xref: /trueos/sys/sys/buf_ring.h (revision b3587236c4ed5131fd2c62c2f7ec705c43624f3a)
1 /*-
2  * Copyright (c) 2007-2009 Kip Macy <kmacy@freebsd.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  *
28  */
29 
30 #ifndef	_SYS_BUF_RING_H_
31 #define	_SYS_BUF_RING_H_
32 
33 #include <machine/cpu.h>
34 
35 #if defined(INVARIANTS) && !defined(DEBUG_BUFRING)
36 #define DEBUG_BUFRING 1
37 #endif
38 
39 #ifdef DEBUG_BUFRING
40 #include <sys/lock.h>
41 #include <sys/mutex.h>
42 #endif
43 
44 struct buf_ring {
45 	volatile uint32_t	br_prod_head;
46 	volatile uint32_t	br_prod_tail;
47 	int              	br_prod_size;
48 	int              	br_prod_mask;
49 	uint64_t		br_drops;
50 	volatile uint32_t	br_cons_head __aligned(CACHE_LINE_SIZE);
51 	volatile uint32_t	br_cons_tail;
52 	int		 	br_cons_size;
53 	int              	br_cons_mask;
54 #ifdef DEBUG_BUFRING
55 	struct mtx		*br_lock;
56 #endif
57 	void			*br_ring[0] __aligned(CACHE_LINE_SIZE);
58 };
59 
60 /*
61  * multi-producer safe lock-free ring buffer enqueue
62  *
63  */
64 static __inline int
buf_ring_enqueue(struct buf_ring * br,void * buf)65 buf_ring_enqueue(struct buf_ring *br, void *buf)
66 {
67 	uint32_t prod_head, prod_next, cons_tail;
68 #ifdef DEBUG_BUFRING
69 	int i;
70 	for (i = br->br_cons_head; i != br->br_prod_head;
71 	     i = ((i + 1) & br->br_cons_mask))
72 		if(br->br_ring[i] == buf)
73 			panic("buf=%p already enqueue at %d prod=%d cons=%d",
74 			    buf, i, br->br_prod_tail, br->br_cons_tail);
75 #endif
76 	critical_enter();
77 	do {
78 		prod_head = br->br_prod_head;
79 		prod_next = (prod_head + 1) & br->br_prod_mask;
80 		cons_tail = br->br_cons_tail;
81 
82 		if (prod_next == cons_tail) {
83 			rmb();
84 			if (prod_head == br->br_prod_head &&
85 			    cons_tail == br->br_cons_tail) {
86 				br->br_drops++;
87 				critical_exit();
88 				return (ENOBUFS);
89 			}
90 			continue;
91 		}
92 	} while (!atomic_cmpset_acq_int(&br->br_prod_head, prod_head, prod_next));
93 #ifdef DEBUG_BUFRING
94 	if (br->br_ring[prod_head] != NULL)
95 		panic("dangling value in enqueue");
96 #endif
97 	br->br_ring[prod_head] = buf;
98 
99 	/*
100 	 * If there are other enqueues in progress
101 	 * that preceeded us, we need to wait for them
102 	 * to complete
103 	 */
104 	while (br->br_prod_tail != prod_head)
105 		cpu_spinwait();
106 	atomic_store_rel_int(&br->br_prod_tail, prod_next);
107 	critical_exit();
108 	return (0);
109 }
110 
111 /*
112  * multi-consumer safe dequeue
113  *
114  */
115 static __inline void *
buf_ring_dequeue_mc(struct buf_ring * br)116 buf_ring_dequeue_mc(struct buf_ring *br)
117 {
118 	uint32_t cons_head, cons_next;
119 	void *buf;
120 
121 	critical_enter();
122 	do {
123 		cons_head = br->br_cons_head;
124 		cons_next = (cons_head + 1) & br->br_cons_mask;
125 
126 		if (cons_head == br->br_prod_tail) {
127 			critical_exit();
128 			return (NULL);
129 		}
130 	} while (!atomic_cmpset_acq_int(&br->br_cons_head, cons_head, cons_next));
131 
132 	buf = br->br_ring[cons_head];
133 #ifdef DEBUG_BUFRING
134 	br->br_ring[cons_head] = NULL;
135 #endif
136 	/*
137 	 * If there are other dequeues in progress
138 	 * that preceeded us, we need to wait for them
139 	 * to complete
140 	 */
141 	while (br->br_cons_tail != cons_head)
142 		cpu_spinwait();
143 
144 	atomic_store_rel_int(&br->br_cons_tail, cons_next);
145 	critical_exit();
146 
147 	return (buf);
148 }
149 
150 /*
151  * single-consumer dequeue
152  * use where dequeue is protected by a lock
153  * e.g. a network driver's tx queue lock
154  */
155 static __inline void *
buf_ring_dequeue_sc(struct buf_ring * br)156 buf_ring_dequeue_sc(struct buf_ring *br)
157 {
158 	uint32_t cons_head, cons_next, cons_next_next;
159 	uint32_t prod_tail;
160 	void *buf;
161 
162 	cons_head = br->br_cons_head;
163 	prod_tail = br->br_prod_tail;
164 
165 	cons_next = (cons_head + 1) & br->br_cons_mask;
166 	cons_next_next = (cons_head + 2) & br->br_cons_mask;
167 
168 	if (cons_head == prod_tail)
169 		return (NULL);
170 
171 #ifdef PREFETCH_DEFINED
172 	if (cons_next != prod_tail) {
173 		prefetch(br->br_ring[cons_next]);
174 		if (cons_next_next != prod_tail)
175 			prefetch(br->br_ring[cons_next_next]);
176 	}
177 #endif
178 	br->br_cons_head = cons_next;
179 	buf = br->br_ring[cons_head];
180 
181 #ifdef DEBUG_BUFRING
182 	br->br_ring[cons_head] = NULL;
183 	if (!mtx_owned(br->br_lock))
184 		panic("lock not held on single consumer dequeue");
185 	if (br->br_cons_tail != cons_head)
186 		panic("inconsistent list cons_tail=%d cons_head=%d",
187 		    br->br_cons_tail, cons_head);
188 #endif
189 	br->br_cons_tail = cons_next;
190 	return (buf);
191 }
192 
193 /*
194  * single-consumer advance after a peek
195  * use where it is protected by a lock
196  * e.g. a network driver's tx queue lock
197  */
198 static __inline void
buf_ring_advance_sc(struct buf_ring * br)199 buf_ring_advance_sc(struct buf_ring *br)
200 {
201 	uint32_t cons_head, cons_next;
202 	uint32_t prod_tail;
203 
204 	cons_head = br->br_cons_head;
205 	prod_tail = br->br_prod_tail;
206 
207 	cons_next = (cons_head + 1) & br->br_cons_mask;
208 	if (cons_head == prod_tail)
209 		return;
210 	br->br_cons_head = cons_next;
211 #ifdef DEBUG_BUFRING
212 	br->br_ring[cons_head] = NULL;
213 #endif
214 	br->br_cons_tail = cons_next;
215 }
216 
217 /*
218  * Used to return a buffer (most likely already there)
219  * to the top od the ring. The caller should *not*
220  * have used any dequeue to pull it out of the ring
221  * but instead should have used the peek() function.
222  * This is normally used where the transmit queue
223  * of a driver is full, and an mubf must be returned.
224  * Most likely whats in the ring-buffer is what
225  * is being put back (since it was not removed), but
226  * sometimes the lower transmit function may have
227  * done a pullup or other function that will have
228  * changed it. As an optimzation we always put it
229  * back (since jhb says the store is probably cheaper),
230  * if we have to do a multi-queue version we will need
231  * the compare and an atomic.
232  */
233 static __inline void
buf_ring_putback_sc(struct buf_ring * br,void * new)234 buf_ring_putback_sc(struct buf_ring *br, void *new)
235 {
236 	KASSERT(br->br_cons_head != br->br_prod_tail,
237 		("Buf-Ring has none in putback")) ;
238 	br->br_ring[br->br_cons_head] = new;
239 }
240 
241 /*
242  * return a pointer to the first entry in the ring
243  * without modifying it, or NULL if the ring is empty
244  * race-prone if not protected by a lock
245  */
246 static __inline void *
buf_ring_peek(struct buf_ring * br)247 buf_ring_peek(struct buf_ring *br)
248 {
249 
250 #ifdef DEBUG_BUFRING
251 	if ((br->br_lock != NULL) && !mtx_owned(br->br_lock))
252 		panic("lock not held on single consumer dequeue");
253 #endif
254 	/*
255 	 * I believe it is safe to not have a memory barrier
256 	 * here because we control cons and tail is worst case
257 	 * a lagging indicator so we worst case we might
258 	 * return NULL immediately after a buffer has been enqueued
259 	 */
260 	if (br->br_cons_head == br->br_prod_tail)
261 		return (NULL);
262 
263 	return (br->br_ring[br->br_cons_head]);
264 }
265 
266 static __inline int
buf_ring_full(struct buf_ring * br)267 buf_ring_full(struct buf_ring *br)
268 {
269 
270 	return (((br->br_prod_head + 1) & br->br_prod_mask) == br->br_cons_tail);
271 }
272 
273 static __inline int
buf_ring_empty(struct buf_ring * br)274 buf_ring_empty(struct buf_ring *br)
275 {
276 
277 	return (br->br_cons_head == br->br_prod_tail);
278 }
279 
280 static __inline int
buf_ring_count(struct buf_ring * br)281 buf_ring_count(struct buf_ring *br)
282 {
283 
284 	return ((br->br_prod_size + br->br_prod_tail - br->br_cons_tail)
285 	    & br->br_prod_mask);
286 }
287 
288 struct buf_ring *buf_ring_alloc(int count, struct malloc_type *type, int flags,
289     struct mtx *);
290 void buf_ring_free(struct buf_ring *br, struct malloc_type *type);
291 
292 
293 
294 #endif
295