]>
Commit | Line | Data |
---|---|---|
9f95a23c TL |
1 | .. SPDX-License-Identifier: BSD-3-Clause |
2 | Copyright(c) 2010-2014 Intel Corporation. | |
7c673cae FG |
3 | |
4 | .. _Ring_Library: | |
5 | ||
6 | Ring Library | |
7 | ============ | |
8 | ||
9 | The ring allows the management of queues. | |
10 | Instead of having a linked list of infinite size, the rte_ring has the following properties: | |
11 | ||
12 | * FIFO | |
13 | ||
14 | * Maximum size is fixed, the pointers are stored in a table | |
15 | ||
16 | * Lockless implementation | |
17 | ||
18 | * Multi-consumer or single-consumer dequeue | |
19 | ||
20 | * Multi-producer or single-producer enqueue | |
21 | ||
22 | * Bulk dequeue - Dequeues the specified count of objects if successful; otherwise fails | |
23 | ||
24 | * Bulk enqueue - Enqueues the specified count of objects if successful; otherwise fails | |
25 | ||
26 | * Burst dequeue - Dequeue the maximum available objects if the specified count cannot be fulfilled | |
27 | ||
28 | * Burst enqueue - Enqueue the maximum available objects if the specified count cannot be fulfilled | |
29 | ||
30 | The advantages of this data structure over a linked list queue are as follows: | |
31 | ||
32 | * Faster; only requires a single Compare-And-Swap instruction of sizeof(void \*) instead of several double-Compare-And-Swap instructions. | |
33 | ||
34 | * Simpler than a full lockless queue. | |
35 | ||
36 | * Adapted to bulk enqueue/dequeue operations. | |
37 | As pointers are stored in a table, a dequeue of several objects will not produce as many cache misses as in a linked queue. | |
38 | Also, a bulk dequeue of many objects does not cost more than a dequeue of a simple object. | |
39 | ||
40 | The disadvantages: | |
41 | ||
42 | * Size is fixed | |
43 | ||
44 | * Having many rings costs more in terms of memory than a linked list queue. An empty ring contains at least N pointers. | |
45 | ||
46 | A simplified representation of a Ring is shown in with consumer and producer head and tail pointers to objects stored in the data structure. | |
47 | ||
48 | .. _figure_ring1: | |
49 | ||
50 | .. figure:: img/ring1.* | |
51 | ||
52 | Ring Structure | |
53 | ||
54 | ||
55 | References for Ring Implementation in FreeBSD* | |
56 | ---------------------------------------------- | |
57 | ||
58 | The following code was added in FreeBSD 8.0, and is used in some network device drivers (at least in Intel drivers): | |
59 | ||
60 | * `bufring.h in FreeBSD <http://svn.freebsd.org/viewvc/base/release/8.0.0/sys/sys/buf_ring.h?revision=199625&view=markup>`_ | |
61 | ||
62 | * `bufring.c in FreeBSD <http://svn.freebsd.org/viewvc/base/release/8.0.0/sys/kern/subr_bufring.c?revision=199625&view=markup>`_ | |
63 | ||
64 | Lockless Ring Buffer in Linux* | |
65 | ------------------------------ | |
66 | ||
67 | The following is a link describing the `Linux Lockless Ring Buffer Design <http://lwn.net/Articles/340400/>`_. | |
68 | ||
69 | Additional Features | |
70 | ------------------- | |
71 | ||
72 | Name | |
73 | ~~~~ | |
74 | ||
75 | A ring is identified by a unique name. | |
76 | It is not possible to create two rings with the same name (rte_ring_create() returns NULL if this is attempted). | |
77 | ||
7c673cae FG |
78 | Use Cases |
79 | --------- | |
80 | ||
81 | Use cases for the Ring library include: | |
82 | ||
83 | * Communication between applications in the DPDK | |
84 | ||
85 | * Used by memory pool allocator | |
86 | ||
87 | Anatomy of a Ring Buffer | |
88 | ------------------------ | |
89 | ||
90 | This section explains how a ring buffer operates. | |
91 | The ring structure is composed of two head and tail couples; one is used by producers and one is used by the consumers. | |
92 | The figures of the following sections refer to them as prod_head, prod_tail, cons_head and cons_tail. | |
93 | ||
94 | Each figure represents a simplified state of the ring, which is a circular buffer. | |
95 | The content of the function local variables is represented on the top of the figure, | |
96 | and the content of ring structure is represented on the bottom of the figure. | |
97 | ||
98 | Single Producer Enqueue | |
99 | ~~~~~~~~~~~~~~~~~~~~~~~ | |
100 | ||
101 | This section explains what occurs when a producer adds an object to the ring. | |
102 | In this example, only the producer head and tail (prod_head and prod_tail) are modified, | |
103 | and there is only one producer. | |
104 | ||
105 | The initial state is to have a prod_head and prod_tail pointing at the same location. | |
106 | ||
107 | Enqueue First Step | |
108 | ^^^^^^^^^^^^^^^^^^ | |
109 | ||
110 | First, *ring->prod_head* and ring->cons_tail are copied in local variables. | |
111 | The prod_next local variable points to the next element of the table, or several elements after in case of bulk enqueue. | |
112 | ||
113 | If there is not enough room in the ring (this is detected by checking cons_tail), it returns an error. | |
114 | ||
115 | ||
116 | .. _figure_ring-enqueue1: | |
117 | ||
118 | .. figure:: img/ring-enqueue1.* | |
119 | ||
120 | Enqueue first step | |
121 | ||
122 | ||
123 | Enqueue Second Step | |
124 | ^^^^^^^^^^^^^^^^^^^ | |
125 | ||
126 | The second step is to modify *ring->prod_head* in ring structure to point to the same location as prod_next. | |
127 | ||
128 | A pointer to the added object is copied in the ring (obj4). | |
129 | ||
130 | ||
131 | .. _figure_ring-enqueue2: | |
132 | ||
133 | .. figure:: img/ring-enqueue2.* | |
134 | ||
135 | Enqueue second step | |
136 | ||
137 | ||
138 | Enqueue Last Step | |
139 | ^^^^^^^^^^^^^^^^^ | |
140 | ||
141 | Once the object is added in the ring, ring->prod_tail in the ring structure is modified to point to the same location as *ring->prod_head*. | |
142 | The enqueue operation is finished. | |
143 | ||
144 | ||
145 | .. _figure_ring-enqueue3: | |
146 | ||
147 | .. figure:: img/ring-enqueue3.* | |
148 | ||
149 | Enqueue last step | |
150 | ||
151 | ||
152 | Single Consumer Dequeue | |
153 | ~~~~~~~~~~~~~~~~~~~~~~~ | |
154 | ||
155 | This section explains what occurs when a consumer dequeues an object from the ring. | |
156 | In this example, only the consumer head and tail (cons_head and cons_tail) are modified and there is only one consumer. | |
157 | ||
158 | The initial state is to have a cons_head and cons_tail pointing at the same location. | |
159 | ||
160 | Dequeue First Step | |
161 | ^^^^^^^^^^^^^^^^^^ | |
162 | ||
163 | First, ring->cons_head and ring->prod_tail are copied in local variables. | |
164 | The cons_next local variable points to the next element of the table, or several elements after in the case of bulk dequeue. | |
165 | ||
166 | If there are not enough objects in the ring (this is detected by checking prod_tail), it returns an error. | |
167 | ||
168 | ||
169 | .. _figure_ring-dequeue1: | |
170 | ||
171 | .. figure:: img/ring-dequeue1.* | |
172 | ||
173 | Dequeue last step | |
174 | ||
175 | ||
176 | Dequeue Second Step | |
177 | ^^^^^^^^^^^^^^^^^^^ | |
178 | ||
179 | The second step is to modify ring->cons_head in the ring structure to point to the same location as cons_next. | |
180 | ||
181 | The pointer to the dequeued object (obj1) is copied in the pointer given by the user. | |
182 | ||
183 | ||
184 | .. _figure_ring-dequeue2: | |
185 | ||
186 | .. figure:: img/ring-dequeue2.* | |
187 | ||
188 | Dequeue second step | |
189 | ||
190 | ||
191 | Dequeue Last Step | |
192 | ^^^^^^^^^^^^^^^^^ | |
193 | ||
194 | Finally, ring->cons_tail in the ring structure is modified to point to the same location as ring->cons_head. | |
195 | The dequeue operation is finished. | |
196 | ||
197 | ||
198 | .. _figure_ring-dequeue3: | |
199 | ||
200 | .. figure:: img/ring-dequeue3.* | |
201 | ||
202 | Dequeue last step | |
203 | ||
204 | ||
205 | Multiple Producers Enqueue | |
206 | ~~~~~~~~~~~~~~~~~~~~~~~~~~ | |
207 | ||
208 | This section explains what occurs when two producers concurrently add an object to the ring. | |
209 | In this example, only the producer head and tail (prod_head and prod_tail) are modified. | |
210 | ||
211 | The initial state is to have a prod_head and prod_tail pointing at the same location. | |
212 | ||
213 | Multiple Producers Enqueue First Step | |
214 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | |
215 | ||
216 | On both cores, *ring->prod_head* and ring->cons_tail are copied in local variables. | |
217 | The prod_next local variable points to the next element of the table, | |
218 | or several elements after in the case of bulk enqueue. | |
219 | ||
220 | If there is not enough room in the ring (this is detected by checking cons_tail), it returns an error. | |
221 | ||
222 | ||
223 | .. _figure_ring-mp-enqueue1: | |
224 | ||
225 | .. figure:: img/ring-mp-enqueue1.* | |
226 | ||
227 | Multiple producer enqueue first step | |
228 | ||
229 | ||
230 | Multiple Producers Enqueue Second Step | |
231 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | |
232 | ||
233 | The second step is to modify ring->prod_head in the ring structure to point to the same location as prod_next. | |
234 | This operation is done using a Compare And Swap (CAS) instruction, which does the following operations atomically: | |
235 | ||
236 | * If ring->prod_head is different to local variable prod_head, | |
237 | the CAS operation fails, and the code restarts at first step. | |
238 | ||
239 | * Otherwise, ring->prod_head is set to local prod_next, | |
240 | the CAS operation is successful, and processing continues. | |
241 | ||
242 | In the figure, the operation succeeded on core 1, and step one restarted on core 2. | |
243 | ||
244 | ||
245 | .. _figure_ring-mp-enqueue2: | |
246 | ||
247 | .. figure:: img/ring-mp-enqueue2.* | |
248 | ||
249 | Multiple producer enqueue second step | |
250 | ||
251 | ||
252 | Multiple Producers Enqueue Third Step | |
253 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | |
254 | ||
255 | The CAS operation is retried on core 2 with success. | |
256 | ||
257 | The core 1 updates one element of the ring(obj4), and the core 2 updates another one (obj5). | |
258 | ||
259 | ||
260 | .. _figure_ring-mp-enqueue3: | |
261 | ||
262 | .. figure:: img/ring-mp-enqueue3.* | |
263 | ||
264 | Multiple producer enqueue third step | |
265 | ||
266 | ||
267 | Multiple Producers Enqueue Fourth Step | |
268 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | |
269 | ||
270 | Each core now wants to update ring->prod_tail. | |
271 | A core can only update it if ring->prod_tail is equal to the prod_head local variable. | |
272 | This is only true on core 1. The operation is finished on core 1. | |
273 | ||
274 | ||
275 | .. _figure_ring-mp-enqueue4: | |
276 | ||
277 | .. figure:: img/ring-mp-enqueue4.* | |
278 | ||
279 | Multiple producer enqueue fourth step | |
280 | ||
281 | ||
282 | Multiple Producers Enqueue Last Step | |
283 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | |
284 | ||
285 | Once ring->prod_tail is updated by core 1, core 2 is allowed to update it too. | |
286 | The operation is also finished on core 2. | |
287 | ||
288 | ||
289 | .. _figure_ring-mp-enqueue5: | |
290 | ||
291 | .. figure:: img/ring-mp-enqueue5.* | |
292 | ||
293 | Multiple producer enqueue last step | |
294 | ||
295 | ||
296 | Modulo 32-bit Indexes | |
297 | ~~~~~~~~~~~~~~~~~~~~~ | |
298 | ||
299 | In the preceding figures, the prod_head, prod_tail, cons_head and cons_tail indexes are represented by arrows. | |
300 | In the actual implementation, these values are not between 0 and size(ring)-1 as would be assumed. | |
301 | The indexes are between 0 and 2^32 -1, and we mask their value when we access the pointer table (the ring itself). | |
302 | 32-bit modulo also implies that operations on indexes (such as, add/subtract) will automatically do 2^32 modulo | |
303 | if the result overflows the 32-bit number range. | |
304 | ||
305 | The following are two examples that help to explain how indexes are used in a ring. | |
306 | ||
307 | .. note:: | |
308 | ||
309 | To simplify the explanation, operations with modulo 16-bit are used instead of modulo 32-bit. | |
310 | In addition, the four indexes are defined as unsigned 16-bit integers, | |
311 | as opposed to unsigned 32-bit integers in the more realistic case. | |
312 | ||
313 | ||
314 | .. _figure_ring-modulo1: | |
315 | ||
316 | .. figure:: img/ring-modulo1.* | |
317 | ||
318 | Modulo 32-bit indexes - Example 1 | |
319 | ||
320 | ||
321 | This ring contains 11000 entries. | |
322 | ||
323 | ||
324 | .. _figure_ring-modulo2: | |
325 | ||
326 | .. figure:: img/ring-modulo2.* | |
327 | ||
328 | Modulo 32-bit indexes - Example 2 | |
329 | ||
330 | ||
331 | This ring contains 12536 entries. | |
332 | ||
333 | .. note:: | |
334 | ||
335 | For ease of understanding, we use modulo 65536 operations in the above examples. | |
336 | In real execution cases, this is redundant for low efficiency, but is done automatically when the result overflows. | |
337 | ||
338 | The code always maintains a distance between producer and consumer between 0 and size(ring)-1. | |
339 | Thanks to this property, we can do subtractions between 2 index values in a modulo-32bit base: | |
340 | that's why the overflow of the indexes is not a problem. | |
341 | ||
342 | At any time, entries and free_entries are between 0 and size(ring)-1, | |
343 | even if only the first term of subtraction has overflowed: | |
344 | ||
345 | .. code-block:: c | |
346 | ||
347 | uint32_t entries = (prod_tail - cons_head); | |
348 | uint32_t free_entries = (mask + cons_tail -prod_head); | |
349 | ||
350 | References | |
351 | ---------- | |
352 | ||
353 | * `bufring.h in FreeBSD <http://svn.freebsd.org/viewvc/base/release/8.0.0/sys/sys/buf_ring.h?revision=199625&view=markup>`_ (version 8) | |
354 | ||
355 | * `bufring.c in FreeBSD <http://svn.freebsd.org/viewvc/base/release/8.0.0/sys/kern/subr_bufring.c?revision=199625&view=markup>`_ (version 8) | |
356 | ||
357 | * `Linux Lockless Ring Buffer Design <http://lwn.net/Articles/340400/>`_ |