sys/kern/kern_sysref.c

   1 /*
   2  * Copyright (c) 2007 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.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 /*
  35  * System resource control module for all cluster-addressable system resource
  36  * structures.
  37  *
  38  * This module implements the core ref counting, sysid registration, and
  39  * objcache-backed allocation mechanism for all major system resource
  40  * structures.
  41  *
  42  * sysid registrations operate via the objcache ctor/dtor mechanism and
  43  * sysids will be reused if the resource is not explicitly accessed via
  44  * its sysid.  This removes all RB tree handling overhead from the critical
  45  * path for locally used resources.
  46  */
  47
  48 #include <sys/param.h>
  49 #include <sys/systm.h>
  50 #include <sys/kernel.h>
  51 #include <sys/malloc.h>
  52 #include <sys/tree.h>
  53 #include <sys/spinlock.h>
  54 #include <machine/atomic.h>
  55 #include <machine/cpufunc.h>
  56
  57 #include <sys/spinlock2.h>
  58 #include <sys/sysref2.h>
  59
  60 static boolean_t sysref_ctor(void *data, void *privdata, int ocflags);
  61 static void sysref_dtor(void *data, void *privdata);
  62
  63 /*
  64  * Red-Black tree support
  65  */
  66 static int rb_sysref_compare(struct sysref *sr1, struct sysref *sr2);
  67 RB_GENERATE2(sysref_rb_tree, sysref, rbnode, rb_sysref_compare, sysid_t, sysid);
  68
  69 static struct srpercpu {
  70         struct sysref_rb_tree rbtree;
  71         struct spinlock spin;
  72 } sysref_array[MAXCPU];
  73
  74 static void
  75 sysrefbootinit(void *dummy __unused)
  76 {
  77         struct srpercpu *sa;
  78         int i;
  79
  80         for (i = 0; i < ncpus; ++i) {
  81                 sa = &sysref_array[i];
  82                 spin_init(&sa->spin, "sysrefbootinit");
  83                 RB_INIT(&sa->rbtree);
  84         }
  85 }
  86
  87 SYSINIT(sysref, SI_BOOT2_MACHDEP, SI_ORDER_ANY, sysrefbootinit, NULL);
  88
  89 static
  90 int
  91 rb_sysref_compare(struct sysref *sr1, struct sysref *sr2)
  92 {
  93         if (sr1->sysid < sr2->sysid)
  94                 return(-1);
  95         if (sr1->sysid > sr2->sysid)
  96                 return(1);
  97         return(0);
  98 }
  99
 100 /*
 101  * Manual initialization of a resource structure's sysref, only used during
 102  * booting to set up certain statically declared resources which cannot
 103  * be deallocated.
 104  */
 105 void
 106 sysref_init(struct sysref *sr, struct sysref_class *srclass)
 107 {
 108         struct srpercpu *sa;
 109         globaldata_t gd;
 110
 111         gd = mycpu;
 112         crit_enter_gd(gd);
 113         gd->gd_sysid_alloc += ncpus_fit; /* next unique sysid */
 114         sr->sysid = gd->gd_sysid_alloc;
 115         KKASSERT(((int)sr->sysid & ncpus_fit_mask) == gd->gd_cpuid);
 116         sr->refcnt = -0x40000000;
 117         sr->flags = 0;
 118         sr->srclass = srclass;
 119
 120         sa = &sysref_array[gd->gd_cpuid];
 121         spin_lock(&sa->spin);
 122         sysref_rb_tree_RB_INSERT(&sa->rbtree, sr);
 123         spin_unlock(&sa->spin);
 124         crit_exit_gd(gd);
 125 }
 126
 127 /*
 128  * Allocate a resource structure of the specified class, initialize a
 129  * sysid and add the resource to the RB tree.  The caller must complete
 130  * initialization of the resource and call sysref_activate() to activate it.
 131  */
 132 void *
 133 sysref_alloc(struct sysref_class *srclass)
 134 {
 135         struct sysref *sr;
 136         char *data;
 137         int n;
 138
 139         /*
 140          * Create the object cache backing store.
 141          */
 142         if (srclass->oc == NULL) {
 143                 KKASSERT(srclass->mtype != NULL);
 144                 srclass->oc = objcache_create_mbacked(
 145                                 srclass->mtype, srclass->objsize,
 146                                 0, srclass->nom_cache,
 147                                 sysref_ctor, sysref_dtor, srclass);
 148         }
 149
 150         /*
 151          * Allocate the resource.
 152          */
 153         data = objcache_get(srclass->oc, M_WAITOK);
 154         sr = (struct sysref *)(data + srclass->offset);
 155         KKASSERT(sr->flags & SRF_PUTAWAY);
 156         sr->flags &= ~SRF_PUTAWAY;
 157
 158         /*
 159          * Refcnt isn't touched while it is zero.  The objcache ctor
 160          * function has already allocated a sysid and emplaced the
 161          * structure in the RB tree.
 162          */
 163         KKASSERT(sr->refcnt == 0);
 164         sr->refcnt = -0x40000000;
 165
 166         /*
 167          * Clean out the structure unless the caller wants to deal with
 168          * it (e.g. like the vmspace code).
 169          */
 170         if ((srclass->flags & SRC_MANAGEDINIT) == 0) {
 171                 if (srclass->offset != 0)
 172                         bzero(data, srclass->offset);
 173                 n = srclass->offset + sizeof(struct sysref);
 174                 KKASSERT(n <= srclass->objsize);
 175                 if (n != srclass->objsize)
 176                         bzero(data + n, srclass->objsize - n);
 177         }
 178         return(data);
 179 }
 180
 181 /*
 182  * Object cache backing store ctor function.
 183  *
 184  * This allocates the sysid and associates the structure with the
 185  * red-black tree, allowing it to be looked up.  The actual resource
 186  * structure has NOT yet been allocated so it is marked free.
 187  *
 188  * If the sysid is not used to access the resource, we will just
 189  * allow the sysid to be reused when the resource structure is reused,
 190  * allowing the RB tree operation to be 'cached'.  This results in
 191  * virtually no performance penalty for using the sysref facility.
 192  */
 193 static
 194 boolean_t
 195 sysref_ctor(void *data, void *privdata, int ocflags)
 196 {
 197         globaldata_t gd;
 198         struct srpercpu *sa;
 199         struct sysref_class *srclass = privdata;
 200         struct sysref *sr = (void *)((char *)data + srclass->offset);
 201
 202         /*
 203          * Resource structures need to be cleared when allocating from
 204          * malloc backing store.  This is different from the zeroing
 205          * that we do in sysref_alloc().
 206          */
 207         bzero(data, srclass->objsize);
 208
 209         /*
 210          * Resources managed by our objcache do the sysid and RB tree
 211          * handling in the objcache ctor/dtor, so we can reuse the
 212          * structure without re-treeing it over and over again.
 213          */
 214         gd = mycpu;
 215         crit_enter_gd(gd);
 216         gd->gd_sysid_alloc += ncpus_fit; /* next unique sysid */
 217         sr->sysid = gd->gd_sysid_alloc;
 218         KKASSERT(((int)sr->sysid & ncpus_fit_mask) == gd->gd_cpuid);
 219         /* sr->refcnt= 0; already zero */
 220         sr->flags = SRF_ALLOCATED | SRF_PUTAWAY;
 221         sr->srclass = srclass;
 222
 223         sa = &sysref_array[gd->gd_cpuid];
 224         spin_lock(&sa->spin);
 225         sysref_rb_tree_RB_INSERT(&sa->rbtree, sr);
 226         spin_unlock(&sa->spin);
 227         crit_exit_gd(gd);
 228
 229         /*
 230          * Execute the class's ctor function, if any.  NOTE: The class
 231          * should not try to zero out the structure, we've already handled
 232          * that and preinitialized the sysref.
 233          *
 234          * XXX ignores return value for now
 235          */
 236         if (srclass->ctor)
 237                 srclass->ctor(data, privdata, ocflags);
 238         return TRUE;
 239 }
 240
 241 /*
 242  * Object cache destructor, allowing the structure to be returned
 243  * to the system memory pool.  The resource structure must be
 244  * removed from the RB tree.  All other references have already
 245  * been destroyed and the RB tree will not create any new references
 246  * to the structure in its current state.
 247  */
 248 static
 249 void
 250 sysref_dtor(void *data, void *privdata)
 251 {
 252         struct srpercpu *sa;
 253         struct sysref_class *srclass = privdata;
 254         struct sysref *sr = (void *)((char *)data + srclass->offset);
 255
 256         KKASSERT(sr->refcnt == 0);
 257         sa = &sysref_array[(int)sr->sysid & ncpus_fit_mask];
 258         spin_lock(&sa->spin);
 259         sysref_rb_tree_RB_REMOVE(&sa->rbtree, sr);
 260         spin_unlock(&sa->spin);
 261         if (srclass->dtor)
 262                 srclass->dtor(data, privdata);
 263 }
 264
 265 /*
 266  * Activate or reactivate a resource. 0x40000001 is added to the ref count
 267  * so -0x40000000 (during initialization) will translate to a ref count of 1.
 268  * Any references made during initialization will translate to additional
 269  * positive ref counts.
 270  *
 271  * MPSAFE
 272  */
 273 void
 274 sysref_activate(struct sysref *sr)
 275 {
 276         int count;
 277
 278         for (;;) {
 279                 count = sr->refcnt;
 280                 KASSERT(count < 0 && count + 0x40000001 > 0,
 281                         ("sysref_activate: bad count %08x", count));
 282                 if (atomic_cmpset_int(&sr->refcnt, count, count + 0x40000001))
 283                         break;
 284                 cpu_pause();
 285         }
 286 }
 287
 288 /*
 289  * Release a reference under special circumstances.  This call is made
 290  * from the sysref_put() inline from sys/sysref2.h for any 1->0 transitions,
 291  * negative->negative 'termination in progress' transitions, and when the
 292  * cmpset instruction fails during a normal transition.
 293  *
 294  * This function is called from the sysref_put() inline in sys/sysref2.h,
 295  * but handles all cases regardless.
 296  */
 297 void
 298 _sysref_put(struct sysref *sr)
 299 {
 300         int count;
 301         void *data;
 302
 303         KKASSERT((sr->flags & SRF_PUTAWAY) == 0);
 304
 305         for (;;) {
 306                 count = sr->refcnt;
 307                 if (count > 1) {
 308                         /*
 309                          * release 1 count, nominal case, active resource
 310                          * structure, no other action required.
 311                          */
 312                         if (atomic_cmpset_int(&sr->refcnt, count, count - 1))
 313                                 break;
 314                 } else if (count == 1) {
 315                         /*
 316                          * 1->0 transitions transition to -0x40000000 instead,
 317                          * placing the resource structure into a termination-
 318                          * in-progress state.  The termination function is
 319                          * then called.
 320                          */
 321                         data = (char *)sr - sr->srclass->offset;
 322                         sr->srclass->ops.lock(data);
 323                         if (atomic_cmpset_int(&sr->refcnt, count, -0x40000000)) {
 324                                 sr->srclass->ops.terminate(data);
 325                                 /* callback unlocks */
 326                                 break;
 327                         }
 328                         sr->srclass->ops.unlock(data);
 329                 } else if (count > -0x40000000) {
 330                         /*
 331                          * release 1 count, nominal case, resource undergoing
 332                          * termination.  The Resource can be ref'd and
 333                          * deref'd while undergoing termination.
 334                          */
 335                         if (atomic_cmpset_int(&sr->refcnt, count, count - 1))
 336                                 break;
 337                 } else {
 338                         /*
 339                          * Final release, set refcnt to 0.
 340                          * Resource must have been allocated.
 341                          *
 342                          * If SRF_SYSIDUSED is not set just objcache_put() the
 343                          * resource, otherwise objcache_dtor() the resource.
 344                          */
 345                         KKASSERT(count == -0x40000000);
 346                         if (atomic_cmpset_int(&sr->refcnt, count, 0)) {
 347                                 KKASSERT(sr->flags & SRF_ALLOCATED);
 348                                 sr->flags |= SRF_PUTAWAY;
 349                                 data = (char *)sr - sr->srclass->offset;
 350                                 if (sr->flags & SRF_SYSIDUSED)
 351                                         objcache_dtor(sr->srclass->oc, data);
 352                                 else
 353                                         objcache_put(sr->srclass->oc, data);
 354                                 break;
 355                         }
 356                 }
 357                 /* loop until the cmpset succeeds */
 358                 cpu_pause();
 359         }
 360 }
 361
 362 sysid_t
 363 allocsysid(void)
 364 {
 365         globaldata_t gd = mycpu;
 366         sysid_t sysid;
 367
 368         crit_enter_gd(gd);
 369         gd->gd_sysid_alloc += ncpus_fit;
 370         sysid = gd->gd_sysid_alloc;
 371         crit_exit_gd(gd);
 372         return(sysid);
 373 }
 374