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