Subject: Re: pool(9) revisited
To: None <tech-kern@NetBSD.ORG>
From: Paul Kranenburg <pk@cs.few.eur.nl>
List: tech-kern
Date: 07/15/1998 00:23:55
Here's a sample for review.

Notes:

	* pool_drain(): called from page daemon
	* there's an upper and a lower bound per pool on the number
	  of pages to hold on to. The upper bound defaults to
	  infinity, e.g. pages are released only on demand.
	* splimp() is used internally for the moment. Originally,
	  interrupt protection was up to the caller..
	* logging code is not in yet


----------------
/*	$NetBSD$	*/

/*-
 * Copyright (c) 1997 The NetBSD Foundation, Inc.
 * All rights reserved.
 *
 * This code is derived from software contributed to The NetBSD Foundation
 * by Paul Kranenburg.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *        This product includes software developed by the NetBSD
 *        Foundation, Inc. and its contributors.
 * 4. Neither the name of The NetBSD Foundation nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 */

#ifndef _SYS_POOL_H_
#define _SYS_POOL_H_

#include <sys/lock.h>
#include <sys/queue.h>

#define PR_HASHTABSIZE		8

typedef struct pool {
	TAILQ_ENTRY(pool)
			pr_poollist;
	TAILQ_HEAD(,pool_item_header)
			pr_pagelist;	/* Allocated pages */
	struct pool_item_header	*pr_curpage;
	unsigned int	pr_size;	/* Size of item */
	unsigned int	pr_align;	/* Requested alignment, must be 2^n */
	unsigned int	pr_minitems;	/* minimum # of items to keep */
	unsigned int	pr_minpages;	/* same in page units */
	unsigned int	pr_maxpages;	/* maximum # of pages to keep */
	unsigned int	pr_npages;	/* # of pages allocated */
	unsigned int	pr_pagesz;	/* page size, must be 2^n */
	unsigned long	pr_pagemask;	/* abbrev. of above */
	unsigned int	pr_pageshift;	/* shift corr. to above */
	unsigned int	pr_itemsperpage;/* # items that fit in a page */
	unsigned int	pr_slack;	/* unused space in a page */
	void		*(*pr_alloc) __P((unsigned long, int, int));
	void		(*pr_free) __P((void *, unsigned long, int));
	int		pr_mtype;	/* memory allocator tag */
	char		*pr_wchan;	/* tsleep(9) identifier */
	unsigned int	pr_flags;
#define PR_MALLOCOK	1
#define PR_WAITOK	2
#define PR_WANTED	4
#define PR_STATIC	8
#define PR_FREEHEADER	16
#define PR_URGENT	32
#define PR_PHINPAGE	64
	struct simplelock	pr_lock;

	LIST_HEAD(,pool_item_header)		/* Off-page page headers */
			pr_hashtab[PR_HASHTABSIZE];

	int		pr_maxcolor;	/* Cache colouring */
	int		pr_curcolor;
	int		pr_phoffset;	/* Offset in page of page header */

	/*
	 * Instrumentation
	 */
	unsigned long	pr_nalloc;	/* # of successful requests */
	unsigned long	pr_nfail;	/* # of unsuccessful requests */
	unsigned long	pr_nfree;	/* # of releases */
	unsigned long	pr_npagealloc;	/* # of pages allocated */
	unsigned long	pr_npagefree;	/* # of pages released */
	unsigned int	pr_hiwat;	/* max # of pages in pool */
} *pool_handle_t;

pool_handle_t	pool_create __P((size_t, int, char *, size_t,
				 void *(*)__P((unsigned long, int, int)),
				 void  (*)__P((void *, unsigned long, int)),
				 int));
void		pool_init __P((struct pool *, size_t, int, char *, size_t,
				 void *(*)__P((unsigned long, int, int)),
				 void  (*)__P((void *, unsigned long, int)),
				 int));
void		pool_destroy __P((pool_handle_t));
void		*pool_get __P((pool_handle_t, int));
void		pool_put __P((pool_handle_t, void *));
int		pool_prime __P((pool_handle_t, int, caddr_t));
void		pool_setlowat __P((pool_handle_t, int));
void		pool_sethiwat __P((pool_handle_t, int));
void		pool_print __P((pool_handle_t, char *));
void		pool_reclaim __P((pool_handle_t));
void		pool_drain __P((void *));

#define POOL_ITEM_STORAGE_SIZE(size, nitems) (ALIGN(size) * nitems)
#define POOL_STORAGE_SIZE(size, nitems) \
	(ALIGN(sizeof(struct pool)) + POOL_ITEM_STORAGE_SIZE(size, nitems))

#endif /* _SYS_POOL_H_ */

---------------------------------------------
/*	$NetBSD$	*/

/*-
 * Copyright (c) 1997 The NetBSD Foundation, Inc.
 * All rights reserved.
 *
 * This code is derived from software contributed to The NetBSD Foundation
 * by Paul Kranenburg.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *        This product includes software developed by the NetBSD
 *        Foundation, Inc. and its contributors.
 * 4. Neither the name of The NetBSD Foundation nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/errno.h>
#include <sys/kernel.h>
#include <sys/malloc.h>
#include <sys/lock.h>
#include <sys/pool.h>

#include <vm/vm.h>
#include <vm/vm_kern.h>

#if defined(UVM)
#include <uvm/uvm.h>
#endif  


/*
 * 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 taken from the allocated pages themselves.
 * 
 */

/* List of all pools */
static TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);

/* Private pool for page header structures */
static struct pool phpool;

/* # of seconds to retain page after last use */
int pool_inactive_time = 10;

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 */
	int			ph_nmissing;	/* # of chunks in use */
	caddr_t			ph_page;	/* this page's address */
	struct timeval		ph_time;	/* last referenced */
};

#define PH_PAGE_DEPLETED(pp, ph) \
	(TAILQ_FIRST(&ph->ph_itemlist) == NULL)

struct pool_item {
	int pi_magic;
#define PI_MAGIC 0xdeadbeef
	/* Other entries use only this list entry */
	TAILQ_ENTRY(pool_item)	pi_list;
};


#define PR_HASH_INDEX(pp,addr) \
	(((u_long)(addr) >> (pp)->pr_pageshift) & (PR_HASHTABSIZE - 1))

static struct pool_item_header
		*pr_find_pagehead __P((struct pool *, caddr_t));
static void	pool_rm_page __P((struct pool *, struct pool_item_header *));
static int	pool_prime_page __P((struct pool *, caddr_t));
static void	*pool_page_alloc __P((unsigned long, int, int));
static void	pool_page_free __P((void *, unsigned long, int));
int pool_chk __P((struct pool *, char *));

static __inline__ struct pool_item_header *
pr_find_pagehead(pp, page)
	struct pool *pp;
	caddr_t page;
{
	struct pool_item_header *ph;

	if ((pp->pr_flags & 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);
}

static __inline__ void
pool_rm_page(pp, ph)
	struct pool *pp;
	struct pool_item_header *ph;
{

	/*
	 * Unlink a page from the pool and release it.
	 */
	TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
	(*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
	pp->pr_npages--;
	pp->pr_npagefree++;

	if ((pp->pr_flags & PR_PHINPAGE) == 0) {
		LIST_REMOVE(ph, ph_hashlist);
		pool_put(&phpool, ph);
	}

	if (pp->pr_curpage == ph) {
		/*
		 * 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.
		 */
		for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
		     ph = TAILQ_NEXT(ph, ph_pagelist))
			if (!PH_PAGE_DEPLETED(pp, ph))
				break;

		pp->pr_curpage = ph;
	}
}

struct pool *
pool_create(size, nitems, wchan, pagesz, alloc, release, mtype)
	size_t	size;
	int	nitems;
	char	*wchan;
	size_t	pagesz;
	void	*(*alloc) __P((unsigned long, int, int));
	void	(*release) __P((void *, unsigned long, int));
	int	mtype;
{
	struct pool *pp;

	pp = (struct pool *)malloc(sizeof(*pp), mtype, M_NOWAIT);
	if (pp == NULL)
		return (NULL);

	pool_init(pp, size, PR_FREEHEADER, wchan, pagesz,
		  alloc, release, mtype);

	if (nitems != 0) {
		if (pool_prime(pp, nitems, NULL) != 0) {
			pool_destroy(pp);
			return (NULL);
		}
	}

	return (pp);
}

void
pool_init(pp, size, flags, wchan, pagesz, alloc, release, mtype)
	struct pool	*pp;
	size_t		size;
	int		flags;
	char		*wchan;
	size_t		pagesz;
	void		*(*alloc) __P((unsigned long, int, int));
	void		(*release) __P((void *, unsigned long, int));
	int		mtype;
{
	int align;
	int off, slack;

	if (!powerof2(pagesz))
		panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);

	if (alloc == NULL)
		alloc = pool_page_alloc;

	if (release == NULL)
		release = pool_page_free;

	if (pagesz == 0)
		pagesz = NBPG;

	align = ALIGN(1); /* XXX - make argument */

	/*
	 * Initialize the pool structure.
	 */
	TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
	TAILQ_INIT(&pp->pr_pagelist);
	pp->pr_curpage = NULL;
	pp->pr_npages = 0;
	pp->pr_minitems = 0;
	pp->pr_minpages = 0;
	pp->pr_maxpages = UINT_MAX;
	pp->pr_flags = flags;
	pp->pr_size = ALIGN(size);
	pp->pr_align = align;
	pp->pr_wchan = wchan;
	pp->pr_mtype = mtype;
	pp->pr_alloc = alloc;
	pp->pr_free = release;
	pp->pr_pagesz = pagesz;
	pp->pr_pagemask = ~(pagesz - 1);
	pp->pr_pageshift = ffs(pagesz) - 1;

	/*
	 * 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.
	 */
	if (pp->pr_size < 256) {
		/* Use the end of the page for the page header */
		pp->pr_flags |= PR_PHINPAGE;
		pp->pr_phoffset = off =
			pagesz - ALIGN(sizeof(struct pool_item_header));
	} else {
		/* The page header will be taken from our page header pool */
		pp->pr_phoffset = 0;
		off = pagesz;
		bzero(pp->pr_hashtab, sizeof(pp->pr_hashtab));
	}
	pp->pr_itemsperpage = off / pp->pr_size;

	/*
	 * 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_nalloc = 0;
	pp->pr_nfail = 0;
	pp->pr_nfree = 0;
	pp->pr_npagealloc = 0;
	pp->pr_npagefree = 0;
	pp->pr_hiwat = 0;

	simple_lock_init(&pp->pr_lock);

	if (phpool.pr_size == 0) {
		/* Initialize private page header pool */
		pool_init(&phpool, sizeof(struct pool_item_header),
			  0, "phpool", 0, 0, 0, 0);
	}

	return;
}

/*
 * De-commision a pool resource.
 */
void
pool_destroy(pp)
	struct pool *pp;
{
	struct pool_item_header *ph;

#ifdef DIAGNOSTIC
	if (pp->pr_nalloc - pp->pr_nfree != 0)
		panic("pool_destroy: pool busy: still out: %lu\n",
		      pp->pr_nalloc - pp->pr_nfree);
#endif
	if ((pp->pr_flags & PR_STATIC) == 0)
		while ((ph = pp->pr_pagelist.tqh_first) != NULL)
			pool_rm_page(pp, ph);

	if (pp->pr_flags & PR_FREEHEADER)
		free(pp, pp->pr_mtype);
}


/*
 * Grab an item from the pool; must be called at appropriate spl level
 */
void *
pool_get(pp, flags)
	struct pool *pp;
	int flags;
{
	void *v;
	struct pool_item *pi;
	struct pool_item_header *ph;
	int s;

#ifdef DIAGNOSTIC
	if ((pp->pr_flags & PR_STATIC) && (flags & PR_MALLOCOK))
		panic("pool_get: static");
#endif

	s = splimp();
	simple_lock(&pp->pr_lock);

	/*
	 * The convention we use is that if `curpage' is not NULL, then
	 * it points at a non-empty bucket. In particular, `curpage'
	 * never points at a page header which has PR_PHINPAGE set and
	 * has no items in its bucket.
	 */
again:
	if ((ph = pp->pr_curpage) == NULL) {
		void *v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
		if (v == NULL) {
			if (flags & PR_URGENT)
				panic("pool_get: urgent");
			if ((flags & PR_WAITOK) == 0) {
				pp->pr_nfail++;
				simple_unlock(&pp->pr_lock);
				splx(s);
				return (NULL);
			}
pool_print(pp, "sleeping");
			pp->pr_flags |= PR_WANTED;
			simple_unlock(&pp->pr_lock);
			tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0);
			simple_lock(&pp->pr_lock);
		} else {
			pp->pr_npagealloc++;
			pool_prime_page(pp, v);
		}

		goto again;
	}

	if ((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)
		panic("pool_get: %s: page empty", pp->pr_wchan);

#ifdef DIAGNOSTIC
	if (pi->pi_magic != PI_MAGIC) {
		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);
	ph->ph_nmissing++;
	if (PH_PAGE_DEPLETED(pp, ph)) {
		/*
		 * 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.
		 */
		for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
		     ph = TAILQ_NEXT(ph, ph_pagelist))
			if (!PH_PAGE_DEPLETED(pp, ph))
				break;

		pp->pr_curpage = ph;
	}

	pp->pr_nalloc++;
	simple_unlock(&pp->pr_lock);
	splx(s);
	return (v);
}

/*
 * Return resource to the pool; must be called at appropriate spl level
 */
void
pool_put(pp, v)
	struct pool *pp;
	void *v;
{
	struct pool_item *pi = v;
	struct pool_item_header *ph;
	caddr_t page;
	int s;

	page = (caddr_t)((u_long)v & pp->pr_pagemask);

	s = splimp();
	simple_lock(&pp->pr_lock);
	pp->pr_nfree++;

	if ((ph = pr_find_pagehead(pp, page)) == NULL)
		panic("pool_put: %s: page header missing", pp->pr_wchan);

	/*
	 * Return to item list.
	 */
#ifdef DIAGNOSTIC
	pi->pi_magic = PI_MAGIC;
#endif
	TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
	ph->ph_nmissing--;

	/* Cancel "pool empty" condition if it exists */
	if (pp->pr_curpage == NULL)
		pp->pr_curpage = ph;

	if (pp->pr_flags & PR_WANTED) {
		pp->pr_flags &= ~PR_WANTED;
		wakeup((caddr_t)pp);
		simple_unlock(&pp->pr_lock);
		splx(s);
		return;
	}

	/*
	 * If this page is now complete, move it to the end of the pagelist.
	 */
	if (ph->ph_nmissing == 0) {
		if (pp->pr_npages > pp->pr_maxpages) {
#if 0
			timeout(pool_reclaim, pp, pool_inactive_time*hz);
#else
			pool_rm_page(pp, ph);
#endif
		} else {
			TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
			TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
			ph->ph_time = time;
			/* XXX - update curpage */
		}
	}

	simple_unlock(&pp->pr_lock);
	splx(s);
}

/*
 * Add N items to the pool.
 */
int
pool_prime(pp, n, storage)
	struct pool *pp;
	int n;
	caddr_t storage;
{
	caddr_t cp;
	int newnitems, newpages;

#ifdef DIAGNOSTIC
	if (storage && !(pp->pr_flags & PR_STATIC))
		panic("pool_prime: static");
	/* !storage && static caught below */
#endif

	newnitems = pp->pr_minitems + n;
	newpages =
		roundup(pp->pr_itemsperpage,newnitems) / pp->pr_itemsperpage
		- pp->pr_minpages;

	simple_lock(&pp->pr_lock);
	while (newpages-- > 0) {

		if (pp->pr_flags & PR_STATIC) {
			cp = storage;
			storage += pp->pr_pagesz;
		} else {
			cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
		}

		if (cp == NULL) {
			simple_unlock(&pp->pr_lock);
			return (ENOMEM);
		}

		pool_prime_page(pp, cp);
		pp->pr_minpages++;
	}

	pp->pr_minitems = newnitems;

	if (pp->pr_minpages >= pp->pr_maxpages)
		pp->pr_maxpages = pp->pr_minpages + 1;	/* XXX */

	simple_unlock(&pp->pr_lock);
	return (0);
}

/*
 * Add a page worth of items to the pool.
 */
int
pool_prime_page(pp, storage)
	struct pool *pp;
	caddr_t storage;
{
	struct pool_item *pi;
	struct pool_item_header *ph;
	caddr_t cp = storage;
	int n;

	if ((pp->pr_flags & PR_PHINPAGE) != 0) {
		ph = (struct pool_item_header *)(cp + pp->pr_phoffset);
	} else {
		ph = pool_get(&phpool, PR_URGENT);
		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;
	ph->ph_time.tv_sec = ph->ph_time.tv_usec = 0;

	/*
	 * Color this page.
	 */
	cp = (caddr_t)(cp + pp->pr_curcolor);
	if ((pp->pr_curcolor += pp->pr_align) > pp->pr_maxcolor)
		pp->pr_curcolor = 0;

	/*
	 * Insert remaining chunks on the bucket list.
	 */
	n = pp->pr_itemsperpage;

	while (n--) {
		pi = (struct pool_item *)cp;

		/* 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;

	return (0);
}

void
pool_setlowat(pp, n)
	pool_handle_t	pp;
	int n;
{
	pp->pr_minitems = n;
	pp->pr_minpages =
		roundup(pp->pr_itemsperpage,n) / pp->pr_itemsperpage;
}

void
pool_sethiwat(pp, n)
	pool_handle_t	pp;
	int n;
{
	pp->pr_maxpages =
		roundup(pp->pr_itemsperpage,n) / pp->pr_itemsperpage;
}


/*
 * Default page allocator.
 */
static void *
pool_page_alloc(sz, flags, mtype)
	unsigned long sz;
	int flags;
	int mtype;
{
	vm_offset_t va;

#if defined(UVM)
	va = uvm_km_kmemalloc(kernel_map, uvm.kernel_object,
			      (vm_size_t)sz, UVM_KMF_NOWAIT);
#else   
	va = kmem_malloc(kmem_map, (vm_size_t)sz, 0);
#endif                       
	return ((void *)va);
}

static void
pool_page_free(v, sz, mtype)
	void *v;
	unsigned long sz;
	int mtype;
{

#if defined(UVM)
	uvm_km_free(kmem_map, (vm_offset_t)v, sz);
#else
	kmem_free(kmem_map, (vm_offset_t)v, sz);
#endif  
}


void
pool_print(pp, label)
	struct pool *pp;
	char *label;
{

	if (label != NULL)
		printf("%s: ", label);

	printf("pool %s: nalloc %lu nfree %lu npagealloc %lu npagefree %lu\n"
	       "         npages %u minitems %u itemsperpage %u\n",
		pp->pr_wchan,
		pp->pr_nalloc,
		pp->pr_nfree,
		pp->pr_npagealloc,
		pp->pr_npagefree,
		pp->pr_npages,
		pp->pr_minitems,
		pp->pr_itemsperpage);
}

int
pool_chk(pp, label)
	struct pool *pp;
	char *label;
{
	struct pool_item_header *ph;

	for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
	     ph = TAILQ_NEXT(ph, ph_pagelist)) {

		struct pool_item *pi;
		int n;
		caddr_t page;

		page = (caddr_t)((u_long)ph & pp->pr_pagemask);
		if (page != ph->ph_page) {
			if (label != NULL)
				printf("%s: ", label);
			printf("pool(%s): page inconsistency: page %p;"
			       " at page head addr %p (p %p)\n",
				pp->pr_wchan, ph->ph_page,
				ph, page);
			return (1);
		}

		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_pagemask);
			if (page == ph->ph_page)
				continue;

			if (label != NULL)
				printf("%s: ", label);
			printf("pool(%s): page inconsistency: page %p;"
			       " item ordinal %d; addr %p (p %p)\n",
				pp->pr_wchan, ph->ph_page,
				n, pi, page);
			return (1);
		}
	}
	return (0);
}

void
pool_reclaim (pp)
	pool_handle_t pp;
{
	struct pool_item_header *ph, *phnext;
	struct timeval curtime = time;
	int s;

	if (pp->pr_flags & PR_STATIC)
		return;

	s = splimp();
	if (simple_lock_try(&pp->pr_lock) == 0) {
		splx(s);
		return;
	}

	for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
		phnext = TAILQ_NEXT(ph, ph_pagelist);
		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;
			}
			pool_rm_page(pp, ph);
		}
	}

	splx(s);
	simple_unlock(&pp->pr_lock);
}


void
pool_drain(arg)
	void *arg;
{
	struct pool *pp;

	for (pp = TAILQ_FIRST(&pool_head); pp != NULL;
	     pp = TAILQ_NEXT(pp, pr_poollist)) {
		pool_reclaim(pp);
		if (uvmexp.free >= uvmexp.freetarg)
			break;
	}
}