2 * Copyright (c) 1998 Michael Smith <msmith@freebsd.org>
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
31 * Simple LRU block cache
34 #include <sys/stdint.h>
38 #include <bitstring.h>
40 #include "bootstrap.h"
42 /* #define BCACHE_DEBUG */
45 #define BCACHE_TIMEOUT 10
46 # define DEBUG(fmt, args...) printf("%s: " fmt "\n" , __func__ , ## args)
48 #define BCACHE_TIMEOUT 2
49 # define DEBUG(fmt, args...)
60 static struct bcachectl *bcache_ctl;
61 static caddr_t bcache_data;
62 static bitstr_t *bcache_miss;
63 static u_int bcache_nblks;
64 static u_int bcache_blksize;
65 static u_int bcache_hits, bcache_misses, bcache_ops, bcache_bypasses;
66 static u_int bcache_flushes;
67 static u_int bcache_bcount;
69 static void bcache_invalidate(daddr_t blkno);
70 static void bcache_insert(caddr_t buf, daddr_t blkno);
71 static int bcache_lookup(caddr_t buf, daddr_t blkno);
74 * Initialise the cache for (nblks) of (bsize).
77 bcache_init(u_int nblks, size_t bsize)
79 /* discard any old contents */
80 if (bcache_data != NULL) {
86 /* Allocate control structures */
88 bcache_blksize = bsize;
89 bcache_data = malloc(bcache_nblks * bcache_blksize);
90 bcache_ctl = (struct bcachectl *)malloc(bcache_nblks * sizeof(struct bcachectl));
91 bcache_miss = bit_alloc((bcache_nblks + 1) / 2);
92 if ((bcache_data == NULL) || (bcache_ctl == NULL) || (bcache_miss == NULL)) {
116 /* Flush the cache */
117 for (i = 0; i < bcache_nblks; i++) {
118 bcache_ctl[i].bc_count = -1;
119 bcache_ctl[i].bc_blkno = -1;
124 * Handle a write request; write directly to the disk, and populate the
125 * cache with the new values.
128 write_strategy(void *devdata, int unit, int rw, daddr_t blk, size_t size,
129 char *buf, size_t *rsize)
131 struct bcache_devdata *dd = (struct bcache_devdata *)devdata;
135 nblk = size / bcache_blksize;
137 /* Invalidate the blocks being written */
138 for (i = 0; i < nblk; i++) {
139 bcache_invalidate(blk + i);
142 /* Write the blocks */
143 err = dd->dv_strategy(dd->dv_devdata, rw, blk, size, buf, rsize);
145 /* Populate the block cache with the new data */
147 for (i = 0; i < nblk; i++) {
148 bcache_insert(buf + (i * bcache_blksize),blk + i);
156 * Handle a read request; fill in parts of the request that can
157 * be satisfied by the cache, use the supplied strategy routine to do
158 * device I/O and then use the I/O results to populate the cache.
161 read_strategy(void *devdata, int unit, int rw, daddr_t blk, size_t size,
162 char *buf, size_t *rsize)
164 struct bcache_devdata *dd = (struct bcache_devdata *)devdata;
166 daddr_t p_blk, i, j, nblk;
169 nblk = size / bcache_blksize;
172 /* Satisfy any cache hits up front */
173 for (i = 0; i < nblk; i++) {
174 if (bcache_lookup(buf + (bcache_blksize * i), blk + i)) {
175 bit_set(bcache_miss, i); /* cache miss */
178 bit_clear(bcache_miss, i); /* cache hit */
183 /* Go back and fill in any misses XXX optimise */
187 for (i = 0; i < nblk; i++) {
188 if (bit_test(bcache_miss, i)) {
189 /* miss, add to pending transfer */
192 p_buf = buf + (bcache_blksize * i);
197 } else if (p_blk != -1) {
198 /* hit, complete pending transfer */
199 result = dd->dv_strategy(dd->dv_devdata, rw, p_blk, p_size * bcache_blksize, p_buf, NULL);
202 for (j = 0; j < p_size; j++)
203 bcache_insert(p_buf + (j * bcache_blksize), p_blk + j);
208 /* pending transfer left */
209 result = dd->dv_strategy(dd->dv_devdata, rw, p_blk, p_size * bcache_blksize, p_buf, NULL);
212 for (j = 0; j < p_size; j++)
213 bcache_insert(p_buf + (j * bcache_blksize), p_blk + j);
217 if ((result == 0) && (rsize != NULL))
223 * Requests larger than 1/2 the cache size will be bypassed and go
224 * directly to the disk. XXX tune this.
227 bcache_strategy(void *devdata, int unit, int rw, daddr_t blk, size_t size,
228 char *buf, size_t *rsize)
230 static int bcache_unit = -1;
231 struct bcache_devdata *dd = (struct bcache_devdata *)devdata;
235 if(bcache_unit != unit) {
240 /* bypass large requests, or when the cache is inactive */
241 if ((bcache_data == NULL) || ((size * 2 / bcache_blksize) > bcache_nblks)) {
242 DEBUG("bypass %d from %d", size / bcache_blksize, blk);
244 return(dd->dv_strategy(dd->dv_devdata, rw, blk, size, buf, rsize));
249 return read_strategy(devdata, unit, rw, blk, size, buf, rsize);
251 return write_strategy(devdata, unit, rw, blk, size, buf, rsize);
258 * Insert a block into the cache. Retire the oldest block to do so, if required.
260 * XXX the LRU algorithm will fail after 2^31 blocks have been transferred.
263 bcache_insert(caddr_t buf, daddr_t blkno)
270 cand = 0; /* assume the first block */
271 ocount = bcache_ctl[0].bc_count;
273 /* find the oldest block */
274 for (i = 1; i < bcache_nblks; i++) {
275 if (bcache_ctl[i].bc_blkno == blkno) {
276 /* reuse old entry */
280 if (bcache_ctl[i].bc_count < ocount) {
281 ocount = bcache_ctl[i].bc_count;
286 DEBUG("insert blk %d -> %d @ %d # %d", blkno, cand, now, bcache_bcount);
287 bcopy(buf, bcache_data + (bcache_blksize * cand), bcache_blksize);
288 bcache_ctl[cand].bc_blkno = blkno;
289 bcache_ctl[cand].bc_stamp = now;
290 bcache_ctl[cand].bc_count = bcache_bcount++;
294 * Look for a block in the cache. Blocks more than BCACHE_TIMEOUT seconds old
295 * may be stale (removable media) and thus are discarded. Copy the block out
296 * if successful and return zero, or return nonzero on failure.
299 bcache_lookup(caddr_t buf, daddr_t blkno)
306 for (i = 0; i < bcache_nblks; i++)
308 if ((bcache_ctl[i].bc_blkno == blkno) && ((bcache_ctl[i].bc_stamp + BCACHE_TIMEOUT) >= now)) {
309 bcopy(bcache_data + (bcache_blksize * i), buf, bcache_blksize);
310 DEBUG("hit blk %d <- %d (now %d then %d)", blkno, i, now, bcache_ctl[i].bc_stamp);
317 * Invalidate a block from the cache.
320 bcache_invalidate(daddr_t blkno)
324 for (i = 0; i < bcache_nblks; i++) {
325 if (bcache_ctl[i].bc_blkno == blkno) {
326 bcache_ctl[i].bc_count = -1;
327 bcache_ctl[i].bc_blkno = -1;
328 DEBUG("invalidate blk %d", blkno);
334 COMMAND_SET(bcachestat, "bcachestat", "get disk block cache stats", command_bcache);
337 command_bcache(int argc, char *argv[])
341 for (i = 0; i < bcache_nblks; i++) {
342 printf("%08jx %04x %04x|", (uintmax_t)bcache_ctl[i].bc_blkno, (unsigned int)bcache_ctl[i].bc_stamp & 0xffff, bcache_ctl[i].bc_count & 0xffff);
343 if (((i + 1) % 4) == 0)
346 printf("\n%d ops %d bypasses %d hits %d misses %d flushes\n", bcache_ops, bcache_bypasses, bcache_hits, bcache_misses, bcache_flushes);