sys/kern/dsched/fq/fq_core.c

   1 /*
   2  * Copyright (c) 2009, 2010 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Alex Hornung <ahornung@gmail.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  */
  34 #include <sys/param.h>
  35 #include <sys/systm.h>
  36 #include <sys/kernel.h>
  37 #include <sys/proc.h>
  38 #include <sys/sysctl.h>
  39 #include <sys/buf.h>
  40 #include <sys/conf.h>
  41 #include <sys/diskslice.h>
  42 #include <sys/disk.h>
  43 #include <machine/atomic.h>
  44 #include <sys/thread.h>
  45 #include <sys/thread2.h>
  46 #include <sys/ctype.h>
  47 #include <sys/buf2.h>
  48 #include <sys/syslog.h>
  49 #include <sys/dsched.h>
  50 #include <machine/param.h>
  51
  52 #include <kern/dsched/fq/fq.h>
  53
  54 static int      dsched_fq_version_maj = 1;
  55 static int      dsched_fq_version_min = 1;
  56
  57 /* Make sure our structs fit */
  58 CTASSERT(sizeof(struct fq_thread_io) <= DSCHED_THREAD_IO_MAX_SZ);
  59 CTASSERT(sizeof(struct fq_disk_ctx) <= DSCHED_DISK_CTX_MAX_SZ);
  60
  61 struct dsched_fq_stats  fq_stats;
  62
  63 extern struct dsched_policy dsched_fq_policy;
  64
  65 void
  66 fq_dispatcher(struct fq_disk_ctx *diskctx)
  67 {
  68         struct dispatch_prep *dispatch_ary;
  69         struct dsched_thread_io *ds_tdio, *ds_tdio2;
  70         struct fq_thread_io     *tdio;
  71         struct bio *bio, *bio2;
  72         int idle;
  73         int i, prepd_io;
  74
  75         /*
  76          * Array is dangerously big for an on-stack declaration, allocate
  77          * it instead.
  78          */
  79         dispatch_ary = kmalloc(sizeof(*dispatch_ary) * FQ_DISPATCH_ARRAY_SZ,
  80                                M_TEMP, M_INTWAIT | M_ZERO);
  81
  82         /*
  83          * We need to manually assign an tdio to the tdctx of this thread
  84          * since it isn't assigned one during fq_prepare, as the disk
  85          * is not set up yet.
  86          */
  87         tdio = (struct fq_thread_io *)dsched_new_policy_thread_tdio(&diskctx->head,
  88             &dsched_fq_policy);
  89
  90         DSCHED_DISK_CTX_LOCK(&diskctx->head);
  91         for(;;) {
  92                 idle = 0;
  93                 /*
  94                  * sleep ~60 ms, failsafe low hz rates.
  95                  */
  96                 if ((lksleep(diskctx, &diskctx->head.lock, 0,
  97                              "fq_dispatcher", (hz + 14) / 15) == 0)) {
  98                         /*
  99                          * We've been woken up; this either means that we are
 100                          * supposed to die away nicely or that the disk is idle.
 101                          */
 102
 103                         if (__predict_false(diskctx->die == 1))
 104                                 break;
 105
 106                         /*
 107                          * We have been awakened because the disk is idle.
 108                          * So let's get ready to dispatch some extra bios.
 109                          */
 110                         idle = 1;
 111                 }
 112
 113                 /* Maybe the disk is idle and we just didn't get the wakeup */
 114                 if (idle == 0)
 115                         idle = diskctx->idle;
 116
 117                 /* Set the number of prepared requests to 0 */
 118                 i = 0;
 119
 120                 /*
 121                  * XXX: further room for improvements here. It would be better
 122                  *      to dispatch a few requests from each tdio as to ensure
 123                  *      real fairness.
 124                  */
 125                 TAILQ_FOREACH_MUTABLE(ds_tdio, &diskctx->head.tdio_list, dlink, ds_tdio2) {
 126                         tdio = (struct fq_thread_io *)ds_tdio;
 127                         if (tdio->head.qlength == 0)
 128                                 continue;
 129
 130                         DSCHED_THREAD_IO_LOCK(&tdio->head);
 131                         if (atomic_cmpset_int(&tdio->rebalance, 1, 0))
 132                                 fq_balance_self(tdio);
 133                         /*
 134                          * XXX: why 5 extra? should probably be dynamic,
 135                          *      relying on information on latency.
 136                          */
 137                         if ((tdio->max_tp > 0) && idle &&
 138                             (tdio->issued >= tdio->max_tp)) {
 139                                 tdio->max_tp += 5;
 140                         }
 141
 142                         prepd_io = 0;
 143                         TAILQ_FOREACH_MUTABLE(bio, &tdio->head.queue, link, bio2) {
 144                                 if (atomic_cmpset_int(&tdio->rebalance, 1, 0))
 145                                         fq_balance_self(tdio);
 146                                 if (((tdio->max_tp > 0) &&
 147                                     (tdio->issued + prepd_io >= tdio->max_tp)) ||
 148                                     (i == FQ_DISPATCH_ARRAY_SZ))
 149                                         break;
 150
 151                                 TAILQ_REMOVE(&tdio->head.queue, bio, link);
 152                                 --tdio->head.qlength;
 153
 154                                 /*
 155                                  * beware that we do have an tdio reference
 156                                  * from the queueing
 157                                  *
 158                                  * XXX: note that here we don't dispatch it yet
 159                                  *      but just prepare it for dispatch so
 160                                  *      that no locks are held when calling
 161                                  *      into the drivers.
 162                                  */
 163                                 dispatch_ary[i].bio = bio;
 164                                 dispatch_ary[i].tdio = tdio;
 165                                 ++i;
 166                                 ++prepd_io;
 167                         }
 168                         DSCHED_THREAD_IO_UNLOCK(&tdio->head);
 169
 170                 }
 171
 172                 dsched_disk_ctx_ref(&diskctx->head);
 173                 DSCHED_DISK_CTX_UNLOCK(&diskctx->head);
 174
 175                 /*
 176                  * Dispatch all the previously prepared bios, now without
 177                  * holding any locks.
 178                  */
 179                 for (--i; i >= 0; i--) {
 180                         bio = dispatch_ary[i].bio;
 181                         tdio = dispatch_ary[i].tdio;
 182                         fq_dispatch(diskctx, bio, tdio);
 183                 }
 184
 185                 DSCHED_DISK_CTX_LOCK(&diskctx->head);
 186                 dsched_disk_ctx_unref(&diskctx->head);
 187         }
 188
 189         /*
 190          * If we are supposed to die, drain all queues, then
 191          * unlock and exit.
 192          */
 193         fq_drain(diskctx, FQ_DRAIN_FLUSH);
 194         DSCHED_DISK_CTX_UNLOCK(&diskctx->head);
 195         kfree(dispatch_ary, M_TEMP);
 196
 197         kprintf("fq_dispatcher is peacefully dying\n");
 198         lwkt_exit();
 199         /* NOTREACHED */
 200 }
 201
 202 void
 203 fq_balance_thread(struct fq_disk_ctx *diskctx)
 204 {
 205         struct dsched_thread_io *ds_tdio;
 206         struct  fq_thread_io    *tdio;
 207         struct timeval tv, old_tv;
 208         int64_t total_budget, product;
 209         int64_t budget[FQ_PRIO_MAX+1];
 210         int     n, i, sum, total_disk_time;
 211         int     lost_bits;
 212
 213         DSCHED_DISK_CTX_LOCK(&diskctx->head);
 214
 215         getmicrotime(&diskctx->start_interval);
 216
 217         for (;;) {
 218                 /* sleep ~1s */
 219                 if ((lksleep(curthread, &diskctx->head.lock, 0, "fq_balancer", hz/2) == 0)) {
 220                         if (__predict_false(diskctx->die)) {
 221                                 DSCHED_DISK_CTX_UNLOCK(&diskctx->head);
 222                                 lwkt_exit();
 223                         }
 224                 }
 225
 226                 bzero(budget, sizeof(budget));
 227                 total_budget = 0;
 228                 n = 0;
 229
 230                 old_tv = diskctx->start_interval;
 231                 getmicrotime(&tv);
 232
 233                 total_disk_time = (int)(1000000*((tv.tv_sec - old_tv.tv_sec)) +
 234                     (tv.tv_usec - old_tv.tv_usec));
 235
 236                 if (total_disk_time == 0)
 237                         total_disk_time = 1;
 238
 239                 dsched_debug(LOG_INFO, "total_disk_time = %d\n", total_disk_time);
 240
 241                 diskctx->start_interval = tv;
 242
 243                 diskctx->disk_busy = (100*(total_disk_time - diskctx->idle_time)) / total_disk_time;
 244                 if (diskctx->disk_busy < 0)
 245                         diskctx->disk_busy = 0;
 246
 247                 diskctx->idle_time = 0;
 248                 lost_bits = 0;
 249
 250                 TAILQ_FOREACH(ds_tdio, &diskctx->head.tdio_list, dlink) {
 251                         tdio = (struct fq_thread_io *)ds_tdio;
 252                         tdio->interval_avg_latency = tdio->avg_latency;
 253                         tdio->interval_transactions = tdio->transactions;
 254                         if (tdio->interval_transactions > 0) {
 255                                 product = (int64_t)tdio->interval_avg_latency *
 256                                     tdio->interval_transactions;
 257                                 product >>= lost_bits;
 258                                 while(total_budget >= INT64_MAX - product) {
 259                                         ++lost_bits;
 260                                         product >>= 1;
 261                                         total_budget >>= 1;
 262                                 }
 263                                 total_budget += product;
 264                                 ++budget[(tdio->head.p) ? tdio->head.p->p_ionice : 0];
 265                                 KKASSERT(total_budget >= 0);
 266                                 dsched_debug(LOG_INFO,
 267                                     "%d) avg_latency = %d, transactions = %d, ioprio = %d\n",
 268                                     n, tdio->interval_avg_latency, tdio->interval_transactions,
 269                                     (tdio->head.p) ? tdio->head.p->p_ionice : 0);
 270                                 ++n;
 271                         } else {
 272                                 tdio->max_tp = 0;
 273                         }
 274                         tdio->rebalance = 0;
 275                         tdio->transactions = 0;
 276                         tdio->avg_latency = 0;
 277                         tdio->issued = 0;
 278                 }
 279
 280                 dsched_debug(LOG_INFO, "%d procs competing for disk\n"
 281                     "total_budget = %jd (lost bits = %d)\n"
 282                     "incomplete tp = %d\n", n, (intmax_t)total_budget,
 283                     lost_bits, diskctx->incomplete_tp);
 284
 285                 if (n == 0)
 286                         continue;
 287
 288                 sum = 0;
 289
 290                 for (i = 0; i < FQ_PRIO_MAX+1; i++) {
 291                         if (budget[i] == 0)
 292                                 continue;
 293                         sum += (FQ_PRIO_BIAS+i)*budget[i];
 294                 }
 295
 296                 if (sum == 0)
 297                         sum = 1;
 298
 299                 dsched_debug(LOG_INFO, "sum = %d\n", sum);
 300
 301                 for (i = 0; i < FQ_PRIO_MAX+1; i++) {
 302                         if (budget[i] == 0)
 303                                 continue;
 304
 305                         /*
 306                          * XXX: if we still overflow here, we really need to switch to
 307                          *      some more advanced mechanism such as compound int128 or
 308                          *      storing the lost bits so they can be used in the
 309                          *      fq_balance_self.
 310                          */
 311                         diskctx->budgetpb[i] = ((FQ_PRIO_BIAS+i)*total_budget/sum) << lost_bits;
 312                         KKASSERT(diskctx->budgetpb[i] >= 0);
 313                 }
 314
 315                 dsched_debug(4, "disk is %d%% busy\n", diskctx->disk_busy);
 316                 TAILQ_FOREACH(ds_tdio, &diskctx->head.tdio_list, dlink) {
 317                         tdio = (struct fq_thread_io *)ds_tdio;
 318                         tdio->rebalance = 1;
 319                 }
 320
 321                 diskctx->prev_full = diskctx->last_full;
 322                 diskctx->last_full = (diskctx->disk_busy >= 90)?1:0;
 323         }
 324 }
 325
 326
 327 /*
 328  * fq_balance_self should be called from all sorts of dispatchers. It basically
 329  * offloads some of the heavier calculations on throttling onto the process that
 330  * wants to do I/O instead of doing it in the fq_balance thread.
 331  * - should be called with diskctx lock held
 332  */
 333 void
 334 fq_balance_self(struct fq_thread_io *tdio) {
 335         struct fq_disk_ctx *diskctx;
 336
 337         int64_t budget, used_budget;
 338         int64_t avg_latency;
 339         int64_t transactions;
 340
 341         transactions = (int64_t)tdio->interval_transactions;
 342         avg_latency = (int64_t)tdio->interval_avg_latency;
 343         diskctx = (struct fq_disk_ctx *)tdio->head.diskctx;
 344
 345 #if 0
 346         /* XXX: do we really require the lock? */
 347         DSCHED_DISK_CTX_LOCK_ASSERT(diskctx);
 348 #endif
 349
 350         used_budget = avg_latency * transactions;
 351         budget = diskctx->budgetpb[(tdio->head.p) ? tdio->head.p->p_ionice : 0];
 352
 353         if (used_budget > 0) {
 354                 dsched_debug(LOG_INFO,
 355                     "info: used_budget = %jd, budget = %jd\n",
 356                     (intmax_t)used_budget, budget);
 357         }
 358
 359         if ((used_budget > budget) && (diskctx->disk_busy >= 90)) {
 360                 KKASSERT(avg_latency != 0);
 361
 362                 tdio->max_tp = budget/(avg_latency);
 363                 atomic_add_int(&fq_stats.procs_limited, 1);
 364
 365                 dsched_debug(LOG_INFO,
 366                     "rate limited to %d transactions\n", tdio->max_tp);
 367
 368         } else if (((used_budget*2 < budget) || (diskctx->disk_busy < 80)) &&
 369             (!diskctx->prev_full && !diskctx->last_full)) {
 370                 tdio->max_tp = 0;
 371         }
 372 }
 373
 374
 375 static int
 376 do_fqstats(SYSCTL_HANDLER_ARGS)
 377 {
 378         return (sysctl_handle_opaque(oidp, &fq_stats, sizeof(struct dsched_fq_stats), req));
 379 }
 380
 381 static int
 382 fq_mod_handler(module_t mod, int type, void *unused)
 383 {
 384         static struct sysctl_ctx_list sysctl_ctx;
 385         static struct sysctl_oid *oid;
 386         static char version[16];
 387         int error;
 388
 389         ksnprintf(version, sizeof(version), "%d.%d",
 390             dsched_fq_version_maj, dsched_fq_version_min);
 391
 392         switch (type) {
 393         case MOD_LOAD:
 394                 bzero(&fq_stats, sizeof(struct dsched_fq_stats));
 395                 if ((error = dsched_register(&dsched_fq_policy)))
 396                         return (error);
 397
 398                 sysctl_ctx_init(&sysctl_ctx);
 399                 oid = SYSCTL_ADD_NODE(&sysctl_ctx,
 400                     SYSCTL_STATIC_CHILDREN(_dsched),
 401                     OID_AUTO,
 402                     "fq",
 403                     CTLFLAG_RD, 0, "");
 404
 405                 SYSCTL_ADD_PROC(&sysctl_ctx, SYSCTL_CHILDREN(oid),
 406                     OID_AUTO, "stats", CTLTYPE_OPAQUE|CTLFLAG_RD,
 407                     0, 0, do_fqstats, "S,dsched_fq_stats", "fq statistics");
 408
 409                 SYSCTL_ADD_STRING(&sysctl_ctx, SYSCTL_CHILDREN(oid),
 410                     OID_AUTO, "version", CTLFLAG_RD, version, 0, "fq version");
 411
 412                 kprintf("FQ scheduler policy version %d.%d loaded\n",
 413                     dsched_fq_version_maj, dsched_fq_version_min);
 414                 break;
 415
 416         case MOD_UNLOAD:
 417                 if ((error = dsched_unregister(&dsched_fq_policy)))
 418                         return (error);
 419                 sysctl_ctx_free(&sysctl_ctx);
 420                 kprintf("FQ scheduler policy unloaded\n");
 421                 break;
 422
 423         default:
 424                 break;
 425         }
 426
 427         return 0;
 428 }
 429
 430 DSCHED_POLICY_MODULE(dsched_fq, fq_mod_handler);