1 // SPDX-FileCopyrightText: 2010-2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
2 // SPDX-FileCopyrightText: 2011 Lai Jiangshan <laijs@cn.fujitsu.com>
4 // SPDX-License-Identifier: LGPL-2.1-or-later
7 * Userspace RCU library - Lock-Free Resizable RCU Hash Table
11 * Based on the following articles:
12 * - Ori Shalev and Nir Shavit. Split-ordered lists: Lock-free
13 * extensible hash tables. J. ACM 53, 3 (May 2006), 379-405.
14 * - Michael, M. M. High performance dynamic lock-free hash tables
15 * and list-based sets. In Proceedings of the fourteenth annual ACM
16 * symposium on Parallel algorithms and architectures, ACM Press,
19 * Some specificities of this Lock-Free Resizable RCU Hash Table
22 * - RCU read-side critical section allows readers to perform hash
23 * table lookups, as well as traversals, and use the returned objects
24 * safely by allowing memory reclaim to take place only after a grace
26 * - Add and remove operations are lock-free, and do not need to
27 * allocate memory. They need to be executed within RCU read-side
28 * critical section to ensure the objects they read are valid and to
29 * deal with the cmpxchg ABA problem.
30 * - add and add_unique operations are supported. add_unique checks if
31 * the node key already exists in the hash table. It ensures not to
32 * populate a duplicate key if the node key already exists in the hash
34 * - The resize operation executes concurrently with
35 * add/add_unique/add_replace/remove/lookup/traversal.
36 * - Hash table nodes are contained within a split-ordered list. This
37 * list is ordered by incrementing reversed-bits-hash value.
38 * - An index of bucket nodes is kept. These bucket nodes are the hash
39 * table "buckets". These buckets are internal nodes that allow to
40 * perform a fast hash lookup, similarly to a skip list. These
41 * buckets are chained together in the split-ordered list, which
42 * allows recursive expansion by inserting new buckets between the
43 * existing buckets. The split-ordered list allows adding new buckets
44 * between existing buckets as the table needs to grow.
45 * - The resize operation for small tables only allows expanding the
46 * hash table. It is triggered automatically by detecting long chains
47 * in the add operation.
48 * - The resize operation for larger tables (and available through an
49 * API) allows both expanding and shrinking the hash table.
50 * - Split-counters are used to keep track of the number of
51 * nodes within the hash table for automatic resize triggering.
52 * - Resize operation initiated by long chain detection is executed by a
53 * worker thread, which keeps lock-freedom of add and remove.
54 * - Resize operations are protected by a mutex.
55 * - The removal operation is split in two parts: first, a "removed"
56 * flag is set in the next pointer within the node to remove. Then,
57 * a "garbage collection" is performed in the bucket containing the
58 * removed node (from the start of the bucket up to the removed node).
59 * All encountered nodes with "removed" flag set in their next
60 * pointers are removed from the linked-list. If the cmpxchg used for
61 * removal fails (due to concurrent garbage-collection or concurrent
62 * add), we retry from the beginning of the bucket. This ensures that
63 * the node with "removed" flag set is removed from the hash table
64 * (not visible to lookups anymore) before the RCU read-side critical
65 * section held across removal ends. Furthermore, this ensures that
66 * the node with "removed" flag set is removed from the linked-list
67 * before its memory is reclaimed. After setting the "removal" flag,
68 * only the thread which removal is the first to set the "removal
69 * owner" flag (with an xchg) into a node's next pointer is considered
70 * to have succeeded its removal (and thus owns the node to reclaim).
71 * Because we garbage-collect starting from an invariant node (the
72 * start-of-bucket bucket node) up to the "removed" node (or find a
73 * reverse-hash that is higher), we are sure that a successful
74 * traversal of the chain leads to a chain that is present in the
75 * linked-list (the start node is never removed) and that it does not
76 * contain the "removed" node anymore, even if concurrent delete/add
77 * operations are changing the structure of the list concurrently.
78 * - The add operations perform garbage collection of buckets if they
79 * encounter nodes with removed flag set in the bucket where they want
80 * to add their new node. This ensures lock-freedom of add operation by
81 * helping the remover unlink nodes from the list rather than to wait
83 * - There are three memory backends for the hash table buckets: the
84 * "order table", the "chunks", and the "mmap".
85 * - These bucket containers contain a compact version of the hash table
87 * - The RCU "order table":
88 * - has a first level table indexed by log2(hash index) which is
89 * copied and expanded by the resize operation. This order table
90 * allows finding the "bucket node" tables.
91 * - There is one bucket node table per hash index order. The size of
92 * each bucket node table is half the number of hashes contained in
93 * this order (except for order 0).
94 * - The RCU "chunks" is best suited for close interaction with a page
95 * allocator. It uses a linear array as index to "chunks" containing
96 * each the same number of buckets.
97 * - The RCU "mmap" memory backend uses a single memory map to hold
99 * - synchronize_rcu is used to garbage-collect the old bucket node table.
101 * Ordering Guarantees:
103 * To discuss these guarantees, we first define "read" operation as any
104 * of the the basic cds_lfht_lookup, cds_lfht_next_duplicate,
105 * cds_lfht_first, cds_lfht_next operation, as well as
106 * cds_lfht_add_unique (failure).
108 * We define "read traversal" operation as any of the following
109 * group of operations
110 * - cds_lfht_lookup followed by iteration with cds_lfht_next_duplicate
111 * (and/or cds_lfht_next, although less common).
112 * - cds_lfht_add_unique (failure) followed by iteration with
113 * cds_lfht_next_duplicate (and/or cds_lfht_next, although less
115 * - cds_lfht_first followed iteration with cds_lfht_next (and/or
116 * cds_lfht_next_duplicate, although less common).
118 * We define "write" operations as any of cds_lfht_add, cds_lfht_replace,
119 * cds_lfht_add_unique (success), cds_lfht_add_replace, cds_lfht_del.
121 * When cds_lfht_add_unique succeeds (returns the node passed as
122 * parameter), it acts as a "write" operation. When cds_lfht_add_unique
123 * fails (returns a node different from the one passed as parameter), it
124 * acts as a "read" operation. A cds_lfht_add_unique failure is a
125 * cds_lfht_lookup "read" operation, therefore, any ordering guarantee
126 * referring to "lookup" imply any of "lookup" or cds_lfht_add_unique
129 * We define "prior" and "later" node as nodes observable by reads and
130 * read traversals respectively before and after a write or sequence of
133 * Hash-table operations are often cascaded, for example, the pointer
134 * returned by a cds_lfht_lookup() might be passed to a cds_lfht_next(),
135 * whose return value might in turn be passed to another hash-table
136 * operation. This entire cascaded series of operations must be enclosed
137 * by a pair of matching rcu_read_lock() and rcu_read_unlock()
140 * The following ordering guarantees are offered by this hash table:
142 * A.1) "read" after "write": if there is ordering between a write and a
143 * later read, then the read is guaranteed to see the write or some
145 * A.2) "read traversal" after "write": given that there is dependency
146 * ordering between reads in a "read traversal", if there is
147 * ordering between a write and the first read of the traversal,
148 * then the "read traversal" is guaranteed to see the write or
150 * B.1) "write" after "read": if there is ordering between a read and a
151 * later write, then the read will never see the write.
152 * B.2) "write" after "read traversal": given that there is dependency
153 * ordering between reads in a "read traversal", if there is
154 * ordering between the last read of the traversal and a later
155 * write, then the "read traversal" will never see the write.
156 * C) "write" while "read traversal": if a write occurs during a "read
157 * traversal", the traversal may, or may not, see the write.
158 * D.1) "write" after "write": if there is ordering between a write and
159 * a later write, then the later write is guaranteed to see the
160 * effects of the first write.
161 * D.2) Concurrent "write" pairs: The system will assign an arbitrary
162 * order to any pair of concurrent conflicting writes.
163 * Non-conflicting writes (for example, to different keys) are
165 * E) If a grace period separates a "del" or "replace" operation
166 * and a subsequent operation, then that subsequent operation is
167 * guaranteed not to see the removed item.
168 * F) Uniqueness guarantee: given a hash table that does not contain
169 * duplicate items for a given key, there will only be one item in
170 * the hash table after an arbitrary sequence of add_unique and/or
171 * add_replace operations. Note, however, that a pair of
172 * concurrent read operations might well access two different items
174 * G.1) If a pair of lookups for a given key are ordered (e.g. by a
175 * memory barrier), then the second lookup will return the same
176 * node as the previous lookup, or some later node.
177 * G.2) A "read traversal" that starts after the end of a prior "read
178 * traversal" (ordered by memory barriers) is guaranteed to see the
179 * same nodes as the previous traversal, or some later nodes.
180 * G.3) Concurrent "read" pairs: concurrent reads are unordered. For
181 * example, if a pair of reads to the same key run concurrently
182 * with an insertion of that same key, the reads remain unordered
183 * regardless of their return values. In other words, you cannot
184 * rely on the values returned by the reads to deduce ordering.
186 * Progress guarantees:
188 * * Reads are wait-free. These operations always move forward in the
189 * hash table linked list, and this list has no loop.
190 * * Writes are lock-free. Any retry loop performed by a write operation
191 * is triggered by progress made within another update operation.
193 * Bucket node tables:
195 * hash table hash table the last all bucket node tables
196 * order size bucket node 0 1 2 3 4 5 6(index)
203 * 5 32 16 1 1 2 4 8 16
204 * 6 64 32 1 1 2 4 8 16 32
206 * When growing/shrinking, we only focus on the last bucket node table
207 * which size is (!order ? 1 : (1 << (order -1))).
209 * Example for growing/shrinking:
210 * grow hash table from order 5 to 6: init the index=6 bucket node table
211 * shrink hash table from order 6 to 5: fini the index=6 bucket node table
213 * A bit of ascii art explanation:
215 * The order index is the off-by-one compared to the actual power of 2
216 * because we use index 0 to deal with the 0 special-case.
218 * This shows the nodes for a small table ordered by reversed bits:
230 * This shows the nodes in order of non-reversed bits, linked by
231 * reversed-bit order.
236 * 2 | | 2 010 010 <- |
237 * | | | 3 011 110 | <- |
238 * 3 -> | | | 4 100 001 | |
254 #include "compat-getcpu.h"
255 #include <urcu/assert.h>
256 #include <urcu/pointer.h>
257 #include <urcu/call-rcu.h>
258 #include <urcu/flavor.h>
259 #include <urcu/arch.h>
260 #include <urcu/uatomic.h>
261 #include <urcu/compiler.h>
262 #include <urcu/rculfhash.h>
266 #include "rculfhash-internal.h"
267 #include "workqueue.h"
268 #include "urcu-die.h"
269 #include "urcu-utils.h"
270 #include "compat-smp.h"
273 * Split-counters lazily update the global counter each 1024
274 * addition/removal. It automatically keeps track of resize required.
275 * We use the bucket length as indicator for need to expand for small
276 * tables and machines lacking per-cpu data support.
278 #define COUNT_COMMIT_ORDER 10
279 #define DEFAULT_SPLIT_COUNT_MASK 0xFUL
280 #define CHAIN_LEN_TARGET 1
281 #define CHAIN_LEN_RESIZE_THRESHOLD 3
284 * Define the minimum table size.
286 #define MIN_TABLE_ORDER 0
287 #define MIN_TABLE_SIZE (1UL << MIN_TABLE_ORDER)
290 * Minimum number of bucket nodes to touch per thread to parallelize grow/shrink.
292 #define MIN_PARTITION_PER_THREAD_ORDER 12
293 #define MIN_PARTITION_PER_THREAD (1UL << MIN_PARTITION_PER_THREAD_ORDER)
296 * The removed flag needs to be updated atomically with the pointer.
297 * It indicates that no node must attach to the node scheduled for
298 * removal, and that node garbage collection must be performed.
299 * The bucket flag does not require to be updated atomically with the
300 * pointer, but it is added as a pointer low bit flag to save space.
301 * The "removal owner" flag is used to detect which of the "del"
302 * operation that has set the "removed flag" gets to return the removed
303 * node to its caller. Note that the replace operation does not need to
304 * iteract with the "removal owner" flag, because it validates that
305 * the "removed" flag is not set before performing its cmpxchg.
307 #define REMOVED_FLAG (1UL << 0)
308 #define BUCKET_FLAG (1UL << 1)
309 #define REMOVAL_OWNER_FLAG (1UL << 2)
310 #define FLAGS_MASK ((1UL << 3) - 1)
312 /* Value of the end pointer. Should not interact with flags. */
313 #define END_VALUE NULL
316 * ht_items_count: Split-counters counting the number of node addition
317 * and removal in the table. Only used if the CDS_LFHT_ACCOUNTING flag
318 * is set at hash table creation.
320 * These are free-running counters, never reset to zero. They count the
321 * number of add/remove, and trigger every (1 << COUNT_COMMIT_ORDER)
322 * operations to update the global counter. We choose a power-of-2 value
323 * for the trigger to deal with 32 or 64-bit overflow of the counter.
325 struct ht_items_count
{
326 unsigned long add
, del
;
327 } __attribute__((aligned(CAA_CACHE_LINE_SIZE
)));
330 * resize_work: Contains arguments passed to worker thread
331 * responsible for performing lazy resize.
334 struct urcu_work work
;
339 * partition_resize_work: Contains arguments passed to worker threads
340 * executing the hash table resize on partitions of the hash table
341 * assigned to each processor's worker thread.
343 struct partition_resize_work
{
346 unsigned long i
, start
, len
;
347 void (*fct
)(struct cds_lfht
*ht
, unsigned long i
,
348 unsigned long start
, unsigned long len
);
351 enum nr_cpus_mask_state
{
352 NR_CPUS_MASK_INIT_FAILED
= -2,
353 NR_CPUS_MASK_UNINITIALIZED
= -1,
356 static struct urcu_workqueue
*cds_lfht_workqueue
;
359 * Mutex ensuring mutual exclusion between workqueue initialization and
360 * fork handlers. cds_lfht_fork_mutex nests inside call_rcu_mutex.
362 static pthread_mutex_t cds_lfht_fork_mutex
= PTHREAD_MUTEX_INITIALIZER
;
364 static struct urcu_atfork cds_lfht_atfork
;
367 * atfork handler nesting counters. Handle being registered to many urcu
368 * flavors, thus being possibly invoked more than once in the
369 * pthread_atfork list of callbacks.
371 static int cds_lfht_workqueue_atfork_nesting
;
373 static void __attribute__((destructor
)) cds_lfht_exit(void);
374 static void cds_lfht_init_worker(const struct rcu_flavor_struct
*flavor
);
376 #ifdef CONFIG_CDS_LFHT_ITER_DEBUG
379 void cds_lfht_iter_debug_set_ht(struct cds_lfht
*ht
, struct cds_lfht_iter
*iter
)
384 #define cds_lfht_iter_debug_assert(...) urcu_posix_assert(__VA_ARGS__)
389 void cds_lfht_iter_debug_set_ht(struct cds_lfht
*ht
__attribute__((unused
)),
390 struct cds_lfht_iter
*iter
__attribute__((unused
)))
394 #define cds_lfht_iter_debug_assert(...)
399 * Algorithm to reverse bits in a word by lookup table, extended to
402 * http://graphics.stanford.edu/~seander/bithacks.html#BitReverseTable
403 * Originally from Public Domain.
406 static const uint8_t BitReverseTable256
[256] =
408 #define R2(n) (n), (n) + 2*64, (n) + 1*64, (n) + 3*64
409 #define R4(n) R2(n), R2((n) + 2*16), R2((n) + 1*16), R2((n) + 3*16)
410 #define R6(n) R4(n), R4((n) + 2*4 ), R4((n) + 1*4 ), R4((n) + 3*4 )
411 R6(0), R6(2), R6(1), R6(3)
418 uint8_t bit_reverse_u8(uint8_t v
)
420 return BitReverseTable256
[v
];
423 #if (CAA_BITS_PER_LONG == 32)
425 uint32_t bit_reverse_u32(uint32_t v
)
427 return ((uint32_t) bit_reverse_u8(v
) << 24) |
428 ((uint32_t) bit_reverse_u8(v
>> 8) << 16) |
429 ((uint32_t) bit_reverse_u8(v
>> 16) << 8) |
430 ((uint32_t) bit_reverse_u8(v
>> 24));
434 uint64_t bit_reverse_u64(uint64_t v
)
436 return ((uint64_t) bit_reverse_u8(v
) << 56) |
437 ((uint64_t) bit_reverse_u8(v
>> 8) << 48) |
438 ((uint64_t) bit_reverse_u8(v
>> 16) << 40) |
439 ((uint64_t) bit_reverse_u8(v
>> 24) << 32) |
440 ((uint64_t) bit_reverse_u8(v
>> 32) << 24) |
441 ((uint64_t) bit_reverse_u8(v
>> 40) << 16) |
442 ((uint64_t) bit_reverse_u8(v
>> 48) << 8) |
443 ((uint64_t) bit_reverse_u8(v
>> 56));
448 unsigned long bit_reverse_ulong(unsigned long v
)
450 #if (CAA_BITS_PER_LONG == 32)
451 return bit_reverse_u32(v
);
453 return bit_reverse_u64(v
);
458 * fls: returns the position of the most significant bit.
459 * Returns 0 if no bit is set, else returns the position of the most
460 * significant bit (from 1 to 32 on 32-bit, from 1 to 64 on 64-bit).
462 #if defined(URCU_ARCH_X86)
464 unsigned int fls_u32(uint32_t x
)
468 __asm__ ("bsrl %1,%0\n\t"
472 : "=r" (r
) : "rm" (x
));
478 #if defined(URCU_ARCH_AMD64)
480 unsigned int fls_u64(uint64_t x
)
484 __asm__ ("bsrq %1,%0\n\t"
488 : "=r" (r
) : "rm" (x
));
495 static __attribute__((unused
))
496 unsigned int fls_u64(uint64_t x
)
503 if (!(x
& 0xFFFFFFFF00000000ULL
)) {
507 if (!(x
& 0xFFFF000000000000ULL
)) {
511 if (!(x
& 0xFF00000000000000ULL
)) {
515 if (!(x
& 0xF000000000000000ULL
)) {
519 if (!(x
& 0xC000000000000000ULL
)) {
523 if (!(x
& 0x8000000000000000ULL
)) {
532 static __attribute__((unused
))
533 unsigned int fls_u32(uint32_t x
)
539 if (!(x
& 0xFFFF0000U
)) {
543 if (!(x
& 0xFF000000U
)) {
547 if (!(x
& 0xF0000000U
)) {
551 if (!(x
& 0xC0000000U
)) {
555 if (!(x
& 0x80000000U
)) {
563 unsigned int cds_lfht_fls_ulong(unsigned long x
)
565 #if (CAA_BITS_PER_LONG == 32)
572 static void *cds_lfht_malloc(void *state
__attribute__((unused
)),
578 static void *cds_lfht_calloc(void *state
__attribute__((unused
)),
579 size_t nmemb
, size_t size
)
581 return calloc(nmemb
, size
);
584 static void *cds_lfht_realloc(void *state
__attribute__((unused
)),
585 void *ptr
, size_t size
)
587 return realloc(ptr
, size
);
590 static void *cds_lfht_aligned_alloc(void *state
__attribute__((unused
)),
591 size_t alignment
, size_t size
)
595 if (posix_memalign(&ptr
, alignment
, size
))
600 static void cds_lfht_free(void *state
__attribute__((unused
)), void *ptr
)
606 /* Default memory allocator */
607 static struct cds_lfht_alloc cds_lfht_default_alloc
= {
608 .malloc
= cds_lfht_malloc
,
609 .calloc
= cds_lfht_calloc
,
610 .realloc
= cds_lfht_realloc
,
611 .aligned_alloc
= cds_lfht_aligned_alloc
,
612 .free
= cds_lfht_free
,
617 * Return the minimum order for which x <= (1UL << order).
618 * Return -1 if x is 0.
621 int cds_lfht_get_count_order_u32(uint32_t x
)
626 return fls_u32(x
- 1);
630 * Return the minimum order for which x <= (1UL << order).
631 * Return -1 if x is 0.
633 int cds_lfht_get_count_order_ulong(unsigned long x
)
638 return cds_lfht_fls_ulong(x
- 1);
642 void cds_lfht_resize_lazy_grow(struct cds_lfht
*ht
, unsigned long size
, int growth
);
645 void cds_lfht_resize_lazy_count(struct cds_lfht
*ht
, unsigned long size
,
646 unsigned long count
);
648 static void mutex_lock(pthread_mutex_t
*mutex
)
652 #ifndef DISTRUST_SIGNALS_EXTREME
653 ret
= pthread_mutex_lock(mutex
);
656 #else /* #ifndef DISTRUST_SIGNALS_EXTREME */
657 while ((ret
= pthread_mutex_trylock(mutex
)) != 0) {
658 if (ret
!= EBUSY
&& ret
!= EINTR
)
660 if (CMM_LOAD_SHARED(URCU_TLS(rcu_reader
).need_mb
)) {
661 uatomic_store(&URCU_TLS(rcu_reader
).need_mb
, 0, CMM_SEQ_CST
);
663 (void) poll(NULL
, 0, 10);
665 #endif /* #else #ifndef DISTRUST_SIGNALS_EXTREME */
668 static void mutex_unlock(pthread_mutex_t
*mutex
)
672 ret
= pthread_mutex_unlock(mutex
);
677 static long nr_cpus_mask
= NR_CPUS_MASK_UNINITIALIZED
;
678 static long split_count_mask
= -1;
679 static int split_count_order
= -1;
681 static void ht_init_nr_cpus_mask(void)
685 maxcpus
= get_possible_cpus_array_len();
687 nr_cpus_mask
= NR_CPUS_MASK_INIT_FAILED
;
691 * round up number of CPUs to next power of two, so we
692 * can use & for modulo.
694 maxcpus
= 1UL << cds_lfht_get_count_order_ulong(maxcpus
);
695 nr_cpus_mask
= maxcpus
- 1;
699 void alloc_split_items_count(struct cds_lfht
*ht
)
701 if (nr_cpus_mask
== NR_CPUS_MASK_UNINITIALIZED
) {
702 ht_init_nr_cpus_mask();
703 if (nr_cpus_mask
< 0)
704 split_count_mask
= DEFAULT_SPLIT_COUNT_MASK
;
706 split_count_mask
= nr_cpus_mask
;
708 cds_lfht_get_count_order_ulong(split_count_mask
+ 1);
711 urcu_posix_assert(split_count_mask
>= 0);
713 if (ht
->flags
& CDS_LFHT_ACCOUNTING
) {
714 ht
->split_count
= ht
->alloc
->calloc(ht
->alloc
->state
, split_count_mask
+ 1,
715 sizeof(struct ht_items_count
));
716 urcu_posix_assert(ht
->split_count
);
718 ht
->split_count
= NULL
;
723 void free_split_items_count(struct cds_lfht
*ht
)
725 poison_free(ht
->alloc
, ht
->split_count
);
729 int ht_get_split_count_index(unsigned long hash
)
733 urcu_posix_assert(split_count_mask
>= 0);
734 cpu
= urcu_sched_getcpu();
735 if (caa_unlikely(cpu
< 0))
736 return hash
& split_count_mask
;
738 return cpu
& split_count_mask
;
742 void ht_count_add(struct cds_lfht
*ht
, unsigned long size
, unsigned long hash
)
744 unsigned long split_count
, count
;
747 if (caa_unlikely(!ht
->split_count
))
749 index
= ht_get_split_count_index(hash
);
750 split_count
= uatomic_add_return(&ht
->split_count
[index
].add
, 1);
751 if (caa_likely(split_count
& ((1UL << COUNT_COMMIT_ORDER
) - 1)))
753 /* Only if number of add multiple of 1UL << COUNT_COMMIT_ORDER */
755 dbg_printf("add split count %lu\n", split_count
);
756 count
= uatomic_add_return(&ht
->count
,
757 1UL << COUNT_COMMIT_ORDER
);
758 if (caa_likely(count
& (count
- 1)))
760 /* Only if global count is power of 2 */
762 if ((count
>> CHAIN_LEN_RESIZE_THRESHOLD
) < size
)
764 dbg_printf("add set global %lu\n", count
);
765 cds_lfht_resize_lazy_count(ht
, size
,
766 count
>> (CHAIN_LEN_TARGET
- 1));
770 void ht_count_del(struct cds_lfht
*ht
, unsigned long size
, unsigned long hash
)
772 unsigned long split_count
, count
;
775 if (caa_unlikely(!ht
->split_count
))
777 index
= ht_get_split_count_index(hash
);
778 split_count
= uatomic_add_return(&ht
->split_count
[index
].del
, 1);
779 if (caa_likely(split_count
& ((1UL << COUNT_COMMIT_ORDER
) - 1)))
781 /* Only if number of deletes multiple of 1UL << COUNT_COMMIT_ORDER */
783 dbg_printf("del split count %lu\n", split_count
);
784 count
= uatomic_add_return(&ht
->count
,
785 -(1UL << COUNT_COMMIT_ORDER
));
786 if (caa_likely(count
& (count
- 1)))
788 /* Only if global count is power of 2 */
790 if ((count
>> CHAIN_LEN_RESIZE_THRESHOLD
) >= size
)
792 dbg_printf("del set global %lu\n", count
);
794 * Don't shrink table if the number of nodes is below a
797 if (count
< (1UL << COUNT_COMMIT_ORDER
) * (split_count_mask
+ 1))
799 cds_lfht_resize_lazy_count(ht
, size
,
800 count
>> (CHAIN_LEN_TARGET
- 1));
804 void check_resize(struct cds_lfht
*ht
, unsigned long size
, uint32_t chain_len
)
808 if (!(ht
->flags
& CDS_LFHT_AUTO_RESIZE
))
810 count
= uatomic_read(&ht
->count
);
812 * Use bucket-local length for small table expand and for
813 * environments lacking per-cpu data support.
815 if (count
>= (1UL << (COUNT_COMMIT_ORDER
+ split_count_order
)))
818 dbg_printf("WARNING: large chain length: %u.\n",
820 if (chain_len
>= CHAIN_LEN_RESIZE_THRESHOLD
) {
824 * Ideal growth calculated based on chain length.
826 growth
= cds_lfht_get_count_order_u32(chain_len
827 - (CHAIN_LEN_TARGET
- 1));
828 if ((ht
->flags
& CDS_LFHT_ACCOUNTING
)
830 >= (1UL << (COUNT_COMMIT_ORDER
831 + split_count_order
))) {
833 * If ideal growth expands the hash table size
834 * beyond the "small hash table" sizes, use the
835 * maximum small hash table size to attempt
836 * expanding the hash table. This only applies
837 * when node accounting is available, otherwise
838 * the chain length is used to expand the hash
839 * table in every case.
841 growth
= COUNT_COMMIT_ORDER
+ split_count_order
842 - cds_lfht_get_count_order_ulong(size
);
846 cds_lfht_resize_lazy_grow(ht
, size
, growth
);
851 struct cds_lfht_node
*clear_flag(struct cds_lfht_node
*node
)
853 return (struct cds_lfht_node
*) (((unsigned long) node
) & ~FLAGS_MASK
);
857 int is_removed(const struct cds_lfht_node
*node
)
859 return ((unsigned long) node
) & REMOVED_FLAG
;
863 int is_bucket(struct cds_lfht_node
*node
)
865 return ((unsigned long) node
) & BUCKET_FLAG
;
869 struct cds_lfht_node
*flag_bucket(struct cds_lfht_node
*node
)
871 return (struct cds_lfht_node
*) (((unsigned long) node
) | BUCKET_FLAG
);
875 int is_removal_owner(struct cds_lfht_node
*node
)
877 return ((unsigned long) node
) & REMOVAL_OWNER_FLAG
;
881 struct cds_lfht_node
*flag_removed(struct cds_lfht_node
*node
)
883 return (struct cds_lfht_node
*) (((unsigned long) node
) | REMOVED_FLAG
);
887 struct cds_lfht_node
*flag_removal_owner(struct cds_lfht_node
*node
)
889 return (struct cds_lfht_node
*) (((unsigned long) node
) | REMOVAL_OWNER_FLAG
);
893 struct cds_lfht_node
*flag_removed_or_removal_owner(struct cds_lfht_node
*node
)
895 return (struct cds_lfht_node
*) (((unsigned long) node
) | REMOVED_FLAG
| REMOVAL_OWNER_FLAG
);
899 struct cds_lfht_node
*get_end(void)
901 return (struct cds_lfht_node
*) END_VALUE
;
905 int is_end(struct cds_lfht_node
*node
)
907 return clear_flag(node
) == (struct cds_lfht_node
*) END_VALUE
;
911 unsigned long _uatomic_xchg_monotonic_increase(unsigned long *ptr
,
914 unsigned long old1
, old2
;
916 old1
= uatomic_read(ptr
);
923 } while ((old1
= uatomic_cmpxchg(ptr
, old2
, v
)) != old2
);
928 void cds_lfht_alloc_bucket_table(struct cds_lfht
*ht
, unsigned long order
)
930 return ht
->mm
->alloc_bucket_table(ht
, order
);
934 * cds_lfht_free_bucket_table() should be called with decreasing order.
935 * When cds_lfht_free_bucket_table(0) is called, it means the whole
939 void cds_lfht_free_bucket_table(struct cds_lfht
*ht
, unsigned long order
)
941 return ht
->mm
->free_bucket_table(ht
, order
);
945 struct cds_lfht_node
*bucket_at(struct cds_lfht
*ht
, unsigned long index
)
947 return ht
->bucket_at(ht
, index
);
951 struct cds_lfht_node
*lookup_bucket(struct cds_lfht
*ht
, unsigned long size
,
954 urcu_posix_assert(size
> 0);
955 return bucket_at(ht
, hash
& (size
- 1));
959 * Remove all logically deleted nodes from a bucket up to a certain node key.
962 void _cds_lfht_gc_bucket(struct cds_lfht_node
*bucket
, struct cds_lfht_node
*node
)
964 struct cds_lfht_node
*iter_prev
, *iter
, *next
, *new_next
;
966 urcu_posix_assert(!is_bucket(bucket
));
967 urcu_posix_assert(!is_removed(bucket
));
968 urcu_posix_assert(!is_removal_owner(bucket
));
969 urcu_posix_assert(!is_bucket(node
));
970 urcu_posix_assert(!is_removed(node
));
971 urcu_posix_assert(!is_removal_owner(node
));
974 /* We can always skip the bucket node initially */
975 iter
= rcu_dereference(iter_prev
->next
);
976 urcu_posix_assert(!is_removed(iter
));
977 urcu_posix_assert(!is_removal_owner(iter
));
978 urcu_posix_assert(iter_prev
->reverse_hash
<= node
->reverse_hash
);
980 * We should never be called with bucket (start of chain)
981 * and logically removed node (end of path compression
982 * marker) being the actual same node. This would be a
983 * bug in the algorithm implementation.
985 urcu_posix_assert(bucket
!= node
);
987 if (caa_unlikely(is_end(iter
)))
989 if (caa_likely(clear_flag(iter
)->reverse_hash
> node
->reverse_hash
))
991 next
= rcu_dereference(clear_flag(iter
)->next
);
992 if (caa_likely(is_removed(next
)))
994 iter_prev
= clear_flag(iter
);
997 urcu_posix_assert(!is_removed(iter
));
998 urcu_posix_assert(!is_removal_owner(iter
));
1000 new_next
= flag_bucket(clear_flag(next
));
1002 new_next
= clear_flag(next
);
1003 (void) uatomic_cmpxchg(&iter_prev
->next
, iter
, new_next
);
1008 int _cds_lfht_replace(struct cds_lfht
*ht
, unsigned long size
,
1009 struct cds_lfht_node
*old_node
,
1010 struct cds_lfht_node
*old_next
,
1011 struct cds_lfht_node
*new_node
)
1013 struct cds_lfht_node
*bucket
, *ret_next
;
1015 if (!old_node
) /* Return -ENOENT if asked to replace NULL node */
1018 urcu_posix_assert(!is_removed(old_node
));
1019 urcu_posix_assert(!is_removal_owner(old_node
));
1020 urcu_posix_assert(!is_bucket(old_node
));
1021 urcu_posix_assert(!is_removed(new_node
));
1022 urcu_posix_assert(!is_removal_owner(new_node
));
1023 urcu_posix_assert(!is_bucket(new_node
));
1024 urcu_posix_assert(new_node
!= old_node
);
1026 /* Insert after node to be replaced */
1027 if (is_removed(old_next
)) {
1029 * Too late, the old node has been removed under us
1030 * between lookup and replace. Fail.
1034 urcu_posix_assert(old_next
== clear_flag(old_next
));
1035 urcu_posix_assert(new_node
!= old_next
);
1037 * REMOVAL_OWNER flag is _NEVER_ set before the REMOVED
1038 * flag. It is either set atomically at the same time
1039 * (replace) or after (del).
1041 urcu_posix_assert(!is_removal_owner(old_next
));
1042 new_node
->next
= old_next
;
1044 * Here is the whole trick for lock-free replace: we add
1045 * the replacement node _after_ the node we want to
1046 * replace by atomically setting its next pointer at the
1047 * same time we set its removal flag. Given that
1048 * the lookups/get next use an iterator aware of the
1049 * next pointer, they will either skip the old node due
1050 * to the removal flag and see the new node, or use
1051 * the old node, but will not see the new one.
1052 * This is a replacement of a node with another node
1053 * that has the same value: we are therefore not
1054 * removing a value from the hash table. We set both the
1055 * REMOVED and REMOVAL_OWNER flags atomically so we own
1056 * the node after successful cmpxchg.
1058 ret_next
= uatomic_cmpxchg(&old_node
->next
,
1059 old_next
, flag_removed_or_removal_owner(new_node
));
1060 if (ret_next
== old_next
)
1061 break; /* We performed the replacement. */
1062 old_next
= ret_next
;
1066 * Ensure that the old node is not visible to readers anymore:
1067 * lookup for the node, and remove it (along with any other
1068 * logically removed node) if found.
1070 bucket
= lookup_bucket(ht
, size
, bit_reverse_ulong(old_node
->reverse_hash
));
1071 _cds_lfht_gc_bucket(bucket
, new_node
);
1073 urcu_posix_assert(is_removed(CMM_LOAD_SHARED(old_node
->next
)));
1078 * A non-NULL unique_ret pointer uses the "add unique" (or uniquify) add
1079 * mode. A NULL unique_ret allows creation of duplicate keys.
1082 void _cds_lfht_add(struct cds_lfht
*ht
,
1084 cds_lfht_match_fct match
,
1087 struct cds_lfht_node
*node
,
1088 struct cds_lfht_iter
*unique_ret
,
1091 struct cds_lfht_node
*iter_prev
, *iter
, *next
, *new_node
, *new_next
,
1093 struct cds_lfht_node
*bucket
;
1095 urcu_posix_assert(!is_bucket(node
));
1096 urcu_posix_assert(!is_removed(node
));
1097 urcu_posix_assert(!is_removal_owner(node
));
1098 bucket
= lookup_bucket(ht
, size
, hash
);
1100 uint32_t chain_len
= 0;
1103 * iter_prev points to the non-removed node prior to the
1107 /* We can always skip the bucket node initially */
1108 iter
= rcu_dereference(iter_prev
->next
);
1109 urcu_posix_assert(iter_prev
->reverse_hash
<= node
->reverse_hash
);
1111 if (caa_unlikely(is_end(iter
)))
1113 if (caa_likely(clear_flag(iter
)->reverse_hash
> node
->reverse_hash
))
1116 /* bucket node is the first node of the identical-hash-value chain */
1117 if (bucket_flag
&& clear_flag(iter
)->reverse_hash
== node
->reverse_hash
)
1120 next
= rcu_dereference(clear_flag(iter
)->next
);
1121 if (caa_unlikely(is_removed(next
)))
1127 && clear_flag(iter
)->reverse_hash
== node
->reverse_hash
) {
1128 struct cds_lfht_iter d_iter
= {
1131 #ifdef CONFIG_CDS_LFHT_ITER_DEBUG
1137 * uniquely adding inserts the node as the first
1138 * node of the identical-hash-value node chain.
1140 * This semantic ensures no duplicated keys
1141 * should ever be observable in the table
1142 * (including traversing the table node by
1143 * node by forward iterations)
1145 cds_lfht_next_duplicate(ht
, match
, key
, &d_iter
);
1149 *unique_ret
= d_iter
;
1153 /* Only account for identical reverse hash once */
1154 if (iter_prev
->reverse_hash
!= clear_flag(iter
)->reverse_hash
1155 && !is_bucket(next
))
1156 check_resize(ht
, size
, ++chain_len
);
1157 iter_prev
= clear_flag(iter
);
1162 urcu_posix_assert(node
!= clear_flag(iter
));
1163 urcu_posix_assert(!is_removed(iter_prev
));
1164 urcu_posix_assert(!is_removal_owner(iter_prev
));
1165 urcu_posix_assert(!is_removed(iter
));
1166 urcu_posix_assert(!is_removal_owner(iter
));
1167 urcu_posix_assert(iter_prev
!= node
);
1169 node
->next
= clear_flag(iter
);
1171 node
->next
= flag_bucket(clear_flag(iter
));
1172 if (is_bucket(iter
))
1173 new_node
= flag_bucket(node
);
1176 if (uatomic_cmpxchg(&iter_prev
->next
, iter
,
1177 new_node
) != iter
) {
1178 continue; /* retry */
1185 urcu_posix_assert(!is_removed(iter
));
1186 urcu_posix_assert(!is_removal_owner(iter
));
1187 if (is_bucket(iter
))
1188 new_next
= flag_bucket(clear_flag(next
));
1190 new_next
= clear_flag(next
);
1191 (void) uatomic_cmpxchg(&iter_prev
->next
, iter
, new_next
);
1196 unique_ret
->node
= return_node
;
1197 /* unique_ret->next left unset, never used. */
1202 int _cds_lfht_del(struct cds_lfht
*ht
, unsigned long size
,
1203 struct cds_lfht_node
*node
)
1205 struct cds_lfht_node
*bucket
, *next
;
1206 uintptr_t *node_next
;
1208 if (!node
) /* Return -ENOENT if asked to delete NULL node */
1211 /* logically delete the node */
1212 urcu_posix_assert(!is_bucket(node
));
1213 urcu_posix_assert(!is_removed(node
));
1214 urcu_posix_assert(!is_removal_owner(node
));
1217 * We are first checking if the node had previously been
1218 * logically removed (this check is not atomic with setting the
1219 * logical removal flag). Return -ENOENT if the node had
1220 * previously been removed.
1222 next
= CMM_LOAD_SHARED(node
->next
); /* next is not dereferenced */
1223 if (caa_unlikely(is_removed(next
)))
1225 urcu_posix_assert(!is_bucket(next
));
1227 * The del operation semantic guarantees a full memory barrier
1228 * before the uatomic_or atomic commit of the deletion flag.
1230 * We set the REMOVED_FLAG unconditionally. Note that there may
1231 * be more than one concurrent thread setting this flag.
1232 * Knowing which wins the race will be known after the garbage
1233 * collection phase, stay tuned!
1235 * NOTE: The node_next variable is present to avoid breaking
1236 * strict-aliasing rules.
1238 node_next
= (uintptr_t*)&node
->next
;
1239 uatomic_or_mo(node_next
, REMOVED_FLAG
, CMM_RELEASE
);
1241 /* We performed the (logical) deletion. */
1244 * Ensure that the node is not visible to readers anymore: lookup for
1245 * the node, and remove it (along with any other logically removed node)
1248 bucket
= lookup_bucket(ht
, size
, bit_reverse_ulong(node
->reverse_hash
));
1249 _cds_lfht_gc_bucket(bucket
, node
);
1251 urcu_posix_assert(is_removed(CMM_LOAD_SHARED(node
->next
)));
1253 * Last phase: atomically exchange node->next with a version
1254 * having "REMOVAL_OWNER_FLAG" set. If the returned node->next
1255 * pointer did _not_ have "REMOVAL_OWNER_FLAG" set, we now own
1256 * the node and win the removal race.
1257 * It is interesting to note that all "add" paths are forbidden
1258 * to change the next pointer starting from the point where the
1259 * REMOVED_FLAG is set, so here using a read, followed by a
1260 * xchg() suffice to guarantee that the xchg() will ever only
1261 * set the "REMOVAL_OWNER_FLAG" (or change nothing if the flag
1264 if (!is_removal_owner(uatomic_xchg(&node
->next
,
1265 flag_removal_owner(uatomic_load(&node
->next
, CMM_RELAXED
)))))
1272 void *partition_resize_thread(void *arg
)
1274 struct partition_resize_work
*work
= arg
;
1276 work
->ht
->flavor
->register_thread();
1277 work
->fct(work
->ht
, work
->i
, work
->start
, work
->len
);
1278 work
->ht
->flavor
->unregister_thread();
1283 void partition_resize_helper(struct cds_lfht
*ht
, unsigned long i
,
1285 void (*fct
)(struct cds_lfht
*ht
, unsigned long i
,
1286 unsigned long start
, unsigned long len
))
1288 unsigned long partition_len
, start
= 0;
1289 struct partition_resize_work
*work
;
1291 unsigned long thread
, nr_threads
;
1292 sigset_t newmask
, oldmask
;
1294 urcu_posix_assert(nr_cpus_mask
!= NR_CPUS_MASK_UNINITIALIZED
);
1295 if (nr_cpus_mask
< 0 || len
< 2 * MIN_PARTITION_PER_THREAD
)
1299 * Note: nr_cpus_mask + 1 is always power of 2.
1300 * We spawn just the number of threads we need to satisfy the minimum
1301 * partition size, up to the number of CPUs in the system.
1303 if (nr_cpus_mask
> 0) {
1304 nr_threads
= min_t(unsigned long, nr_cpus_mask
+ 1,
1305 len
>> MIN_PARTITION_PER_THREAD_ORDER
);
1309 partition_len
= len
>> cds_lfht_get_count_order_ulong(nr_threads
);
1310 work
= ht
->alloc
->calloc(ht
->alloc
->state
, nr_threads
, sizeof(*work
));
1312 dbg_printf("error allocating for resize, single-threading\n");
1316 ret
= sigfillset(&newmask
);
1317 urcu_posix_assert(!ret
);
1318 ret
= pthread_sigmask(SIG_BLOCK
, &newmask
, &oldmask
);
1319 urcu_posix_assert(!ret
);
1321 for (thread
= 0; thread
< nr_threads
; thread
++) {
1322 work
[thread
].ht
= ht
;
1324 work
[thread
].len
= partition_len
;
1325 work
[thread
].start
= thread
* partition_len
;
1326 work
[thread
].fct
= fct
;
1327 ret
= pthread_create(&(work
[thread
].thread_id
),
1328 ht
->caller_resize_attr
? &ht
->resize_attr
: NULL
,
1329 partition_resize_thread
, &work
[thread
]);
1330 if (ret
== EAGAIN
) {
1332 * Out of resources: wait and join the threads
1333 * we've created, then handle leftovers.
1335 dbg_printf("error spawning for resize, single-threading\n");
1336 start
= work
[thread
].start
;
1338 nr_threads
= thread
;
1341 urcu_posix_assert(!ret
);
1344 ret
= pthread_sigmask(SIG_SETMASK
, &oldmask
, NULL
);
1345 urcu_posix_assert(!ret
);
1347 for (thread
= 0; thread
< nr_threads
; thread
++) {
1348 ret
= pthread_join(work
[thread
].thread_id
, NULL
);
1349 urcu_posix_assert(!ret
);
1351 ht
->alloc
->free(ht
->alloc
->state
, work
);
1354 * A pthread_create failure above will either lead in us having
1355 * no threads to join or starting at a non-zero offset,
1356 * fallback to single thread processing of leftovers.
1358 if (start
== 0 && nr_threads
> 0)
1361 fct(ht
, i
, start
, len
);
1365 * Holding RCU read lock to protect _cds_lfht_add against memory
1366 * reclaim that could be performed by other worker threads (ABA
1369 * When we reach a certain length, we can split this population phase over
1370 * many worker threads, based on the number of CPUs available in the system.
1371 * This should therefore take care of not having the expand lagging behind too
1372 * many concurrent insertion threads by using the scheduler's ability to
1373 * schedule bucket node population fairly with insertions.
1376 void init_table_populate_partition(struct cds_lfht
*ht
, unsigned long i
,
1377 unsigned long start
, unsigned long len
)
1379 unsigned long j
, size
= 1UL << (i
- 1);
1381 urcu_posix_assert(i
> MIN_TABLE_ORDER
);
1382 ht
->flavor
->read_lock();
1383 for (j
= size
+ start
; j
< size
+ start
+ len
; j
++) {
1384 struct cds_lfht_node
*new_node
= bucket_at(ht
, j
);
1386 urcu_posix_assert(j
>= size
&& j
< (size
<< 1));
1387 dbg_printf("init populate: order %lu index %lu hash %lu\n",
1389 new_node
->reverse_hash
= bit_reverse_ulong(j
);
1390 _cds_lfht_add(ht
, j
, NULL
, NULL
, size
, new_node
, NULL
, 1);
1392 ht
->flavor
->read_unlock();
1396 void init_table_populate(struct cds_lfht
*ht
, unsigned long i
,
1399 partition_resize_helper(ht
, i
, len
, init_table_populate_partition
);
1403 void init_table(struct cds_lfht
*ht
,
1404 unsigned long first_order
, unsigned long last_order
)
1408 dbg_printf("init table: first_order %lu last_order %lu\n",
1409 first_order
, last_order
);
1410 urcu_posix_assert(first_order
> MIN_TABLE_ORDER
);
1411 for (i
= first_order
; i
<= last_order
; i
++) {
1414 len
= 1UL << (i
- 1);
1415 dbg_printf("init order %lu len: %lu\n", i
, len
);
1417 /* Stop expand if the resize target changes under us */
1418 if (CMM_LOAD_SHARED(ht
->resize_target
) < (1UL << i
))
1421 cds_lfht_alloc_bucket_table(ht
, i
);
1424 * Set all bucket nodes reverse hash values for a level and
1425 * link all bucket nodes into the table.
1427 init_table_populate(ht
, i
, len
);
1430 * Update table size.
1432 * Populate data before RCU size.
1434 uatomic_store(&ht
->size
, 1UL << i
, CMM_RELEASE
);
1436 dbg_printf("init new size: %lu\n", 1UL << i
);
1437 if (CMM_LOAD_SHARED(ht
->in_progress_destroy
))
1443 * Holding RCU read lock to protect _cds_lfht_remove against memory
1444 * reclaim that could be performed by other worker threads (ABA
1446 * For a single level, we logically remove and garbage collect each node.
1448 * As a design choice, we perform logical removal and garbage collection on a
1449 * node-per-node basis to simplify this algorithm. We also assume keeping good
1450 * cache locality of the operation would overweight possible performance gain
1451 * that could be achieved by batching garbage collection for multiple levels.
1452 * However, this would have to be justified by benchmarks.
1454 * Concurrent removal and add operations are helping us perform garbage
1455 * collection of logically removed nodes. We guarantee that all logically
1456 * removed nodes have been garbage-collected (unlinked) before work
1457 * enqueue is invoked to free a hole level of bucket nodes (after a
1460 * Logical removal and garbage collection can therefore be done in batch
1461 * or on a node-per-node basis, as long as the guarantee above holds.
1463 * When we reach a certain length, we can split this removal over many worker
1464 * threads, based on the number of CPUs available in the system. This should
1465 * take care of not letting resize process lag behind too many concurrent
1466 * updater threads actively inserting into the hash table.
1469 void remove_table_partition(struct cds_lfht
*ht
, unsigned long i
,
1470 unsigned long start
, unsigned long len
)
1472 unsigned long j
, size
= 1UL << (i
- 1);
1474 urcu_posix_assert(i
> MIN_TABLE_ORDER
);
1475 ht
->flavor
->read_lock();
1476 for (j
= size
+ start
; j
< size
+ start
+ len
; j
++) {
1477 struct cds_lfht_node
*fini_bucket
= bucket_at(ht
, j
);
1478 struct cds_lfht_node
*parent_bucket
= bucket_at(ht
, j
- size
);
1479 uintptr_t *fini_bucket_next
;
1481 urcu_posix_assert(j
>= size
&& j
< (size
<< 1));
1482 dbg_printf("remove entry: order %lu index %lu hash %lu\n",
1484 /* Set the REMOVED_FLAG to freeze the ->next for gc.
1486 * NOTE: The fini_bucket_next variable is present to
1487 * avoid breaking strict-aliasing rules.
1489 fini_bucket_next
= (uintptr_t*)&fini_bucket
->next
;
1490 uatomic_or(fini_bucket_next
, REMOVED_FLAG
);
1491 _cds_lfht_gc_bucket(parent_bucket
, fini_bucket
);
1493 ht
->flavor
->read_unlock();
1497 void remove_table(struct cds_lfht
*ht
, unsigned long i
, unsigned long len
)
1499 partition_resize_helper(ht
, i
, len
, remove_table_partition
);
1503 * fini_table() is never called for first_order == 0, which is why
1504 * free_by_rcu_order == 0 can be used as criterion to know if free must
1508 void fini_table(struct cds_lfht
*ht
,
1509 unsigned long first_order
, unsigned long last_order
)
1511 unsigned long free_by_rcu_order
= 0, i
;
1513 dbg_printf("fini table: first_order %lu last_order %lu\n",
1514 first_order
, last_order
);
1515 urcu_posix_assert(first_order
> MIN_TABLE_ORDER
);
1516 for (i
= last_order
; i
>= first_order
; i
--) {
1519 len
= 1UL << (i
- 1);
1520 dbg_printf("fini order %ld len: %lu\n", i
, len
);
1522 /* Stop shrink if the resize target changes under us */
1523 if (CMM_LOAD_SHARED(ht
->resize_target
) > (1UL << (i
- 1)))
1526 cmm_smp_wmb(); /* populate data before RCU size */
1527 CMM_STORE_SHARED(ht
->size
, 1UL << (i
- 1));
1530 * We need to wait for all add operations to reach Q.S. (and
1531 * thus use the new table for lookups) before we can start
1532 * releasing the old bucket nodes. Otherwise their lookup will
1533 * return a logically removed node as insert position.
1535 ht
->flavor
->update_synchronize_rcu();
1536 if (free_by_rcu_order
)
1537 cds_lfht_free_bucket_table(ht
, free_by_rcu_order
);
1540 * Set "removed" flag in bucket nodes about to be removed.
1541 * Unlink all now-logically-removed bucket node pointers.
1542 * Concurrent add/remove operation are helping us doing
1545 remove_table(ht
, i
, len
);
1547 free_by_rcu_order
= i
;
1549 dbg_printf("fini new size: %lu\n", 1UL << i
);
1550 if (CMM_LOAD_SHARED(ht
->in_progress_destroy
))
1554 if (free_by_rcu_order
) {
1555 ht
->flavor
->update_synchronize_rcu();
1556 cds_lfht_free_bucket_table(ht
, free_by_rcu_order
);
1561 * Never called with size < 1.
1564 void cds_lfht_create_bucket(struct cds_lfht
*ht
, unsigned long size
)
1566 struct cds_lfht_node
*prev
, *node
;
1567 unsigned long order
, len
, i
;
1570 cds_lfht_alloc_bucket_table(ht
, 0);
1572 dbg_printf("create bucket: order 0 index 0 hash 0\n");
1573 node
= bucket_at(ht
, 0);
1574 node
->next
= flag_bucket(get_end());
1575 node
->reverse_hash
= 0;
1577 bucket_order
= cds_lfht_get_count_order_ulong(size
);
1578 urcu_posix_assert(bucket_order
>= 0);
1580 for (order
= 1; order
< (unsigned long) bucket_order
+ 1; order
++) {
1581 len
= 1UL << (order
- 1);
1582 cds_lfht_alloc_bucket_table(ht
, order
);
1584 for (i
= 0; i
< len
; i
++) {
1586 * Now, we are trying to init the node with the
1587 * hash=(len+i) (which is also a bucket with the
1588 * index=(len+i)) and insert it into the hash table,
1589 * so this node has to be inserted after the bucket
1590 * with the index=(len+i)&(len-1)=i. And because there
1591 * is no other non-bucket node nor bucket node with
1592 * larger index/hash inserted, so the bucket node
1593 * being inserted should be inserted directly linked
1594 * after the bucket node with index=i.
1596 prev
= bucket_at(ht
, i
);
1597 node
= bucket_at(ht
, len
+ i
);
1599 dbg_printf("create bucket: order %lu index %lu hash %lu\n",
1600 order
, len
+ i
, len
+ i
);
1601 node
->reverse_hash
= bit_reverse_ulong(len
+ i
);
1603 /* insert after prev */
1604 urcu_posix_assert(is_bucket(prev
->next
));
1605 node
->next
= prev
->next
;
1606 prev
->next
= flag_bucket(node
);
1611 #if (CAA_BITS_PER_LONG > 32)
1613 * For 64-bit architectures, with max number of buckets small enough not to
1614 * use the entire 64-bit memory mapping space (and allowing a fair number of
1615 * hash table instances), use the mmap allocator, which is faster. Otherwise,
1616 * fallback to the order allocator.
1619 const struct cds_lfht_mm_type
*get_mm_type(unsigned long max_nr_buckets
)
1621 if (max_nr_buckets
&& max_nr_buckets
<= (1ULL << 32))
1622 return &cds_lfht_mm_mmap
;
1624 return &cds_lfht_mm_order
;
1628 * For 32-bit architectures, use the order allocator.
1631 const struct cds_lfht_mm_type
*get_mm_type(
1632 unsigned long max_nr_buckets
__attribute__((unused
)))
1634 return &cds_lfht_mm_order
;
1638 void cds_lfht_node_init_deleted(struct cds_lfht_node
*node
)
1640 cds_lfht_node_init(node
);
1641 node
->next
= flag_removed(NULL
);
1644 struct cds_lfht
*_cds_lfht_new_with_alloc(unsigned long init_size
,
1645 unsigned long min_nr_alloc_buckets
,
1646 unsigned long max_nr_buckets
,
1648 const struct cds_lfht_mm_type
*mm
,
1649 const struct rcu_flavor_struct
*flavor
,
1650 const struct cds_lfht_alloc
*alloc
,
1651 pthread_attr_t
*attr
)
1653 struct cds_lfht
*ht
;
1654 unsigned long order
;
1656 /* min_nr_alloc_buckets must be power of two */
1657 if (!min_nr_alloc_buckets
|| (min_nr_alloc_buckets
& (min_nr_alloc_buckets
- 1)))
1660 /* init_size must be power of two */
1661 if (!init_size
|| (init_size
& (init_size
- 1)))
1665 * Memory management plugin default.
1668 mm
= get_mm_type(max_nr_buckets
);
1670 /* max_nr_buckets == 0 for order based mm means infinite */
1671 if (mm
== &cds_lfht_mm_order
&& !max_nr_buckets
)
1672 max_nr_buckets
= 1UL << (MAX_TABLE_ORDER
- 1);
1674 /* max_nr_buckets must be power of two */
1675 if (!max_nr_buckets
|| (max_nr_buckets
& (max_nr_buckets
- 1)))
1678 if (flags
& CDS_LFHT_AUTO_RESIZE
)
1679 cds_lfht_init_worker(flavor
);
1681 min_nr_alloc_buckets
= max(min_nr_alloc_buckets
, MIN_TABLE_SIZE
);
1682 init_size
= max(init_size
, MIN_TABLE_SIZE
);
1683 max_nr_buckets
= max(max_nr_buckets
, min_nr_alloc_buckets
);
1684 init_size
= min(init_size
, max_nr_buckets
);
1686 ht
= mm
->alloc_cds_lfht(min_nr_alloc_buckets
, max_nr_buckets
, alloc
? : &cds_lfht_default_alloc
);
1688 urcu_posix_assert(ht
);
1689 urcu_posix_assert(ht
->mm
== mm
);
1690 urcu_posix_assert(ht
->bucket_at
== mm
->bucket_at
);
1693 ht
->flavor
= flavor
;
1694 ht
->caller_resize_attr
= attr
;
1696 ht
->resize_attr
= *attr
;
1697 alloc_split_items_count(ht
);
1698 /* this mutex should not nest in read-side C.S. */
1699 pthread_mutex_init(&ht
->resize_mutex
, NULL
);
1700 order
= cds_lfht_get_count_order_ulong(init_size
);
1701 ht
->resize_target
= 1UL << order
;
1702 cds_lfht_create_bucket(ht
, 1UL << order
);
1703 ht
->size
= 1UL << order
;
1707 struct cds_lfht
*_cds_lfht_new(unsigned long init_size
,
1708 unsigned long min_nr_alloc_buckets
,
1709 unsigned long max_nr_buckets
,
1711 const struct cds_lfht_mm_type
*mm
,
1712 const struct rcu_flavor_struct
*flavor
,
1713 pthread_attr_t
*attr
)
1715 return _cds_lfht_new_with_alloc(init_size
,
1716 min_nr_alloc_buckets
, max_nr_buckets
,
1717 flags
, mm
, flavor
, NULL
, attr
);
1720 void cds_lfht_lookup(struct cds_lfht
*ht
, unsigned long hash
,
1721 cds_lfht_match_fct match
, const void *key
,
1722 struct cds_lfht_iter
*iter
)
1724 struct cds_lfht_node
*node
, *next
, *bucket
;
1725 unsigned long reverse_hash
, size
;
1727 cds_lfht_iter_debug_set_ht(ht
, iter
);
1729 reverse_hash
= bit_reverse_ulong(hash
);
1732 * Use load acquire instead of rcu_dereference because there is no
1733 * dependency between the table size and the dereference of the bucket
1736 * This acquire is paired with the store release in init_table().
1738 size
= uatomic_load(&ht
->size
, CMM_ACQUIRE
);
1739 bucket
= lookup_bucket(ht
, size
, hash
);
1740 /* We can always skip the bucket node initially */
1741 node
= rcu_dereference(bucket
->next
);
1742 node
= clear_flag(node
);
1744 if (caa_unlikely(is_end(node
))) {
1748 if (caa_unlikely(node
->reverse_hash
> reverse_hash
)) {
1752 next
= rcu_dereference(node
->next
);
1753 urcu_posix_assert(node
== clear_flag(node
));
1754 if (caa_likely(!is_removed(next
))
1756 && node
->reverse_hash
== reverse_hash
1757 && caa_likely(match(node
, key
))) {
1760 node
= clear_flag(next
);
1762 urcu_posix_assert(!node
|| !is_bucket(CMM_LOAD_SHARED(node
->next
)));
1767 void cds_lfht_next_duplicate(struct cds_lfht
*ht
__attribute__((unused
)),
1768 cds_lfht_match_fct match
,
1769 const void *key
, struct cds_lfht_iter
*iter
)
1771 struct cds_lfht_node
*node
, *next
;
1772 unsigned long reverse_hash
;
1774 cds_lfht_iter_debug_assert(ht
== iter
->lfht
);
1776 reverse_hash
= node
->reverse_hash
;
1778 node
= clear_flag(next
);
1781 if (caa_unlikely(is_end(node
))) {
1785 if (caa_unlikely(node
->reverse_hash
> reverse_hash
)) {
1789 next
= rcu_dereference(node
->next
);
1790 if (caa_likely(!is_removed(next
))
1792 && caa_likely(match(node
, key
))) {
1795 node
= clear_flag(next
);
1797 urcu_posix_assert(!node
|| !is_bucket(uatomic_load(&node
->next
, CMM_RELAXED
)));
1802 void cds_lfht_next(struct cds_lfht
*ht
__attribute__((unused
)),
1803 struct cds_lfht_iter
*iter
)
1805 struct cds_lfht_node
*node
, *next
;
1807 cds_lfht_iter_debug_assert(ht
== iter
->lfht
);
1808 node
= clear_flag(iter
->next
);
1810 if (caa_unlikely(is_end(node
))) {
1814 next
= rcu_dereference(node
->next
);
1815 if (caa_likely(!is_removed(next
))
1816 && !is_bucket(next
)) {
1819 node
= clear_flag(next
);
1821 urcu_posix_assert(!node
|| !is_bucket(uatomic_load(&node
->next
, CMM_RELAXED
)));
1826 void cds_lfht_first(struct cds_lfht
*ht
, struct cds_lfht_iter
*iter
)
1828 cds_lfht_iter_debug_set_ht(ht
, iter
);
1830 * Get next after first bucket node. The first bucket node is the
1831 * first node of the linked list.
1833 iter
->next
= uatomic_load(&bucket_at(ht
, 0)->next
, CMM_CONSUME
);
1834 cds_lfht_next(ht
, iter
);
1837 void cds_lfht_add(struct cds_lfht
*ht
, unsigned long hash
,
1838 struct cds_lfht_node
*node
)
1842 node
->reverse_hash
= bit_reverse_ulong(hash
);
1843 size
= uatomic_load(&ht
->size
, CMM_ACQUIRE
);
1844 _cds_lfht_add(ht
, hash
, NULL
, NULL
, size
, node
, NULL
, 0);
1845 ht_count_add(ht
, size
, hash
);
1848 struct cds_lfht_node
*cds_lfht_add_unique(struct cds_lfht
*ht
,
1850 cds_lfht_match_fct match
,
1852 struct cds_lfht_node
*node
)
1855 struct cds_lfht_iter iter
;
1857 node
->reverse_hash
= bit_reverse_ulong(hash
);
1858 size
= uatomic_load(&ht
->size
, CMM_ACQUIRE
);
1859 _cds_lfht_add(ht
, hash
, match
, key
, size
, node
, &iter
, 0);
1860 if (iter
.node
== node
)
1861 ht_count_add(ht
, size
, hash
);
1865 struct cds_lfht_node
*cds_lfht_add_replace(struct cds_lfht
*ht
,
1867 cds_lfht_match_fct match
,
1869 struct cds_lfht_node
*node
)
1872 struct cds_lfht_iter iter
;
1874 node
->reverse_hash
= bit_reverse_ulong(hash
);
1875 size
= uatomic_load(&ht
->size
, CMM_ACQUIRE
);
1877 _cds_lfht_add(ht
, hash
, match
, key
, size
, node
, &iter
, 0);
1878 if (iter
.node
== node
) {
1879 ht_count_add(ht
, size
, hash
);
1883 if (!_cds_lfht_replace(ht
, size
, iter
.node
, iter
.next
, node
))
1888 int cds_lfht_replace(struct cds_lfht
*ht
,
1889 struct cds_lfht_iter
*old_iter
,
1891 cds_lfht_match_fct match
,
1893 struct cds_lfht_node
*new_node
)
1897 new_node
->reverse_hash
= bit_reverse_ulong(hash
);
1898 if (!old_iter
->node
)
1900 if (caa_unlikely(old_iter
->node
->reverse_hash
!= new_node
->reverse_hash
))
1902 if (caa_unlikely(!match(old_iter
->node
, key
)))
1904 size
= uatomic_load(&ht
->size
, CMM_ACQUIRE
);
1905 return _cds_lfht_replace(ht
, size
, old_iter
->node
, old_iter
->next
,
1909 int cds_lfht_del(struct cds_lfht
*ht
, struct cds_lfht_node
*node
)
1914 size
= uatomic_load(&ht
->size
, CMM_ACQUIRE
);
1915 ret
= _cds_lfht_del(ht
, size
, node
);
1919 hash
= bit_reverse_ulong(node
->reverse_hash
);
1920 ht_count_del(ht
, size
, hash
);
1925 int cds_lfht_is_node_deleted(const struct cds_lfht_node
*node
)
1927 return is_removed(CMM_LOAD_SHARED(node
->next
));
1931 bool cds_lfht_is_empty(struct cds_lfht
*ht
)
1933 struct cds_lfht_node
*node
, *next
;
1937 was_online
= ht
->flavor
->read_ongoing();
1939 ht
->flavor
->thread_online();
1940 ht
->flavor
->read_lock();
1942 /* Check that the table is empty */
1943 node
= bucket_at(ht
, 0);
1945 next
= rcu_dereference(node
->next
);
1946 if (!is_bucket(next
)) {
1950 node
= clear_flag(next
);
1951 } while (!is_end(node
));
1953 ht
->flavor
->read_unlock();
1954 ht
->flavor
->thread_offline();
1960 int cds_lfht_delete_bucket(struct cds_lfht
*ht
)
1962 struct cds_lfht_node
*node
;
1963 unsigned long order
, i
, size
;
1965 /* Check that the table is empty */
1966 node
= bucket_at(ht
, 0);
1968 node
= clear_flag(node
)->next
;
1969 if (!is_bucket(node
))
1971 urcu_posix_assert(!is_removed(node
));
1972 urcu_posix_assert(!is_removal_owner(node
));
1973 } while (!is_end(node
));
1975 * size accessed without rcu_dereference because hash table is
1979 /* Internal sanity check: all nodes left should be buckets */
1980 for (i
= 0; i
< size
; i
++) {
1981 node
= bucket_at(ht
, i
);
1982 dbg_printf("delete bucket: index %lu expected hash %lu hash %lu\n",
1983 i
, i
, bit_reverse_ulong(node
->reverse_hash
));
1984 urcu_posix_assert(is_bucket(node
->next
));
1987 for (order
= cds_lfht_get_count_order_ulong(size
); (long)order
>= 0; order
--)
1988 cds_lfht_free_bucket_table(ht
, order
);
1994 void do_auto_resize_destroy_cb(struct urcu_work
*work
)
1996 struct cds_lfht
*ht
= caa_container_of(work
, struct cds_lfht
, destroy_work
);
1999 ht
->flavor
->register_thread();
2000 ret
= cds_lfht_delete_bucket(ht
);
2003 free_split_items_count(ht
);
2004 ret
= pthread_mutex_destroy(&ht
->resize_mutex
);
2007 ht
->flavor
->unregister_thread();
2008 poison_free(ht
->alloc
, ht
);
2012 * Should only be called when no more concurrent readers nor writers can
2013 * possibly access the table.
2015 int cds_lfht_destroy(struct cds_lfht
*ht
, pthread_attr_t
**attr
)
2019 if (ht
->flags
& CDS_LFHT_AUTO_RESIZE
) {
2021 * Perform error-checking for emptiness before queuing
2022 * work, so we can return error to the caller. This runs
2023 * concurrently with ongoing resize.
2025 if (!cds_lfht_is_empty(ht
))
2027 /* Cancel ongoing resize operations. */
2028 uatomic_store(&ht
->in_progress_destroy
, 1, CMM_RELAXED
);
2030 *attr
= ht
->caller_resize_attr
;
2031 ht
->caller_resize_attr
= NULL
;
2034 * Queue destroy work after prior queued resize
2035 * operations. Given there are no concurrent writers
2036 * accessing the hash table at this point, no resize
2037 * operations can be queued after this destroy work.
2039 urcu_workqueue_queue_work(cds_lfht_workqueue
,
2040 &ht
->destroy_work
, do_auto_resize_destroy_cb
);
2043 ret
= cds_lfht_delete_bucket(ht
);
2046 free_split_items_count(ht
);
2048 *attr
= ht
->caller_resize_attr
;
2049 ret
= pthread_mutex_destroy(&ht
->resize_mutex
);
2052 poison_free(ht
->alloc
, ht
);
2056 void cds_lfht_count_nodes(struct cds_lfht
*ht
,
2057 long *approx_before
,
2058 unsigned long *count
,
2061 struct cds_lfht_node
*node
, *next
;
2062 unsigned long nr_bucket
= 0, nr_removed
= 0;
2065 if (ht
->split_count
) {
2068 for (i
= 0; i
< split_count_mask
+ 1; i
++) {
2069 *approx_before
+= uatomic_read(&ht
->split_count
[i
].add
);
2070 *approx_before
-= uatomic_read(&ht
->split_count
[i
].del
);
2076 /* Count non-bucket nodes in the table */
2077 node
= bucket_at(ht
, 0);
2079 next
= rcu_dereference(node
->next
);
2080 if (is_removed(next
)) {
2081 if (!is_bucket(next
))
2085 } else if (!is_bucket(next
))
2089 node
= clear_flag(next
);
2090 } while (!is_end(node
));
2091 dbg_printf("number of logically removed nodes: %lu\n", nr_removed
);
2092 dbg_printf("number of bucket nodes: %lu\n", nr_bucket
);
2094 if (ht
->split_count
) {
2097 for (i
= 0; i
< split_count_mask
+ 1; i
++) {
2098 *approx_after
+= uatomic_read(&ht
->split_count
[i
].add
);
2099 *approx_after
-= uatomic_read(&ht
->split_count
[i
].del
);
2104 /* called with resize mutex held */
2106 void _do_cds_lfht_grow(struct cds_lfht
*ht
,
2107 unsigned long old_size
, unsigned long new_size
)
2109 unsigned long old_order
, new_order
;
2111 old_order
= cds_lfht_get_count_order_ulong(old_size
);
2112 new_order
= cds_lfht_get_count_order_ulong(new_size
);
2113 dbg_printf("resize from %lu (order %lu) to %lu (order %lu) buckets\n",
2114 old_size
, old_order
, new_size
, new_order
);
2115 urcu_posix_assert(new_size
> old_size
);
2116 init_table(ht
, old_order
+ 1, new_order
);
2119 /* called with resize mutex held */
2121 void _do_cds_lfht_shrink(struct cds_lfht
*ht
,
2122 unsigned long old_size
, unsigned long new_size
)
2124 unsigned long old_order
, new_order
;
2126 new_size
= max(new_size
, MIN_TABLE_SIZE
);
2127 old_order
= cds_lfht_get_count_order_ulong(old_size
);
2128 new_order
= cds_lfht_get_count_order_ulong(new_size
);
2129 dbg_printf("resize from %lu (order %lu) to %lu (order %lu) buckets\n",
2130 old_size
, old_order
, new_size
, new_order
);
2131 urcu_posix_assert(new_size
< old_size
);
2133 /* Remove and unlink all bucket nodes to remove. */
2134 fini_table(ht
, new_order
+ 1, old_order
);
2138 /* called with resize mutex held */
2140 void _do_cds_lfht_resize(struct cds_lfht
*ht
)
2142 unsigned long new_size
, old_size
;
2145 * Resize table, re-do if the target size has changed under us.
2148 if (uatomic_load(&ht
->in_progress_destroy
, CMM_RELAXED
))
2151 uatomic_store(&ht
->resize_initiated
, 1, CMM_RELAXED
);
2153 old_size
= ht
->size
;
2154 new_size
= uatomic_load(&ht
->resize_target
, CMM_RELAXED
);
2155 if (old_size
< new_size
)
2156 _do_cds_lfht_grow(ht
, old_size
, new_size
);
2157 else if (old_size
> new_size
)
2158 _do_cds_lfht_shrink(ht
, old_size
, new_size
);
2160 uatomic_store(&ht
->resize_initiated
, 0, CMM_RELAXED
);
2161 /* write resize_initiated before read resize_target */
2163 } while (ht
->size
!= uatomic_load(&ht
->resize_target
, CMM_RELAXED
));
2167 unsigned long resize_target_grow(struct cds_lfht
*ht
, unsigned long new_size
)
2169 return _uatomic_xchg_monotonic_increase(&ht
->resize_target
, new_size
);
2173 void resize_target_update_count(struct cds_lfht
*ht
,
2174 unsigned long count
)
2176 count
= max(count
, MIN_TABLE_SIZE
);
2177 count
= min(count
, ht
->max_nr_buckets
);
2178 uatomic_set(&ht
->resize_target
, count
);
2181 void cds_lfht_resize(struct cds_lfht
*ht
, unsigned long new_size
)
2183 resize_target_update_count(ht
, new_size
);
2186 * Set flags has early as possible even in contention case.
2188 uatomic_store(&ht
->resize_initiated
, 1, CMM_RELAXED
);
2190 mutex_lock(&ht
->resize_mutex
);
2191 _do_cds_lfht_resize(ht
);
2192 mutex_unlock(&ht
->resize_mutex
);
2196 void do_resize_cb(struct urcu_work
*work
)
2198 struct resize_work
*resize_work
=
2199 caa_container_of(work
, struct resize_work
, work
);
2200 struct cds_lfht
*ht
= resize_work
->ht
;
2202 ht
->flavor
->register_thread();
2203 mutex_lock(&ht
->resize_mutex
);
2204 _do_cds_lfht_resize(ht
);
2205 mutex_unlock(&ht
->resize_mutex
);
2206 ht
->flavor
->unregister_thread();
2207 poison_free(ht
->alloc
, work
);
2211 void __cds_lfht_resize_lazy_launch(struct cds_lfht
*ht
)
2213 struct resize_work
*work
;
2216 * Store to resize_target is before read resize_initiated as guaranteed
2217 * by either cmpxchg or _uatomic_xchg_monotonic_increase.
2219 if (!uatomic_load(&ht
->resize_initiated
, CMM_RELAXED
)) {
2220 if (uatomic_load(&ht
->in_progress_destroy
, CMM_RELAXED
)) {
2223 work
= ht
->alloc
->malloc(ht
->alloc
->state
, sizeof(*work
));
2225 dbg_printf("error allocating resize work, bailing out\n");
2229 urcu_workqueue_queue_work(cds_lfht_workqueue
,
2230 &work
->work
, do_resize_cb
);
2231 uatomic_store(&ht
->resize_initiated
, 1, CMM_RELAXED
);
2236 void cds_lfht_resize_lazy_grow(struct cds_lfht
*ht
, unsigned long size
, int growth
)
2238 unsigned long target_size
= size
<< growth
;
2240 target_size
= min(target_size
, ht
->max_nr_buckets
);
2241 if (resize_target_grow(ht
, target_size
) >= target_size
)
2244 __cds_lfht_resize_lazy_launch(ht
);
2248 * We favor grow operations over shrink. A shrink operation never occurs
2249 * if a grow operation is queued for lazy execution. A grow operation
2250 * cancels any pending shrink lazy execution.
2253 void cds_lfht_resize_lazy_count(struct cds_lfht
*ht
, unsigned long size
,
2254 unsigned long count
)
2256 if (!(ht
->flags
& CDS_LFHT_AUTO_RESIZE
))
2258 count
= max(count
, MIN_TABLE_SIZE
);
2259 count
= min(count
, ht
->max_nr_buckets
);
2261 return; /* Already the right size, no resize needed */
2262 if (count
> size
) { /* lazy grow */
2263 if (resize_target_grow(ht
, count
) >= count
)
2265 } else { /* lazy shrink */
2269 s
= uatomic_cmpxchg(&ht
->resize_target
, size
, count
);
2271 break; /* no resize needed */
2273 return; /* growing is/(was just) in progress */
2275 return; /* some other thread do shrink */
2279 __cds_lfht_resize_lazy_launch(ht
);
2282 static void cds_lfht_before_fork(void *priv
__attribute__((unused
)))
2284 if (cds_lfht_workqueue_atfork_nesting
++)
2286 mutex_lock(&cds_lfht_fork_mutex
);
2287 if (!cds_lfht_workqueue
)
2289 urcu_workqueue_pause_worker(cds_lfht_workqueue
);
2292 static void cds_lfht_after_fork_parent(void *priv
__attribute__((unused
)))
2294 if (--cds_lfht_workqueue_atfork_nesting
)
2296 if (!cds_lfht_workqueue
)
2298 urcu_workqueue_resume_worker(cds_lfht_workqueue
);
2300 mutex_unlock(&cds_lfht_fork_mutex
);
2303 static void cds_lfht_after_fork_child(void *priv
__attribute__((unused
)))
2305 if (--cds_lfht_workqueue_atfork_nesting
)
2307 if (!cds_lfht_workqueue
)
2309 urcu_workqueue_create_worker(cds_lfht_workqueue
);
2311 mutex_unlock(&cds_lfht_fork_mutex
);
2314 static struct urcu_atfork cds_lfht_atfork
= {
2315 .before_fork
= cds_lfht_before_fork
,
2316 .after_fork_parent
= cds_lfht_after_fork_parent
,
2317 .after_fork_child
= cds_lfht_after_fork_child
,
2320 static void cds_lfht_init_worker(const struct rcu_flavor_struct
*flavor
)
2322 flavor
->register_rculfhash_atfork(&cds_lfht_atfork
);
2324 mutex_lock(&cds_lfht_fork_mutex
);
2325 if (!cds_lfht_workqueue
)
2326 cds_lfht_workqueue
= urcu_workqueue_create(0, -1, NULL
,
2327 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
);
2328 mutex_unlock(&cds_lfht_fork_mutex
);
2331 static void cds_lfht_exit(void)
2333 mutex_lock(&cds_lfht_fork_mutex
);
2334 if (cds_lfht_workqueue
) {
2335 urcu_workqueue_flush_queued_work(cds_lfht_workqueue
);
2336 urcu_workqueue_destroy(cds_lfht_workqueue
);
2337 cds_lfht_workqueue
= NULL
;
2339 mutex_unlock(&cds_lfht_fork_mutex
);