[BACK]Return to subr_pool.c CVS log [TXT][DIR] Up to [cvs.NetBSD.org] / src / sys / kern

Please note that diffs are not public domain; they are subject to the copyright notices on the relevant files.

Diff for /src/sys/kern/subr_pool.c between version 1.2 and 1.84

version 1.2, 1998/02/19 23:52:14 version 1.84, 2003/01/18 10:06:33
Line 1 
Line 1 
 /*      $NetBSD$        */  /*      $NetBSD$        */
   
 /*-  /*-
  * Copyright (c) 1997 The NetBSD Foundation, Inc.   * Copyright (c) 1997, 1999, 2000 The NetBSD Foundation, Inc.
  * All rights reserved.   * All rights reserved.
  *   *
  * This code is derived from software contributed to The NetBSD Foundation   * This code is derived from software contributed to The NetBSD Foundation
  * by Paul Kranenburg.   * by Paul Kranenburg; by Jason R. Thorpe of the Numerical Aerospace
    * Simulation Facility, NASA Ames Research Center.
  *   *
  * Redistribution and use in source and binary forms, with or without   * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions   * modification, are permitted provided that the following conditions
Line 17 
Line 18 
  *    documentation and/or other materials provided with the distribution.   *    documentation and/or other materials provided with the distribution.
  * 3. All advertising materials mentioning features or use of this software   * 3. All advertising materials mentioning features or use of this software
  *    must display the following acknowledgement:   *    must display the following acknowledgement:
  *        This product includes software developed by the NetBSD   *      This product includes software developed by the NetBSD
  *        Foundation, Inc. and its contributors.   *      Foundation, Inc. and its contributors.
  * 4. Neither the name of The NetBSD Foundation nor the names of its   * 4. Neither the name of The NetBSD Foundation nor the names of its
  *    contributors may be used to endorse or promote products derived   *    contributors may be used to endorse or promote products derived
  *    from this software without specific prior written permission.   *    from this software without specific prior written permission.
Line 36 
Line 37 
  * POSSIBILITY OF SUCH DAMAGE.   * POSSIBILITY OF SUCH DAMAGE.
  */   */
   
   #include <sys/cdefs.h>
   __KERNEL_RCSID(0, "$NetBSD$");
   
   #include "opt_pool.h"
   #include "opt_poollog.h"
   #include "opt_lockdebug.h"
   
 #include <sys/param.h>  #include <sys/param.h>
 #include <sys/systm.h>  #include <sys/systm.h>
 #include <sys/proc.h>  #include <sys/proc.h>
Line 44 
Line 52 
 #include <sys/malloc.h>  #include <sys/malloc.h>
 #include <sys/lock.h>  #include <sys/lock.h>
 #include <sys/pool.h>  #include <sys/pool.h>
   #include <sys/syslog.h>
   
   #include <uvm/uvm.h>
   
 /*  /*
  * Pool resource management utility.   * Pool resource management utility.
    *
    * Memory is allocated in pages which are split into pieces according
    * to the pool item size. Each page is kept on a list headed by `pr_pagelist'
    * in the pool structure and the individual pool items are on a linked list
    * headed by `ph_itemlist' in each page header. The memory for building
    * the page list is either taken from the allocated pages themselves (for
    * small pool items) or taken from an internal pool of page headers (`phpool').
  */   */
   
   /* List of all pools */
   TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
   
   /* Private pool for page header structures */
   static struct pool phpool;
   
   #ifdef POOL_SUBPAGE
   /* Pool of subpages for use by normal pools. */
   static struct pool psppool;
   #endif
   
   /* # of seconds to retain page after last use */
   int pool_inactive_time = 10;
   
   /* Next candidate for drainage (see pool_drain()) */
   static struct pool      *drainpp;
   
   /* This spin lock protects both pool_head and drainpp. */
   struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER;
   
   struct pool_item_header {
           /* Page headers */
           TAILQ_ENTRY(pool_item_header)
                                   ph_pagelist;    /* pool page list */
           TAILQ_HEAD(,pool_item)  ph_itemlist;    /* chunk list for this page */
           LIST_ENTRY(pool_item_header)
                                   ph_hashlist;    /* Off-page page headers */
           unsigned int            ph_nmissing;    /* # of chunks in use */
           caddr_t                 ph_page;        /* this page's address */
           struct timeval          ph_time;        /* last referenced */
   };
   TAILQ_HEAD(pool_pagelist,pool_item_header);
   
 struct pool_item {  struct pool_item {
         struct pool_item        *pi_next;  #ifdef DIAGNOSTIC
           u_int pi_magic;
   #endif
   #define PI_MAGIC 0xdeadbeefU
           /* Other entries use only this list entry */
           TAILQ_ENTRY(pool_item)  pi_list;
   };
   
   #define PR_HASH_INDEX(pp,addr) \
           (((u_long)(addr) >> (pp)->pr_alloc->pa_pageshift) & \
            (PR_HASHTABSIZE - 1))
   
   #define POOL_NEEDS_CATCHUP(pp)                                          \
           ((pp)->pr_nitems < (pp)->pr_minitems)
   
   /*
    * Pool cache management.
    *
    * Pool caches provide a way for constructed objects to be cached by the
    * pool subsystem.  This can lead to performance improvements by avoiding
    * needless object construction/destruction; it is deferred until absolutely
    * necessary.
    *
    * Caches are grouped into cache groups.  Each cache group references
    * up to 16 constructed objects.  When a cache allocates an object
    * from the pool, it calls the object's constructor and places it into
    * a cache group.  When a cache group frees an object back to the pool,
    * it first calls the object's destructor.  This allows the object to
    * persist in constructed form while freed to the cache.
    *
    * Multiple caches may exist for each pool.  This allows a single
    * object type to have multiple constructed forms.  The pool references
    * each cache, so that when a pool is drained by the pagedaemon, it can
    * drain each individual cache as well.  Each time a cache is drained,
    * the most idle cache group is freed to the pool in its entirety.
    *
    * Pool caches are layed on top of pools.  By layering them, we can avoid
    * the complexity of cache management for pools which would not benefit
    * from it.
    */
   
   /* The cache group pool. */
   static struct pool pcgpool;
   
   static void     pool_cache_reclaim(struct pool_cache *);
   
   static int      pool_catchup(struct pool *);
   static void     pool_prime_page(struct pool *, caddr_t,
                       struct pool_item_header *);
   
   void            *pool_allocator_alloc(struct pool *, int);
   void            pool_allocator_free(struct pool *, void *);
   
   static void pool_print1(struct pool *, const char *,
           void (*)(const char *, ...));
   
   /*
    * Pool log entry. An array of these is allocated in pool_init().
    */
   struct pool_log {
           const char      *pl_file;
           long            pl_line;
           int             pl_action;
   #define PRLOG_GET       1
   #define PRLOG_PUT       2
           void            *pl_addr;
 };  };
   
   /* Number of entries in pool log buffers */
   #ifndef POOL_LOGSIZE
   #define POOL_LOGSIZE    10
   #endif
   
   int pool_logsize = POOL_LOGSIZE;
   
 struct pool *  #ifdef POOL_DIAGNOSTIC
 pool_create(size, nitems, wchan, mtype, storage)  static __inline void
         size_t  size;  pr_log(struct pool *pp, void *v, int action, const char *file, long line)
         int     nitems;  
         char    *wchan;  
         int     mtype;  
         caddr_t storage;  
 {  {
         struct pool *pp;          int n = pp->pr_curlogentry;
         caddr_t cp;          struct pool_log *pl;
   
         if (size < sizeof(struct pool_item)) {          if ((pp->pr_roflags & PR_LOGGING) == 0)
                 printf("pool_create: size %lu too small\n", (u_long)size);                  return;
                 return (NULL);  
           /*
            * Fill in the current entry. Wrap around and overwrite
            * the oldest entry if necessary.
            */
           pl = &pp->pr_log[n];
           pl->pl_file = file;
           pl->pl_line = line;
           pl->pl_action = action;
           pl->pl_addr = v;
           if (++n >= pp->pr_logsize)
                   n = 0;
           pp->pr_curlogentry = n;
   }
   
   static void
   pr_printlog(struct pool *pp, struct pool_item *pi,
       void (*pr)(const char *, ...))
   {
           int i = pp->pr_logsize;
           int n = pp->pr_curlogentry;
   
           if ((pp->pr_roflags & PR_LOGGING) == 0)
                   return;
   
           /*
            * Print all entries in this pool's log.
            */
           while (i-- > 0) {
                   struct pool_log *pl = &pp->pr_log[n];
                   if (pl->pl_action != 0) {
                           if (pi == NULL || pi == pl->pl_addr) {
                                   (*pr)("\tlog entry %d:\n", i);
                                   (*pr)("\t\taction = %s, addr = %p\n",
                                       pl->pl_action == PRLOG_GET ? "get" : "put",
                                       pl->pl_addr);
                                   (*pr)("\t\tfile: %s at line %lu\n",
                                       pl->pl_file, pl->pl_line);
                           }
                   }
                   if (++n >= pp->pr_logsize)
                           n = 0;
           }
   }
   
   static __inline void
   pr_enter(struct pool *pp, const char *file, long line)
   {
   
           if (__predict_false(pp->pr_entered_file != NULL)) {
                   printf("pool %s: reentrancy at file %s line %ld\n",
                       pp->pr_wchan, file, line);
                   printf("         previous entry at file %s line %ld\n",
                       pp->pr_entered_file, pp->pr_entered_line);
                   panic("pr_enter");
           }
   
           pp->pr_entered_file = file;
           pp->pr_entered_line = line;
   }
   
   static __inline void
   pr_leave(struct pool *pp)
   {
   
           if (__predict_false(pp->pr_entered_file == NULL)) {
                   printf("pool %s not entered?\n", pp->pr_wchan);
                   panic("pr_leave");
           }
   
           pp->pr_entered_file = NULL;
           pp->pr_entered_line = 0;
   }
   
   static __inline void
   pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
   {
   
           if (pp->pr_entered_file != NULL)
                   (*pr)("\n\tcurrently entered from file %s line %ld\n",
                       pp->pr_entered_file, pp->pr_entered_line);
   }
   #else
   #define pr_log(pp, v, action, file, line)
   #define pr_printlog(pp, pi, pr)
   #define pr_enter(pp, file, line)
   #define pr_leave(pp)
   #define pr_enter_check(pp, pr)
   #endif /* POOL_DIAGNOSTIC */
   
   /*
    * Return the pool page header based on page address.
    */
   static __inline struct pool_item_header *
   pr_find_pagehead(struct pool *pp, caddr_t page)
   {
           struct pool_item_header *ph;
   
           if ((pp->pr_roflags & PR_PHINPAGE) != 0)
                   return ((struct pool_item_header *)(page + pp->pr_phoffset));
   
           for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]);
                ph != NULL;
                ph = LIST_NEXT(ph, ph_hashlist)) {
                   if (ph->ph_page == page)
                           return (ph);
           }
           return (NULL);
   }
   
   /*
    * Remove a page from the pool.
    */
   static __inline void
   pr_rmpage(struct pool *pp, struct pool_item_header *ph,
        struct pool_pagelist *pq)
   {
           int s;
   
           /*
            * If the page was idle, decrement the idle page count.
            */
           if (ph->ph_nmissing == 0) {
   #ifdef DIAGNOSTIC
                   if (pp->pr_nidle == 0)
                           panic("pr_rmpage: nidle inconsistent");
                   if (pp->pr_nitems < pp->pr_itemsperpage)
                           panic("pr_rmpage: nitems inconsistent");
   #endif
                   pp->pr_nidle--;
         }          }
   
         if (storage) {          pp->pr_nitems -= pp->pr_itemsperpage;
                 pp = (struct pool *)storage;  
                 cp = (caddr_t)ALIGN(pp + 1);          /*
            * Unlink a page from the pool and release it (or queue it for release).
            */
           TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
           if (pq) {
                   TAILQ_INSERT_HEAD(pq, ph, ph_pagelist);
         } else {          } else {
                 pp = (struct pool *)malloc(sizeof(*pp), mtype, M_NOWAIT);                  pool_allocator_free(pp, ph->ph_page);
                 if (pp == NULL)                  if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                         return (NULL);                          LIST_REMOVE(ph, ph_hashlist);
                 cp = NULL;                          s = splhigh();
                           pool_put(&phpool, ph);
                           splx(s);
                   }
         }          }
           pp->pr_npages--;
           pp->pr_npagefree++;
   
         pp->pr_freelist = NULL;          if (pp->pr_curpage == ph) {
         pp->pr_freecount = 0;                  /*
         pp->pr_hiwat = 0;                   * Find a new non-empty page header, if any.
         pp->pr_flags = (storage ? PR_STATIC : 0);                   * Start search from the page head, to increase the
                    * chance for "high water" pages to be freed.
                    */
                   TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist)
                           if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                                   break;
   
                   pp->pr_curpage = ph;
           }
   }
   
   /*
    * Initialize the given pool resource structure.
    *
    * We export this routine to allow other kernel parts to declare
    * static pools that must be initialized before malloc() is available.
    */
   void
   pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
       const char *wchan, struct pool_allocator *palloc)
   {
           int off, slack, i;
   
   #ifdef POOL_DIAGNOSTIC
           /*
            * Always log if POOL_DIAGNOSTIC is defined.
            */
           if (pool_logsize != 0)
                   flags |= PR_LOGGING;
   #endif
   
   #ifdef POOL_SUBPAGE
           /*
            * XXX We don't provide a real `nointr' back-end
            * yet; all sub-pages come from a kmem back-end.
            * maybe some day...
            */
           if (palloc == NULL) {
                   extern struct pool_allocator pool_allocator_kmem_subpage;
                   palloc = &pool_allocator_kmem_subpage;
           }
           /*
            * We'll assume any user-specified back-end allocator
            * will deal with sub-pages, or simply don't care.
            */
   #else
           if (palloc == NULL)
                   palloc = &pool_allocator_kmem;
   #endif /* POOL_SUBPAGE */
           if ((palloc->pa_flags & PA_INITIALIZED) == 0) {
                   if (palloc->pa_pagesz == 0) {
   #ifdef POOL_SUBPAGE
                           if (palloc == &pool_allocator_kmem)
                                   palloc->pa_pagesz = PAGE_SIZE;
                           else
                                   palloc->pa_pagesz = POOL_SUBPAGE;
   #else
                           palloc->pa_pagesz = PAGE_SIZE;
   #endif /* POOL_SUBPAGE */
                   }
   
                   TAILQ_INIT(&palloc->pa_list);
   
                   simple_lock_init(&palloc->pa_slock);
                   palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                   palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
                   palloc->pa_flags |= PA_INITIALIZED;
           }
   
           if (align == 0)
                   align = ALIGN(1);
   
           if (size < sizeof(struct pool_item))
                   size = sizeof(struct pool_item);
   
           size = roundup(size, align);
   #ifdef DIAGNOSTIC
           if (size > palloc->pa_pagesz)
                   panic("pool_init: pool item size (%lu) too large",
                         (u_long)size);
   #endif
   
           /*
            * Initialize the pool structure.
            */
           TAILQ_INIT(&pp->pr_pagelist);
           TAILQ_INIT(&pp->pr_cachelist);
           pp->pr_curpage = NULL;
           pp->pr_npages = 0;
           pp->pr_minitems = 0;
           pp->pr_minpages = 0;
           pp->pr_maxpages = UINT_MAX;
           pp->pr_roflags = flags;
           pp->pr_flags = 0;
         pp->pr_size = size;          pp->pr_size = size;
           pp->pr_align = align;
         pp->pr_wchan = wchan;          pp->pr_wchan = wchan;
         pp->pr_mtype = mtype;          pp->pr_alloc = palloc;
         simple_lock_init(&pp->pr_lock);          pp->pr_nitems = 0;
           pp->pr_nout = 0;
         if (nitems != 0) {          pp->pr_hardlimit = UINT_MAX;
                 if (pool_prime(pp, nitems, cp) != 0) {          pp->pr_hardlimit_warning = NULL;
                         pool_destroy(pp);          pp->pr_hardlimit_ratecap.tv_sec = 0;
                         return (NULL);          pp->pr_hardlimit_ratecap.tv_usec = 0;
           pp->pr_hardlimit_warning_last.tv_sec = 0;
           pp->pr_hardlimit_warning_last.tv_usec = 0;
           pp->pr_drain_hook = NULL;
           pp->pr_drain_hook_arg = NULL;
   
           /*
            * Decide whether to put the page header off page to avoid
            * wasting too large a part of the page. Off-page page headers
            * go on a hash table, so we can match a returned item
            * with its header based on the page address.
            * We use 1/16 of the page size as the threshold (XXX: tune)
            */
           if (pp->pr_size < palloc->pa_pagesz/16) {
                   /* Use the end of the page for the page header */
                   pp->pr_roflags |= PR_PHINPAGE;
                   pp->pr_phoffset = off = palloc->pa_pagesz -
                       ALIGN(sizeof(struct pool_item_header));
           } else {
                   /* The page header will be taken from our page header pool */
                   pp->pr_phoffset = 0;
                   off = palloc->pa_pagesz;
                   for (i = 0; i < PR_HASHTABSIZE; i++) {
                           LIST_INIT(&pp->pr_hashtab[i]);
                 }                  }
         }          }
   
         return (pp);          /*
            * Alignment is to take place at `ioff' within the item. This means
            * we must reserve up to `align - 1' bytes on the page to allow
            * appropriate positioning of each item.
            *
            * Silently enforce `0 <= ioff < align'.
            */
           pp->pr_itemoffset = ioff = ioff % align;
           pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
           KASSERT(pp->pr_itemsperpage != 0);
   
           /*
            * Use the slack between the chunks and the page header
            * for "cache coloring".
            */
           slack = off - pp->pr_itemsperpage * pp->pr_size;
           pp->pr_maxcolor = (slack / align) * align;
           pp->pr_curcolor = 0;
   
           pp->pr_nget = 0;
           pp->pr_nfail = 0;
           pp->pr_nput = 0;
           pp->pr_npagealloc = 0;
           pp->pr_npagefree = 0;
           pp->pr_hiwat = 0;
           pp->pr_nidle = 0;
   
   #ifdef POOL_DIAGNOSTIC
           if (flags & PR_LOGGING) {
                   if (kmem_map == NULL ||
                       (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                        M_TEMP, M_NOWAIT)) == NULL)
                           pp->pr_roflags &= ~PR_LOGGING;
                   pp->pr_curlogentry = 0;
                   pp->pr_logsize = pool_logsize;
           }
   #endif
   
           pp->pr_entered_file = NULL;
           pp->pr_entered_line = 0;
   
           simple_lock_init(&pp->pr_slock);
   
           /*
            * Initialize private page header pool and cache magazine pool if we
            * haven't done so yet.
            * XXX LOCKING.
            */
           if (phpool.pr_size == 0) {
   #ifdef POOL_SUBPAGE
                   pool_init(&phpool, sizeof(struct pool_item_header), 0, 0, 0,
                       "phpool", &pool_allocator_kmem);
                   pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
                       PR_RECURSIVE, "psppool", &pool_allocator_kmem);
   #else
                   pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
                       0, "phpool", NULL);
   #endif
                   pool_init(&pcgpool, sizeof(struct pool_cache_group), 0, 0,
                       0, "pcgpool", NULL);
           }
   
           /* Insert into the list of all pools. */
           simple_lock(&pool_head_slock);
           TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
           simple_unlock(&pool_head_slock);
   
           /* Insert this into the list of pools using this allocator. */
           simple_lock(&palloc->pa_slock);
           TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
           simple_unlock(&palloc->pa_slock);
 }  }
   
 /*  /*
  * De-commision a pool resource.   * De-commision a pool resource.
  */   */
 void  void
 pool_destroy(pp)  pool_destroy(struct pool *pp)
         struct pool *pp;  
 {  {
         struct pool_item *pi;          struct pool_item_header *ph;
           struct pool_cache *pc;
   
         if (pp->pr_flags & PR_STATIC)          /* Locking order: pool_allocator -> pool */
                 return;          simple_lock(&pp->pr_alloc->pa_slock);
           TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
           simple_unlock(&pp->pr_alloc->pa_slock);
   
           /* Destroy all caches for this pool. */
           while ((pc = TAILQ_FIRST(&pp->pr_cachelist)) != NULL)
                   pool_cache_destroy(pc);
   
   #ifdef DIAGNOSTIC
           if (pp->pr_nout != 0) {
                   pr_printlog(pp, NULL, printf);
                   panic("pool_destroy: pool busy: still out: %u",
                       pp->pr_nout);
           }
   #endif
   
         while ((pi = pp->pr_freelist) != NULL) {          /* Remove all pages */
                 pp->pr_freelist = pi->pi_next;          while ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
                 free(pi, pp->pr_mtype);                  pr_rmpage(pp, ph, NULL);
   
           /* Remove from global pool list */
           simple_lock(&pool_head_slock);
           TAILQ_REMOVE(&pool_head, pp, pr_poollist);
           if (drainpp == pp) {
                   drainpp = NULL;
         }          }
         free(pp, pp->pr_mtype);          simple_unlock(&pool_head_slock);
   
   #ifdef POOL_DIAGNOSTIC
           if ((pp->pr_roflags & PR_LOGGING) != 0)
                   free(pp->pr_log, M_TEMP);
   #endif
   }
   
   void
   pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
   {
   
           /* XXX no locking -- must be used just after pool_init() */
   #ifdef DIAGNOSTIC
           if (pp->pr_drain_hook != NULL)
                   panic("pool_set_drain_hook(%s): already set", pp->pr_wchan);
   #endif
           pp->pr_drain_hook = fn;
           pp->pr_drain_hook_arg = arg;
 }  }
   
   static __inline struct pool_item_header *
   pool_alloc_item_header(struct pool *pp, caddr_t storage, int flags)
   {
           struct pool_item_header *ph;
           int s;
   
           LOCK_ASSERT(simple_lock_held(&pp->pr_slock) == 0);
   
           if ((pp->pr_roflags & PR_PHINPAGE) != 0)
                   ph = (struct pool_item_header *) (storage + pp->pr_phoffset);
           else {
                   s = splhigh();
                   ph = pool_get(&phpool, flags);
                   splx(s);
           }
   
           return (ph);
   }
   
 /*  /*
  * Grab an item from the pool; must be called at splbio   * Grab an item from the pool; must be called at appropriate spl level
  */   */
 void *  void *
 pool_get(pp, flags)  #ifdef POOL_DIAGNOSTIC
         struct pool *pp;  _pool_get(struct pool *pp, int flags, const char *file, long line)
         int flags;  #else
   pool_get(struct pool *pp, int flags)
   #endif
 {  {
         void *v;  
         struct pool_item *pi;          struct pool_item *pi;
           struct pool_item_header *ph;
           void *v;
   
   #ifdef DIAGNOSTIC
           if (__predict_false(curlwp == NULL && doing_shutdown == 0 &&
                               (flags & PR_WAITOK) != 0))
                   panic("pool_get: %s: must have NOWAIT", pp->pr_wchan);
   
   #ifdef LOCKDEBUG
           if (flags & PR_WAITOK)
                   simple_lock_only_held(NULL, "pool_get(PR_WAITOK)");
   #endif
   #endif /* DIAGNOSTIC */
   
           simple_lock(&pp->pr_slock);
           pr_enter(pp, file, line);
   
    startover:
           /*
            * Check to see if we've reached the hard limit.  If we have,
            * and we can wait, then wait until an item has been returned to
            * the pool.
            */
 #ifdef DIAGNOSTIC  #ifdef DIAGNOSTIC
         if ((pp->pr_flags & PR_STATIC) && (flags & PR_MALLOCOK))          if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
                 panic("pool_get: static");                  pr_leave(pp);
                   simple_unlock(&pp->pr_slock);
                   panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
           }
 #endif  #endif
           if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
                   if (pp->pr_drain_hook != NULL) {
                           /*
                            * Since the drain hook is going to free things
                            * back to the pool, unlock, call the hook, re-lock,
                            * and check the hardlimit condition again.
                            */
                           pr_leave(pp);
                           simple_unlock(&pp->pr_slock);
                           (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                           simple_lock(&pp->pr_slock);
                           pr_enter(pp, file, line);
                           if (pp->pr_nout < pp->pr_hardlimit)
                                   goto startover;
                   }
   
                   if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
                           /*
                            * XXX: A warning isn't logged in this case.  Should
                            * it be?
                            */
                           pp->pr_flags |= PR_WANTED;
                           pr_leave(pp);
                           ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
                           pr_enter(pp, file, line);
                           goto startover;
                   }
   
                   /*
                    * Log a message that the hard limit has been hit.
                    */
                   if (pp->pr_hardlimit_warning != NULL &&
                       ratecheck(&pp->pr_hardlimit_warning_last,
                                 &pp->pr_hardlimit_ratecap))
                           log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
   
                   pp->pr_nfail++;
   
                   pr_leave(pp);
                   simple_unlock(&pp->pr_slock);
                   return (NULL);
           }
   
 again:          /*
         simple_lock(&pp->pr_lock);           * The convention we use is that if `curpage' is not NULL, then
         if ((v = pp->pr_freelist) == NULL) {           * it points at a non-empty bucket. In particular, `curpage'
                 if (flags & PR_MALLOCOK)           * never points at a page header which has PR_PHINPAGE set and
                         v = (void *)malloc(pp->pr_size, pp->pr_mtype, M_NOWAIT);           * has no items in its bucket.
            */
           if ((ph = pp->pr_curpage) == NULL) {
   #ifdef DIAGNOSTIC
                   if (pp->pr_nitems != 0) {
                           simple_unlock(&pp->pr_slock);
                           printf("pool_get: %s: curpage NULL, nitems %u\n",
                               pp->pr_wchan, pp->pr_nitems);
                           panic("pool_get: nitems inconsistent");
                   }
   #endif
   
                 if (v == NULL) {                  /*
                         if ((flags & PR_WAITOK) == 0)                   * Call the back-end page allocator for more memory.
                    * Release the pool lock, as the back-end page allocator
                    * may block.
                    */
                   pr_leave(pp);
                   simple_unlock(&pp->pr_slock);
                   v = pool_allocator_alloc(pp, flags);
                   if (__predict_true(v != NULL))
                           ph = pool_alloc_item_header(pp, v, flags);
                   simple_lock(&pp->pr_slock);
                   pr_enter(pp, file, line);
   
                   if (__predict_false(v == NULL || ph == NULL)) {
                           if (v != NULL)
                                   pool_allocator_free(pp, v);
   
                           /*
                            * We were unable to allocate a page or item
                            * header, but we released the lock during
                            * allocation, so perhaps items were freed
                            * back to the pool.  Check for this case.
                            */
                           if (pp->pr_curpage != NULL)
                                   goto startover;
   
                           if ((flags & PR_WAITOK) == 0) {
                                   pp->pr_nfail++;
                                   pr_leave(pp);
                                   simple_unlock(&pp->pr_slock);
                                 return (NULL);                                  return (NULL);
                           }
   
                           /*
                            * Wait for items to be returned to this pool.
                            *
                            * XXX: maybe we should wake up once a second and
                            * try again?
                            */
                         pp->pr_flags |= PR_WANTED;                          pp->pr_flags |= PR_WANTED;
                         simple_unlock(&pp->pr_lock);                          /* PA_WANTED is already set on the allocator. */
                         tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0);                          pr_leave(pp);
                         goto again;                          ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
                           pr_enter(pp, file, line);
                           goto startover;
                 }                  }
         } else {  
                 pi = v;                  /* We have more memory; add it to the pool */
                 pp->pr_freelist = pi->pi_next;                  pool_prime_page(pp, v, ph);
                 pp->pr_freecount--;                  pp->pr_npagealloc++;
   
                   /* Start the allocation process over. */
                   goto startover;
           }
   
           if (__predict_false((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)) {
                   pr_leave(pp);
                   simple_unlock(&pp->pr_slock);
                   panic("pool_get: %s: page empty", pp->pr_wchan);
           }
   #ifdef DIAGNOSTIC
           if (__predict_false(pp->pr_nitems == 0)) {
                   pr_leave(pp);
                   simple_unlock(&pp->pr_slock);
                   printf("pool_get: %s: items on itemlist, nitems %u\n",
                       pp->pr_wchan, pp->pr_nitems);
                   panic("pool_get: nitems inconsistent");
           }
   #endif
   
   #ifdef POOL_DIAGNOSTIC
           pr_log(pp, v, PRLOG_GET, file, line);
   #endif
   
   #ifdef DIAGNOSTIC
           if (__predict_false(pi->pi_magic != PI_MAGIC)) {
                   pr_printlog(pp, pi, printf);
                   panic("pool_get(%s): free list modified: magic=%x; page %p;"
                          " item addr %p\n",
                           pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
           }
   #endif
   
           /*
            * Remove from item list.
            */
           TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
           pp->pr_nitems--;
           pp->pr_nout++;
           if (ph->ph_nmissing == 0) {
   #ifdef DIAGNOSTIC
                   if (__predict_false(pp->pr_nidle == 0))
                           panic("pool_get: nidle inconsistent");
   #endif
                   pp->pr_nidle--;
           }
           ph->ph_nmissing++;
           if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
   #ifdef DIAGNOSTIC
                   if (__predict_false(ph->ph_nmissing != pp->pr_itemsperpage)) {
                           pr_leave(pp);
                           simple_unlock(&pp->pr_slock);
                           panic("pool_get: %s: nmissing inconsistent",
                               pp->pr_wchan);
                   }
   #endif
                   /*
                    * Find a new non-empty page header, if any.
                    * Start search from the page head, to increase
                    * the chance for "high water" pages to be freed.
                    *
                    * Migrate empty pages to the end of the list.  This
                    * will speed the update of curpage as pages become
                    * idle.  Empty pages intermingled with idle pages
                    * is no big deal.  As soon as a page becomes un-empty,
                    * it will move back to the head of the list.
                    */
                   TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                   TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                   TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist)
                           if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                                   break;
   
                   pp->pr_curpage = ph;
           }
   
           pp->pr_nget++;
   
           /*
            * If we have a low water mark and we are now below that low
            * water mark, add more items to the pool.
            */
           if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
                   /*
                    * XXX: Should we log a warning?  Should we set up a timeout
                    * to try again in a second or so?  The latter could break
                    * a caller's assumptions about interrupt protection, etc.
                    */
         }          }
         simple_unlock(&pp->pr_lock);  
           pr_leave(pp);
           simple_unlock(&pp->pr_slock);
         return (v);          return (v);
 }  }
   
 /*  /*
  * Return resource to the pool; must be called at splbio   * Internal version of pool_put().  Pool is already locked/entered.
  */   */
 void  static void
 pool_put(pp, v)  pool_do_put(struct pool *pp, void *v)
         struct pool *pp;  
         void *v;  
 {  {
         struct pool_item *pi = v;          struct pool_item *pi = v;
           struct pool_item_header *ph;
           caddr_t page;
           int s;
   
         simple_lock(&pp->pr_lock);          LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
         if ((pp->pr_flags & PR_WANTED) || pp->pr_freecount < pp->pr_hiwat) {  
                 /* Return to pool */          page = (caddr_t)((u_long)v & pp->pr_alloc->pa_pagemask);
                 pi->pi_next = pp->pr_freelist;  
                 pp->pr_freelist = pi;  #ifdef DIAGNOSTIC
                 pp->pr_freecount++;          if (__predict_false(pp->pr_nout == 0)) {
                 if (pp->pr_flags & PR_WANTED) {                  printf("pool %s: putting with none out\n",
                         pp->pr_flags &= ~PR_WANTED;                      pp->pr_wchan);
                         wakeup((caddr_t)pp);                  panic("pool_put");
                 }          }
         } else {  #endif
   
           if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) {
                   pr_printlog(pp, NULL, printf);
                   panic("pool_put: %s: page header missing", pp->pr_wchan);
           }
   
   #ifdef LOCKDEBUG
           /*
            * Check if we're freeing a locked simple lock.
            */
           simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size);
   #endif
   
           /*
            * Return to item list.
            */
 #ifdef DIAGNOSTIC  #ifdef DIAGNOSTIC
                 if (pp->pr_flags & PR_STATIC) {          pi->pi_magic = PI_MAGIC;
                         /* can't happen because hiwat > freecount */  #endif
                         panic("pool_put: static");  #ifdef DEBUG
           {
                   int i, *ip = v;
   
                   for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                           *ip++ = PI_MAGIC;
                 }                  }
           }
 #endif  #endif
                 /* Return to system */  
                 free(v, M_DEVBUF);  
   
                 /*          TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
                  * Return any excess items allocated during periods of          KDASSERT(ph->ph_nmissing != 0);
                  * contention.          ph->ph_nmissing--;
                  */          pp->pr_nput++;
                 while (pp->pr_freecount > pp->pr_hiwat) {          pp->pr_nitems++;
                         pi = pp->pr_freelist;          pp->pr_nout--;
                         pp->pr_freelist = pi->pi_next;  
                         pp->pr_freecount--;          /* Cancel "pool empty" condition if it exists */
                         free(pi, M_DEVBUF);          if (pp->pr_curpage == NULL)
                   pp->pr_curpage = ph;
   
           if (pp->pr_flags & PR_WANTED) {
                   pp->pr_flags &= ~PR_WANTED;
                   if (ph->ph_nmissing == 0)
                           pp->pr_nidle++;
                   wakeup((caddr_t)pp);
                   return;
           }
   
           /*
            * If this page is now complete, do one of two things:
            *
            *      (1) If we have more pages than the page high water
            *          mark, free the page back to the system.
            *
            *      (2) Move it to the end of the page list, so that
            *          we minimize our chances of fragmenting the
            *          pool.  Idle pages migrate to the end (along with
            *          completely empty pages, so that we find un-empty
            *          pages more quickly when we update curpage) of the
            *          list so they can be more easily swept up by
            *          the pagedaemon when pages are scarce.
            */
           if (ph->ph_nmissing == 0) {
                   pp->pr_nidle++;
                   if (pp->pr_npages > pp->pr_maxpages ||
                       (pp->pr_alloc->pa_flags & PA_WANT) != 0) {
                           pr_rmpage(pp, ph, NULL);
                   } else {
                           TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                           TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
   
                           /*
                            * Update the timestamp on the page.  A page must
                            * be idle for some period of time before it can
                            * be reclaimed by the pagedaemon.  This minimizes
                            * ping-pong'ing for memory.
                            */
                           s = splclock();
                           ph->ph_time = mono_time;
                           splx(s);
   
                           /*
                            * Update the current page pointer.  Just look for
                            * the first page with any free items.
                            *
                            * XXX: Maybe we want an option to look for the
                            * page with the fewest available items, to minimize
                            * fragmentation?
                            */
                           TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist)
                                   if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                                           break;
   
                           pp->pr_curpage = ph;
                 }                  }
         }          }
         simple_unlock(&pp->pr_lock);          /*
            * If the page has just become un-empty, move it to the head of
            * the list, and make it the current page.  The next allocation
            * will get the item from this page, instead of further fragmenting
            * the pool.
            */
           else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
                   TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                   TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                   pp->pr_curpage = ph;
           }
 }  }
   
 /*  /*
  * Add N items to the pool   * Return resource to the pool; must be called at appropriate spl level
  */   */
 int  #ifdef POOL_DIAGNOSTIC
 pool_prime(pp, n, storage)  void
         struct pool *pp;  _pool_put(struct pool *pp, void *v, const char *file, long line)
         int n;  
         caddr_t storage;  
 {  {
         struct pool_item *pi;  
         caddr_t cp = storage;  
   
 #ifdef DIAGNOSTIC          simple_lock(&pp->pr_slock);
         if (storage && !(pp->pr_flags & PR_STATIC))          pr_enter(pp, file, line);
                 panic("pool_prime: static");  
         /* !storage && static caught below */          pr_log(pp, v, PRLOG_PUT, file, line);
   
           pool_do_put(pp, v);
   
           pr_leave(pp);
           simple_unlock(&pp->pr_slock);
   }
   #undef pool_put
   #endif /* POOL_DIAGNOSTIC */
   
   void
   pool_put(struct pool *pp, void *v)
   {
   
           simple_lock(&pp->pr_slock);
   
           pool_do_put(pp, v);
   
           simple_unlock(&pp->pr_slock);
   }
   
   #ifdef POOL_DIAGNOSTIC
   #define         pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
 #endif  #endif
   
         simple_lock(&pp->pr_lock);  /*
         pp->pr_hiwat += n;   * Add N items to the pool.
         while (n--) {   */
                 if (pp->pr_flags & PR_STATIC) {  int
                         pi = (struct pool_item *)cp;  pool_prime(struct pool *pp, int n)
                         cp = (caddr_t)ALIGN(cp + pp->pr_size);  {
                 } else          struct pool_item_header *ph = NULL;
                         pi = malloc(pp->pr_size, pp->pr_mtype, M_NOWAIT);          caddr_t cp;
           int newpages;
                 if (pi == NULL) {  
                         simple_unlock(&pp->pr_lock);          simple_lock(&pp->pr_slock);
                         return (ENOMEM);  
                 }          newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
   
                 pi->pi_next = pp->pr_freelist;          while (newpages-- > 0) {
                 pp->pr_freelist = pi;                  simple_unlock(&pp->pr_slock);
                 pp->pr_freecount++;                  cp = pool_allocator_alloc(pp, PR_NOWAIT);
                   if (__predict_true(cp != NULL))
                           ph = pool_alloc_item_header(pp, cp, PR_NOWAIT);
                   simple_lock(&pp->pr_slock);
   
                   if (__predict_false(cp == NULL || ph == NULL)) {
                           if (cp != NULL)
                                   pool_allocator_free(pp, cp);
                           break;
                   }
   
                   pool_prime_page(pp, cp, ph);
                   pp->pr_npagealloc++;
                   pp->pr_minpages++;
         }          }
         simple_unlock(&pp->pr_lock);  
         return (0);          if (pp->pr_minpages >= pp->pr_maxpages)
                   pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
   
           simple_unlock(&pp->pr_slock);
           return (0);
   }
   
   /*
    * Add a page worth of items to the pool.
    *
    * Note, we must be called with the pool descriptor LOCKED.
    */
   static void
   pool_prime_page(struct pool *pp, caddr_t storage, struct pool_item_header *ph)
   {
           struct pool_item *pi;
           caddr_t cp = storage;
           unsigned int align = pp->pr_align;
           unsigned int ioff = pp->pr_itemoffset;
           int n;
   
   #ifdef DIAGNOSTIC
           if (((u_long)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0)
                   panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
   #endif
   
           if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                   LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
                       ph, ph_hashlist);
   
           /*
            * Insert page header.
            */
           TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
           TAILQ_INIT(&ph->ph_itemlist);
           ph->ph_page = storage;
           ph->ph_nmissing = 0;
           memset(&ph->ph_time, 0, sizeof(ph->ph_time));
   
           pp->pr_nidle++;
   
           /*
            * Color this page.
            */
           cp = (caddr_t)(cp + pp->pr_curcolor);
           if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   pp->pr_curcolor = 0;
   
           /*
            * Adjust storage to apply aligment to `pr_itemoffset' in each item.
            */
           if (ioff != 0)
                   cp = (caddr_t)(cp + (align - ioff));
   
           /*
            * Insert remaining chunks on the bucket list.
            */
           n = pp->pr_itemsperpage;
           pp->pr_nitems += n;
   
           while (n--) {
                   pi = (struct pool_item *)cp;
   
                   KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0);
   
                   /* Insert on page list */
                   TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
   #ifdef DIAGNOSTIC
                   pi->pi_magic = PI_MAGIC;
   #endif
                   cp = (caddr_t)(cp + pp->pr_size);
           }
   
           /*
            * If the pool was depleted, point at the new page.
            */
           if (pp->pr_curpage == NULL)
                   pp->pr_curpage = ph;
   
           if (++pp->pr_npages > pp->pr_hiwat)
                   pp->pr_hiwat = pp->pr_npages;
   }
   
   /*
    * Used by pool_get() when nitems drops below the low water mark.  This
    * is used to catch up nitmes with the low water mark.
    *
    * Note 1, we never wait for memory here, we let the caller decide what to do.
    *
    * Note 2, we must be called with the pool already locked, and we return
    * with it locked.
    */
   static int
   pool_catchup(struct pool *pp)
   {
           struct pool_item_header *ph = NULL;
           caddr_t cp;
           int error = 0;
   
           while (POOL_NEEDS_CATCHUP(pp)) {
                   /*
                    * Call the page back-end allocator for more memory.
                    *
                    * XXX: We never wait, so should we bother unlocking
                    * the pool descriptor?
                    */
                   simple_unlock(&pp->pr_slock);
                   cp = pool_allocator_alloc(pp, PR_NOWAIT);
                   if (__predict_true(cp != NULL))
                           ph = pool_alloc_item_header(pp, cp, PR_NOWAIT);
                   simple_lock(&pp->pr_slock);
                   if (__predict_false(cp == NULL || ph == NULL)) {
                           if (cp != NULL)
                                   pool_allocator_free(pp, cp);
                           error = ENOMEM;
                           break;
                   }
                   pool_prime_page(pp, cp, ph);
                   pp->pr_npagealloc++;
           }
   
           return (error);
   }
   
   void
   pool_setlowat(struct pool *pp, int n)
   {
   
           simple_lock(&pp->pr_slock);
   
           pp->pr_minitems = n;
           pp->pr_minpages = (n == 0)
                   ? 0
                   : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
   
           /* Make sure we're caught up with the newly-set low water mark. */
           if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
                   /*
                    * XXX: Should we log a warning?  Should we set up a timeout
                    * to try again in a second or so?  The latter could break
                    * a caller's assumptions about interrupt protection, etc.
                    */
           }
   
           simple_unlock(&pp->pr_slock);
   }
   
   void
   pool_sethiwat(struct pool *pp, int n)
   {
   
           simple_lock(&pp->pr_slock);
   
           pp->pr_maxpages = (n == 0)
                   ? 0
                   : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
   
           simple_unlock(&pp->pr_slock);
   }
   
   void
   pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
   {
   
           simple_lock(&pp->pr_slock);
   
           pp->pr_hardlimit = n;
           pp->pr_hardlimit_warning = warnmess;
           pp->pr_hardlimit_ratecap.tv_sec = ratecap;
           pp->pr_hardlimit_warning_last.tv_sec = 0;
           pp->pr_hardlimit_warning_last.tv_usec = 0;
   
           /*
            * In-line version of pool_sethiwat(), because we don't want to
            * release the lock.
            */
           pp->pr_maxpages = (n == 0)
                   ? 0
                   : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
   
           simple_unlock(&pp->pr_slock);
   }
   
   /*
    * Release all complete pages that have not been used recently.
    */
   int
   #ifdef POOL_DIAGNOSTIC
   _pool_reclaim(struct pool *pp, const char *file, long line)
   #else
   pool_reclaim(struct pool *pp)
   #endif
   {
           struct pool_item_header *ph, *phnext;
           struct pool_cache *pc;
           struct timeval curtime;
           struct pool_pagelist pq;
           int s;
   
           if (pp->pr_drain_hook != NULL) {
                   /*
                    * The drain hook must be called with the pool unlocked.
                    */
                   (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
           }
   
           if (simple_lock_try(&pp->pr_slock) == 0)
                   return (0);
           pr_enter(pp, file, line);
   
           TAILQ_INIT(&pq);
   
           /*
            * Reclaim items from the pool's caches.
            */
           TAILQ_FOREACH(pc, &pp->pr_cachelist, pc_poollist)
                   pool_cache_reclaim(pc);
   
           s = splclock();
           curtime = mono_time;
           splx(s);
   
           for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
                   phnext = TAILQ_NEXT(ph, ph_pagelist);
   
                   /* Check our minimum page claim */
                   if (pp->pr_npages <= pp->pr_minpages)
                           break;
   
                   if (ph->ph_nmissing == 0) {
                           struct timeval diff;
                           timersub(&curtime, &ph->ph_time, &diff);
                           if (diff.tv_sec < pool_inactive_time)
                                   continue;
   
                           /*
                            * If freeing this page would put us below
                            * the low water mark, stop now.
                            */
                           if ((pp->pr_nitems - pp->pr_itemsperpage) <
                               pp->pr_minitems)
                                   break;
   
                           pr_rmpage(pp, ph, &pq);
                   }
           }
   
           pr_leave(pp);
           simple_unlock(&pp->pr_slock);
           if (TAILQ_EMPTY(&pq))
                   return (0);
   
           while ((ph = TAILQ_FIRST(&pq)) != NULL) {
                   TAILQ_REMOVE(&pq, ph, ph_pagelist);
                   pool_allocator_free(pp, ph->ph_page);
                   if (pp->pr_roflags & PR_PHINPAGE) {
                           continue;
                   }
                   LIST_REMOVE(ph, ph_hashlist);
                   s = splhigh();
                   pool_put(&phpool, ph);
                   splx(s);
           }
   
           return (1);
   }
   
   /*
    * Drain pools, one at a time.
    *
    * Note, we must never be called from an interrupt context.
    */
   void
   pool_drain(void *arg)
   {
           struct pool *pp;
           int s;
   
           pp = NULL;
           s = splvm();
           simple_lock(&pool_head_slock);
           if (drainpp == NULL) {
                   drainpp = TAILQ_FIRST(&pool_head);
           }
           if (drainpp) {
                   pp = drainpp;
                   drainpp = TAILQ_NEXT(pp, pr_poollist);
           }
           simple_unlock(&pool_head_slock);
           pool_reclaim(pp);
           splx(s);
   }
   
   /*
    * Diagnostic helpers.
    */
   void
   pool_print(struct pool *pp, const char *modif)
   {
           int s;
   
           s = splvm();
           if (simple_lock_try(&pp->pr_slock) == 0) {
                   printf("pool %s is locked; try again later\n",
                       pp->pr_wchan);
                   splx(s);
                   return;
           }
           pool_print1(pp, modif, printf);
           simple_unlock(&pp->pr_slock);
           splx(s);
   }
   
   void
   pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
   {
           int didlock = 0;
   
           if (pp == NULL) {
                   (*pr)("Must specify a pool to print.\n");
                   return;
           }
   
           /*
            * Called from DDB; interrupts should be blocked, and all
            * other processors should be paused.  We can skip locking
            * the pool in this case.
            *
            * We do a simple_lock_try() just to print the lock
            * status, however.
            */
   
           if (simple_lock_try(&pp->pr_slock) == 0)
                   (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
           else
                   didlock = 1;
   
           pool_print1(pp, modif, pr);
   
           if (didlock)
                   simple_unlock(&pp->pr_slock);
   }
   
   static void
   pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
   {
           struct pool_item_header *ph;
           struct pool_cache *pc;
           struct pool_cache_group *pcg;
   #ifdef DIAGNOSTIC
           struct pool_item *pi;
   #endif
           int i, print_log = 0, print_pagelist = 0, print_cache = 0;
           char c;
   
           while ((c = *modif++) != '\0') {
                   if (c == 'l')
                           print_log = 1;
                   if (c == 'p')
                           print_pagelist = 1;
                   if (c == 'c')
                           print_cache = 1;
           }
   
           (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
               pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
               pp->pr_roflags);
           (*pr)("\talloc %p\n", pp->pr_alloc);
           (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
               pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
           (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
               pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
   
           (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
               pp->pr_nget, pp->pr_nfail, pp->pr_nput);
           (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
               pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
   
           if (print_pagelist == 0)
                   goto skip_pagelist;
   
           if ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
                   (*pr)("\n\tpage list:\n");
           for (; ph != NULL; ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                       ph->ph_page, ph->ph_nmissing,
                       (u_long)ph->ph_time.tv_sec,
                       (u_long)ph->ph_time.tv_usec);
   #ifdef DIAGNOSTIC
                   TAILQ_FOREACH(pi, &ph->ph_itemlist, pi_list) {
                           if (pi->pi_magic != PI_MAGIC) {
                                   (*pr)("\t\t\titem %p, magic 0x%x\n",
                                       pi, pi->pi_magic);
                           }
                   }
   #endif
           }
           if (pp->pr_curpage == NULL)
                   (*pr)("\tno current page\n");
           else
                   (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
   
    skip_pagelist:
   
           if (print_log == 0)
                   goto skip_log;
   
           (*pr)("\n");
           if ((pp->pr_roflags & PR_LOGGING) == 0)
                   (*pr)("\tno log\n");
           else
                   pr_printlog(pp, NULL, pr);
   
    skip_log:
   
           if (print_cache == 0)
                   goto skip_cache;
   
           TAILQ_FOREACH(pc, &pp->pr_cachelist, pc_poollist) {
                   (*pr)("\tcache %p: allocfrom %p freeto %p\n", pc,
                       pc->pc_allocfrom, pc->pc_freeto);
                   (*pr)("\t    hits %lu misses %lu ngroups %lu nitems %lu\n",
                       pc->pc_hits, pc->pc_misses, pc->pc_ngroups, pc->pc_nitems);
                   TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
                           (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);
                           for (i = 0; i < PCG_NOBJECTS; i++)
                                   (*pr)("\t\t\t%p\n", pcg->pcg_objects[i]);
                   }
           }
   
    skip_cache:
   
           pr_enter_check(pp, pr);
   }
   
   int
   pool_chk(struct pool *pp, const char *label)
   {
           struct pool_item_header *ph;
           int r = 0;
   
           simple_lock(&pp->pr_slock);
   
           TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist) {
                   struct pool_item *pi;
                   int n;
                   caddr_t page;
   
                   page = (caddr_t)((u_long)ph & pp->pr_alloc->pa_pagemask);
                   if (page != ph->ph_page &&
                       (pp->pr_roflags & PR_PHINPAGE) != 0) {
                           if (label != NULL)
                                   printf("%s: ", label);
                           printf("pool(%p:%s): page inconsistency: page %p;"
                                  " at page head addr %p (p %p)\n", pp,
                                   pp->pr_wchan, ph->ph_page,
                                   ph, page);
                           r++;
                           goto out;
                   }
   
                   for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
                        pi != NULL;
                        pi = TAILQ_NEXT(pi,pi_list), n++) {
   
   #ifdef DIAGNOSTIC
                           if (pi->pi_magic != PI_MAGIC) {
                                   if (label != NULL)
                                           printf("%s: ", label);
                                   printf("pool(%s): free list modified: magic=%x;"
                                          " page %p; item ordinal %d;"
                                          " addr %p (p %p)\n",
                                           pp->pr_wchan, pi->pi_magic, ph->ph_page,
                                           n, pi, page);
                                   panic("pool");
                           }
   #endif
                           page =
                               (caddr_t)((u_long)pi & pp->pr_alloc->pa_pagemask);
                           if (page == ph->ph_page)
                                   continue;
   
                           if (label != NULL)
                                   printf("%s: ", label);
                           printf("pool(%p:%s): page inconsistency: page %p;"
                                  " item ordinal %d; addr %p (p %p)\n", pp,
                                   pp->pr_wchan, ph->ph_page,
                                   n, pi, page);
                           r++;
                           goto out;
                   }
           }
   out:
           simple_unlock(&pp->pr_slock);
           return (r);
   }
   
   /*
    * pool_cache_init:
    *
    *      Initialize a pool cache.
    *
    *      NOTE: If the pool must be protected from interrupts, we expect
    *      to be called at the appropriate interrupt priority level.
    */
   void
   pool_cache_init(struct pool_cache *pc, struct pool *pp,
       int (*ctor)(void *, void *, int),
       void (*dtor)(void *, void *),
       void *arg)
   {
   
           TAILQ_INIT(&pc->pc_grouplist);
           simple_lock_init(&pc->pc_slock);
   
           pc->pc_allocfrom = NULL;
           pc->pc_freeto = NULL;
           pc->pc_pool = pp;
   
           pc->pc_ctor = ctor;
           pc->pc_dtor = dtor;
           pc->pc_arg  = arg;
   
           pc->pc_hits   = 0;
           pc->pc_misses = 0;
   
           pc->pc_ngroups = 0;
   
           pc->pc_nitems = 0;
   
           simple_lock(&pp->pr_slock);
           TAILQ_INSERT_TAIL(&pp->pr_cachelist, pc, pc_poollist);
           simple_unlock(&pp->pr_slock);
   }
   
   /*
    * pool_cache_destroy:
    *
    *      Destroy a pool cache.
    */
   void
   pool_cache_destroy(struct pool_cache *pc)
   {
           struct pool *pp = pc->pc_pool;
   
           /* First, invalidate the entire cache. */
           pool_cache_invalidate(pc);
   
           /* ...and remove it from the pool's cache list. */
           simple_lock(&pp->pr_slock);
           TAILQ_REMOVE(&pp->pr_cachelist, pc, pc_poollist);
           simple_unlock(&pp->pr_slock);
   }
   
   static __inline void *
   pcg_get(struct pool_cache_group *pcg)
   {
           void *object;
           u_int idx;
   
           KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
           KASSERT(pcg->pcg_avail != 0);
           idx = --pcg->pcg_avail;
   
           KASSERT(pcg->pcg_objects[idx] != NULL);
           object = pcg->pcg_objects[idx];
           pcg->pcg_objects[idx] = NULL;
   
           return (object);
   }
   
   static __inline void
   pcg_put(struct pool_cache_group *pcg, void *object)
   {
           u_int idx;
   
           KASSERT(pcg->pcg_avail < PCG_NOBJECTS);
           idx = pcg->pcg_avail++;
   
           KASSERT(pcg->pcg_objects[idx] == NULL);
           pcg->pcg_objects[idx] = object;
   }
   
   /*
    * pool_cache_get:
    *
    *      Get an object from a pool cache.
    */
   void *
   pool_cache_get(struct pool_cache *pc, int flags)
   {
           struct pool_cache_group *pcg;
           void *object;
   
   #ifdef LOCKDEBUG
           if (flags & PR_WAITOK)
                   simple_lock_only_held(NULL, "pool_cache_get(PR_WAITOK)");
   #endif
   
           simple_lock(&pc->pc_slock);
   
           if ((pcg = pc->pc_allocfrom) == NULL) {
                   TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
                           if (pcg->pcg_avail != 0) {
                                   pc->pc_allocfrom = pcg;
                                   goto have_group;
                           }
                   }
   
                   /*
                    * No groups with any available objects.  Allocate
                    * a new object, construct it, and return it to
                    * the caller.  We will allocate a group, if necessary,
                    * when the object is freed back to the cache.
                    */
                   pc->pc_misses++;
                   simple_unlock(&pc->pc_slock);
                   object = pool_get(pc->pc_pool, flags);
                   if (object != NULL && pc->pc_ctor != NULL) {
                           if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                                   pool_put(pc->pc_pool, object);
                                   return (NULL);
                           }
                   }
                   return (object);
           }
   
    have_group:
           pc->pc_hits++;
           pc->pc_nitems--;
           object = pcg_get(pcg);
   
           if (pcg->pcg_avail == 0)
                   pc->pc_allocfrom = NULL;
   
           simple_unlock(&pc->pc_slock);
   
           return (object);
   }
   
   /*
    * pool_cache_put:
    *
    *      Put an object back to the pool cache.
    */
   void
   pool_cache_put(struct pool_cache *pc, void *object)
   {
           struct pool_cache_group *pcg;
           int s;
   
           simple_lock(&pc->pc_slock);
   
           if ((pcg = pc->pc_freeto) == NULL) {
                   TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
                           if (pcg->pcg_avail != PCG_NOBJECTS) {
                                   pc->pc_freeto = pcg;
                                   goto have_group;
                           }
                   }
   
                   /*
                    * No empty groups to free the object to.  Attempt to
                    * allocate one.
                    */
                   simple_unlock(&pc->pc_slock);
                   s = splvm();
                   pcg = pool_get(&pcgpool, PR_NOWAIT);
                   splx(s);
                   if (pcg != NULL) {
                           memset(pcg, 0, sizeof(*pcg));
                           simple_lock(&pc->pc_slock);
                           pc->pc_ngroups++;
                           TAILQ_INSERT_TAIL(&pc->pc_grouplist, pcg, pcg_list);
                           if (pc->pc_freeto == NULL)
                                   pc->pc_freeto = pcg;
                           goto have_group;
                   }
   
                   /*
                    * Unable to allocate a cache group; destruct the object
                    * and free it back to the pool.
                    */
                   pool_cache_destruct_object(pc, object);
                   return;
           }
   
    have_group:
           pc->pc_nitems++;
           pcg_put(pcg, object);
   
           if (pcg->pcg_avail == PCG_NOBJECTS)
                   pc->pc_freeto = NULL;
   
           simple_unlock(&pc->pc_slock);
   }
   
   /*
    * pool_cache_destruct_object:
    *
    *      Force destruction of an object and its release back into
    *      the pool.
    */
   void
   pool_cache_destruct_object(struct pool_cache *pc, void *object)
   {
   
           if (pc->pc_dtor != NULL)
                   (*pc->pc_dtor)(pc->pc_arg, object);
           pool_put(pc->pc_pool, object);
   }
   
   /*
    * pool_cache_do_invalidate:
    *
    *      This internal function implements pool_cache_invalidate() and
    *      pool_cache_reclaim().
    */
   static void
   pool_cache_do_invalidate(struct pool_cache *pc, int free_groups,
       void (*putit)(struct pool *, void *))
   {
           struct pool_cache_group *pcg, *npcg;
           void *object;
           int s;
   
           for (pcg = TAILQ_FIRST(&pc->pc_grouplist); pcg != NULL;
                pcg = npcg) {
                   npcg = TAILQ_NEXT(pcg, pcg_list);
                   while (pcg->pcg_avail != 0) {
                           pc->pc_nitems--;
                           object = pcg_get(pcg);
                           if (pcg->pcg_avail == 0 && pc->pc_allocfrom == pcg)
                                   pc->pc_allocfrom = NULL;
                           if (pc->pc_dtor != NULL)
                                   (*pc->pc_dtor)(pc->pc_arg, object);
                           (*putit)(pc->pc_pool, object);
                   }
                   if (free_groups) {
                           pc->pc_ngroups--;
                           TAILQ_REMOVE(&pc->pc_grouplist, pcg, pcg_list);
                           if (pc->pc_freeto == pcg)
                                   pc->pc_freeto = NULL;
                           s = splvm();
                           pool_put(&pcgpool, pcg);
                           splx(s);
                   }
           }
   }
   
   /*
    * pool_cache_invalidate:
    *
    *      Invalidate a pool cache (destruct and release all of the
    *      cached objects).
    */
   void
   pool_cache_invalidate(struct pool_cache *pc)
   {
   
           simple_lock(&pc->pc_slock);
           pool_cache_do_invalidate(pc, 0, pool_put);
           simple_unlock(&pc->pc_slock);
   }
   
   /*
    * pool_cache_reclaim:
    *
    *      Reclaim a pool cache for pool_reclaim().
    */
   static void
   pool_cache_reclaim(struct pool_cache *pc)
   {
   
           simple_lock(&pc->pc_slock);
           pool_cache_do_invalidate(pc, 1, pool_do_put);
           simple_unlock(&pc->pc_slock);
   }
   
   /*
    * Pool backend allocators.
    *
    * Each pool has a backend allocator that handles allocation, deallocation,
    * and any additional draining that might be needed.
    *
    * We provide two standard allocators:
    *
    *      pool_allocator_kmem - the default when no allocator is specified
    *
    *      pool_allocator_nointr - used for pools that will not be accessed
    *      in interrupt context.
    */
   void    *pool_page_alloc(struct pool *, int);
   void    pool_page_free(struct pool *, void *);
   
   struct pool_allocator pool_allocator_kmem = {
           pool_page_alloc, pool_page_free, 0,
   };
   
   void    *pool_page_alloc_nointr(struct pool *, int);
   void    pool_page_free_nointr(struct pool *, void *);
   
   struct pool_allocator pool_allocator_nointr = {
           pool_page_alloc_nointr, pool_page_free_nointr, 0,
   };
   
   #ifdef POOL_SUBPAGE
   void    *pool_subpage_alloc(struct pool *, int);
   void    pool_subpage_free(struct pool *, void *);
   
   struct pool_allocator pool_allocator_kmem_subpage = {
           pool_subpage_alloc, pool_subpage_free, 0,
   };
   #endif /* POOL_SUBPAGE */
   
   /*
    * We have at least three different resources for the same allocation and
    * each resource can be depleted.  First, we have the ready elements in the
    * pool.  Then we have the resource (typically a vm_map) for this allocator.
    * Finally, we have physical memory.  Waiting for any of these can be
    * unnecessary when any other is freed, but the kernel doesn't support
    * sleeping on multiple wait channels, so we have to employ another strategy.
    *
    * The caller sleeps on the pool (so that it can be awakened when an item
    * is returned to the pool), but we set PA_WANT on the allocator.  When a
    * page is returned to the allocator and PA_WANT is set, pool_allocator_free
    * will wake up all sleeping pools belonging to this allocator.
    *
    * XXX Thundering herd.
    */
   void *
   pool_allocator_alloc(struct pool *org, int flags)
   {
           struct pool_allocator *pa = org->pr_alloc;
           struct pool *pp, *start;
           int s, freed;
           void *res;
   
           do {
                   if ((res = (*pa->pa_alloc)(org, flags)) != NULL)
                           return (res);
                   if ((flags & PR_WAITOK) == 0) {
                           /*
                            * We only run the drain hookhere if PR_NOWAIT.
                            * In other cases, the hook will be run in
                            * pool_reclaim().
                            */
                           if (org->pr_drain_hook != NULL) {
                                   (*org->pr_drain_hook)(org->pr_drain_hook_arg,
                                       flags);
                                   if ((res = (*pa->pa_alloc)(org, flags)) != NULL)
                                           return (res);
                           }
                           break;
                   }
   
                   /*
                    * Drain all pools, except "org", that use this
                    * allocator.  We do this to reclaim VA space.
                    * pa_alloc is responsible for waiting for
                    * physical memory.
                    *
                    * XXX We risk looping forever if start if someone
                    * calls pool_destroy on "start".  But there is no
                    * other way to have potentially sleeping pool_reclaim,
                    * non-sleeping locks on pool_allocator, and some
                    * stirring of drained pools in the allocator.
                    *
                    * XXX Maybe we should use pool_head_slock for locking
                    * the allocators?
                    */
                   freed = 0;
   
                   s = splvm();
                   simple_lock(&pa->pa_slock);
                   pp = start = TAILQ_FIRST(&pa->pa_list);
                   do {
                           TAILQ_REMOVE(&pa->pa_list, pp, pr_alloc_list);
                           TAILQ_INSERT_TAIL(&pa->pa_list, pp, pr_alloc_list);
                           if (pp == org)
                                   continue;
                           simple_unlock(&pa->pa_slock);
                           freed = pool_reclaim(pp);
                           simple_lock(&pa->pa_slock);
                   } while ((pp = TAILQ_FIRST(&pa->pa_list)) != start &&
                            freed == 0);
   
                   if (freed == 0) {
                           /*
                            * We set PA_WANT here, the caller will most likely
                            * sleep waiting for pages (if not, this won't hurt
                            * that much), and there is no way to set this in
                            * the caller without violating locking order.
                            */
                           pa->pa_flags |= PA_WANT;
                   }
                   simple_unlock(&pa->pa_slock);
                   splx(s);
           } while (freed);
           return (NULL);
   }
   
   void
   pool_allocator_free(struct pool *pp, void *v)
   {
           struct pool_allocator *pa = pp->pr_alloc;
           int s;
   
           (*pa->pa_free)(pp, v);
   
           s = splvm();
           simple_lock(&pa->pa_slock);
           if ((pa->pa_flags & PA_WANT) == 0) {
                   simple_unlock(&pa->pa_slock);
                   splx(s);
                   return;
           }
   
           TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
                   simple_lock(&pp->pr_slock);
                   if ((pp->pr_flags & PR_WANTED) != 0) {
                           pp->pr_flags &= ~PR_WANTED;
                           wakeup(pp);
                   }
                   simple_unlock(&pp->pr_slock);
           }
           pa->pa_flags &= ~PA_WANT;
           simple_unlock(&pa->pa_slock);
           splx(s);
   }
   
   void *
   pool_page_alloc(struct pool *pp, int flags)
   {
           boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
   
           return ((void *) uvm_km_alloc_poolpage(waitok));
   }
   
   void
   pool_page_free(struct pool *pp, void *v)
   {
   
           uvm_km_free_poolpage((vaddr_t) v);
   }
   
   #ifdef POOL_SUBPAGE
   /* Sub-page allocator, for machines with large hardware pages. */
   void *
   pool_subpage_alloc(struct pool *pp, int flags)
   {
   
           return (pool_get(&psppool, flags));
   }
   
   void
   pool_subpage_free(struct pool *pp, void *v)
   {
   
           pool_put(&psppool, v);
   }
   
   /* We don't provide a real nointr allocator.  Maybe later. */
   void *
   pool_page_alloc_nointr(struct pool *pp, int flags)
   {
   
           return (pool_subpage_alloc(pp, flags));
   }
   
   void
   pool_page_free_nointr(struct pool *pp, void *v)
   {
   
           pool_subpage_free(pp, v);
   }
   #else
   void *
   pool_page_alloc_nointr(struct pool *pp, int flags)
   {
           boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
   
           return ((void *) uvm_km_alloc_poolpage1(kernel_map,
               uvm.kernel_object, waitok));
   }
   
   void
   pool_page_free_nointr(struct pool *pp, void *v)
   {
   
           uvm_km_free_poolpage1(kernel_map, (vaddr_t) v);
 }  }
   #endif /* POOL_SUBPAGE */

Legend:
Removed from v.1.2  
changed lines
  Added in v.1.84

CVSweb <webmaster@jp.NetBSD.org>