!is_removed(new_node)); + assert(!is_bucket(new_node)); + assert(new_node != old_node); + for (;;) { + /* Insert after node to be replaced */ + if (is_removed(old_next)) { + /* + * Too late, the old node has been removed under us + * between lookup and replace. Fail. + */ + return -ENOENT; + } + assert(!is_bucket(old_next)); + assert(new_node != clear_flag(old_next)); + new_node->next = clear_flag(old_next); + /* + * Here is the whole trick for lock-free replace: we add + * the replacement node _after_ the node we want to + * replace by atomically setting its next pointer at the + * same time we set its removal flag. Given that + * the lookups/get next use an iterator aware of the + * next pointer, they will either skip the old node due + * to the removal flag and see the new node, or use + * the old node, but will not see the new one. + */ + ret_next = uatomic_cmpxchg(&old_node->next, + old_next, flag_removed(new_node)); + if (ret_next == old_next) + break; /* We performed the replacement. */ + old_next = ret_next; + } + + /* + * Ensure that the old node is not visible to readers anymore: + * lookup for the node, and remove it (along with any other + * logically removed node) if found. + */ + bucket = lookup_bucket(ht, size, bit_reverse_ulong(old_node->reverse_hash)); + _cds_lfht_gc_bucket(bucket, new_node); + + assert(is_removed(rcu_dereference(old_node->next))); + return 0; +} + +/* + * A non-NULL unique_ret pointer uses the "add unique" (or uniquify) add + * mode. A NULL unique_ret allows creation of duplicate keys. + */ +static +void _cds_lfht_add(struct cds_lfht *ht, + cds_lfht_match_fct match, + const void *key, + unsigned long size, + struct cds_lfht_node *node, + struct cds_lfht_iter *unique_ret, + int bucket_flag) +{ + struct cds_lfht_node *iter_prev, *iter, *next, *new_node, *new_next, + *return_node; + struct cds_lfht_node *bucket; + + assert(!is_bucket(node)); + assert(!is_removed(node)); + bucket = lookup_bucket(ht, size, bit_reverse_ulong(node->reverse_hash)); + for (;;) { + uint32_t chain_len = 0; + + /* + * iter_prev points to the non-removed node prior to the + * insert location. + */ + iter_prev = bucket; + /* We can always skip the bucket node initially */ + iter = rcu_dereference(iter_prev->next); + assert(iter_prev->reverse_hash <= node->reverse_hash); + for (;;) { + if (caa_unlikely(is_end(iter))) + goto insert; + if (caa_likely(clear_flag(iter)->reverse_hash > node->reverse_hash)) + goto insert; + + /* bucket node is the first node of the identical-hash-value chain */ + if (bucket_flag && clear_flag(iter)->reverse_hash == node->reverse_hash) + goto insert; + + next = rcu_dereference(clear_flag(iter)->next); + if (caa_unlikely(is_removed(next))) + goto gc_node; + + /* uniquely add */ + if (unique_ret + && !is_bucket(next) + && clear_flag(iter)->reverse_hash == node->reverse_hash) { + struct cds_lfht_iter d_iter = { .node = node, .next = iter, }; + + /* + * uniquely adding inserts the node as the first + * node of the identical-hash-value node chain. + * + * This semantic ensures no duplicated keys + * should ever be observable in the table + * (including observe one node by one node + * by forward iterations) + */ + cds_lfht_next_duplicate(ht, match, key, &d_iter); + if (!d_iter.node) + goto insert; + + *unique_ret = d_iter; + return; + } + + /* Only account for identical reverse hash once */ + if (iter_prev->reverse_hash != clear_flag(iter)->reverse_hash + && !is_bucket(next)) + check_resize(ht, size, ++chain_len); + iter_prev = clear_flag(iter); + iter = next; + } + + insert: + assert(node != clear_flag(iter)); + assert(!is_removed(iter_prev)); + assert(!is_removed(iter)); + assert(iter_prev != node); + if (!bucket_flag) + node->next = clear_flag(iter); + else + node->next = flag_bucket(clear_flag(iter)); + if (is_bucket(iter)) + new_node = flag_bucket(node); + else + new_node = node; + if (uatomic_cmpxchg(&iter_prev->next, iter, + new_node) != iter) { + continue; /* retry */ + } else { + return_node = node; + goto end; + } + + gc_node: + assert(!is_removed(iter)); + if (is_bucket(iter)) + new_next = flag_bucket(clear_flag(next)); + else + new_next = clear_flag(next); + (void) uatomic_cmpxchg(&iter_prev->next, iter, new_next); + /* retry */ + } +end: + if (unique_ret) { + unique_ret->node = return_node; + /* unique_ret->next left unset, never used. */ + } +} + +static +int _cds_lfht_del(struct cds_lfht *ht, unsigned long size, + struct cds_lfht_node *node, + int bucket_removal) +{ + struct cds_lfht_node *bucket, *next, *old; + + if (!node) /* Return -ENOENT if asked to delete NULL node */ + return -ENOENT; + + /* logically delete the node */ + assert(!is_bucket(node)); + assert(!is_removed(node)); + old = rcu_dereference(node->next); + do { + struct cds_lfht_node *new_next; + + next = old; + if (caa_unlikely(is_removed(next))) + return -ENOENT; + if (bucket_removal) + assert(is_bucket(next)); + else + assert(!is_bucket(next)); + new_next = flag_removed(next); + old = uatomic_cmpxchg(&node->next, next, new_next); + } while (old != next); + /* We performed the (logical) deletion. */ + + /* + * Ensure that the node is not visible to readers anymore: lookup for + * the node, and remove it (along with any other logically removed node) + * if found. + */ + bucket = lookup_bucket(ht, size, bit_reverse_ulong(node->reverse_hash)); + _cds_lfht_gc_bucket(bucket, node); + + assert(is_removed(rcu_dereference(node->next))); + return 0; +} + +static +void *partition_resize_thread(void *arg) +{ + struct partition_resize_work *work = arg; + + work->ht->flavor->register_thread(); + work->fct(work->ht, work->i, work->start, work->len); + work->ht->flavor->unregister_thread(); + return NULL; +} + +static +void partition_resize_helper(struct cds_lfht *ht, unsigned long i, + unsigned long len, + void (*fct)(struct cds_lfht *ht, unsigned long i, + unsigned long start, unsigned long len)) +{ + unsigned long partition_len; + struct partition_resize_work *work; + int thread, ret; + unsigned long nr_threads; + + /* + * Note: nr_cpus_mask + 1 is always power of 2. + * We spawn just the number of threads we need to satisfy the minimum + * partition size, up to the number of CPUs in the system. + */ + if (nr_cpus_mask > 0) { + nr_threads = min(nr_cpus_mask + 1, + len >> MIN_PARTITION_PER_THREAD_ORDER); + } else { + nr_threads = 1; + } + partition_len = len >> get_count_order_ulong(nr_threads); + work = calloc(nr_threads, sizeof(*work)); + assert(work); + for (thread = 0; thread < nr_threads; thread++) { + work[thread].ht = ht; + work[thread].i = i; + work[thread].len = partition_len; + work[thread].start = thread * partition_len; + work[thread].fct = fct; + ret = pthread_create(&(work[thread].thread_id), ht->resize_attr, + partition_resize_thread, &work[thread]); + assert(!ret); + } + for (thread = 0; thread < nr_threads; thread++) { + ret = pthread_join(work[thread].thread_id, NULL); + assert(!ret); + } + free(work); +} + +/* + * Holding RCU read lock to protect _cds_lfht_add against memory + * reclaim that could be performed by other call_rcu worker threads (ABA + * problem). + * + * When we reach a certain length, we can split this population phase over + * many worker threads, based on the number of CPUs available in the system. + * This should therefore take care of not having the expand lagging behind too + * many concurrent insertion threads by using the scheduler's ability to + * schedule bucket node population fairly with insertions. + */ +static +void init_table_populate_partition(struct cds_lfht *ht, unsigned long i, + unsigned long start, unsigned long len) +{ + unsigned long j, size = 1UL << (i - 1); + + assert(i > MIN_TABLE_ORDER); + ht->flavor->read_lock(); + for (j = size + start; j < size + start + len; j++) { + struct cds_lfht_node *new_node = bucket_at(ht, j); + + assert(j >= size && j < (size << 1)); + dbg_printf("init populate: order %lu index %lu hash %lu\n", + i, j, j); + new_node->reverse_hash = bit_reverse_ulong(j); + _cds_lfht_add(ht, NULL, NULL, size, new_node, NULL, 1); + } + ht->flavor->read_unlock(); +} + +static +void init_table_populate(struct cds_lfht *ht, unsigned long i, + unsigned long len) +{ + assert(nr_cpus_mask != -1); + if (nr_cpus_mask < 0 || len < 2 * MIN_PARTITION_PER_THREAD) { + ht->flavor->thread_online(); + init_table_populate_partition(ht, i, 0, len); + ht->flavor->thread_offline(); + return; + } + partition_resize_helper(ht, i, len, init_table_populate_partition); +} + +static +void init_table(struct cds_lfht *ht, + unsigned long first_order, unsigned long last_order) +{ + unsigned long i; + + dbg_printf("init table: first_order %lu last_order %lu\n", + first_order, last_order); + assert(first_order > MIN_TABLE_ORDER); + for (i = first_order; i <= last_order; i++) { + unsigned long len; + + len = 1UL << (i - 1); + dbg_printf("init order %lu len: %lu\n", i, len); + + /* Stop expand if the resize target changes under us */ + if (CMM_LOAD_SHARED(ht->resize_target) < (1UL << i)) + break; + + cds_lfht_alloc_bucket_table(ht, i); + + /* + * Set all bucket nodes reverse hash values for a level and + * link all bucket nodes into the table. + */ + init_table_populate(ht, i, len); + + /* + * Update table size. + */ + cmm_smp_wmb(); /* populate data before RCU size */ + CMM_STORE_SHARED(ht->size, 1UL << i); + + dbg_printf("init new size: %lu\n", 1UL << i); + if (CMM_LOAD_SHARED(ht->in_progress_destroy)) + break; + } +} + +/* + * Holding RCU read lock to protect _cds_lfht_remove against memory + * reclaim that could be performed by other call_rcu worker threads (ABA + * problem). + * For a single level, we logically remove and garbage collect each node. + * + * As a design choice, we perform logical removal and garbage collection on a + * node-per-node basis to simplify this algorithm. We also assume keeping good + * cache locality of the operation would overweight possible performance gain + * that could be achieved by batching garbage collection for multiple levels. + * However, this would have to be justified by benchmarks. + * + * Concurrent removal and add operations are helping us perform garbage + * collection of logically removed nodes. We guarantee that all logically + * removed nodes have been garbage-collected (unlinked) before call_rcu is + * invoked to free a hole level of bucket nodes (after a grace period). + * + * Logical removal and garbage collection can therefore be done in batch or on a + * node-per-node basis, as long as the guarantee above holds. + * + * When we reach a certain length, we can split this removal over many worker + * threads, based on the number of CPUs available in the system. This should + * take care of not letting resize process lag behind too many concurrent + * updater threads actively inserting into the hash table. + */ +static +void remove_table_partition(struct cds_lfht *ht, unsigned long i, + unsigned long start, unsigned long len) +{ + unsigned long j, size = 1UL << (i - 1); + + assert(i > MIN_TABLE_ORDER); + ht->flavor->read_lock(); + for (j = size + start; j < size + start + len; j++) { + struct cds_lfht_node *fini_node = bucket_at(ht, j); + + assert(j >= size && j < (size << 1)); + dbg_printf("remove entry: order %lu index %lu hash %lu\n", + i, j, j); + fini_node->reverse_hash = bit_reverse_ulong(j); + (void) _cds_lfht_del(ht, size, fini_node, 1); + } + ht->flavor->read_unlock(); +} + +static +void remove_table(struct cds_lfht *ht, unsigned long i, unsigned long len) +{ + + assert(nr_cpus_mask != -1); + if (nr_cpus_mask < 0 || len < 2 * MIN_PARTITION_PER_THREAD) { + ht->flavor->thread_online(); + remove_table_partition(ht, i, 0, len); + ht->flavor->thread_offline(); + return; + } + partition_resize_helper(ht, i, len, remove_table_partition); +} + +/* + * fini_table() is never called for first_order == 0, which is why + * free_by_rcu_order == 0 can be used as criterion to know if free must + * be called. + */ +static +void fini_table(struct cds_lfht *ht, + unsigned long first_order, unsigned long last_order) +{ + long i; + unsigned long free_by_rcu_order = 0; + + dbg_printf("fini table: first_order %lu last_order %lu\n", + first_order, last_order); + assert(first_order > MIN_TABLE_ORDER); + for (i = last_order; i >= first_order; i--) { + unsigned long len; + + len = 1UL << (i - 1); + dbg_printf("fini order %lu len: %lu\n", i, len); + + /* Stop shrink if the resize target changes under us */ + if (CMM_LOAD_SHARED(ht->resize_target) > (1UL << (i - 1))) + break; + + cmm_smp_wmb(); /* populate data before RCU size */ + CMM_STORE_SHARED(ht->size, 1UL << (i - 1)); + + /* + * We need to wait for all add operations to reach Q.S. (and + * thus use the new table for lookups) before we can start + * releasing the old bucket nodes. Otherwise their lookup will + * return a logically removed node as insert position. + */ + ht->flavor->update_synchronize_rcu(); + if (free_by_rcu_order) + cds_lfht_free_bucket_table(ht, free_by_rcu_order); + + /* + * Set "removed" flag in bucket nodes about to be removed. + * Unlink all now-logically-removed bucket node pointers. + * Concurrent add/remove operation are helping us doing + * the gc. + */ + remove_table(ht, i, len); + + free_by_rcu_order = i; + + dbg_printf("fini new size: %lu\n", 1UL << i); + if (CMM_LOAD_SHARED(ht->in_progress_destroy)) + break; + } + + if (free_by_rcu_order) { + ht->flavor->update_synchronize_rcu(); + cds_lfht_free_bucket_table(ht, free_by_rcu_order); + } +} + +static +void cds_lfht_create_bucket(struct cds_lfht *ht, unsigned long size) +{ + struct cds_lfht_node *prev, *node; + unsigned long order, len, i; + + cds_lfht_alloc_bucket_table(ht, 0); + + dbg_printf("create bucket: order 0 index 0 hash 0\n"); + node = bucket_at(ht, 0); + node->next = flag_bucket(get_end()); + node->reverse_hash = 0; + + for (order = 1; order < get_count_order_ulong(size) + 1; order++) { + len = 1UL << (order - 1); + cds_lfht_alloc_bucket_table(ht, order); + + for (i = 0; i < len; i++) { + /* + * Now, we are trying to init the node with the + * hash=(len+i) (which is also a bucket with the + * index=(len+i)) and insert it into the hash table, + * so this node has to be inserted after the bucket + * with the index=(len+i)&(len-1)=i. And because there + * is no other non-bucket node nor bucket node with + * larger index/hash inserted, so the bucket node + * being inserted should be inserted directly linked + * after the bucket node with index=i. + */ + prev = bucket_at(ht, i); + node = bucket_at(ht, len + i); + + dbg_printf("create bucket: order %lu index %lu hash %lu\n", + order, len + i, len + i); + node->reverse_hash = bit_reverse_ulong(len + i); + + /* insert after prev */ + assert(is_bucket(prev->next)); + node->next = prev->next; + prev->next = flag_bucket(node); + } + } +} + +struct cds_lfht *_cds_lfht_new(unsigned long init_size, + unsigned long min_nr_alloc_buckets, + unsigned long max_nr_buckets, + int flags, + const struct cds_lfht_mm_type *mm, + const struct rcu_flavor_struct *flavor, + pthread_attr_t *attr) +{ + struct cds_lfht *ht; + unsigned long order; + + /* min_nr_alloc_buckets must be power of two */ + if (!min_nr_alloc_buckets || (min_nr_alloc_buckets & (min_nr_alloc_buckets - 1))) + return NULL; + + /* init_size must be power of two */ + if (!init_size || (init_size & (init_size - 1))) + return NULL; + + /* + * Memory management plugin default. + */ + if (!mm) { + if (CAA_BITS_PER_LONG > 32 + && max_nr_buckets + && max_nr_buckets <= (1ULL << 32)) { + /* + * For 64-bit architectures, with max number of + * buckets small enough not to use the entire + * 64-bit memory mapping space (and allowing a + * fair number of hash table instances), use the + * mmap allocator, which is faster than the + * order allocator. + */ + mm = &cds_lfht_mm_mmap; + } else { + /* + * The fallback is to use the order allocator. + */ + mm = &cds_lfht_mm_order; + } + } + + /* max_nr_buckets == 0 for order based mm means infinite */ + if (mm == &cds_lfht_mm_order && !max_nr_buckets) + max_nr_buckets = 1UL << (MAX_TABLE_ORDER - 1); + + /* max_nr_buckets must be power of two */ + if (!max_nr_buckets || (max_nr_buckets & (max_nr_buckets - 1))) + return NULL; + + min_nr_alloc_buckets = max(min_nr_alloc_buckets, MIN_TABLE_SIZE); + init_size = max(init_size, MIN_TABLE_SIZE); + max_nr_buckets = max(max_nr_buckets, min_nr_alloc_buckets); + init_size = min(init_size, max_nr_buckets); + + ht = mm->alloc_cds_lfht(min_nr_alloc_buckets, max_nr_buckets); + assert(ht); + assert(ht->mm == mm); + assert(ht->bucket_at == mm->bucket_at); + + ht->flags = flags; + ht->flavor = flavor; + ht->resize_attr = attr; + alloc_split_items_count(ht); + /* this mutex should not nest in read-side C.S. */ + pthread_mutex_init(&ht->resize_mutex, NULL); + order = get_count_order_ulong(init_size); + ht->resize_target = 1UL << order; + cds_lfht_create_bucket(ht, 1UL << order); + ht->size = 1UL << order; + return ht; +} + +void cds_lfht_lookup(struct cds_lfht *ht, unsigned long hash, + cds_lfht_match_fct match, const void *key, + struct cds_lfht_iter *iter) +{ + struct cds_lfht_node *node, *next, *bucket; + unsigned long reverse_hash, size; + + reverse_hash = bit_reverse_ulong(hash); + + size = rcu_dereference(ht->size); + bucket = lookup_bucket(ht, size, hash); + /* We can always skip the bucket node initially */ + node = rcu_dereference(bucket->next); + node = clear_flag(node); + for (;;) { + if (caa_unlikely(is_end(node))) { + node = next = NULL; + break; + } + if (caa_unlikely(node->reverse_hash > reverse_hash)) { + node = next = NULL; + break; + } + next = rcu_dereference(node->next); + assert(node == clear_flag(node)); + if (caa_likely(!is_removed(next)) + && !is_bucket(next) + && node->reverse_hash == reverse_hash + && caa_likely(match(node, key))) { + break; + } + node = clear_flag(next); + } + assert(!node || !is_bucket(rcu_dereference(node->next))); + iter->node = node; + iter->next = next; +} + +void cds_lfht_next_duplicate(struct cds_lfht *ht, cds_lfht_match_fct match, + const void *key, struct cds_lfht_iter *iter) +{ + struct cds_lfht_node *node, *next; + unsigned long reverse_hash; + + node = iter->node; + reverse_hash = node->reverse_hash; + next = iter->next; + node = clear_flag(next); + + for (;;) { + if (caa_unlikely(is_end(node))) { + node = next = NULL; + break; + } + if (caa_unlikely(node->reverse_hash > reverse_hash)) { + node = next = NULL; + break; + } + next = rcu_dereference(node->next); + if (caa_likely(!is_removed(next)) + && !is_bucket(next) + && caa_likely(match(node, key))) { + break; + } + node = clear_flag(next); + } + assert(!node || !is_bucket(rcu_dereference(node->next))); + iter->node = node; + iter->next = next; +} + +void cds_lfht_next(struct cds_lfht *ht, struct cds_lfht_iter *iter) +{ + struct cds_lfht_node *node, *next; + + node = clear_flag(iter->next); + for (;;) { + if (caa_unlikely(is_end(node))) { + node = next = NULL; + break; + } + next = rcu_dereference(node->next); + if (caa_likely(!is_removed(next)) + && !is_bucket(next)) { + break; + } + node = clear_flag(next); + } + assert(!node || !is_bucket(rcu_dereference(node->next))); + iter->node = node; + iter->next = next; +} + +void cds_lfht_first(struct cds_lfht *ht, struct cds_lfht_iter *iter) +{ + /* + * Get next after first bucket node. The first bucket node is the + * first node of the linked list. + */ + iter->next = bucket_at(ht, 0)->next; + cds_lfht_next(ht, iter); +} + +void cds_lfht_add(struct cds_lfht *ht, unsigned long hash, + struct cds_lfht_node *node) +{ + unsigned long size; + + node->reverse_hash = bit_reverse_ulong((unsigned long) hash); + size = rcu_dereference(ht->size); + _cds_lfht_add(ht, NULL, NULL, size, node, NULL, 0); + ht_count_add(ht, size, hash); +} + +struct cds_lfht_node *cds_lfht_add_unique(struct cds_lfht *ht, + unsigned long hash, + cds_lfht_match_fct match, + const void *key, + struct cds_lfht_node *node) +{ + unsigned long size; + struct cds_lfht_iter iter; + + node->reverse_hash = bit_reverse_ulong((unsigned long) hash); + size = rcu_dereference(ht->size); + _cds_lfht_add(ht, match, key, size, node, &iter, 0); + if (iter.node == node) + ht_count_add(ht, size, hash); + return iter.node; +} + +struct cds_lfht_node *cds_lfht_add_replace(struct cds_lfht *ht, + unsigned long hash, + cds_lfht_match_fct match, + const void *key, + struct cds_lfht_node *node) +{ + unsigned long size; + struct cds_lfht_iter iter; + + node->reverse_hash = bit_reverse_ulong((unsigned long) hash); + size = rcu_dereference(ht->size); + for (;;) { + _cds_lfht_add(ht, match, key, size, node, &iter, 0); + if (iter.node == node) { + ht_count_add(ht, size, hash); + return NULL; + } + + if (!_cds_lfht_replace(ht, size, iter.node, iter.next, node)) + return iter.node; + } +} + +int cds_lfht_replace(struct cds_lfht *ht, struct cds_lfht_iter *old_iter, + struct cds_lfht_node *new_node) +{ + unsigned long size; + + size = rcu_dereference(ht->size); + return _cds_lfht_replace(ht, size, old_iter->node, old_iter->next, + new_node); +} + +int cds_lfht_del(struct cds_lfht *ht, struct cds_lfht_iter *iter) +{ + unsigned long size, hash; + int ret; + + size = rcu_dereference(ht->size); + ret = _cds_lfht_del(ht, size, iter->node, 0); + if (!ret) { + hash = bit_reverse_ulong(iter->node->reverse_hash); + ht_count_del(ht, size, hash); + } + return ret; +} + +static +int cds_lfht_delete_bucket(struct cds_lfht *ht) +{ + struct cds_lfht_node *node; + unsigned long order, i, size; + + /* Check that the table is empty */ + node = bucket_at(ht, 0); + do { + node = clear_flag(node)->next; + if (!is_bucket(node)) + return -EPERM; + assert(!is_removed(node)); + } while (!is_end(node)); + /* + * size accessed without rcu_dereference because hash table is + * being destroyed. + */ + size = ht->size; + /* Internal sanity check: all nodes left should be bucket */ + for (i = 0; i < size; i++) { + node = bucket_at(ht, i); + dbg_printf("delete bucket: index %lu expected hash %lu hash %lu\n", + i, i, bit_reverse_ulong(node->reverse_hash)); + assert(is_bucket(node->next)); + } + + for (order = get_count_order_ulong(size); (long)order >= 0; order--) + cds_lfht_free_bucket_table(ht, order); + + return 0; +} + +/* + * Should only be called when no more concurrent readers nor writers can + * possibly access the table. + */ +int cds_lfht_destroy(struct cds_lfht *ht, pthread_attr_t **attr) +{ + int ret; + + /* Wait for in-flight resize operations to complete */ + _CMM_STORE_SHARED(ht->in_progress_destroy, 1); + cmm_smp_mb(); /* Store destroy before load resize */ + while (uatomic_read(&ht->in_progress_resize)) + poll(NULL, 0, 100); /* wait for 100ms */ + ret = cds_lfht_delete_bucket(ht); + if (ret) + return ret; + free_split_items_count(ht); + if (attr) + *attr = ht->resize_attr; + poison_free(ht); + return ret; +} + +void cds_lfht_count_nodes(struct cds_lfht *ht, + long *approx_before, + unsigned long *count, + unsigned long *removed, + long *approx_after) +{ + struct cds_lfht_node *node, *next; + unsigned long nr_bucket = 0; + + *approx_before = 0; + if (ht->split_count) { + int i; + + for (i = 0; i < split_count_mask + 1; i++) { + *approx_before += uatomic_read(&ht->split_count[i].add); + *approx_before -= uatomic_read(&ht->split_count[i].del); + } + } + + *count = 0; + *removed = 0; + + /* Count non-bucket nodes in the table */ + node = bucket_at(ht, 0); + do { + next = rcu_dereference(node->next); + if (is_removed(next)) { + if (!is_bucket(next)) + (*removed)++; + else + (nr_bucket)++; + } else if (!is_bucket(next)) + (*count)++; + else + (nr_bucket)++; + node = clear_flag(next); + } while (!is_end(node)); + dbg_printf("number of bucket nodes: %lu\n", nr_bucket); + *approx_after = 0; + if (ht->split_count) { + int i; + + for (i = 0; i < split_count_mask + 1; i++) { + *approx_after += uatomic_read(&ht->split_count[i].add); + *approx_after -= uatomic_read(&ht->split_count[i].del); + } + } +} + +/* called with resize mutex held */ +static +void _do_cds_lfht_grow(struct cds_lfht *ht, + unsigned long old_size, unsigned long new_size) +{ + unsigned long old_order, new_order; + + old_order = get_count_order_ulong(old_size); + new_order = get_count_order_ulong(new_size); + dbg_printf("resize from %lu (order %lu) to %lu (order %lu) buckets\n", + old_size, old_order, new_size, new_order); + assert(new_size > old_size); + init_table(ht, old_order + 1, new_order); +} + +/* called with resize mutex held */ +static +void _do_cds_lfht_shrink(struct cds_lfht *ht, + unsigned long old_size, unsigned long new_size) +{ + unsigned long old_order, new_order; + + new_size = max(new_size, MIN_TABLE_SIZE); + old_order = get_count_order_ulong(old_size); + new_order = get_count_order_ulong(new_size); + dbg_printf("resize from %lu (order %lu) to %lu (order %lu) buckets\n", + old_size, old_order, new_size, new_order); + assert(new_size < old_size); + + /* Remove and unlink all bucket nodes to remove. */ + fini_table(ht, new_order + 1, old_order); +} + + +/* called with resize mutex held */ +static +void _do_cds_lfht_resize(struct cds_lfht *ht) +{ + unsigned long new_size, old_size; + + /* + * Resize table, re-do if the target size has changed under us. + */ + do { + assert(uatomic_read(&ht->in_progress_resize)); + if (CMM_LOAD_SHARED(ht->in_progress_destroy)) + break; + ht->resize_initiated = 1; + old_size = ht->size; + new_size = CMM_LOAD_SHARED(ht->resize_target); + if (old_size < new_size) + _do_cds_lfht_grow(ht, old_size, new_size); + else if (old_size > new_size) + _do_cds_lfht_shrink(ht, old_size, new_size); + ht->resize_initiated = 0; + /* write resize_initiated before read resize_target */ + cmm_smp_mb(); + } while (ht->size != CMM_LOAD_SHARED(ht->resize_target)); +} + +static +unsigned long resize_target_grow(struct cds_lfht *ht, unsigned long new_size) +{ + return _uatomic_xchg_monotonic_increase(&ht->resize_target, new_size); +} + +static +void resize_target_update_count(struct cds_lfht *ht, + unsigned long count) +{ + count = max(count, MIN_TABLE_SIZE); + count = min(count, ht->max_nr_buckets); + uatomic_set(&ht->resize_target, count); +} + +void cds_lfht_resize(struct cds_lfht *ht, unsigned long new_size) +{ + resize_target_update_count(ht, new_size); + CMM_STORE_SHARED(ht->resize_initiated, 1); + ht->flavor->thread_offline(); + pthread_mutex_lock(&ht->resize_mutex); + _do_cds_lfht_resize(ht); + pthread_mutex_unlock(&ht->resize_mutex); + ht->flavor->thread_online(); +} + +static +void do_resize_cb(struct rcu_head *head) +{ + struct rcu_resize_work *work = + caa_container_of(head, struct rcu_resize_work, head); + struct cds_lfht *ht = work->ht; + + ht->flavor->thread_offline(); + pthread_mutex_lock(&ht->resize_mutex); + _do_cds_lfht_resize(ht); + pthread_mutex_unlock(&ht->resize_mutex); + ht->flavor->thread_online(); + poison_free(work); + cmm_smp_mb(); /* finish resize before decrement */ + uatomic_dec(&ht->in_progress_resize); +} + +static +void __cds_lfht_resize_lazy_launch(struct cds_lfht *ht) +{ + struct rcu_resize_work *work; + + /* Store resize_target before read resize_initiated */ + cmm_smp_mb(); + if (!CMM_LOAD_SHARED(ht->resize_initiated)) { + uatomic_inc(&ht->in_progress_resize); + cmm_smp_mb(); /* increment resize count before load destroy */ + if (CMM_LOAD_SHARED(ht->in_progress_destroy)) { + uatomic_dec(&ht->in_progress_resize); + return; + } + work = malloc(sizeof(*work)); + work->ht = ht; + ht->flavor->update_call_rcu(&work->head, do_resize_cb); + CMM_STORE_SHARED(ht->resize_initiated, 1); + } +} + +static +void cds_lfht_resize_lazy_grow(struct cds_lfht *ht, unsigned long size, int growth) +{ + unsigned long target_size = size << growth; + + target_size = min(target_size, ht->max_nr_buckets); + if (resize_target_grow(ht, target_size) >= target_size) + return; + + __cds_lfht_resize_lazy_launch(ht); +} + +/* + * We favor grow operations over shrink. A shrink operation never occurs + * if a grow operation is queued for lazy execution. A grow operation + * cancels any pending shrink lazy execution. + */ +static +void cds_lfht_resize_lazy_count(struct cds_lfht *ht, unsigned long size, + unsigned long count) +{ + if (!(ht->flags & CDS_LFHT_AUTO_RESIZE)) + return; + count = max(count, MIN_TABLE_SIZE); + count = min(count, ht->max_nr_buckets); + if (count == size) + return; /* Already the right size, no resize needed */ + if (count > size) { /* lazy grow */ + if (resize_target_grow(ht, count) >= count) + return; + } else { /* lazy shrink */ + for (;;) { + unsigned long s; + + s = uatomic_cmpxchg(&ht->resize_target, size, count); + if (s == size) + break; /* no resize needed */ + if (s > size) + return; /* growing is/(was just) in progress */ + if (s <= count) + return; /* some other thread do shrink */ + size = s; + } + } + __cds_lfht_resize_lazy_launch(ht); +} diff --git a/tests/Makefile.am b/tests/Makefile.am index 44fac55..05a8024 100644 --- a/tests/Makefile.am +++ b/tests/Makefile.am @@ -15,7 +15,7 @@ noinst_PROGRAMS = test_urcu test_urcu_dynamic_link test_urcu_timing \ test_urcu_bp test_urcu_bp_dynamic_link test_cycles_per_loop \ test_urcu_lfq test_urcu_wfq test_urcu_lfs test_urcu_wfs \ test_urcu_wfq_dynlink test_urcu_wfs_dynlink \ - test_urcu_lfq_dynlink test_urcu_lfs_dynlink + test_urcu_lfq_dynlink test_urcu_lfs_dynlink test_urcu_hash noinst_HEADERS = rcutorture.h if COMPAT_ARCH @@ -175,6 +175,10 @@ test_urcu_wfs_dynlink_SOURCES = test_urcu_wfs.c test_urcu_wfs_dynlink_CFLAGS = -DDYNAMIC_LINK_TEST $(AM_CFLAGS) test_urcu_wfs_dynlink_LDADD = $(URCU_COMMON_LIB) +test_urcu_hash_SOURCES = test_urcu_hash.c $(COMPAT) +test_urcu_hash_CFLAGS = -DRCU_QSBR $(AM_CFLAGS) +test_urcu_hash_LDADD = $(URCU_QSBR_LIB) $(URCU_CDS_LIB) + urcutorture.c: api.h check-am: diff --git a/tests/test_urcu_hash.c b/tests/test_urcu_hash.c new file mode 100644 index 0000000..c696533 --- /dev/null +++ b/tests/test_urcu_hash.c @@ -0,0 +1,1126 @@ +/* + * test_urcu_hash.c + * + * Userspace RCU library - test program + * + * Copyright February 2009 - Mathieu Desnoyers + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License along + * with this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. + */ + +#define _GNU_SOURCE +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#ifdef __linux__ +#include +#endif + +#define DEFAULT_HASH_SIZE 32 +#define DEFAULT_MIN_ALLOC_SIZE 1 +#define DEFAULT_RAND_POOL 1000000 + +#define TEST_HASH_SEED 0x42UL + +/* Make this big enough to include the POWER5+ L3 cacheline size of 256B */ +#define CACHE_LINE_SIZE 4096 + +/* hardcoded number of CPUs */ +#define NR_CPUS 16384 + +#ifdef POISON_FREE +#define poison_free(ptr) \ + do { \ + memset(ptr, 0x42, sizeof(*(ptr))); \ + free(ptr); \ + } while (0) +#else +#define poison_free(ptr) free(ptr) +#endif + + + +#if defined(_syscall0) +_syscall0(pid_t, gettid) +#elif defined(__NR_gettid) +static inline pid_t gettid(void) +{ + return syscall(__NR_gettid); +} +#else +#warning "use pid as tid" +static inline pid_t gettid(void) +{ + return getpid(); +} +#endif + +#ifndef DYNAMIC_LINK_TEST +#define _LGPL_SOURCE +#else +#define debug_yield_read() +#endif +#include +#include +#include + +struct wr_count { + unsigned long update_ops; + unsigned long add; + unsigned long add_exist; + unsigned long remove; +}; + +static unsigned int __thread rand_lookup; +static unsigned long __thread nr_add; +static unsigned long __thread nr_addexist; +static unsigned long __thread nr_del; +static unsigned long __thread nr_delnoent; +static unsigned long __thread lookup_fail; +static unsigned long __thread lookup_ok; + +static struct cds_lfht *test_ht; + +struct test_data { + int a; + int b; +}; + +struct lfht_test_node { + struct cds_lfht_node node; + void *key; + unsigned int key_len; + /* cache-cold for iteration */ + struct rcu_head head; +}; + +static inline struct lfht_test_node * +to_test_node(struct cds_lfht_node *node) +{ + return caa_container_of(node, struct lfht_test_node, node); +} + +static inline +void lfht_test_node_init(struct lfht_test_node *node, void *key, + size_t key_len) +{ + cds_lfht_node_init(&node->node); + node->key = key; + node->key_len = key_len; +} + +static inline struct lfht_test_node * +cds_lfht_iter_get_test_node(struct cds_lfht_iter *iter) +{ + return to_test_node(cds_lfht_iter_get_node(iter)); +} + +static volatile int test_go, test_stop; + +static unsigned long wdelay; + +static unsigned long duration; + +/* read-side C.S. duration, in loops */ +static unsigned long rduration; + +static unsigned long init_hash_size = DEFAULT_HASH_SIZE; +static unsigned long min_hash_alloc_size = DEFAULT_MIN_ALLOC_SIZE; +static unsigned long max_hash_buckets_size = (1UL << 20); +static unsigned long init_populate; +static int opt_auto_resize; +static int add_only, add_unique, add_replace; +static const struct cds_lfht_mm_type *memory_backend; + +static unsigned long init_pool_offset, lookup_pool_offset, write_pool_offset; +static unsigned long init_pool_size = DEFAULT_RAND_POOL, + lookup_pool_size = DEFAULT_RAND_POOL, + write_pool_size = DEFAULT_RAND_POOL; +static int validate_lookup; + +static int count_pipe[2]; + +static inline void loop_sleep(unsigned long l) +{ + while(l-- != 0) + caa_cpu_relax(); +} + +static int verbose_mode; + +#define printf_verbose(fmt, args...) \ + do { \ + if (verbose_mode) \ + printf(fmt, ## args); \ + } while (0) + +static unsigned int cpu_affinities[NR_CPUS]; +static unsigned int next_aff = 0; +static int use_affinity = 0; + +pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER; + +static void set_affinity(void) +{ + cpu_set_t mask; + int cpu; + int ret; + + if (!use_affinity) + return; + + ret = pthread_mutex_lock(&affinity_mutex); + if (ret) { + perror("Error in pthread mutex lock"); + exit(-1); + } + cpu = cpu_affinities[next_aff++]; + ret = pthread_mutex_unlock(&affinity_mutex); + if (ret) { + perror("Error in pthread mutex unlock"); + exit(-1); + } + CPU_ZERO(&mask); + CPU_SET(cpu, &mask); + sched_setaffinity(0, sizeof(mask), &mask); +} + +static enum { + AR_RANDOM = 0, + AR_ADD = 1, + AR_REMOVE = -1, +} addremove; /* 1: add, -1 remove, 0: random */ + +static +void sigusr1_handler(int signo) +{ + switch (addremove) { + case AR_ADD: + printf("Add/Remove: random.\n"); + addremove = AR_RANDOM; + break; + case AR_RANDOM: + printf("Add/Remove: remove only.\n"); + addremove = AR_REMOVE; + break; + case AR_REMOVE: + printf("Add/Remove: add only.\n"); + addremove = AR_ADD; + break; + } +} + +static +void sigusr2_handler(int signo) +{ + char msg[1] = { 0x42 }; + ssize_t ret; + + do { + ret = write(count_pipe[1], msg, 1); /* wakeup thread */ + } while (ret == -1L && errno == EINTR); +} + +/* + * returns 0 if test should end. + */ +static int test_duration_write(void) +{ + return !test_stop; +} + +static int test_duration_read(void) +{ + return !test_stop; +} + +static unsigned long long __thread nr_writes; +static unsigned long long __thread nr_reads; + +static unsigned int nr_readers; +static unsigned int nr_writers; + +pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER; + +void rcu_copy_mutex_lock(void) +{ + int ret; + ret = pthread_mutex_lock(&rcu_copy_mutex); + if (ret) { + perror("Error in pthread mutex lock"); + exit(-1); + } +} + +void rcu_copy_mutex_unlock(void) +{ + int ret; + + ret = pthread_mutex_unlock(&rcu_copy_mutex); + if (ret) { + perror("Error in pthread mutex unlock"); + exit(-1); + } +} + +/* + * Hash function + * Source: http://burtleburtle.net/bob/c/lookup3.c + * Originally Public Domain + */ + +#define rot(x, k) (((x) << (k)) | ((x) >> (32 - (k)))) + +#define mix(a, b, c) \ +do { \ + a -= c; a ^= rot(c, 4); c += b; \ + b -= a; b ^= rot(a, 6); a += c; \ + c -= b; c ^= rot(b, 8); b += a; \ + a -= c; a ^= rot(c, 16); c += b; \ + b -= a; b ^= rot(a, 19); a += c; \ + c -= b; c ^= rot(b, 4); b += a; \ +} while (0) + +#define final(a, b, c) \ +{ \ + c ^= b; c -= rot(b, 14); \ + a ^= c; a -= rot(c, 11); \ + b ^= a; b -= rot(a, 25); \ + c ^= b; c -= rot(b, 16); \ + a ^= c; a -= rot(c, 4);\ + b ^= a; b -= rot(a, 14); \ + c ^= b; c -= rot(b, 24); \ +} + +static __attribute__((unused)) +uint32_t hash_u32( + const uint32_t *k, /* the key, an array of uint32_t values */ + size_t length, /* the length of the key, in uint32_ts */ + uint32_t initval) /* the previous hash, or an arbitrary value */ +{ + uint32_t a, b, c; + + /* Set up the internal state */ + a = b = c = 0xdeadbeef + (((uint32_t) length) << 2) + initval; + + /*----------------------------------------- handle most of the key */ + while (length > 3) { + a += k[0]; + b += k[1]; + c += k[2]; + mix(a, b, c); + length -= 3; + k += 3; + } + + /*----------------------------------- handle the last 3 uint32_t's */ + switch (length) { /* all the case statements fall through */ + case 3: c += k[2]; + case 2: b += k[1]; + case 1: a += k[0]; + final(a, b, c); + case 0: /* case 0: nothing left to add */ + break; + } + /*---------------------------------------------- report the result */ + return c; +} + +static +void hashword2( + const uint32_t *k, /* the key, an array of uint32_t values */ + size_t length, /* the length of the key, in uint32_ts */ + uint32_t *pc, /* IN: seed OUT: primary hash value */ + uint32_t *pb) /* IN: more seed OUT: secondary hash value */ +{ + uint32_t a, b, c; + + /* Set up the internal state */ + a = b = c = 0xdeadbeef + ((uint32_t) (length << 2)) + *pc; + c += *pb; + + /*----------------------------------------- handle most of the key */ + while (length > 3) { + a += k[0]; + b += k[1]; + c += k[2]; + mix(a, b, c); + length -= 3; + k += 3; + } + + /*----------------------------------- handle the last 3 uint32_t's */ + switch (length) { /* all the case statements fall through */ + case 3: c += k[2]; + case 2: b += k[1]; + case 1: a += k[0]; + final(a, b, c); + case 0: /* case 0: nothing left to add */ + break; + } + /*---------------------------------------------- report the result */ + *pc = c; + *pb = b; +} + +#if (CAA_BITS_PER_LONG == 32) +static +unsigned long test_hash(const void *_key, size_t length, unsigned long seed) +{ + unsigned int key = (unsigned int) _key; + + assert(length == sizeof(unsigned int)); + return hash_u32(&key, 1, seed); +} +#else +static +unsigned long test_hash(const void *_key, size_t length, unsigned long seed) +{ + union { + uint64_t v64; + uint32_t v32[2]; + } v; + union { + uint64_t v64; + uint32_t v32[2]; + } key; + + assert(length == sizeof(unsigned long)); + v.v64 = (uint64_t) seed; + key.v64 = (uint64_t) _key; + hashword2(key.v32, 2, &v.v32[0], &v.v32[1]); + return v.v64; +} +#endif + +static +unsigned long test_compare(const void *key1, size_t key1_len, + const void *key2, size_t key2_len) +{ + if (caa_unlikely(key1_len != key2_len)) + return -1; + assert(key1_len == sizeof(unsigned long)); + if (key1 == key2) + return 0; + else + return 1; +} + +static +int test_match(struct cds_lfht_node *node, const void *key) +{ + struct lfht_test_node *test_node = to_test_node(node); + + return !test_compare(test_node->key, test_node->key_len, + key, sizeof(unsigned long)); +} + +static inline +void cds_lfht_test_lookup(struct cds_lfht *ht, void *key, size_t key_len, + struct cds_lfht_iter *iter) +{ + assert(key_len == sizeof(unsigned long)); + + cds_lfht_lookup(ht, test_hash(key, key_len, TEST_HASH_SEED), + test_match, key, iter); +} + +void *thr_count(void *arg) +{ + printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", + "counter", pthread_self(), (unsigned long)gettid()); + + rcu_register_thread(); + + for (;;) { + unsigned long count, removed; + long approx_before, approx_after; + ssize_t len; + char buf[1]; + + rcu_thread_offline(); + len = read(count_pipe[0], buf, 1); + rcu_thread_online(); + if (caa_unlikely(!test_duration_read())) + break; + if (len != 1) + continue; + /* Accounting */ + printf("Counting nodes... "); + fflush(stdout); + rcu_read_lock(); + cds_lfht_count_nodes(test_ht, &approx_before, &count, &removed, + &approx_after); + rcu_read_unlock(); + printf("done.\n"); + printf("Approximation before node accounting: %ld nodes.\n", + approx_before); + printf("Accounting of nodes in the hash table: " + "%lu nodes + %lu logically removed.\n", + count, removed); + printf("Approximation after node accounting: %ld nodes.\n", + approx_after); + } + rcu_unregister_thread(); + return NULL; +} + +void *thr_reader(void *_count) +{ + unsigned long long *count = _count; + struct lfht_test_node *node; + struct cds_lfht_iter iter; + + printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", + "reader", pthread_self(), (unsigned long)gettid()); + + set_affinity(); + + rcu_register_thread(); + + while (!test_go) + { + } + cmm_smp_mb(); + + for (;;) { + rcu_read_lock(); + cds_lfht_test_lookup(test_ht, + (void *)(((unsigned long) rand_r(&rand_lookup) % lookup_pool_size) + lookup_pool_offset), + sizeof(void *), &iter); + node = cds_lfht_iter_get_test_node(&iter); + if (node == NULL) { + if (validate_lookup) { + printf("[ERROR] Lookup cannot find initial node.\n"); + exit(-1); + } + lookup_fail++; + } else { + lookup_ok++; + } + debug_yield_read(); + if (caa_unlikely(rduration)) + loop_sleep(rduration); + rcu_read_unlock(); + nr_reads++; + if (caa_unlikely(!test_duration_read())) + break; + if (caa_unlikely((nr_reads & ((1 << 10) - 1)) == 0)) + rcu_quiescent_state(); + } + + rcu_unregister_thread(); + + *count = nr_reads; + printf_verbose("thread_end %s, thread id : %lx, tid %lu\n", + "reader", pthread_self(), (unsigned long)gettid()); + printf_verbose("readid : %lx, lookupfail %lu, lookupok %lu\n", + pthread_self(), lookup_fail, lookup_ok); + return ((void*)1); + +} + +static +void free_node_cb(struct rcu_head *head) +{ + struct lfht_test_node *node = + caa_container_of(head, struct lfht_test_node, head); + free(node); +} + +void *thr_writer(void *_count) +{ + struct lfht_test_node *node; + struct cds_lfht_node *ret_node; + struct cds_lfht_iter iter; + struct wr_count *count = _count; + int ret; + + printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", + "writer", pthread_self(), (unsigned long)gettid()); + + set_affinity(); + + rcu_register_thread(); + + while (!test_go) + { + } + cmm_smp_mb(); + + for (;;) { + if ((addremove == AR_ADD || add_only) + || (addremove == AR_RANDOM && rand_r(&rand_lookup) & 1)) { + node = malloc(sizeof(struct lfht_test_node)); + lfht_test_node_init(node, + (void *)(((unsigned long) rand_r(&rand_lookup) % write_pool_size) + write_pool_offset), + sizeof(void *)); + rcu_read_lock(); + if (add_unique) { + ret_node = cds_lfht_add_unique(test_ht, + test_hash(node->key, node->key_len, TEST_HASH_SEED), + test_match, node->key, &node->node); + } else { + if (add_replace) + ret_node = cds_lfht_add_replace(test_ht, + test_hash(node->key, node->key_len, TEST_HASH_SEED), + test_match, node->key, &node->node); + else + cds_lfht_add(test_ht, + test_hash(node->key, node->key_len, TEST_HASH_SEED), + &node->node); + } + rcu_read_unlock(); + if (add_unique && ret_node != &node->node) { + free(node); + nr_addexist++; + } else { + if (add_replace && ret_node) { + call_rcu(&to_test_node(ret_node)->head, + free_node_cb); + nr_addexist++; + } else { + nr_add++; + } + } + } else { + /* May delete */ + rcu_read_lock(); + cds_lfht_test_lookup(test_ht, + (void *)(((unsigned long) rand_r(&rand_lookup) % write_pool_size) + write_pool_offset), + sizeof(void *), &iter); + ret = cds_lfht_del(test_ht, &iter); + rcu_read_unlock(); + if (ret == 0) { + node = cds_lfht_iter_get_test_node(&iter); + call_rcu(&node->head, free_node_cb); + nr_del++; + } else + nr_delnoent++; + } +#if 0 + //if (nr_writes % 100000 == 0) { + if (nr_writes % 1000 == 0) { + rcu_read_lock(); + if (rand_r(&rand_lookup) & 1) { + ht_resize(test_ht, 1); + } else { + ht_resize(test_ht, -1); + } + rcu_read_unlock(); + } +#endif //0 + nr_writes++; + if (caa_unlikely(!test_duration_write())) + break; + if (caa_unlikely(wdelay)) + loop_sleep(wdelay); + if (caa_unlikely((nr_writes & ((1 << 10) - 1)) == 0)) + rcu_quiescent_state(); + } + + rcu_unregister_thread(); + + printf_verbose("thread_end %s, thread id : %lx, tid %lu\n", + "writer", pthread_self(), (unsigned long)gettid()); + printf_verbose("info id %lx: nr_add %lu, nr_addexist %lu, nr_del %lu, " + "nr_delnoent %lu\n", pthread_self(), nr_add, + nr_addexist, nr_del, nr_delnoent); + count->update_ops = nr_writes; + count->add = nr_add; + count->add_exist = nr_addexist; + count->remove = nr_del; + return ((void*)2); +} + +static int populate_hash(void) +{ + struct lfht_test_node *node; + struct cds_lfht_node *ret_node; + + if (!init_populate) + return 0; + + if ((add_unique || add_replace) && init_populate * 10 > init_pool_size) { + printf("WARNING: required to populate %lu nodes (-k), but random " +"pool is quite small (%lu values) and we are in add_unique (-u) or add_replace (-s) mode. Try with a " +"larger random pool (-p option). This may take a while...\n", init_populate, init_pool_size); + } + + while (nr_add < init_populate) { + node = malloc(sizeof(struct lfht_test_node)); + lfht_test_node_init(node, + (void *)(((unsigned long) rand_r(&rand_lookup) % init_pool_size) + init_pool_offset), + sizeof(void *)); + rcu_read_lock(); + if (add_unique) { + ret_node = cds_lfht_add_unique(test_ht, + test_hash(node->key, node->key_len, TEST_HASH_SEED), + test_match, node->key, &node->node); + } else { + if (add_replace) + ret_node = cds_lfht_add_replace(test_ht, + test_hash(node->key, node->key_len, TEST_HASH_SEED), + test_match, node->key, &node->node); + else + cds_lfht_add(test_ht, + test_hash(node->key, node->key_len, TEST_HASH_SEED), + &node->node); + } + rcu_read_unlock(); + if (add_unique && ret_node != &node->node) { + free(node); + nr_addexist++; + } else { + if (add_replace && ret_node) { + call_rcu(&to_test_node(ret_node)->head, free_node_cb); + nr_addexist++; + } else { + nr_add++; + } + } + nr_writes++; + } + return 0; +} + +static +void test_delete_all_nodes(struct cds_lfht *ht) +{ + struct cds_lfht_iter iter; + struct lfht_test_node *node; + unsigned long count = 0; + + cds_lfht_for_each_entry(ht, &iter, node, node) { + int ret; + + ret = cds_lfht_del(test_ht, &iter); + assert(!ret); + call_rcu(&node->head, free_node_cb); + count++; + } + printf("deleted %lu nodes.\n", count); +} + +void show_usage(int argc, char **argv) +{ + printf("Usage : %s nr_readers nr_writers duration (s)\n", argv[0]); +#ifdef DEBUG_YIELD + printf(" [-r] [-w] (yield reader and/or writer)\n"); +#endif + printf(" [-d delay] (writer period (us))\n"); + printf(" [-c duration] (reader C.S. duration (in loops))\n"); + printf(" [-v] (verbose output)\n"); + printf(" [-a cpu#] [-a cpu#]... (affinity)\n"); + printf(" [-h size] (initial number of buckets)\n"); + printf(" [-m size] (minimum number of allocated buckets)\n"); + printf(" [-n size] (maximum number of buckets)\n"); + printf(" [not -u nor -s] Add entries (supports redundant keys).\n"); + printf(" [-u] Uniquify add (no redundant keys).\n"); + printf(" [-s] Replace (swap) entries.\n"); + printf(" [-i] Add only (no removal).\n"); + printf(" [-k nr_nodes] Number of nodes to insert initially.\n"); + printf(" [-A] Automatically resize hash table.\n"); + printf(" [-B order|chunk|mmap] Specify the memory backend.\n"); + printf(" [-R offset] Lookup pool offset.\n"); + printf(" [-S offset] Write pool offset.\n"); + printf(" [-T offset] Init pool offset.\n"); + printf(" [-M size] Lookup pool size.\n"); + printf(" [-N size] Write pool size.\n"); + printf(" [-O size] Init pool size.\n"); + printf(" [-V] Validate lookups of init values (use with filled init pool, same lookup range, with different write range).\n"); + printf("\n\n"); +} + +int main(int argc, char **argv) +{ + int err; + pthread_t *tid_reader, *tid_writer; + pthread_t tid_count; + void *tret; + unsigned long long *count_reader; + struct wr_count *count_writer; + unsigned long long tot_reads = 0, tot_writes = 0, + tot_add = 0, tot_add_exist = 0, tot_remove = 0; + unsigned long count, removed; + long approx_before, approx_after; + int i, a, ret; + struct sigaction act; + unsigned int remain; + + if (argc < 4) { + show_usage(argc, argv); + return -1; + } + + err = sscanf(argv[1], "%u", &nr_readers); + if (err != 1) { + show_usage(argc, argv); + return -1; + } + + err = sscanf(argv[2], "%u", &nr_writers); + if (err != 1) { + show_usage(argc, argv); + return -1; + } + + err = sscanf(argv[3], "%lu", &duration); + if (err != 1) { + show_usage(argc, argv); + return -1; + } + + for (i = 4; i < argc; i++) { + if (argv[i][0] != '-') + continue; + switch (argv[i][1]) { +#ifdef DEBUG_YIELD + case 'r': + yield_active |= YIELD_READ; + break; + case 'w': + yield_active |= YIELD_WRITE; + break; +#endif + case 'a': + if (argc < i + 2) { + show_usage(argc, argv); + return -1; + } + a = atoi(argv[++i]); + cpu_affinities[next_aff++] = a; + use_affinity = 1; + printf_verbose("Adding CPU %d affinity\n", a); + break; + case 'c': + if (argc < i + 2) { + show_usage(argc, argv); + return -1; + } + rduration = atol(argv[++i]); + break; + case 'd': + if (argc < i + 2) { + show_usage(argc, argv); + return -1; + } + wdelay = atol(argv[++i]); + break; + case 'v': + verbose_mode = 1; + break; + case 'h': + if (argc < i + 2) { + show_usage(argc, argv); + return -1; + } + init_hash_size = atol(argv[++i]); + break; + case 'm': + if (argc < i + 2) { + show_usage(argc, argv); + return -1; + } + min_hash_alloc_size = atol(argv[++i]); + break; + case 'n': + if (argc < i + 2) { + show_usage(argc, argv); + return -1; + } + max_hash_buckets_size = atol(argv[++i]); + break; + case 'u': + if (add_replace) { + printf("Please specify at most one of -s or -u.\n"); + exit(-1); + } + add_unique = 1; + break; + case 's': + if (add_unique) { + printf("Please specify at most one of -s or -u.\n"); + exit(-1); + } + add_replace = 1; + break; + case 'i': + add_only = 1; + break; + case 'k': + init_populate = atol(argv[++i]); + break; + case 'A': + opt_auto_resize = 1; + break; + case 'B': + if (argc < i + 2) { + show_usage(argc, argv); + return -1; + } + i++; + if (!strcmp("order", argv[i])) + memory_backend = &cds_lfht_mm_order; + else if (!strcmp("chunk", argv[i])) + memory_backend = &cds_lfht_mm_chunk; + else if (!strcmp("mmap", argv[i])) + memory_backend = &cds_lfht_mm_mmap; + else { + printf("Please specify memory backend with order|chunk|mmap.\n"); + exit(-1); + } + break; + case 'R': + lookup_pool_offset = atol(argv[++i]); + break; + case 'S': + write_pool_offset = atol(argv[++i]); + break; + case 'T': + init_pool_offset = atol(argv[++i]); + break; + case 'M': + lookup_pool_size = atol(argv[++i]); + break; + case 'N': + write_pool_size = atol(argv[++i]); + break; + case 'O': + init_pool_size = atol(argv[++i]); + break; + case 'V': + validate_lookup = 1; + break; + + } + } + + /* Check if hash size is power of 2 */ + if (init_hash_size && init_hash_size & (init_hash_size - 1)) { + printf("Error: Initial number of buckets (%lu) is not a power of 2.\n", + init_hash_size); + return -1; + } + + if (min_hash_alloc_size && min_hash_alloc_size & (min_hash_alloc_size - 1)) { + printf("Error: Minimum number of allocated buckets (%lu) is not a power of 2.\n", + min_hash_alloc_size); + return -1; + } + + if (max_hash_buckets_size && max_hash_buckets_size & (max_hash_buckets_size - 1)) { + printf("Error: Maximum number of buckets (%lu) is not a power of 2.\n", + max_hash_buckets_size); + return -1; + } + + memset(&act, 0, sizeof(act)); + ret = sigemptyset(&act.sa_mask); + if (ret == -1) { + perror("sigemptyset"); + return -1; + } + act.sa_handler = sigusr1_handler; + act.sa_flags = SA_RESTART; + ret = sigaction(SIGUSR1, &act, NULL); + if (ret == -1) { + perror("sigaction"); + return -1; + } + + ret = pipe(count_pipe); + if (ret == -1) { + perror("pipe"); + return -1; + } + + /* spawn counter thread */ + err = pthread_create(&tid_count, NULL, thr_count, + NULL); + if (err != 0) + exit(1); + + act.sa_handler = sigusr2_handler; + act.sa_flags = SA_RESTART; + ret = sigaction(SIGUSR2, &act, NULL); + if (ret == -1) { + perror("sigaction"); + return -1; + } + + printf_verbose("running test for %lu seconds, %u readers, %u writers.\n", + duration, nr_readers, nr_writers); + printf_verbose("Writer delay : %lu loops.\n", wdelay); + printf_verbose("Reader duration : %lu loops.\n", rduration); + printf_verbose("Mode:%s%s.\n", + add_only ? " add only" : " add/remove", + add_unique ? " uniquify" : ( add_replace ? " replace" : " insert")); + printf_verbose("Initial number of buckets: %lu buckets.\n", init_hash_size); + printf_verbose("Minimum number of allocated buckets: %lu buckets.\n", min_hash_alloc_size); + printf_verbose("Maximum number of buckets: %lu buckets.\n", max_hash_buckets_size); + printf_verbose("Init pool size offset %lu size %lu.\n", + init_pool_offset, init_pool_size); + printf_verbose("Lookup pool size offset %lu size %lu.\n", + lookup_pool_offset, lookup_pool_size); + printf_verbose("Update pool size offset %lu size %lu.\n", + write_pool_offset, write_pool_size); + printf_verbose("thread %-6s, thread id : %lx, tid %lu\n", + "main", pthread_self(), (unsigned long)gettid()); + + tid_reader = malloc(sizeof(*tid_reader) * nr_readers); + tid_writer = malloc(sizeof(*tid_writer) * nr_writers); + count_reader = malloc(sizeof(*count_reader) * nr_readers); + count_writer = malloc(sizeof(*count_writer) * nr_writers); + + err = create_all_cpu_call_rcu_data(0); + assert(!err); + + if (memory_backend) { + test_ht = _cds_lfht_new(init_hash_size, min_hash_alloc_size, + max_hash_buckets_size, + (opt_auto_resize ? CDS_LFHT_AUTO_RESIZE : 0) | + CDS_LFHT_ACCOUNTING, memory_backend, + &rcu_flavor, NULL); + } else { + test_ht = cds_lfht_new(init_hash_size, min_hash_alloc_size, + max_hash_buckets_size, + (opt_auto_resize ? CDS_LFHT_AUTO_RESIZE : 0) | + CDS_LFHT_ACCOUNTING, NULL); + } + + /* + * Hash Population needs to be seen as a RCU reader + * thread from the point of view of resize. + */ + rcu_register_thread(); + ret = populate_hash(); + assert(!ret); + + rcu_thread_offline(); + + next_aff = 0; + + for (i = 0; i < nr_readers; i++) { + err = pthread_create(&tid_reader[i], NULL, thr_reader, + &count_reader[i]); + if (err != 0) + exit(1); + } + for (i = 0; i < nr_writers; i++) { + err = pthread_create(&tid_writer[i], NULL, thr_writer, + &count_writer[i]); + if (err != 0) + exit(1); + } + + cmm_smp_mb(); + + test_go = 1; + + remain = duration; + do { + remain = sleep(remain); + } while (remain > 0); + + test_stop = 1; + + for (i = 0; i < nr_readers; i++) { + err = pthread_join(tid_reader[i], &tret); + if (err != 0) + exit(1); + tot_reads += count_reader[i]; + } + for (i = 0; i < nr_writers; i++) { + err = pthread_join(tid_writer[i], &tret); + if (err != 0) + exit(1); + tot_writes += count_writer[i].update_ops; + tot_add += count_writer[i].add; + tot_add_exist += count_writer[i].add_exist; + tot_remove += count_writer[i].remove; + } + + /* teardown counter thread */ + act.sa_handler = SIG_IGN; + act.sa_flags = SA_RESTART; + ret = sigaction(SIGUSR2, &act, NULL); + if (ret == -1) { + perror("sigaction"); + return -1; + } + { + char msg[1] = { 0x42 }; + ssize_t ret; + + do { + ret = write(count_pipe[1], msg, 1); /* wakeup thread */ + } while (ret == -1L && errno == EINTR); + } + err = pthread_join(tid_count, &tret); + if (err != 0) + exit(1); + + fflush(stdout); + rcu_thread_online(); + rcu_read_lock(); + printf("Counting nodes... "); + cds_lfht_count_nodes(test_ht, &approx_before, &count, &removed, + &approx_after); + printf("done.\n"); + test_delete_all_nodes(test_ht); + rcu_read_unlock(); + rcu_thread_offline(); + if (count || removed) { + printf("Approximation before node accounting: %ld nodes.\n", + approx_before); + printf("Nodes deleted from hash table before destroy: " + "%lu nodes + %lu logically removed.\n", + count, removed); + printf("Approximation after node accounting: %ld nodes.\n", + approx_after); + } + ret = cds_lfht_destroy(test_ht, NULL); + if (ret) + printf_verbose("final delete aborted\n"); + else + printf_verbose("final delete success\n"); + printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads, + tot_writes); + printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu " + "nr_writers %3u " + "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu " + "nr_add %12llu nr_add_fail %12llu nr_remove %12llu nr_leaked %12lld\n", + argv[0], duration, nr_readers, rduration, + nr_writers, wdelay, tot_reads, tot_writes, + tot_reads + tot_writes, tot_add, tot_add_exist, tot_remove, + (long long) tot_add + init_populate - tot_remove - count); + rcu_unregister_thread(); + free_all_cpu_call_rcu_data(); + free(tid_reader); + free(tid_writer); + free(count_reader); + free(count_writer); + return 0; +} diff --git a/urcu-bp.c b/urcu-bp.c index f3249b4..912e9b2 100644 --- a/urcu-bp.c +++ b/urcu-bp.c @@ -444,5 +444,7 @@ void *rcu_cmpxchg_pointer_sym_bp(void **p, void *old, void *_new) return uatomic_cmpxchg(p, old, _new); } +DEFINE_RCU_FLAVOR() + #include "urcu-call-rcu-impl.h" #include "urcu-defer-impl.h" diff --git a/urcu-bp.h b/urcu-bp.h index 34f7a5c..3623f35 100644 --- a/urcu-bp.h +++ b/urcu-bp.h @@ -186,5 +186,6 @@ static inline void rcu_thread_online(void) #include #include +#include #endif /* _URCU_BP_H */ diff --git a/urcu-flavor.h b/urcu-flavor.h new file mode 100644 index 0000000..e46c9a1 --- /dev/null +++ b/urcu-flavor.h @@ -0,0 +1,65 @@ +#ifndef _URCU_FLAVOR_H +#define _URCU_FLAVOR_H + +/* + * urcu-flavor.h + * + * Userspace RCU header - rcu flavor declarations + * + * Copyright (c) 2011 Lai Jiangshan + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifdef __cplusplus +extern "C" { +#endif + +struct rcu_flavor_struct { + void (*read_lock)(void); + void (*read_unlock)(void); + void (*read_quiescent_state)(void); + void (*update_call_rcu)(struct rcu_head *head, + void (*func)(struct rcu_head *head)); + void (*update_synchronize_rcu)(void); + void (*update_defer_rcu)(void (*fct)(void *p), void *p); + + void (*thread_offline)(void); + void (*thread_online)(void); + void (*register_thread)(void); + void (*unregister_thread)(void); +}; + +#define DEFINE_RCU_FLAVOR() \ +const struct rcu_flavor_struct rcu_flavor = { \ + .read_lock = rcu_read_lock, \ + .read_unlock = rcu_read_unlock, \ + .read_quiescent_state = rcu_quiescent_state, \ + .update_call_rcu = call_rcu, \ + .update_synchronize_rcu = synchronize_rcu, \ + .update_defer_rcu = defer_rcu, \ + .thread_offline = rcu_thread_offline, \ + .thread_online = rcu_thread_online, \ + .register_thread = rcu_register_thread, \ + .unregister_thread = rcu_unregister_thread,\ +}; + +extern const struct rcu_flavor_struct rcu_flavor; + +#ifdef __cplusplus +} +#endif + +#endif /* _URCU_FLAVOR_H */ diff --git a/urcu-qsbr.c b/urcu-qsbr.c index 5530295..06e81c7 100644 --- a/urcu-qsbr.c +++ b/urcu-qsbr.c @@ -356,5 +356,7 @@ void rcu_exit(void) */ } +DEFINE_RCU_FLAVOR() + #include "urcu-call-rcu-impl.h" #include "urcu-defer-impl.h" diff --git a/urcu-qsbr.h b/urcu-qsbr.h index b362304..a2f6575 100644 --- a/urcu-qsbr.h +++ b/urcu-qsbr.h @@ -127,5 +127,6 @@ extern void rcu_unregister_thread(void); #include #include +#include #endif /* _URCU_QSBR_H */ diff --git a/urcu.c b/urcu.c index ba013d9..b434655 100644 --- a/urcu.c +++ b/urcu.c @@ -452,5 +452,7 @@ void rcu_exit(void) #endif /* #ifdef RCU_SIGNAL */ +DEFINE_RCU_FLAVOR() + #include "urcu-call-rcu-impl.h" #include "urcu-defer-impl.h" diff --git a/urcu.h b/urcu.h index 1ad971c..cd4fcbb 100644 --- a/urcu.h +++ b/urcu.h @@ -128,5 +128,6 @@ static inline void rcu_thread_online(void) #include #include +#include #endif /* _URCU_H */ diff --git a/urcu/cds.h b/urcu/cds.h index f37a63a..ea74cf1 100644 --- a/urcu/cds.h +++ b/urcu/cds.h @@ -29,6 +29,7 @@ #include #include #include +#include #include #include diff --git a/urcu/jhash.h b/urcu/jhash.h new file mode 100644 index 0000000..def03b8 --- /dev/null +++ b/urcu/jhash.h @@ -0,0 +1,89 @@ +#ifndef _KCOMPAT_JHASH_H +#define _KCOMPAT_JHASH_H + +/* jhash.h: Jenkins hash support. + * + * Copyright (C) 1996 Bob Jenkins (bob_jenkins@burtleburtle.net) + * + * http://burtleburtle.net/bob/hash/ + * + * These are the credits from Bob's sources: + * + * lookup2.c, by Bob Jenkins, December 1996, Public Domain. + * hash(), hash2(), hash3, and mix() are externally useful functions. + * Routines to test the hash are included if SELF_TEST is defined. + * You can use this free for any purpose. It has no warranty. + * + * Copyright (C) 2003 David S. Miller (davem@redhat.com) + * + * I've modified Bob's hash to be useful in the Linux kernel, and + * any bugs present are surely my fault. -DaveM + */ + +#include + +typedef uint8_t u8; +typedef uint32_t u32; + +/* NOTE: Arguments are modified. */ +#define __jhash_mix(a, b, c) \ +{ \ + a -= b; a -= c; a ^= (c>>13); \ + b -= c; b -= a; b ^= (a<<8); \ + c -= a; c -= b; c ^= (b>>13); \ + a -= b; a -= c; a ^= (c>>12); \ + b -= c; b -= a; b ^= (a<<16); \ + c -= a; c -= b; c ^= (b>>5); \ + a -= b; a -= c; a ^= (c>>3); \ + b -= c; b -= a; b ^= (a<<10); \ + c -= a; c -= b; c ^= (b>>15); \ +} + +/* The golden ration: an arbitrary value */ +#define JHASH_GOLDEN_RATIO 0x9e3779b9 + +/* The most generic version, hashes an arbitrary sequence + * of bytes. No alignment or length assumptions are made about + * the input key. + */ +static inline u32 jhash(const void *key, u32 length, u32 initval) +{ + u32 a, b, c, len; + const u8 *k = key; + + len = length; + a = b = JHASH_GOLDEN_RATIO; + c = initval; + + while (len >= 12) { + a += (k[0] +((u32)k[1]<<8) +((u32)k[2]<<16) +((u32)k[3]<<24)); + b += (k[4] +((u32)k[5]<<8) +((u32)k[6]<<16) +((u32)k[7]<<24)); + c += (k[8] +((u32)k[9]<<8) +((u32)k[10]<<16)+((u32)k[11]<<24)); + + __jhash_mix(a,b,c); + + k += 12; + len -= 12; + } + + c += length; + switch (len) { + case 11: c += ((u32)k[10]<<24); + case 10: c += ((u32)k[9]<<16); + case 9 : c += ((u32)k[8]<<8); + case 8 : b += ((u32)k[7]<<24); + case 7 : b += ((u32)k[6]<<16); + case 6 : b += ((u32)k[5]<<8); + case 5 : b += k[4]; + case 4 : a += ((u32)k[3]<<24); + case 3 : a += ((u32)k[2]<<16); + case 2 : a += ((u32)k[1]<<8); + case 1 : a += k[0]; + }; + + __jhash_mix(a,b,c); + + return c; +} + +#endif /* _KCOMPAT_JHASH_H */ diff --git a/urcu/map/urcu-bp.h b/urcu/map/urcu-bp.h index 4abe8dc..16601a1 100644 --- a/urcu/map/urcu-bp.h +++ b/urcu/map/urcu-bp.h @@ -64,4 +64,6 @@ #define rcu_defer_barrier rcu_defer_barrier_bp #define rcu_defer_barrier_thread rcu_defer_barrier_thread_bp +#define rcu_flavor rcu_flavor_bp + #endif /* _URCU_BP_MAP_H */ diff --git a/urcu/map/urcu-qsbr.h b/urcu/map/urcu-qsbr.h index 0d88d83..a95441d 100644 --- a/urcu/map/urcu-qsbr.h +++ b/urcu/map/urcu-qsbr.h @@ -66,4 +66,6 @@ #define rcu_defer_barrier rcu_defer_barrier_qsbr #define rcu_defer_barrier_thread rcu_defer_barrier_thread_qsbr +#define rcu_flavor rcu_flavor_qsbr + #endif /* _URCU_QSBR_MAP_H */ diff --git a/urcu/map/urcu.h b/urcu/map/urcu.h index 3f436a7..dd7a691 100644 --- a/urcu/map/urcu.h +++ b/urcu/map/urcu.h @@ -95,6 +95,8 @@ #define rcu_defer_barrier rcu_defer_barrier_memb #define rcu_defer_barrier_thread rcu_defer_barrier_thread_memb +#define rcu_flavor rcu_flavor_memb + #elif defined(RCU_SIGNAL) #define rcu_read_lock rcu_read_lock_sig @@ -127,6 +129,8 @@ #define rcu_defer_barrier rcu_defer_barrier_sig #define rcu_defer_barrier_thread rcu_defer_barrier_thread_sig +#define rcu_flavor rcu_flavor_sig + #elif defined(RCU_MB) #define rcu_read_lock rcu_read_lock_mb @@ -159,6 +163,8 @@ #define rcu_defer_barrier rcu_defer_barrier_mb #define rcu_defer_barrier_thread rcu_defer_barrier_thread_mb +#define rcu_flavor rcu_flavor_mb + #else #error "Undefined selection" diff --git a/urcu/rculfhash.h b/urcu/rculfhash.h new file mode 100644 index 0000000..de31526 --- /dev/null +++ b/urcu/rculfhash.h @@ -0,0 +1,419 @@ +#ifndef _URCU_RCULFHASH_H +#define _URCU_RCULFHASH_H + +/* + * urcu/rculfhash.h + * + * Userspace RCU library - Lock-Free RCU Hash Table + * + * Copyright 2011 - Mathieu Desnoyers + * Copyright 2011 - Lai Jiangshan + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + * Include this file _after_ including your URCU flavor. + */ + +#include +#include +#include +#include + +#ifdef __cplusplus +extern "C" { +#endif + +/* + * cds_lfht_node: Contains the next pointers and reverse-hash + * value required for lookup and traversal of the hash table. + * + * struct cds_lfht_node should be aligned on 4-bytes boundaries because + * the two lower bits are used as flags. + * + * struct cds_lfht_node can be embedded into a structure (as a field). + * caa_container_of() can be used to get the structure from the struct + * cds_lfht_node after a lookup. + * + * The structure which embeds it typically holds the key (or key-value + * pair) of the object. The caller code is responsible for calculation + * of the hash value for cds_lfht APIs. + */ +struct cds_lfht_node { + struct cds_lfht_node *next; /* ptr | BUCKET_FLAG | REMOVED_FLAG */ + unsigned long reverse_hash; +} __attribute__((aligned(4))); + +/* cds_lfht_iter: Used to track state while traversing a hash chain. */ +struct cds_lfht_iter { + struct cds_lfht_node *node, *next; +}; + +static inline +struct cds_lfht_node *cds_lfht_iter_get_node(struct cds_lfht_iter *iter) +{ + return iter->node; +} + +struct cds_lfht; + +/* + * Caution ! + * Ensure reader and writer threads are registered as urcu readers. + */ + +typedef int (*cds_lfht_match_fct)(struct cds_lfht_node *node, const void *key); + +/* + * cds_lfht_node_init - initialize a hash table node + * @node: the node to initialize. + * + * This function is kept to be eventually used for debugging purposes + * (detection of memory corruption). + */ +static inline +void cds_lfht_node_init(struct cds_lfht_node *node) +{ +} + +/* + * Hash table creation flags. + */ +enum { + CDS_LFHT_AUTO_RESIZE = (1U << 0), + CDS_LFHT_ACCOUNTING = (1U << 1), +}; + +struct cds_lfht_mm_type { + struct cds_lfht *(*alloc_cds_lfht)(unsigned long min_nr_alloc_buckets, + unsigned long max_nr_buckets); + void (*alloc_bucket_table)(struct cds_lfht *ht, unsigned long order); + void (*free_bucket_table)(struct cds_lfht *ht, unsigned long order); + struct cds_lfht_node *(*bucket_at)(struct cds_lfht *ht, + unsigned long index); +}; + +extern const struct cds_lfht_mm_type cds_lfht_mm_order; +extern const struct cds_lfht_mm_type cds_lfht_mm_chunk; +extern const struct cds_lfht_mm_type cds_lfht_mm_mmap; + +/* + * _cds_lfht_new - API used by cds_lfht_new wrapper. Do not use directly. + */ +struct cds_lfht *_cds_lfht_new(unsigned long init_size, + unsigned long min_nr_alloc_buckets, + unsigned long max_nr_buckets, + int flags, + const struct cds_lfht_mm_type *mm, + const struct rcu_flavor_struct *flavor, + pthread_attr_t *attr); + +/* + * cds_lfht_new - allocate a hash table. + * @init_size: number of buckets to allocate initially. Must be power of two. + * @min_nr_alloc_buckets: the minimum number of allocated buckets. + * (must be power of two) + * @max_nr_buckets: the maximum number of hash table buckets allowed. + * (must be power of two) + * @flags: hash table creation flags (can be combined with bitwise or: '|'). + * 0: no flags. + * CDS_LFHT_AUTO_RESIZE: automatically resize hash table. + * CDS_LFHT_ACCOUNTING: count the number of node addition + * and removal in the table + * @attr: optional resize worker thread attributes. NULL for default. + * + * Return NULL on error. + * Note: the RCU flavor must be already included before the hash table header. + * + * The programmer is responsible for ensuring that resize operation has a + * priority equal to hash table updater threads. It should be performed by + * specifying the appropriate priority in the pthread "attr" argument, and, + * for CDS_LFHT_AUTO_RESIZE, by ensuring that call_rcu worker threads also have + * this priority level. Having lower priority for call_rcu and resize threads + * does not pose any correctness issue, but the resize operations could be + * starved by updates, thus leading to long hash table bucket chains. + * Threads calling this API are NOT required to be registered RCU read-side + * threads. It can be called very early.(before rcu is initialized ...etc.) + */ +static inline +struct cds_lfht *cds_lfht_new(unsigned long init_size, + unsigned long min_nr_alloc_buckets, + unsigned long max_nr_buckets, + int flags, + pthread_attr_t *attr) +{ + return _cds_lfht_new(init_size, min_nr_alloc_buckets, max_nr_buckets, + flags, NULL, &rcu_flavor, attr); +} + +/* + * cds_lfht_destroy - destroy a hash table. + * @ht: the hash table to destroy. + * @attr: (output) resize worker thread attributes, as received by cds_lfht_new. + * The caller will typically want to free this pointer if dynamically + * allocated. The attr point can be NULL if the caller does not + * need to be informed of the value passed to cds_lfht_new(). + * + * Return 0 on success, negative error value on error. + * Threads calling this API need to be registered RCU read-side threads. + */ +int cds_lfht_destroy(struct cds_lfht *ht, pthread_attr_t **attr); + +/* + * cds_lfht_count_nodes - count the number of nodes in the hash table. + * @ht: the hash table. + * @split_count_before: Sample the node count split-counter before traversal. + * @count: Traverse the hash table, count the number of nodes observed. + * @removed: Number of logically removed nodes observed during traversal. + * @split_count_after: Sample the node count split-counter after traversal. + * + * Call with rcu_read_lock held. + * Threads calling this API need to be registered RCU read-side threads. + */ +void cds_lfht_count_nodes(struct cds_lfht *ht, + long *split_count_before, + unsigned long *count, + unsigned long *removed, + long *split_count_after); + +/* + * cds_lfht_lookup - lookup a node by key. + * @ht: the hash table. + * @hash: the key hash. + * @match: the key match function. + * @key: the current node key. + * @iter: Node, if found (output). *iter->node set to NULL if not found. + * + * Call with rcu_read_lock held. + * Threads calling this API need to be registered RCU read-side threads. + */ +void cds_lfht_lookup(struct cds_lfht *ht, unsigned long hash, + cds_lfht_match_fct match, const void *key, + struct cds_lfht_iter *iter); + +/* + * cds_lfht_next_duplicate - get the next item with same key (after a lookup). + * @ht: the hash table. + * @match: the key match function. + * @key: the current node key. + * @iter: Node, if found (output). *iter->node set to NULL if not found. + * + * Uses an iterator initialized by a lookup. + * Sets *iter-node to the following node with same key. + * Sets *iter->node to NULL if no following node exists with same key. + * RCU read-side lock must be held across cds_lfht_lookup and + * cds_lfht_next calls, and also between cds_lfht_next calls using the + * node returned by a previous cds_lfht_next. + * Call with rcu_read_lock held. + * Threads calling this API need to be registered RCU read-side threads. + */ +void cds_lfht_next_duplicate(struct cds_lfht *ht, + cds_lfht_match_fct match, const void *key, + struct cds_lfht_iter *iter); + +/* + * cds_lfht_first - get the first node in the table. + * @ht: the hash table. + * @iter: First node, if exists (output). *iter->node set to NULL if not found. + * + * Output in "*iter". *iter->node set to NULL if table is empty. + * Call with rcu_read_lock held. + * Threads calling this API need to be registered RCU read-side threads. + */ +void cds_lfht_first(struct cds_lfht *ht, struct cds_lfht_iter *iter); + +/* + * cds_lfht_next - get the next node in the table. + * @ht: the hash table. + * @iter: Next node, if exists (output). *iter->node set to NULL if not found. + * + * Input/Output in "*iter". *iter->node set to NULL if *iter was + * pointing to the last table node. + * Call with rcu_read_lock held. + * Threads calling this API need to be registered RCU read-side threads. + */ +void cds_lfht_next(struct cds_lfht *ht, struct cds_lfht_iter *iter); + +/* + * cds_lfht_add - add a node to the hash table. + * @ht: the hash table. + * @hash: the key hash. + * @node: the node to add. + * + * This function supports adding redundant keys into the table. + * Call with rcu_read_lock held. + * Threads calling this API need to be registered RCU read-side threads. + */ +void cds_lfht_add(struct cds_lfht *ht, unsigned long hash, + struct cds_lfht_node *node); + +/* + * cds_lfht_add_unique - add a node to hash table, if key is not present. + * @ht: the hash table. + * @hash: the node's hash. + * @match: the key match function. + * @key: the node's key. + * @node: the node to try adding. + * + * Return the node added upon success. + * Return the unique node already present upon failure. If + * cds_lfht_add_unique fails, the node passed as parameter should be + * freed by the caller. + * Call with rcu_read_lock held. + * Threads calling this API need to be registered RCU read-side threads. + * + * The semantic of this function is that if only this function is used + * to add keys into the table, no duplicated keys should ever be + * observable in the table. The same guarantee apply for combination of + * add_unique and add_replace (see below). + */ +struct cds_lfht_node *cds_lfht_add_unique(struct cds_lfht *ht, + unsigned long hash, + cds_lfht_match_fct match, + const void *key, + struct cds_lfht_node *node); + +/* + * cds_lfht_add_replace - replace or add a node within hash table. + * @ht: the hash table. + * @hash: the node's hash. + * @match: the key match function. + * @key: the node's key. + * @node: the node to add. + * + * Return the node replaced upon success. If no node matching the key + * was present, return NULL, which also means the operation succeeded. + * This replacement operation should never fail. + * Call with rcu_read_lock held. + * Threads calling this API need to be registered RCU read-side threads. + * After successful replacement, a grace period must be waited for before + * freeing the memory reserved for the returned node. + * + * The semantic of replacement vs lookups is the following: if lookups + * are performed between a key unique insertion and its removal, we + * guarantee that the lookups and get next will always find exactly one + * instance of the key if it is replaced concurrently with the lookups. + * + * Providing this semantic allows us to ensure that replacement-only + * schemes will never generate duplicated keys. It also allows us to + * guarantee that a combination of add_replace and add_unique updates + * will never generate duplicated keys. + */ +struct cds_lfht_node *cds_lfht_add_replace(struct cds_lfht *ht, + unsigned long hash, + cds_lfht_match_fct match, + const void *key, + struct cds_lfht_node *node); + +/* + * cds_lfht_replace - replace a node pointer to by iter within hash table. + * @ht: the hash table. + * @old_iter: the iterator position of the node to replace. + * @now_node: the new node to try using for replacement. + * + * Return 0 if replacement is successful, negative value otherwise. + * Replacing a NULL old node or an already removed node will fail with a + * negative value. + * Old node can be looked up with cds_lfht_lookup and cds_lfht_next. + * RCU read-side lock must be held between lookup and replacement. + * Call with rcu_read_lock held. + * Threads calling this API need to be registered RCU read-side threads. + * After successful replacement, a grace period must be waited for before + * freeing the memory reserved for the old node (which can be accessed + * with cds_lfht_iter_get_node). + * + * The semantic of replacement vs lookups is the following: if lookups + * are performed between a key unique insertion and its removal, we + * guarantee that the lookups and get next will always find exactly one + * instance of the key if it is replaced concurrently with the lookups. + * + * Providing this semantic allows us to ensure that replacement-only + * schemes will never generate duplicated keys. It also allows us to + * guarantee that a combination of add_replace and add_unique updates + * will never generate duplicated keys. + */ +int cds_lfht_replace(struct cds_lfht *ht, struct cds_lfht_iter *old_iter, + struct cds_lfht_node *new_node); + +/* + * cds_lfht_del - remove node pointed to by iterator from hash table. + * @ht: the hash table. + * @iter: the iterator position of the node to delete. + * + * Return 0 if the node is successfully removed, negative value + * otherwise. + * Replacing a NULL node or an already removed node will fail with a + * negative value. + * Node can be looked up with cds_lfht_lookup and cds_lfht_next. + * cds_lfht_iter_get_node. + * RCU read-side lock must be held between lookup and removal. + * Call with rcu_read_lock held. + * Threads calling this API need to be registered RCU read-side threads. + * After successful removal, a grace period must be waited for before + * freeing the memory reserved for old node (which can be accessed with + * cds_lfht_iter_get_node). + */ +int cds_lfht_del(struct cds_lfht *ht, struct cds_lfht_iter *iter); + +/* + * cds_lfht_resize - Force a hash table resize + * @ht: the hash table. + * @new_size: update to this hash table size. + * + * Threads calling this API need to be registered RCU read-side threads. + */ +void cds_lfht_resize(struct cds_lfht *ht, unsigned long new_size); + +/* + * Note: cds_lfht_for_each are safe for element removal during + * iteration. + */ +#define cds_lfht_for_each(ht, iter, node) \ + for (cds_lfht_first(ht, iter), \ + node = cds_lfht_iter_get_node(iter); \ + node != NULL; \ + cds_lfht_next(ht, iter), \ + node = cds_lfht_iter_get_node(iter)) + +#define cds_lfht_for_each_duplicate(ht, hash, match, key, iter, node) \ + for (cds_lfht_lookup(ht, hash, match, key, iter), \ + node = cds_lfht_iter_get_node(iter); \ + node != NULL; \ + cds_lfht_next_duplicate(ht, match, key, iter), \ + node = cds_lfht_iter_get_node(iter)) + +#define cds_lfht_for_each_entry(ht, iter, pos, member) \ + for (cds_lfht_first(ht, iter), \ + pos = caa_container_of(cds_lfht_iter_get_node(iter), \ + typeof(*(pos)), member); \ + &(pos)->member != NULL; \ + cds_lfht_next(ht, iter), \ + pos = caa_container_of(cds_lfht_iter_get_node(iter), \ + typeof(*(pos)), member)) + +#define cds_lfht_for_each_entry_duplicate(ht, hash, match, key, \ + iter, pos, member) \ + for (cds_lfht_lookup(ht, hash, match, key, iter), \ + pos = caa_container_of(cds_lfht_iter_get_node(iter), \ + typeof(*(pos)), member); \ + &(pos)->member != NULL; \ + cds_lfht_next_duplicate(ht, match, key, iter), \ + pos = caa_container_of(cds_lfht_iter_get_node(iter), \ + typeof(*(pos)), member)) + +#ifdef __cplusplus +} +#endif + +#endif /* _URCU_RCULFHASH_H */