sys/kern/kern_dsched.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 <sys/malloc.h>
  44 #include <machine/md_var.h>
  45 #include <sys/ctype.h>
  46 #include <sys/syslog.h>
  47 #include <sys/device.h>
  48 #include <sys/msgport.h>
  49 #include <sys/msgport2.h>
  50 #include <sys/buf2.h>
  51 #include <sys/dsched.h>
  52 #include <sys/fcntl.h>
  53 #include <machine/varargs.h>
  54
  55 MALLOC_DEFINE(M_DSCHED, "dsched", "dsched allocs");
  56
  57 static dsched_prepare_t         noop_prepare;
  58 static dsched_teardown_t        noop_teardown;
  59 static dsched_cancel_t          noop_cancel;
  60 static dsched_queue_t           noop_queue;
  61
  62 static void dsched_sysctl_add_disk(struct dsched_disk_ctx *diskctx, char *name);
  63
  64 static int      dsched_inited = 0;
  65 static int      default_set = 0;
  66
  67 struct lock     dsched_lock;
  68 static int      dsched_debug_enable = 0;
  69
  70 struct dsched_stats     dsched_stats;
  71
  72 struct objcache_malloc_args dsched_disk_ctx_malloc_args = {
  73         DSCHED_DISK_CTX_MAX_SZ, M_DSCHED };
  74 struct objcache_malloc_args dsched_thread_io_malloc_args = {
  75         DSCHED_THREAD_IO_MAX_SZ, M_DSCHED };
  76 struct objcache_malloc_args dsched_thread_ctx_malloc_args = {
  77         DSCHED_THREAD_CTX_MAX_SZ, M_DSCHED };
  78
  79 static struct objcache  *dsched_diskctx_cache;
  80 static struct objcache  *dsched_tdctx_cache;
  81 static struct objcache  *dsched_tdio_cache;
  82
  83 TAILQ_HEAD(, dsched_thread_ctx) dsched_tdctx_list =
  84                 TAILQ_HEAD_INITIALIZER(dsched_tdctx_list);
  85
  86 struct lock     dsched_tdctx_lock;
  87
  88 static struct dsched_policy_head dsched_policy_list =
  89                 TAILQ_HEAD_INITIALIZER(dsched_policy_list);
  90
  91 static struct dsched_policy dsched_noop_policy = {
  92         .name = "noop",
  93
  94         .prepare = noop_prepare,
  95         .teardown = noop_teardown,
  96         .cancel_all = noop_cancel,
  97         .bio_queue = noop_queue
  98 };
  99
 100 static struct dsched_policy *default_policy = &dsched_noop_policy;
 101
 102 /*
 103  * dsched_debug() is a SYSCTL and TUNABLE controlled debug output function
 104  * using kvprintf
 105  */
 106 int
 107 dsched_debug(int level, char *fmt, ...)
 108 {
 109         __va_list ap;
 110
 111         __va_start(ap, fmt);
 112         if (level <= dsched_debug_enable)
 113                 kvprintf(fmt, ap);
 114         __va_end(ap);
 115
 116         return 0;
 117 }
 118
 119 /*
 120  * Called on disk_create()
 121  * tries to read which policy to use from loader.conf, if there's
 122  * none specified, the default policy is used.
 123  */
 124 void
 125 dsched_disk_create_callback(struct disk *dp, const char *head_name, int unit)
 126 {
 127         char tunable_key[SPECNAMELEN + 48];
 128         char sched_policy[DSCHED_POLICY_NAME_LENGTH];
 129         char *ptr;
 130         struct dsched_policy *policy = NULL;
 131
 132         /* Also look for serno stuff? */
 133         /* kprintf("dsched_disk_create_callback() for disk %s%d\n", head_name, unit); */
 134         lockmgr(&dsched_lock, LK_EXCLUSIVE);
 135
 136         ksnprintf(tunable_key, sizeof(tunable_key), "dsched.policy.%s%d",
 137             head_name, unit);
 138         if (TUNABLE_STR_FETCH(tunable_key, sched_policy,
 139             sizeof(sched_policy)) != 0) {
 140                 policy = dsched_find_policy(sched_policy);
 141         }
 142
 143         ksnprintf(tunable_key, sizeof(tunable_key), "dsched.policy.%s",
 144             head_name);
 145         for (ptr = tunable_key; *ptr; ptr++) {
 146                 if (*ptr == '/')
 147                         *ptr = '-';
 148         }
 149         if (!policy && (TUNABLE_STR_FETCH(tunable_key, sched_policy,
 150             sizeof(sched_policy)) != 0)) {
 151                 policy = dsched_find_policy(sched_policy);
 152         }
 153
 154         ksnprintf(tunable_key, sizeof(tunable_key), "dsched.policy.default");
 155         if (!policy && !default_set && (TUNABLE_STR_FETCH(tunable_key, sched_policy,
 156             sizeof(sched_policy)) != 0)) {
 157                 policy = dsched_find_policy(sched_policy);
 158         }
 159
 160         if (!policy) {
 161                 if (!default_set) {
 162                         dsched_debug(0, "No policy for %s%d specified, "
 163                             "or policy not found\n", head_name, unit);
 164                 }
 165                 dsched_set_policy(dp, default_policy);
 166         } else {
 167                 dsched_set_policy(dp, policy);
 168         }
 169
 170         if (strncmp(head_name, "mapper/", strlen("mapper/")) == 0)
 171                 ksnprintf(tunable_key, sizeof(tunable_key), "%s", head_name);
 172         else
 173                 ksnprintf(tunable_key, sizeof(tunable_key), "%s%d", head_name, unit);
 174         for (ptr = tunable_key; *ptr; ptr++) {
 175                 if (*ptr == '/')
 176                         *ptr = '-';
 177         }
 178         dsched_sysctl_add_disk(
 179             (struct dsched_disk_ctx *)dsched_get_disk_priv(dp),
 180             tunable_key);
 181
 182         lockmgr(&dsched_lock, LK_RELEASE);
 183 }
 184
 185 /*
 186  * Called from disk_setdiskinfo (or rather _setdiskinfo). This will check if
 187  * there's any policy associated with the serial number of the device.
 188  */
 189 void
 190 dsched_disk_update_callback(struct disk *dp, struct disk_info *info)
 191 {
 192         char tunable_key[SPECNAMELEN + 48];
 193         char sched_policy[DSCHED_POLICY_NAME_LENGTH];
 194         struct dsched_policy *policy = NULL;
 195
 196         if (info->d_serialno == NULL)
 197                 return;
 198
 199         lockmgr(&dsched_lock, LK_EXCLUSIVE);
 200
 201         ksnprintf(tunable_key, sizeof(tunable_key), "dsched.policy.%s",
 202             info->d_serialno);
 203
 204         if((TUNABLE_STR_FETCH(tunable_key, sched_policy,
 205             sizeof(sched_policy)) != 0)) {
 206                 policy = dsched_find_policy(sched_policy);
 207         }
 208
 209         if (policy) {
 210                 dsched_switch(dp, policy);
 211         }
 212
 213         dsched_sysctl_add_disk(
 214             (struct dsched_disk_ctx *)dsched_get_disk_priv(dp),
 215             info->d_serialno);
 216
 217         lockmgr(&dsched_lock, LK_RELEASE);
 218 }
 219
 220 /*
 221  * Called on disk_destroy()
 222  * shuts down the scheduler core and cancels all remaining bios
 223  */
 224 void
 225 dsched_disk_destroy_callback(struct disk *dp)
 226 {
 227         struct dsched_policy *old_policy;
 228         struct dsched_disk_ctx *diskctx;
 229
 230         lockmgr(&dsched_lock, LK_EXCLUSIVE);
 231
 232         diskctx = dsched_get_disk_priv(dp);
 233
 234         old_policy = dp->d_sched_policy;
 235         dp->d_sched_policy = &dsched_noop_policy;
 236         old_policy->cancel_all(dsched_get_disk_priv(dp));
 237         old_policy->teardown(dsched_get_disk_priv(dp));
 238
 239         if (diskctx->flags & DSCHED_SYSCTL_CTX_INITED)
 240                 sysctl_ctx_free(&diskctx->sysctl_ctx);
 241
 242         policy_destroy(dp);
 243         atomic_subtract_int(&old_policy->ref_count, 1);
 244         KKASSERT(old_policy->ref_count >= 0);
 245
 246         lockmgr(&dsched_lock, LK_RELEASE);
 247 }
 248
 249
 250 void
 251 dsched_queue(struct disk *dp, struct bio *bio)
 252 {
 253         struct dsched_thread_ctx        *tdctx;
 254         struct dsched_thread_io         *tdio;
 255         struct dsched_disk_ctx          *diskctx;
 256
 257         int found = 0, error = 0;
 258
 259         tdctx = dsched_get_buf_priv(bio->bio_buf);
 260         if (tdctx == NULL) {
 261                 /* We don't handle this case, let dsched dispatch */
 262                 atomic_add_int(&dsched_stats.no_tdctx, 1);
 263                 dsched_strategy_raw(dp, bio);
 264                 return;
 265         }
 266
 267         DSCHED_THREAD_CTX_LOCK(tdctx);
 268
 269         KKASSERT(!TAILQ_EMPTY(&tdctx->tdio_list));
 270         TAILQ_FOREACH(tdio, &tdctx->tdio_list, link) {
 271                 if (tdio->dp == dp) {
 272                         dsched_thread_io_ref(tdio);
 273                         found = 1;
 274                         break;
 275                 }
 276         }
 277
 278         DSCHED_THREAD_CTX_UNLOCK(tdctx);
 279         dsched_clr_buf_priv(bio->bio_buf);
 280         dsched_thread_ctx_unref(tdctx); /* acquired on new_buf */
 281
 282         KKASSERT(found == 1);
 283         diskctx = dsched_get_disk_priv(dp);
 284         dsched_disk_ctx_ref(diskctx);
 285         error = dp->d_sched_policy->bio_queue(diskctx, tdio, bio);
 286
 287         if (error) {
 288                 dsched_strategy_raw(dp, bio);
 289         }
 290         dsched_disk_ctx_unref(diskctx);
 291         dsched_thread_io_unref(tdio);
 292 }
 293
 294
 295 /*
 296  * Called from each module_init or module_attach of each policy
 297  * registers the policy in the local policy list.
 298  */
 299 int
 300 dsched_register(struct dsched_policy *d_policy)
 301 {
 302         struct dsched_policy *policy;
 303         int error = 0;
 304
 305         lockmgr(&dsched_lock, LK_EXCLUSIVE);
 306
 307         policy = dsched_find_policy(d_policy->name);
 308
 309         if (!policy) {
 310                 TAILQ_INSERT_TAIL(&dsched_policy_list, d_policy, link);
 311                 atomic_add_int(&d_policy->ref_count, 1);
 312         } else {
 313                 dsched_debug(LOG_ERR, "Policy with name %s already registered!\n",
 314                     d_policy->name);
 315                 error = EEXIST;
 316         }
 317
 318         lockmgr(&dsched_lock, LK_RELEASE);
 319         return error;
 320 }
 321
 322 /*
 323  * Called from each module_detach of each policy
 324  * unregisters the policy
 325  */
 326 int
 327 dsched_unregister(struct dsched_policy *d_policy)
 328 {
 329         struct dsched_policy *policy;
 330
 331         lockmgr(&dsched_lock, LK_EXCLUSIVE);
 332         policy = dsched_find_policy(d_policy->name);
 333
 334         if (policy) {
 335                 if (policy->ref_count > 1) {
 336                         lockmgr(&dsched_lock, LK_RELEASE);
 337                         return EBUSY;
 338                 }
 339                 TAILQ_REMOVE(&dsched_policy_list, policy, link);
 340                 atomic_subtract_int(&policy->ref_count, 1);
 341                 KKASSERT(policy->ref_count == 0);
 342         }
 343         lockmgr(&dsched_lock, LK_RELEASE);
 344         return 0;
 345 }
 346
 347
 348 /*
 349  * switches the policy by first removing the old one and then
 350  * enabling the new one.
 351  */
 352 int
 353 dsched_switch(struct disk *dp, struct dsched_policy *new_policy)
 354 {
 355         struct dsched_policy *old_policy;
 356
 357         /* If we are asked to set the same policy, do nothing */
 358         if (dp->d_sched_policy == new_policy)
 359                 return 0;
 360
 361         /* lock everything down, diskwise */
 362         lockmgr(&dsched_lock, LK_EXCLUSIVE);
 363         old_policy = dp->d_sched_policy;
 364
 365         atomic_subtract_int(&old_policy->ref_count, 1);
 366         KKASSERT(old_policy->ref_count >= 0);
 367
 368         dp->d_sched_policy = &dsched_noop_policy;
 369         old_policy->teardown(dsched_get_disk_priv(dp));
 370         policy_destroy(dp);
 371
 372         /* Bring everything back to life */
 373         dsched_set_policy(dp, new_policy);
 374         lockmgr(&dsched_lock, LK_RELEASE);
 375         return 0;
 376 }
 377
 378
 379 /*
 380  * Loads a given policy and attaches it to the specified disk.
 381  * Also initializes the core for the policy
 382  */
 383 void
 384 dsched_set_policy(struct disk *dp, struct dsched_policy *new_policy)
 385 {
 386         int locked = 0;
 387
 388         /* Check if it is locked already. if not, we acquire the devfs lock */
 389         if (!(lockstatus(&dsched_lock, curthread)) == LK_EXCLUSIVE) {
 390                 lockmgr(&dsched_lock, LK_EXCLUSIVE);
 391                 locked = 1;
 392         }
 393
 394         policy_new(dp, new_policy);
 395         new_policy->prepare(dsched_get_disk_priv(dp));
 396         dp->d_sched_policy = new_policy;
 397         atomic_add_int(&new_policy->ref_count, 1);
 398         kprintf("disk scheduler: set policy of %s to %s\n", dp->d_cdev->si_name,
 399             new_policy->name);
 400
 401         /* If we acquired the lock, we also get rid of it */
 402         if (locked)
 403                 lockmgr(&dsched_lock, LK_RELEASE);
 404 }
 405
 406 struct dsched_policy*
 407 dsched_find_policy(char *search)
 408 {
 409         struct dsched_policy *policy;
 410         struct dsched_policy *policy_found = NULL;
 411         int locked = 0;
 412
 413         /* Check if it is locked already. if not, we acquire the devfs lock */
 414         if (!(lockstatus(&dsched_lock, curthread)) == LK_EXCLUSIVE) {
 415                 lockmgr(&dsched_lock, LK_EXCLUSIVE);
 416                 locked = 1;
 417         }
 418
 419         TAILQ_FOREACH(policy, &dsched_policy_list, link) {
 420                 if (!strcmp(policy->name, search)) {
 421                         policy_found = policy;
 422                         break;
 423                 }
 424         }
 425
 426         /* If we acquired the lock, we also get rid of it */
 427         if (locked)
 428                 lockmgr(&dsched_lock, LK_RELEASE);
 429
 430         return policy_found;
 431 }
 432
 433 struct disk*
 434 dsched_find_disk(char *search)
 435 {
 436         struct disk *dp_found = NULL;
 437         struct disk *dp = NULL;
 438
 439         while((dp = disk_enumerate(dp))) {
 440                 if (!strcmp(dp->d_cdev->si_name, search)) {
 441                         dp_found = dp;
 442                         break;
 443                 }
 444         }
 445
 446         return dp_found;
 447 }
 448
 449 struct disk*
 450 dsched_disk_enumerate(struct disk *dp, struct dsched_policy *policy)
 451 {
 452         while ((dp = disk_enumerate(dp))) {
 453                 if (dp->d_sched_policy == policy)
 454                         return dp;
 455         }
 456
 457         return NULL;
 458 }
 459
 460 struct dsched_policy *
 461 dsched_policy_enumerate(struct dsched_policy *pol)
 462 {
 463         if (!pol)
 464                 return (TAILQ_FIRST(&dsched_policy_list));
 465         else
 466                 return (TAILQ_NEXT(pol, link));
 467 }
 468
 469 void
 470 dsched_cancel_bio(struct bio *bp)
 471 {
 472         bp->bio_buf->b_error = ENXIO;
 473         bp->bio_buf->b_flags |= B_ERROR;
 474         bp->bio_buf->b_resid = bp->bio_buf->b_bcount;
 475
 476         biodone(bp);
 477 }
 478
 479 void
 480 dsched_strategy_raw(struct disk *dp, struct bio *bp)
 481 {
 482         /*
 483          * Ideally, this stuff shouldn't be needed... but just in case, we leave it in
 484          * to avoid panics
 485          */
 486         KASSERT(dp->d_rawdev != NULL, ("dsched_strategy_raw sees NULL d_rawdev!!"));
 487         if(bp->bio_track != NULL) {
 488                 dsched_debug(LOG_INFO,
 489                     "dsched_strategy_raw sees non-NULL bio_track!! "
 490                     "bio: %p\n", bp);
 491                 bp->bio_track = NULL;
 492         }
 493         dev_dstrategy(dp->d_rawdev, bp);
 494 }
 495
 496 void
 497 dsched_strategy_sync(struct disk *dp, struct bio *bio)
 498 {
 499         struct buf *bp, *nbp;
 500         struct bio *nbio;
 501
 502         bp = bio->bio_buf;
 503
 504         nbp = getpbuf(NULL);
 505         nbio = &nbp->b_bio1;
 506
 507         nbp->b_cmd = bp->b_cmd;
 508         nbp->b_bufsize = bp->b_bufsize;
 509         nbp->b_runningbufspace = bp->b_runningbufspace;
 510         nbp->b_bcount = bp->b_bcount;
 511         nbp->b_resid = bp->b_resid;
 512         nbp->b_data = bp->b_data;
 513 #if 0
 514         /*
 515          * Buffers undergoing device I/O do not need a kvabase/size.
 516          */
 517         nbp->b_kvabase = bp->b_kvabase;
 518         nbp->b_kvasize = bp->b_kvasize;
 519 #endif
 520         nbp->b_dirtyend = bp->b_dirtyend;
 521
 522         nbio->bio_done = biodone_sync;
 523         nbio->bio_flags |= BIO_SYNC;
 524         nbio->bio_track = NULL;
 525
 526         nbio->bio_caller_info1.ptr = dp;
 527         nbio->bio_offset = bio->bio_offset;
 528
 529         dev_dstrategy(dp->d_rawdev, nbio);
 530         biowait(nbio, "dschedsync");
 531         bp->b_resid = nbp->b_resid;
 532         bp->b_error = nbp->b_error;
 533         biodone(bio);
 534 #if 0
 535         nbp->b_kvabase = NULL;
 536         nbp->b_kvasize = 0;
 537 #endif
 538         relpbuf(nbp, NULL);
 539 }
 540
 541 void
 542 dsched_strategy_async(struct disk *dp, struct bio *bio, biodone_t *done, void *priv)
 543 {
 544         struct bio *nbio;
 545
 546         nbio = push_bio(bio);
 547         nbio->bio_done = done;
 548         nbio->bio_offset = bio->bio_offset;
 549
 550         dsched_set_bio_dp(nbio, dp);
 551         dsched_set_bio_priv(nbio, priv);
 552
 553         getmicrotime(&nbio->bio_caller_info3.tv);
 554         dev_dstrategy(dp->d_rawdev, nbio);
 555 }
 556
 557 void
 558 dsched_disk_ctx_ref(struct dsched_disk_ctx *diskctx)
 559 {
 560         int refcount;
 561
 562         refcount = atomic_fetchadd_int(&diskctx->refcount, 1);
 563
 564         KKASSERT(refcount >= 0);
 565 }
 566
 567 void
 568 dsched_thread_io_ref(struct dsched_thread_io *tdio)
 569 {
 570         int refcount;
 571
 572         refcount = atomic_fetchadd_int(&tdio->refcount, 1);
 573
 574         KKASSERT(refcount >= 0);
 575 }
 576
 577 void
 578 dsched_thread_ctx_ref(struct dsched_thread_ctx *tdctx)
 579 {
 580         int refcount;
 581
 582         refcount = atomic_fetchadd_int(&tdctx->refcount, 1);
 583
 584         KKASSERT(refcount >= 0);
 585 }
 586
 587 void
 588 dsched_disk_ctx_unref(struct dsched_disk_ctx *diskctx)
 589 {
 590         struct dsched_thread_io *tdio, *tdio2;
 591         int refcount;
 592
 593         refcount = atomic_fetchadd_int(&diskctx->refcount, -1);
 594
 595
 596         KKASSERT(refcount >= 0 || refcount <= -0x400);
 597
 598         if (refcount == 1) {
 599                 atomic_subtract_int(&diskctx->refcount, 0x400); /* mark as: in destruction */
 600 #if 0
 601                 kprintf("diskctx (%p) destruction started, trace:\n", diskctx);
 602                 print_backtrace(4);
 603 #endif
 604                 lockmgr(&diskctx->lock, LK_EXCLUSIVE);
 605                 TAILQ_FOREACH_MUTABLE(tdio, &diskctx->tdio_list, dlink, tdio2) {
 606                         TAILQ_REMOVE(&diskctx->tdio_list, tdio, dlink);
 607                         tdio->flags &= ~DSCHED_LINKED_DISK_CTX;
 608                         dsched_thread_io_unref(tdio);
 609                 }
 610                 lockmgr(&diskctx->lock, LK_RELEASE);
 611                 if (diskctx->dp->d_sched_policy->destroy_diskctx)
 612                         diskctx->dp->d_sched_policy->destroy_diskctx(diskctx);
 613                 objcache_put(dsched_diskctx_cache, diskctx);
 614                 atomic_subtract_int(&dsched_stats.diskctx_allocations, 1);
 615         }
 616 }
 617
 618 void
 619 dsched_thread_io_unref(struct dsched_thread_io *tdio)
 620 {
 621         struct dsched_thread_ctx        *tdctx;
 622         struct dsched_disk_ctx  *diskctx;
 623         int refcount;
 624
 625         refcount = atomic_fetchadd_int(&tdio->refcount, -1);
 626
 627         KKASSERT(refcount >= 0 || refcount <= -0x400);
 628
 629         if (refcount == 1) {
 630                 atomic_subtract_int(&tdio->refcount, 0x400); /* mark as: in destruction */
 631 #if 0
 632                 kprintf("tdio (%p) destruction started, trace:\n", tdio);
 633                 print_backtrace(8);
 634 #endif
 635                 diskctx = tdio->diskctx;
 636                 KKASSERT(diskctx != NULL);
 637                 KKASSERT(tdio->qlength == 0);
 638
 639                 if (tdio->flags & DSCHED_LINKED_DISK_CTX) {
 640                         lockmgr(&diskctx->lock, LK_EXCLUSIVE);
 641
 642                         TAILQ_REMOVE(&diskctx->tdio_list, tdio, dlink);
 643                         tdio->flags &= ~DSCHED_LINKED_DISK_CTX;
 644
 645                         lockmgr(&diskctx->lock, LK_RELEASE);
 646                 }
 647
 648                 if (tdio->flags & DSCHED_LINKED_THREAD_CTX) {
 649                         tdctx = tdio->tdctx;
 650                         KKASSERT(tdctx != NULL);
 651
 652                         lockmgr(&tdctx->lock, LK_EXCLUSIVE);
 653
 654                         TAILQ_REMOVE(&tdctx->tdio_list, tdio, link);
 655                         tdio->flags &= ~DSCHED_LINKED_THREAD_CTX;
 656
 657                         lockmgr(&tdctx->lock, LK_RELEASE);
 658                 }
 659                 if (tdio->diskctx->dp->d_sched_policy->destroy_tdio)
 660                         tdio->diskctx->dp->d_sched_policy->destroy_tdio(tdio);
 661                 objcache_put(dsched_tdio_cache, tdio);
 662                 atomic_subtract_int(&dsched_stats.tdio_allocations, 1);
 663 #if 0
 664                 dsched_disk_ctx_unref(diskctx);
 665 #endif
 666         }
 667 }
 668
 669 void
 670 dsched_thread_ctx_unref(struct dsched_thread_ctx *tdctx)
 671 {
 672         struct dsched_thread_io *tdio, *tdio2;
 673         int refcount;
 674
 675         refcount = atomic_fetchadd_int(&tdctx->refcount, -1);
 676
 677         KKASSERT(refcount >= 0 || refcount <= -0x400);
 678
 679         if (refcount == 1) {
 680                 atomic_subtract_int(&tdctx->refcount, 0x400); /* mark as: in destruction */
 681 #if 0
 682                 kprintf("tdctx (%p) destruction started, trace:\n", tdctx);
 683                 print_backtrace(8);
 684 #endif
 685                 DSCHED_GLOBAL_THREAD_CTX_LOCK();
 686
 687                 TAILQ_FOREACH_MUTABLE(tdio, &tdctx->tdio_list, link, tdio2) {
 688                         TAILQ_REMOVE(&tdctx->tdio_list, tdio, link);
 689                         tdio->flags &= ~DSCHED_LINKED_THREAD_CTX;
 690                         dsched_thread_io_unref(tdio);
 691                 }
 692                 TAILQ_REMOVE(&dsched_tdctx_list, tdctx, link);
 693
 694                 DSCHED_GLOBAL_THREAD_CTX_UNLOCK();
 695
 696                 objcache_put(dsched_tdctx_cache, tdctx);
 697                 atomic_subtract_int(&dsched_stats.tdctx_allocations, 1);
 698         }
 699 }
 700
 701
 702 struct dsched_thread_io *
 703 dsched_thread_io_alloc(struct disk *dp, struct dsched_thread_ctx *tdctx,
 704     struct dsched_policy *pol)
 705 {
 706         struct dsched_thread_io *tdio;
 707 #if 0
 708         dsched_disk_ctx_ref(dsched_get_disk_priv(dp));
 709 #endif
 710         tdio = objcache_get(dsched_tdio_cache, M_WAITOK);
 711         bzero(tdio, DSCHED_THREAD_IO_MAX_SZ);
 712
 713         /* XXX: maybe we do need another ref for the disk list for tdio */
 714         dsched_thread_io_ref(tdio);
 715
 716         DSCHED_THREAD_IO_LOCKINIT(tdio);
 717         tdio->dp = dp;
 718
 719         tdio->diskctx = dsched_get_disk_priv(dp);
 720         TAILQ_INIT(&tdio->queue);
 721
 722         if (pol->new_tdio)
 723                 pol->new_tdio(tdio);
 724
 725         TAILQ_INSERT_TAIL(&tdio->diskctx->tdio_list, tdio, dlink);
 726         tdio->flags |= DSCHED_LINKED_DISK_CTX;
 727
 728         if (tdctx) {
 729                 tdio->tdctx = tdctx;
 730                 tdio->p = tdctx->p;
 731
 732                 /* Put the tdio in the tdctx list */
 733                 DSCHED_THREAD_CTX_LOCK(tdctx);
 734                 TAILQ_INSERT_TAIL(&tdctx->tdio_list, tdio, link);
 735                 DSCHED_THREAD_CTX_UNLOCK(tdctx);
 736                 tdio->flags |= DSCHED_LINKED_THREAD_CTX;
 737         }
 738
 739         atomic_add_int(&dsched_stats.tdio_allocations, 1);
 740         return tdio;
 741 }
 742
 743
 744 struct dsched_disk_ctx *
 745 dsched_disk_ctx_alloc(struct disk *dp, struct dsched_policy *pol)
 746 {
 747         struct dsched_disk_ctx *diskctx;
 748
 749         diskctx = objcache_get(dsched_diskctx_cache, M_WAITOK);
 750         bzero(diskctx, DSCHED_DISK_CTX_MAX_SZ);
 751         dsched_disk_ctx_ref(diskctx);
 752         diskctx->dp = dp;
 753         DSCHED_DISK_CTX_LOCKINIT(diskctx);
 754         TAILQ_INIT(&diskctx->tdio_list);
 755
 756         atomic_add_int(&dsched_stats.diskctx_allocations, 1);
 757         if (pol->new_diskctx)
 758                 pol->new_diskctx(diskctx);
 759         return diskctx;
 760 }
 761
 762
 763 struct dsched_thread_ctx *
 764 dsched_thread_ctx_alloc(struct proc *p)
 765 {
 766         struct dsched_thread_ctx        *tdctx;
 767         struct dsched_thread_io *tdio;
 768         struct disk     *dp = NULL;
 769
 770         tdctx = objcache_get(dsched_tdctx_cache, M_WAITOK);
 771         bzero(tdctx, DSCHED_THREAD_CTX_MAX_SZ);
 772         dsched_thread_ctx_ref(tdctx);
 773 #if 0
 774         kprintf("dsched_thread_ctx_alloc, new tdctx = %p\n", tdctx);
 775 #endif
 776         DSCHED_THREAD_CTX_LOCKINIT(tdctx);
 777         TAILQ_INIT(&tdctx->tdio_list);
 778         tdctx->p = p;
 779
 780         /* XXX */
 781         while ((dp = disk_enumerate(dp))) {
 782                 tdio = dsched_thread_io_alloc(dp, tdctx, dp->d_sched_policy);
 783         }
 784
 785         DSCHED_GLOBAL_THREAD_CTX_LOCK();
 786         TAILQ_INSERT_TAIL(&dsched_tdctx_list, tdctx, link);
 787         DSCHED_GLOBAL_THREAD_CTX_UNLOCK();
 788
 789         atomic_add_int(&dsched_stats.tdctx_allocations, 1);
 790         /* XXX: no callback here */
 791         return tdctx;
 792 }
 793
 794 void
 795 policy_new(struct disk *dp, struct dsched_policy *pol) {
 796         struct dsched_thread_ctx *tdctx;
 797         struct dsched_disk_ctx *diskctx;
 798         struct dsched_thread_io *tdio;
 799
 800         diskctx = dsched_disk_ctx_alloc(dp, pol);
 801         dsched_disk_ctx_ref(diskctx);
 802         dsched_set_disk_priv(dp, diskctx);
 803
 804         DSCHED_GLOBAL_THREAD_CTX_LOCK();
 805         TAILQ_FOREACH(tdctx, &dsched_tdctx_list, link) {
 806                 tdio = dsched_thread_io_alloc(dp, tdctx, pol);
 807         }
 808         DSCHED_GLOBAL_THREAD_CTX_UNLOCK();
 809
 810 }
 811
 812 void
 813 policy_destroy(struct disk *dp) {
 814         struct dsched_disk_ctx *diskctx;
 815
 816         diskctx = dsched_get_disk_priv(dp);
 817         KKASSERT(diskctx != NULL);
 818
 819         dsched_disk_ctx_unref(diskctx); /* from prepare */
 820         dsched_disk_ctx_unref(diskctx); /* from alloc */
 821
 822         dsched_set_disk_priv(dp, NULL);
 823 }
 824
 825 void
 826 dsched_new_buf(struct buf *bp)
 827 {
 828         struct dsched_thread_ctx        *tdctx = NULL;
 829
 830         if (dsched_inited == 0)
 831                 return;
 832
 833         if (curproc != NULL) {
 834                 tdctx = dsched_get_proc_priv(curproc);
 835         } else {
 836                 /* This is a kernel thread, so no proc info is available */
 837                 tdctx = dsched_get_thread_priv(curthread);
 838         }
 839
 840 #if 0
 841         /*
 842          * XXX: hack. we don't want this assert because we aren't catching all
 843          *      threads. mi_startup() is still getting away without an tdctx.
 844          */
 845
 846         /* by now we should have an tdctx. if not, something bad is going on */
 847         KKASSERT(tdctx != NULL);
 848 #endif
 849
 850         if (tdctx) {
 851                 dsched_thread_ctx_ref(tdctx);
 852         }
 853         dsched_set_buf_priv(bp, tdctx);
 854 }
 855
 856 void
 857 dsched_exit_buf(struct buf *bp)
 858 {
 859         struct dsched_thread_ctx        *tdctx;
 860
 861         tdctx = dsched_get_buf_priv(bp);
 862         if (tdctx != NULL) {
 863                 dsched_clr_buf_priv(bp);
 864                 dsched_thread_ctx_unref(tdctx);
 865         }
 866 }
 867
 868 void
 869 dsched_new_proc(struct proc *p)
 870 {
 871         struct dsched_thread_ctx        *tdctx;
 872
 873         if (dsched_inited == 0)
 874                 return;
 875
 876         KKASSERT(p != NULL);
 877
 878         tdctx = dsched_thread_ctx_alloc(p);
 879         tdctx->p = p;
 880         dsched_thread_ctx_ref(tdctx);
 881
 882         dsched_set_proc_priv(p, tdctx);
 883         atomic_add_int(&dsched_stats.nprocs, 1);
 884 }
 885
 886
 887 void
 888 dsched_new_thread(struct thread *td)
 889 {
 890         struct dsched_thread_ctx        *tdctx;
 891
 892         if (dsched_inited == 0)
 893                 return;
 894
 895         KKASSERT(td != NULL);
 896
 897         tdctx = dsched_thread_ctx_alloc(NULL);
 898         tdctx->td = td;
 899         dsched_thread_ctx_ref(tdctx);
 900
 901         dsched_set_thread_priv(td, tdctx);
 902         atomic_add_int(&dsched_stats.nthreads, 1);
 903 }
 904
 905 void
 906 dsched_exit_proc(struct proc *p)
 907 {
 908         struct dsched_thread_ctx        *tdctx;
 909
 910         if (dsched_inited == 0)
 911                 return;
 912
 913         KKASSERT(p != NULL);
 914
 915         tdctx = dsched_get_proc_priv(p);
 916         KKASSERT(tdctx != NULL);
 917
 918         tdctx->dead = 0xDEAD;
 919         dsched_set_proc_priv(p, NULL);
 920
 921         dsched_thread_ctx_unref(tdctx); /* one for alloc, */
 922         dsched_thread_ctx_unref(tdctx); /* one for ref */
 923         atomic_subtract_int(&dsched_stats.nprocs, 1);
 924 }
 925
 926
 927 void
 928 dsched_exit_thread(struct thread *td)
 929 {
 930         struct dsched_thread_ctx        *tdctx;
 931
 932         if (dsched_inited == 0)
 933                 return;
 934
 935         KKASSERT(td != NULL);
 936
 937         tdctx = dsched_get_thread_priv(td);
 938         KKASSERT(tdctx != NULL);
 939
 940         tdctx->dead = 0xDEAD;
 941         dsched_set_thread_priv(td, 0);
 942
 943         dsched_thread_ctx_unref(tdctx); /* one for alloc, */
 944         dsched_thread_ctx_unref(tdctx); /* one for ref */
 945         atomic_subtract_int(&dsched_stats.nthreads, 1);
 946 }
 947
 948 struct dsched_thread_io *
 949 dsched_new_policy_thread_tdio(struct dsched_disk_ctx *diskctx,
 950     struct dsched_policy *pol) {
 951         struct dsched_thread_ctx *tdctx;
 952         struct dsched_thread_io *tdio;
 953
 954         tdctx = dsched_get_thread_priv(curthread);
 955         KKASSERT(tdctx != NULL);
 956
 957         tdio = dsched_thread_io_alloc(diskctx->dp, tdctx, pol);
 958         return tdio;
 959 }
 960
 961 /* DEFAULT NOOP POLICY */
 962
 963 static int
 964 noop_prepare(struct dsched_disk_ctx *diskctx)
 965 {
 966         return 0;
 967 }
 968
 969 static void
 970 noop_teardown(struct dsched_disk_ctx *diskctx)
 971 {
 972
 973 }
 974
 975 static void
 976 noop_cancel(struct dsched_disk_ctx *diskctx)
 977 {
 978
 979 }
 980
 981 static int
 982 noop_queue(struct dsched_disk_ctx *diskctx, struct dsched_thread_io *tdio,
 983     struct bio *bio)
 984 {
 985         dsched_strategy_raw(diskctx->dp, bio);
 986 #if 0
 987         dsched_strategy_async(diskctx->dp, bio, noop_completed, NULL);
 988 #endif
 989         return 0;
 990 }
 991
 992 /*
 993  * SYSINIT stuff
 994  */
 995 static void
 996 dsched_init(void)
 997 {
 998         dsched_tdio_cache = objcache_create("dsched-tdio-cache", 0, 0,
 999                                            NULL, NULL, NULL,
1000                                            objcache_malloc_alloc,
1001                                            objcache_malloc_free,
1002                                            &dsched_thread_io_malloc_args );
1003
1004         dsched_tdctx_cache = objcache_create("dsched-tdctx-cache", 0, 0,
1005                                            NULL, NULL, NULL,
1006                                            objcache_malloc_alloc,
1007                                            objcache_malloc_free,
1008                                            &dsched_thread_ctx_malloc_args );
1009
1010         dsched_diskctx_cache = objcache_create("dsched-diskctx-cache", 0, 0,
1011                                            NULL, NULL, NULL,
1012                                            objcache_malloc_alloc,
1013                                            objcache_malloc_free,
1014                                            &dsched_disk_ctx_malloc_args );
1015
1016         bzero(&dsched_stats, sizeof(struct dsched_stats));
1017
1018         lockinit(&dsched_lock, "dsched lock", 0, LK_CANRECURSE);
1019         DSCHED_GLOBAL_THREAD_CTX_LOCKINIT();
1020
1021         dsched_register(&dsched_noop_policy);
1022
1023         dsched_inited = 1;
1024 }
1025
1026 static void
1027 dsched_uninit(void)
1028 {
1029 }
1030
1031 SYSINIT(subr_dsched_register, SI_SUB_CREATE_INIT-1, SI_ORDER_FIRST, dsched_init, NULL);
1032 SYSUNINIT(subr_dsched_register, SI_SUB_CREATE_INIT-1, SI_ORDER_ANY, dsched_uninit, NULL);
1033
1034 /*
1035  * SYSCTL stuff
1036  */
1037 static int
1038 sysctl_dsched_stats(SYSCTL_HANDLER_ARGS)
1039 {
1040         return (sysctl_handle_opaque(oidp, &dsched_stats, sizeof(struct dsched_stats), req));
1041 }
1042
1043 static int
1044 sysctl_dsched_list_policies(SYSCTL_HANDLER_ARGS)
1045 {
1046         struct dsched_policy *pol = NULL;
1047         int error, first = 1;
1048
1049         lockmgr(&dsched_lock, LK_EXCLUSIVE);
1050
1051         while ((pol = dsched_policy_enumerate(pol))) {
1052                 if (!first) {
1053                         error = SYSCTL_OUT(req, " ", 1);
1054                         if (error)
1055                                 break;
1056                 } else {
1057                         first = 0;
1058                 }
1059                 error = SYSCTL_OUT(req, pol->name, strlen(pol->name));
1060                 if (error)
1061                         break;
1062
1063         }
1064
1065         lockmgr(&dsched_lock, LK_RELEASE);
1066
1067         error = SYSCTL_OUT(req, "", 1);
1068
1069         return error;
1070 }
1071
1072 static int
1073 sysctl_dsched_policy(SYSCTL_HANDLER_ARGS)
1074 {
1075         char buf[DSCHED_POLICY_NAME_LENGTH];
1076         struct dsched_disk_ctx *diskctx = arg1;
1077         struct dsched_policy *pol = NULL;
1078         int error;
1079
1080         if (diskctx == NULL) {
1081                 return 0;
1082         }
1083
1084         lockmgr(&dsched_lock, LK_EXCLUSIVE);
1085
1086         pol = diskctx->dp->d_sched_policy;
1087         memcpy(buf, pol->name, DSCHED_POLICY_NAME_LENGTH);
1088
1089         error = sysctl_handle_string(oidp, buf, DSCHED_POLICY_NAME_LENGTH, req);
1090         if (error || req->newptr == NULL) {
1091                 lockmgr(&dsched_lock, LK_RELEASE);
1092                 return (error);
1093         }
1094
1095         pol = dsched_find_policy(buf);
1096         if (pol == NULL) {
1097                 lockmgr(&dsched_lock, LK_RELEASE);
1098                 return 0;
1099         }
1100
1101         dsched_switch(diskctx->dp, pol);
1102
1103         lockmgr(&dsched_lock, LK_RELEASE);
1104
1105         return error;
1106 }
1107
1108 static int
1109 sysctl_dsched_default_policy(SYSCTL_HANDLER_ARGS)
1110 {
1111         char buf[DSCHED_POLICY_NAME_LENGTH];
1112         struct dsched_policy *pol = NULL;
1113         int error;
1114
1115         lockmgr(&dsched_lock, LK_EXCLUSIVE);
1116
1117         pol = default_policy;
1118         memcpy(buf, pol->name, DSCHED_POLICY_NAME_LENGTH);
1119
1120         error = sysctl_handle_string(oidp, buf, DSCHED_POLICY_NAME_LENGTH, req);
1121         if (error || req->newptr == NULL) {
1122                 lockmgr(&dsched_lock, LK_RELEASE);
1123                 return (error);
1124         }
1125
1126         pol = dsched_find_policy(buf);
1127         if (pol == NULL) {
1128                 lockmgr(&dsched_lock, LK_RELEASE);
1129                 return 0;
1130         }
1131
1132         default_set = 1;
1133         default_policy = pol;
1134
1135         lockmgr(&dsched_lock, LK_RELEASE);
1136
1137         return error;
1138 }
1139
1140 SYSCTL_NODE(, OID_AUTO, dsched, CTLFLAG_RD, NULL,
1141     "Disk Scheduler Framework (dsched) magic");
1142 SYSCTL_NODE(_dsched, OID_AUTO, policy, CTLFLAG_RW, NULL,
1143     "List of disks and their policies");
1144 SYSCTL_INT(_dsched, OID_AUTO, debug, CTLFLAG_RW, &dsched_debug_enable,
1145     0, "Enable dsched debugging");
1146 SYSCTL_PROC(_dsched, OID_AUTO, stats, CTLTYPE_OPAQUE|CTLFLAG_RD,
1147     0, sizeof(struct dsched_stats), sysctl_dsched_stats, "dsched_stats",
1148     "dsched statistics");
1149 SYSCTL_PROC(_dsched, OID_AUTO, policies, CTLTYPE_STRING|CTLFLAG_RD,
1150     NULL, 0, sysctl_dsched_list_policies, "A", "names of available policies");
1151 SYSCTL_PROC(_dsched_policy, OID_AUTO, default, CTLTYPE_STRING|CTLFLAG_RW,
1152     NULL, 0, sysctl_dsched_default_policy, "A", "default dsched policy");
1153
1154 static void
1155 dsched_sysctl_add_disk(struct dsched_disk_ctx *diskctx, char *name)
1156 {
1157         if (!(diskctx->flags & DSCHED_SYSCTL_CTX_INITED)) {
1158                 diskctx->flags |= DSCHED_SYSCTL_CTX_INITED;
1159                 sysctl_ctx_init(&diskctx->sysctl_ctx);
1160         }
1161
1162         SYSCTL_ADD_PROC(&diskctx->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_dsched_policy),
1163             OID_AUTO, name, CTLTYPE_STRING|CTLFLAG_RW,
1164             diskctx, 0, sysctl_dsched_policy, "A", "policy");
1165 }