1 /* $NetBSD: bufq_readprio.c,v 1.16 2017/05/04 11:03:27 kamil Exp $ */
2 /* NetBSD: subr_disk.c,v 1.61 2004/09/25 03:30:44 thorpej Exp */
3
4 /*-
5 * Copyright (c) 1996, 1997, 1999, 2000 The NetBSD Foundation, Inc.
6 * All rights reserved.
7 *
8 * This code is derived from software contributed to The NetBSD Foundation
9 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
10 * NASA Ames Research Center.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
23 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
24 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
25 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31 * POSSIBILITY OF SUCH DAMAGE.
32 */
33
34 /*
35 * Copyright (c) 1982, 1986, 1988, 1993
36 * The Regents of the University of California. All rights reserved.
37 * (c) UNIX System Laboratories, Inc.
38 * All or some portions of this file are derived from material licensed
39 * to the University of California by American Telephone and Telegraph
40 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
41 * the permission of UNIX System Laboratories, Inc.
42 *
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
45 * are met:
46 * 1. Redistributions of source code must retain the above copyright
47 * notice, this list of conditions and the following disclaimer.
48 * 2. Redistributions in binary form must reproduce the above copyright
49 * notice, this list of conditions and the following disclaimer in the
50 * documentation and/or other materials provided with the distribution.
51 * 3. Neither the name of the University nor the names of its contributors
52 * may be used to endorse or promote products derived from this software
53 * without specific prior written permission.
54 *
55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65 * SUCH DAMAGE.
66 *
67 * @(#)ufs_disksubr.c 8.5 (Berkeley) 1/21/94
68 */
69
70 #include <sys/cdefs.h>
71 __KERNEL_RCSID(0, "$NetBSD: bufq_readprio.c,v 1.16 2017/05/04 11:03:27 kamil Exp $");
72
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/buf.h>
76 #include <sys/bufq.h>
77 #include <sys/bufq_impl.h>
78 #include <sys/kmem.h>
79 #include <sys/module.h>
80
81 /*
82 * Seek sort for disks.
83 *
84 * There are two queues. The first queue holds read requests; the second
85 * holds write requests. The read queue is first-come first-served; the
86 * write queue is sorted in ascendening block order.
87 * The read queue is processed first. After PRIO_READ_BURST consecutive
88 * read requests with non-empty write queue PRIO_WRITE_REQ requests from
89 * the write queue will be processed.
90 */
91
92 #define PRIO_READ_BURST 48
93 #define PRIO_WRITE_REQ 16
94
95 struct bufq_prio {
96 TAILQ_HEAD(, buf) bq_read, bq_write; /* actual list of buffers */
97 struct buf *bq_write_next; /* next request in bq_write */
98 struct buf *bq_next; /* current request */
99 int bq_read_burst; /* # of consecutive reads */
100 };
101
102 static void bufq_readprio_init(struct bufq_state *);
103 static void bufq_prio_put(struct bufq_state *, struct buf *);
104 static struct buf *bufq_prio_get(struct bufq_state *, int);
105
106 BUFQ_DEFINE(readprio, 30, bufq_readprio_init);
107
108 static void
bufq_prio_put(struct bufq_state * bufq,struct buf * bp)109 bufq_prio_put(struct bufq_state *bufq, struct buf *bp)
110 {
111 struct bufq_prio *prio = bufq_private(bufq);
112 struct buf *bq;
113 int sortby;
114
115 sortby = bufq->bq_flags & BUFQ_SORT_MASK;
116
117 /*
118 * If it's a read request append it to the list.
119 */
120 if ((bp->b_flags & B_READ) == B_READ) {
121 TAILQ_INSERT_TAIL(&prio->bq_read, bp, b_actq);
122 return;
123 }
124
125 bq = TAILQ_FIRST(&prio->bq_write);
126
127 /*
128 * If the write list is empty, simply append it to the list.
129 */
130 if (bq == NULL) {
131 TAILQ_INSERT_TAIL(&prio->bq_write, bp, b_actq);
132 prio->bq_write_next = bp;
133 return;
134 }
135
136 /*
137 * If we lie after the next request, insert after this request.
138 */
139 if (buf_inorder(prio->bq_write_next, bp, sortby))
140 bq = prio->bq_write_next;
141
142 /*
143 * Search for the first request at a larger block number.
144 * We go before this request if it exists.
145 */
146 while (bq != NULL && buf_inorder(bq, bp, sortby))
147 bq = TAILQ_NEXT(bq, b_actq);
148
149 if (bq != NULL)
150 TAILQ_INSERT_BEFORE(bq, bp, b_actq);
151 else
152 TAILQ_INSERT_TAIL(&prio->bq_write, bp, b_actq);
153 }
154
155 static struct buf *
bufq_prio_get(struct bufq_state * bufq,int remove)156 bufq_prio_get(struct bufq_state *bufq, int remove)
157 {
158 struct bufq_prio *prio = bufq_private(bufq);
159 struct buf *bp;
160
161 /*
162 * If no current request, get next from the lists.
163 */
164 if (prio->bq_next == NULL) {
165 /*
166 * If at least one list is empty, select the other.
167 */
168 if (TAILQ_FIRST(&prio->bq_read) == NULL) {
169 prio->bq_next = prio->bq_write_next;
170 prio->bq_read_burst = 0;
171 } else if (prio->bq_write_next == NULL) {
172 bp = prio->bq_next = TAILQ_FIRST(&prio->bq_read);
173 prio->bq_read_burst = 0;
174 KASSERT((bp == NULL) ||
175 ((bp->b_flags & B_READ) == B_READ));
176 } else {
177 /*
178 * Both list have requests. Select the read list up
179 * to PRIO_READ_BURST times, then select the write
180 * list PRIO_WRITE_REQ times.
181 */
182 if (prio->bq_read_burst++ < PRIO_READ_BURST)
183 prio->bq_next = TAILQ_FIRST(&prio->bq_read);
184 else if (prio->bq_read_burst <
185 PRIO_READ_BURST + PRIO_WRITE_REQ)
186 prio->bq_next = prio->bq_write_next;
187 else {
188 prio->bq_next = TAILQ_FIRST(&prio->bq_read);
189 prio->bq_read_burst = 0;
190 }
191 }
192 }
193
194 bp = prio->bq_next;
195
196 if (bp != NULL && remove) {
197 if ((bp->b_flags & B_READ) == B_READ)
198 TAILQ_REMOVE(&prio->bq_read, bp, b_actq);
199 else {
200 /*
201 * Advance the write pointer before removing
202 * bp since it is actually prio->bq_write_next.
203 */
204 prio->bq_write_next =
205 TAILQ_NEXT(prio->bq_write_next, b_actq);
206 TAILQ_REMOVE(&prio->bq_write, bp, b_actq);
207 if (prio->bq_write_next == NULL)
208 prio->bq_write_next =
209 TAILQ_FIRST(&prio->bq_write);
210 }
211
212 prio->bq_next = NULL;
213 }
214
215 return (bp);
216 }
217
218 static struct buf *
bufq_prio_cancel(struct bufq_state * bufq,struct buf * buf)219 bufq_prio_cancel(struct bufq_state *bufq, struct buf *buf)
220 {
221 struct bufq_prio *prio = bufq_private(bufq);
222 struct buf *bq;
223
224 /* search read queue */
225 TAILQ_FOREACH(bq, &prio->bq_read, b_actq) {
226 if (bq == buf) {
227 TAILQ_REMOVE(&prio->bq_read, bq, b_actq);
228 /* force new section */
229 prio->bq_next = NULL;
230 return buf;
231 }
232 }
233
234 /* not found in read queue, search write queue */
235 TAILQ_FOREACH(bq, &prio->bq_write, b_actq) {
236 if (bq == buf) {
237 if (bq == prio->bq_write_next) {
238 /*
239 * Advance the write pointer before removing
240 * bp since it is actually prio->bq_write_next.
241 */
242 prio->bq_write_next =
243 TAILQ_NEXT(prio->bq_write_next, b_actq);
244 TAILQ_REMOVE(&prio->bq_write, bq, b_actq);
245 if (prio->bq_write_next == NULL)
246 prio->bq_write_next =
247 TAILQ_FIRST(&prio->bq_write);
248 } else {
249 TAILQ_REMOVE(&prio->bq_write, bq, b_actq);
250 }
251 /* force new section */
252 prio->bq_next = NULL;
253 return buf;
254 }
255 }
256
257 /* still not found */
258 return NULL;
259 }
260
261 static void
bufq_prio_fini(struct bufq_state * bufq)262 bufq_prio_fini(struct bufq_state *bufq)
263 {
264
265 KASSERT(bufq->bq_private != NULL);
266 kmem_free(bufq->bq_private, sizeof(struct bufq_prio));
267 }
268
269 static void
bufq_readprio_init(struct bufq_state * bufq)270 bufq_readprio_init(struct bufq_state *bufq)
271 {
272 struct bufq_prio *prio;
273
274 bufq->bq_get = bufq_prio_get;
275 bufq->bq_put = bufq_prio_put;
276 bufq->bq_cancel = bufq_prio_cancel;
277 bufq->bq_fini = bufq_prio_fini;
278 bufq->bq_private = kmem_zalloc(sizeof(struct bufq_prio), KM_SLEEP);
279 prio = (struct bufq_prio *)bufq->bq_private;
280 TAILQ_INIT(&prio->bq_read);
281 TAILQ_INIT(&prio->bq_write);
282 }
283
284 MODULE(MODULE_CLASS_BUFQ, bufq_readprio, NULL);
285
286 static int
bufq_readprio_modcmd(modcmd_t cmd,void * opaque)287 bufq_readprio_modcmd(modcmd_t cmd, void *opaque)
288 {
289
290 switch (cmd) {
291 case MODULE_CMD_INIT:
292 return bufq_register(&bufq_strat_readprio);
293 case MODULE_CMD_FINI:
294 return bufq_unregister(&bufq_strat_readprio);
295 default:
296 return ENOTTY;
297 }
298 }
299