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 44struct 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 uint64_t br_prod_bufs; 51 /* 52 * Pad out to next L2 cache line 53 */ 54 uint64_t _pad0[11]; 55 56 volatile uint32_t br_cons_head; 57 volatile uint32_t br_cons_tail; 58 int br_cons_size; 59 int br_cons_mask; 60 61 /* 62 * Pad out to next L2 cache line 63 */ 64 uint64_t _pad1[14]; 65#ifdef DEBUG_BUFRING 66 struct mtx *br_lock; 67#endif 68 void *br_ring[0]; 69}; 70 71/* 72 * multi-producer safe lock-free ring buffer enqueue 73 * 74 */ 75static __inline int 76buf_ring_enqueue(struct buf_ring *br, void *buf) 77{ 78 uint32_t prod_head, prod_next; 79 uint32_t cons_tail; 80 int success; 81#ifdef DEBUG_BUFRING 82 int i; 83 for (i = br->br_cons_head; i != br->br_prod_head; 84 i = ((i + 1) & br->br_cons_mask)) 85 if(br->br_ring[i] == buf) 86 panic("buf=%p already enqueue at %d prod=%d cons=%d", 87 buf, i, br->br_prod_tail, br->br_cons_tail); 88#endif 89 critical_enter(); 90 do { 91 prod_head = br->br_prod_head; 92 cons_tail = br->br_cons_tail; 93 94 prod_next = (prod_head + 1) & br->br_prod_mask; 95 96 if (prod_next == cons_tail) { 97 br->br_drops++; 98 critical_exit(); 99 return (ENOBUFS); 100 } 101 102 success = atomic_cmpset_int(&br->br_prod_head, prod_head, 103 prod_next); 104 } while (success == 0); 105#ifdef DEBUG_BUFRING 106 if (br->br_ring[prod_head] != NULL) 107 panic("dangling value in enqueue"); 108#endif 109 br->br_ring[prod_head] = buf; 110 wmb(); 111 112 /* 113 * If there are other enqueues in progress 114 * that preceeded us, we need to wait for them 115 * to complete 116 */ 117 while (br->br_prod_tail != prod_head) 118 cpu_spinwait(); 119 br->br_prod_bufs++; 120 br->br_prod_tail = prod_next; 121 critical_exit(); 122 return (0); 123} 124 125/* 126 * multi-consumer safe dequeue 127 * 128 */ 129static __inline void * 130buf_ring_dequeue_mc(struct buf_ring *br) 131{ 132 uint32_t cons_head, cons_next; 133 uint32_t prod_tail; 134 void *buf; 135 int success; 136 137 critical_enter(); 138 do { 139 cons_head = br->br_cons_head; 140 prod_tail = br->br_prod_tail; 141 142 cons_next = (cons_head + 1) & br->br_cons_mask; 143 144 if (cons_head == prod_tail) { 145 critical_exit(); 146 return (NULL); 147 } 148 149 success = atomic_cmpset_int(&br->br_cons_head, cons_head, 150 cons_next); 151 } while (success == 0); 152 153 buf = br->br_ring[cons_head]; 154#ifdef DEBUG_BUFRING 155 br->br_ring[cons_head] = NULL; 156#endif 157 rmb(); 158 159 /* 160 * If there are other dequeues in progress 161 * that preceeded us, we need to wait for them 162 * to complete 163 */ 164 while (br->br_cons_tail != cons_head) 165 cpu_spinwait(); 166 167 br->br_cons_tail = cons_next; 168 critical_exit(); 169 170 return (buf); 171} 172 173/* 174 * single-consumer dequeue 175 * use where dequeue is protected by a lock 176 * e.g. a network driver's tx queue lock 177 */ 178static __inline void * 179buf_ring_dequeue_sc(struct buf_ring *br) 180{ 181 uint32_t cons_head, cons_next, cons_next_next; 182 uint32_t prod_tail; 183 void *buf; 184 185 cons_head = br->br_cons_head; 186 prod_tail = br->br_prod_tail; 187 188 cons_next = (cons_head + 1) & br->br_cons_mask; 189 cons_next_next = (cons_head + 2) & br->br_cons_mask; 190 191 if (cons_head == prod_tail) 192 return (NULL); 193 194#ifdef PREFETCH_DEFINED 195 if (cons_next != prod_tail) { 196 prefetch(br->br_ring[cons_next]); 197 if (cons_next_next != prod_tail) 198 prefetch(br->br_ring[cons_next_next]); 199 } 200#endif 201 br->br_cons_head = cons_next; 202 buf = br->br_ring[cons_head]; 203 204#ifdef DEBUG_BUFRING 205 br->br_ring[cons_head] = NULL; 206 if (!mtx_owned(br->br_lock)) 207 panic("lock not held on single consumer dequeue"); 208 if (br->br_cons_tail != cons_head) 209 panic("inconsistent list cons_tail=%d cons_head=%d", 210 br->br_cons_tail, cons_head); 211#endif 212 br->br_cons_tail = cons_next; 213 return (buf); 214} 215 216/* 217 * single-consumer advance after a peek 218 * use where it is protected by a lock 219 * e.g. a network driver's tx queue lock 220 */ 221static __inline void 222buf_ring_advance_sc(struct buf_ring *br) 223{ 224 uint32_t cons_head, cons_next; 225 uint32_t prod_tail; 226 227 cons_head = br->br_cons_head; 228 prod_tail = br->br_prod_tail; 229 230 cons_next = (cons_head + 1) & br->br_cons_mask; 231 if (cons_head == prod_tail) 232 return; 233 br->br_cons_head = cons_next; 234#ifdef DEBUG_BUFRING 235 br->br_ring[cons_head] = NULL; 236#endif 237 br->br_cons_tail = cons_next; 238} 239 240/* 241 * Used to return a buffer (most likely already there) 242 * to the top od the ring. The caller should *not* 243 * have used any dequeue to pull it out of the ring 244 * but instead should have used the peek() function. 245 * This is normally used where the transmit queue 246 * of a driver is full, and an mubf must be returned. 247 * Most likely whats in the ring-buffer is what 248 * is being put back (since it was not removed), but 249 * sometimes the lower transmit function may have 250 * done a pullup or other function that will have 251 * changed it. As an optimzation we always put it 252 * back (since jhb says the store is probably cheaper), 253 * if we have to do a multi-queue version we will need 254 * the compare and an atomic. 255 */ 256static __inline void 257buf_ring_putback_sc(struct buf_ring *br, void *new) 258{ 259 KASSERT(br->br_cons_head != br->br_prod_tail, 260 ("Buf-Ring has none in putback")) ; 261 br->br_ring[br->br_cons_head] = new; 262} 263 264/* 265 * return a pointer to the first entry in the ring 266 * without modifying it, or NULL if the ring is empty 267 * race-prone if not protected by a lock 268 */ 269static __inline void * 270buf_ring_peek(struct buf_ring *br) 271{ 272 273#ifdef DEBUG_BUFRING 274 if ((br->br_lock != NULL) && !mtx_owned(br->br_lock)) 275 panic("lock not held on single consumer dequeue"); 276#endif 277 /* 278 * I believe it is safe to not have a memory barrier 279 * here because we control cons and tail is worst case 280 * a lagging indicator so we worst case we might 281 * return NULL immediately after a buffer has been enqueued 282 */ 283 if (br->br_cons_head == br->br_prod_tail) 284 return (NULL); 285 286 return (br->br_ring[br->br_cons_head]); 287} 288 289static __inline int 290buf_ring_full(struct buf_ring *br) 291{ 292 293 return (((br->br_prod_head + 1) & br->br_prod_mask) == br->br_cons_tail); 294} 295 296static __inline int 297buf_ring_empty(struct buf_ring *br) 298{ 299 300 return (br->br_cons_head == br->br_prod_tail); 301} 302 303static __inline int 304buf_ring_count(struct buf_ring *br) 305{ 306 307 return ((br->br_prod_size + br->br_prod_tail - br->br_cons_tail) 308 & br->br_prod_mask); 309} 310 311struct buf_ring *buf_ring_alloc(int count, struct malloc_type *type, int flags, 312 struct mtx *); 313void buf_ring_free(struct buf_ring *br, struct malloc_type *type); 314 315 316 317#endif 318