bcache.c revision 119483
1/*- 2 * Copyright (c) 1998 Michael Smith <msmith@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 27#include <sys/cdefs.h> 28__FBSDID("$FreeBSD: head/sys/boot/common/bcache.c 119483 2003-08-25 23:30:41Z obrien $"); 29 30/* 31 * Simple LRU block cache 32 */ 33 34#include <stand.h> 35#include <string.h> 36#include <bitstring.h> 37 38#include "bootstrap.h" 39 40/* #define BCACHE_DEBUG */ 41 42#ifdef BCACHE_DEBUG 43#define BCACHE_TIMEOUT 10 44# define DEBUG(fmt, args...) printf("%s: " fmt "\n" , __func__ , ## args) 45#else 46#define BCACHE_TIMEOUT 2 47# define DEBUG(fmt, args...) 48#endif 49 50 51struct bcachectl 52{ 53 daddr_t bc_blkno; 54 time_t bc_stamp; 55 int bc_count; 56}; 57 58static struct bcachectl *bcache_ctl; 59static caddr_t bcache_data; 60static bitstr_t *bcache_miss; 61static u_int bcache_nblks; 62static u_int bcache_blksize; 63static u_int bcache_hits, bcache_misses, bcache_ops, bcache_bypasses; 64static u_int bcache_flushes; 65static u_int bcache_bcount; 66 67static void bcache_invalidate(daddr_t blkno); 68static void bcache_insert(caddr_t buf, daddr_t blkno); 69static int bcache_lookup(caddr_t buf, daddr_t blkno); 70 71/* 72 * Initialise the cache for (nblks) of (bsize). 73 */ 74int 75bcache_init(u_int nblks, size_t bsize) 76{ 77 /* discard any old contents */ 78 if (bcache_data != NULL) { 79 free(bcache_data); 80 bcache_data = NULL; 81 free(bcache_ctl); 82 } 83 84 /* Allocate control structures */ 85 bcache_nblks = nblks; 86 bcache_blksize = bsize; 87 bcache_data = malloc(bcache_nblks * bcache_blksize); 88 bcache_ctl = (struct bcachectl *)malloc(bcache_nblks * sizeof(struct bcachectl)); 89 bcache_miss = bit_alloc((bcache_nblks + 1) / 2); 90 if ((bcache_data == NULL) || (bcache_ctl == NULL) || (bcache_miss == NULL)) { 91 if (bcache_miss) 92 free(bcache_miss); 93 if (bcache_ctl) 94 free(bcache_ctl); 95 if (bcache_data) 96 free(bcache_data); 97 bcache_data = NULL; 98 return(ENOMEM); 99 } 100 101 return(0); 102} 103 104/* 105 * Flush the cache 106 */ 107void 108bcache_flush(void) 109{ 110 u_int i; 111 112 bcache_flushes++; 113 114 /* Flush the cache */ 115 for (i = 0; i < bcache_nblks; i++) { 116 bcache_ctl[i].bc_count = -1; 117 bcache_ctl[i].bc_blkno = -1; 118 } 119} 120 121/* 122 * Handle a write request; write directly to the disk, and populate the 123 * cache with the new values. 124 */ 125static int 126write_strategy(void *devdata, int unit, int rw, daddr_t blk, size_t size, 127 char *buf, size_t *rsize) 128{ 129 struct bcache_devdata *dd = (struct bcache_devdata *)devdata; 130 daddr_t i, nblk; 131 int err; 132 133 nblk = size / bcache_blksize; 134 135 /* Invalidate the blocks being written */ 136 for (i = 0; i < nblk; i++) { 137 bcache_invalidate(blk + i); 138 } 139 140 /* Write the blocks */ 141 err = dd->dv_strategy(dd->dv_devdata, rw, blk, size, buf, rsize); 142 143 /* Populate the block cache with the new data */ 144 if (err == 0) { 145 for (i = 0; i < nblk; i++) { 146 bcache_insert(buf + (i * bcache_blksize),blk + i); 147 } 148 } 149 150 return err; 151} 152 153/* 154 * Handle a read request; fill in parts of the request that can 155 * be satisfied by the cache, use the supplied strategy routine to do 156 * device I/O and then use the I/O results to populate the cache. 157 */ 158static int 159read_strategy(void *devdata, int unit, int rw, daddr_t blk, size_t size, 160 char *buf, size_t *rsize) 161{ 162 struct bcache_devdata *dd = (struct bcache_devdata *)devdata; 163 int p_size, result; 164 daddr_t p_blk, i, j, nblk; 165 caddr_t p_buf; 166 167 nblk = size / bcache_blksize; 168 result = 0; 169 170 /* Satisfy any cache hits up front */ 171 for (i = 0; i < nblk; i++) { 172 if (bcache_lookup(buf + (bcache_blksize * i), blk + i)) { 173 bit_set(bcache_miss, i); /* cache miss */ 174 bcache_misses++; 175 } else { 176 bit_clear(bcache_miss, i); /* cache hit */ 177 bcache_hits++; 178 } 179 } 180 181 /* Go back and fill in any misses XXX optimise */ 182 p_blk = -1; 183 p_buf = NULL; 184 p_size = 0; 185 for (i = 0; i < nblk; i++) { 186 if (bit_test(bcache_miss, i)) { 187 /* miss, add to pending transfer */ 188 if (p_blk == -1) { 189 p_blk = blk + i; 190 p_buf = buf + (bcache_blksize * i); 191 p_size = 1; 192 } else { 193 p_size++; 194 } 195 } else if (p_blk != -1) { 196 /* hit, complete pending transfer */ 197 result = dd->dv_strategy(dd->dv_devdata, rw, p_blk, p_size * bcache_blksize, p_buf, NULL); 198 if (result != 0) 199 goto done; 200 for (j = 0; j < p_size; j++) 201 bcache_insert(p_buf + (j * bcache_blksize), p_blk + j); 202 p_blk = -1; 203 } 204 } 205 if (p_blk != -1) { 206 /* pending transfer left */ 207 result = dd->dv_strategy(dd->dv_devdata, rw, p_blk, p_size * bcache_blksize, p_buf, NULL); 208 if (result != 0) 209 goto done; 210 for (j = 0; j < p_size; j++) 211 bcache_insert(p_buf + (j * bcache_blksize), p_blk + j); 212 } 213 214 done: 215 if ((result == 0) && (rsize != NULL)) 216 *rsize = size; 217 return(result); 218} 219 220/* 221 * Requests larger than 1/2 the cache size will be bypassed and go 222 * directly to the disk. XXX tune this. 223 */ 224int 225bcache_strategy(void *devdata, int unit, int rw, daddr_t blk, size_t size, 226 char *buf, size_t *rsize) 227{ 228 static int bcache_unit = -1; 229 struct bcache_devdata *dd = (struct bcache_devdata *)devdata; 230 231 bcache_ops++; 232 233 if(bcache_unit != unit) { 234 bcache_flush(); 235 bcache_unit = unit; 236 } 237 238 /* bypass large requests, or when the cache is inactive */ 239 if ((bcache_data == NULL) || ((size * 2 / bcache_blksize) > bcache_nblks)) { 240 DEBUG("bypass %d from %d", size / bcache_blksize, blk); 241 bcache_bypasses++; 242 return(dd->dv_strategy(dd->dv_devdata, rw, blk, size, buf, rsize)); 243 } 244 245 switch (rw) { 246 case F_READ: 247 return read_strategy(devdata, unit, rw, blk, size, buf, rsize); 248 case F_WRITE: 249 return write_strategy(devdata, unit, rw, blk, size, buf, rsize); 250 } 251 return -1; 252} 253 254 255/* 256 * Insert a block into the cache. Retire the oldest block to do so, if required. 257 * 258 * XXX the LRU algorithm will fail after 2^31 blocks have been transferred. 259 */ 260static void 261bcache_insert(caddr_t buf, daddr_t blkno) 262{ 263 time_t now; 264 int cand, ocount; 265 u_int i; 266 267 time(&now); 268 cand = 0; /* assume the first block */ 269 ocount = bcache_ctl[0].bc_count; 270 271 /* find the oldest block */ 272 for (i = 1; i < bcache_nblks; i++) { 273 if (bcache_ctl[i].bc_blkno == blkno) { 274 /* reuse old entry */ 275 cand = i; 276 break; 277 } 278 if (bcache_ctl[i].bc_count < ocount) { 279 ocount = bcache_ctl[i].bc_count; 280 cand = i; 281 } 282 } 283 284 DEBUG("insert blk %d -> %d @ %d # %d", blkno, cand, now, bcache_bcount); 285 bcopy(buf, bcache_data + (bcache_blksize * cand), bcache_blksize); 286 bcache_ctl[cand].bc_blkno = blkno; 287 bcache_ctl[cand].bc_stamp = now; 288 bcache_ctl[cand].bc_count = bcache_bcount++; 289} 290 291/* 292 * Look for a block in the cache. Blocks more than BCACHE_TIMEOUT seconds old 293 * may be stale (removable media) and thus are discarded. Copy the block out 294 * if successful and return zero, or return nonzero on failure. 295 */ 296static int 297bcache_lookup(caddr_t buf, daddr_t blkno) 298{ 299 time_t now; 300 u_int i; 301 302 time(&now); 303 304 for (i = 0; i < bcache_nblks; i++) 305 /* cache hit? */ 306 if ((bcache_ctl[i].bc_blkno == blkno) && ((bcache_ctl[i].bc_stamp + BCACHE_TIMEOUT) >= now)) { 307 bcopy(bcache_data + (bcache_blksize * i), buf, bcache_blksize); 308 DEBUG("hit blk %d <- %d (now %d then %d)", blkno, i, now, bcache_ctl[i].bc_stamp); 309 return(0); 310 } 311 return(ENOENT); 312} 313 314/* 315 * Invalidate a block from the cache. 316 */ 317static void 318bcache_invalidate(daddr_t blkno) 319{ 320 u_int i; 321 322 for (i = 0; i < bcache_nblks; i++) { 323 if (bcache_ctl[i].bc_blkno == blkno) { 324 bcache_ctl[i].bc_count = -1; 325 bcache_ctl[i].bc_blkno = -1; 326 DEBUG("invalidate blk %d", blkno); 327 break; 328 } 329 } 330} 331 332COMMAND_SET(bcachestat, "bcachestat", "get disk block cache stats", command_bcache); 333 334static int 335command_bcache(int argc, char *argv[]) 336{ 337 u_int i; 338 339 for (i = 0; i < bcache_nblks; i++) { 340 printf("%08x %04x %04x|", bcache_ctl[i].bc_blkno, (unsigned int)bcache_ctl[i].bc_stamp & 0xffff, bcache_ctl[i].bc_count & 0xffff); 341 if (((i + 1) % 4) == 0) 342 printf("\n"); 343 } 344 printf("\n%d ops %d bypasses %d hits %d misses %d flushes\n", bcache_ops, bcache_bypasses, bcache_hits, bcache_misses, bcache_flushes); 345 return(CMD_OK); 346} 347 348