aboutsummaryrefslogtreecommitdiffstats
path: root/libparc/parc/concurrent/parc_RingBuffer_1x1.c
blob: a023f4d748d5bfcaff9fb526cf532feeb8526c76 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
/*
 * Copyright (c) 2017 Cisco and/or its affiliates.
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at:
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

/**
 * A thread-safe fixed size ring buffer.
 *
 * A single-producer/single-consumer version is lock-free, along the lines of Lamport, "Proving the
 * Correctness of Multiprocess Programs," IEEE Trans on Software Engineering 3(2), Mar 1977, which
 * is based on reading/writing native types upto the data bus width being atomic operations.
 *
 * It can hold (elements-1) data items.  elements must be a power of 2.
 *
 * The writer_head is where the next element should be inserted.  The reader_tail is where the next element
 * should be read.
 *
 * All index variables are unbounded uint32_t.  This means they just keep counting up.  To get the actual
 * index in the ring, we mask with (elements-1).  For example, a ring with 16 elements will be masked with
 * 0x0000000F.  We call this the "ring_mask".
 *
 * Because we never let the writer_head and reader_tail differ by more than (elements-1), this technique of
 * masking works just the same as taking the modulus.  There's no problems at the uint32_t wraparound either.
 * The only math operation we are doing is "+1" which works just fine to wrap a uint32_t.
 *
 * Let's look at some exampls.  I'm going to use a uint16_t so its easier to write the numbers.  Let's assume
 * that the ring size is 16, so the first ring is (0 - 15).
 *              head     tail
 * initialize      0        0
 * put x 3         3        0
 * get x 2         3        2
 * put x 13       16        2
 * put x 1        17        2
 * put x 1      blocks          # (0x11 + 1) & 0x0F  == tail & 0x0F
 * get x 14       17       16
 * get x 1        17       17   # ring is now empty
 * ...
 * empty       65534    65534   # 0xFFFE  0xFFFE masked =  14   14
 * put x1      65535    65534   # 0xFFFF  0xFFFE masked =  15   14
 * put x1          0    65534   # 0x0000  0xFFFE masked =   0   14
 * ...
 *
 * The number of remaining available items is (ring_mask + reader_tail - writer_head) & ring_mask.
 *              head     tail   remaining
 * initialize      0        0   15 + 0 - 0 = 15
 * put x 3         3        0   15 + 0 - 3 = 12
 * get x 2         3        2
 * put x 13       16        2   15 + 2 - 16 = 1
 * put x 1        17        2   15 + 2 - 17 = 0
 * put x 1      blocks
 * get x 14       17       16   15 + 16 - 17 = 14
 * get x 1        17       17   15 + 17 - 17 = 15
 * ...
 * empty       65534    65534   15 + 65534 - 65534 = 13 - 65534 = 13 - (-2) = 15
 * put x1      65535    65534   15 + 65534 - 65535 = 13 - 65535 = 13 - (-1) = 14
 * put x1          0    65534   15 + 65534 -     0 = 13 - 65535 = 13 - ( 0) = 13
 * ...
 *
 * If (writer_head + 1) & ring_mask == reader_tail, then the ring is full.
 * If writer_head == reader_tail, then the ring is empty.
 *
 */

#include <config.h>
#include <stdio.h>
#include <stdlib.h>
#include <pthread.h>
#include <errno.h>
#include <string.h>
#include <unistd.h>

#include <parc/algol/parc_Memory.h>
#include <parc/algol/parc_Object.h>
#include <LongBow/runtime.h>

#include <parc/concurrent/parc_RingBuffer_1x1.h>

#ifdef __GNUC__

// on x86 or x86_64, simple assignment will work
#if (__x86_64__ || __i386__)
#define ATOMIC_ADD_AND_FETCH(ptr, increment)      __sync_add_and_fetch(ptr, increment)
#define ATOMIC_BOOL_CAS(ptr, oldvalue, newvalue)  __sync_bool_compare_and_swap(ptr, oldvalue, newvalue)
#define ATOMIC_FETCH(ptr)                        *(ptr)
#define ATOMIC_SET(ptr, oldvalue, newvalue)      *(ptr) = newvalue
#else
#define ATOMIC_ADD_AND_FETCH(ptr, increment)      __sync_add_and_fetch(ptr, increment)
#define ATOMIC_BOOL_CAS(ptr, oldvalue, newvalue)  __sync_bool_compare_and_swap(ptr, oldvalue, newvalue)
#define ATOMIC_FETCH(ptr)                         ATOMIC_ADD_AND_FETCH(ptr, 0)
#define ATOMIC_SET(ptr, oldvalue, newvalue)       ATOMIC_BOOL_CAS(ptr, oldvalue, newvalue)
#endif

#else
#error "Only GNUC supported, we need atomic operations"
#endif

struct parc_ringbuffer_1x1 {
    // LP64     LP32
    volatile uint32_t writer_head;          //  0- 3       0
    volatile uint32_t reader_tail;          //  4- 7       4
    uint32_t elements;                      //  8-11       8
    uint32_t ring_mask;                     // 12-15      12

    RingBufferEntryDestroyer *destroyer;    // 16-23      16
    void **buffer;                          // 24-31      24
};

static bool
_isPowerOfTwo(uint32_t x)
{
    return ((x != 0) && !(x & (x - 1)));
}

static void
_destroy(PARCRingBuffer1x1 **ringptr)
{
    PARCRingBuffer1x1 *ring = *ringptr;

    if (ring->destroyer) {
        void *ptr = NULL;
        while (parcRingBuffer1x1_Get(ring, &ptr)) {
            ring->destroyer(&ptr);
        }
    }
    parcMemory_Deallocate((void **) &(ring->buffer));
}

parcObject_ExtendPARCObject(PARCRingBuffer1x1, _destroy, NULL, NULL, NULL, NULL, NULL, NULL);

static PARCRingBuffer1x1 *
_create(uint32_t elements, RingBufferEntryDestroyer *destroyer)
{
    PARCRingBuffer1x1 *ring = parcObject_CreateInstance(PARCRingBuffer1x1);
    assertNotNull(ring, "parcObject_Create returned NULL");

    ring->buffer = parcMemory_AllocateAndClear(sizeof(void *) * elements);
    assertNotNull((ring->buffer), "parcMemory_AllocateAndClear() failed to allocate array of %u pointers", elements);

    ring->writer_head = 0;
    ring->reader_tail = 0;
    ring->elements = elements;
    ring->destroyer = destroyer;
    ring->ring_mask = elements - 1;

    return ring;
}

PARCRingBuffer1x1 *
parcRingBuffer1x1_Create(uint32_t elements, RingBufferEntryDestroyer *destroyer)
{
    assertTrue(_isPowerOfTwo(elements), "Parameter elements must be a power of 2, got %u", elements);
    return _create(elements, destroyer);
}


parcObject_ImplementAcquire(parcRingBuffer1x1, PARCRingBuffer1x1);

parcObject_ImplementRelease(parcRingBuffer1x1, PARCRingBuffer1x1);

/**
 * Put is protected by the writer mutex.  This means that the tail mutex could
 * actually increase while this is happening. That's ok.  Increasing the tail
 * just means there is _more_ room in the ring.  We only modify writer_head.
 */
bool
parcRingBuffer1x1_Put(PARCRingBuffer1x1 *ring, void *data)
{
    // Our speculative operation
    // The consumer modifies reader_tail, so make sure that's an atomic read.
    // only the prodcuer modifies writer_head, so there's only us

    uint32_t writer_head = ring->writer_head;
    uint32_t reader_tail = ATOMIC_FETCH(&ring->reader_tail);

    uint32_t writer_next = (writer_head + 1) & ring->ring_mask;

    // ring is full
    if (writer_next == reader_tail) {
        return false;
    }

    assertNull(ring->buffer[writer_head], "Ring index %u is not null!", writer_head);
    ring->buffer[writer_head] = data;

    // we're using this just for atomic write to the integer
    ATOMIC_SET(&ring->writer_head, writer_head, writer_next);

    return true;
}

bool
parcRingBuffer1x1_Get(PARCRingBuffer1x1 *ring, void **outputDataPtr)
{
    // do our speculative operation.
    // The producer modifies writer_head, so make sure that's an atomic read.
    // only the consumer modifies reader_tail, so there's only us

    uint32_t writer_head = ATOMIC_FETCH(&ring->writer_head); // native type assignment is atomic
    uint32_t reader_tail = ring->reader_tail;
    uint32_t reader_next = (reader_tail + 1) & ring->ring_mask;

    // ring is empty
    if (writer_head == reader_tail) {
        return false;
    }

    // now try to commit it
    ATOMIC_SET(&ring->reader_tail, reader_tail, reader_next);

    *outputDataPtr = ring->buffer[reader_tail];

    // for sanity's sake
    ring->buffer[reader_tail] = NULL;

    return true;
}

uint32_t
parcRingBuffer1x1_Remaining(PARCRingBuffer1x1 *ring)
{
    uint32_t writer_head = ATOMIC_FETCH(&ring->writer_head);
    uint32_t reader_tail = ATOMIC_FETCH(&ring->reader_tail);

    return (ring->ring_mask + reader_tail - writer_head) & ring->ring_mask;
}