Home | History | Annotate | Line # | Download | only in kern
subr_pool.c revision 1.4.2.2
      1 /*	$NetBSD: subr_pool.c,v 1.4.2.2 1998/08/08 03:06:56 eeh Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1997 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Paul Kranenburg.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  * 3. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *        This product includes software developed by the NetBSD
     21  *        Foundation, Inc. and its contributors.
     22  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23  *    contributors may be used to endorse or promote products derived
     24  *    from this software without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36  * POSSIBILITY OF SUCH DAMAGE.
     37  */
     38 
     39 #include <sys/param.h>
     40 #include <sys/systm.h>
     41 #include <sys/proc.h>
     42 #include <sys/errno.h>
     43 #include <sys/kernel.h>
     44 #include <sys/malloc.h>
     45 #include <sys/lock.h>
     46 #include <sys/pool.h>
     47 
     48 #include <vm/vm.h>
     49 #include <vm/vm_kern.h>
     50 
     51 #if defined(UVM)
     52 #include <uvm/uvm.h>
     53 #endif
     54 
     55 /*
     56  * Pool resource management utility.
     57  *
     58  * Memory is allocated in pages which are split into pieces according
     59  * to the pool item size. Each page is kept on a list headed by `pr_pagelist'
     60  * in the pool structure and the individual pool items are on a linked list
     61  * headed by `ph_itemlist' in each page header. The memory for building
     62  * the page list is either taken from the allocated pages themselves (for
     63  * small pool items) or taken from an internal pool of page headers (`phpool').
     64  *
     65  */
     66 
     67 /* List of all pools */
     68 TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
     69 
     70 /* Private pool for page header structures */
     71 static struct pool phpool;
     72 
     73 /* # of seconds to retain page after last use */
     74 int pool_inactive_time = 10;
     75 
     76 /* Next candidate for drainage (see pool_drain()) */
     77 static struct pool	*drainpp = NULL;
     78 
     79 struct pool_item_header {
     80 	/* Page headers */
     81 	TAILQ_ENTRY(pool_item_header)
     82 				ph_pagelist;	/* pool page list */
     83 	TAILQ_HEAD(,pool_item)	ph_itemlist;	/* chunk list for this page */
     84 	LIST_ENTRY(pool_item_header)
     85 				ph_hashlist;	/* Off-page page headers */
     86 	int			ph_nmissing;	/* # of chunks in use */
     87 	caddr_t			ph_page;	/* this page's address */
     88 	struct timeval		ph_time;	/* last referenced */
     89 };
     90 
     91 struct pool_item {
     92 #ifdef DIAGNOSTIC
     93 	int pi_magic;
     94 #define PI_MAGIC 0xdeadbeef
     95 #endif
     96 	/* Other entries use only this list entry */
     97 	TAILQ_ENTRY(pool_item)	pi_list;
     98 };
     99 
    100 
    101 #define PR_HASH_INDEX(pp,addr) \
    102 	(((u_long)(addr) >> (pp)->pr_pageshift) & (PR_HASHTABSIZE - 1))
    103 
    104 
    105 
    106 static struct pool_item_header
    107 		*pr_find_pagehead __P((struct pool *, caddr_t));
    108 static void	pr_rmpage __P((struct pool *, struct pool_item_header *));
    109 static int	pool_prime_page __P((struct pool *, caddr_t));
    110 static void	*pool_page_alloc __P((unsigned long, int, int));
    111 static void	pool_page_free __P((void *, unsigned long, int));
    112 int pool_chk __P((struct pool *, char *));
    113 
    114 
    115 #ifdef POOL_DIAGNOSTIC
    116 /*
    117  * Pool log entry. An array of these is allocated in pool_create().
    118  */
    119 struct pool_log {
    120 	const char	*pl_file;
    121 	long		pl_line;
    122 	int		pl_action;
    123 #define PRLOG_GET	1
    124 #define PRLOG_PUT	2
    125 	void		*pl_addr;
    126 };
    127 
    128 /* Number of entries in pool log buffers */
    129 int pool_logsize = 10;
    130 
    131 static void	pr_log __P((struct pool *, void *, int, const char *, long));
    132 static void	pr_printlog __P((struct pool *));
    133 
    134 static __inline__ void
    135 pr_log(pp, v, action, file, line)
    136 	struct pool	*pp;
    137 	void		*v;
    138 	int		action;
    139 	const char	*file;
    140 	long		line;
    141 {
    142 	int n = pp->pr_curlogentry;
    143 	struct pool_log *pl;
    144 
    145 	if ((pp->pr_flags & PR_LOGGING) == 0)
    146 		return;
    147 
    148 	/*
    149 	 * Fill in the current entry. Wrap around and overwrite
    150 	 * the oldest entry if necessary.
    151 	 */
    152 	pl = &pp->pr_log[n];
    153 	pl->pl_file = file;
    154 	pl->pl_line = line;
    155 	pl->pl_action = action;
    156 	pl->pl_addr = v;
    157 	if (++n >= pp->pr_logsize)
    158 		n = 0;
    159 	pp->pr_curlogentry = n;
    160 }
    161 
    162 static void
    163 pr_printlog(pp)
    164 	struct pool *pp;
    165 {
    166 	int i = pp->pr_logsize;
    167 	int n = pp->pr_curlogentry;
    168 
    169 	if ((pp->pr_flags & PR_LOGGING) == 0)
    170 		return;
    171 
    172 	pool_print(pp, "printlog");
    173 
    174 	/*
    175 	 * Print all entries in this pool's log.
    176 	 */
    177 	while (i-- > 0) {
    178 		struct pool_log *pl = &pp->pr_log[n];
    179 		if (pl->pl_action != 0) {
    180 			printf("log entry %d:\n", i);
    181 			printf("\taction = %s, addr = %p\n",
    182 				pl->pl_action == PRLOG_GET ? "get" : "put",
    183 				pl->pl_addr);
    184 			printf("\tfile: %s at line %lu\n",
    185 				pl->pl_file, pl->pl_line);
    186 		}
    187 		if (++n >= pp->pr_logsize)
    188 			n = 0;
    189 	}
    190 }
    191 #else
    192 #define pr_log(pp, v, action, file, line)
    193 #define pr_printlog(pp)
    194 #endif
    195 
    196 
    197 /*
    198  * Return the pool page header based on page address.
    199  */
    200 static __inline__ struct pool_item_header *
    201 pr_find_pagehead(pp, page)
    202 	struct pool *pp;
    203 	caddr_t page;
    204 {
    205 	struct pool_item_header *ph;
    206 
    207 	if ((pp->pr_flags & PR_PHINPAGE) != 0)
    208 		return ((struct pool_item_header *)(page + pp->pr_phoffset));
    209 
    210 	for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]);
    211 	     ph != NULL;
    212 	     ph = LIST_NEXT(ph, ph_hashlist)) {
    213 		if (ph->ph_page == page)
    214 			return (ph);
    215 	}
    216 	return (NULL);
    217 }
    218 
    219 /*
    220  * Remove a page from the pool.
    221  */
    222 static __inline__ void
    223 pr_rmpage(pp, ph)
    224 	struct pool *pp;
    225 	struct pool_item_header *ph;
    226 {
    227 
    228 	/*
    229 	 * Unlink a page from the pool and release it.
    230 	 */
    231 	TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
    232 	(*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
    233 	pp->pr_npages--;
    234 	pp->pr_npagefree++;
    235 
    236 	if ((pp->pr_flags & PR_PHINPAGE) == 0) {
    237 		LIST_REMOVE(ph, ph_hashlist);
    238 		pool_put(&phpool, ph);
    239 	}
    240 
    241 	if (pp->pr_curpage == ph) {
    242 		/*
    243 		 * Find a new non-empty page header, if any.
    244 		 * Start search from the page head, to increase the
    245 		 * chance for "high water" pages to be freed.
    246 		 */
    247 		for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
    248 		     ph = TAILQ_NEXT(ph, ph_pagelist))
    249 			if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
    250 				break;
    251 
    252 		pp->pr_curpage = ph;
    253 	}
    254 }
    255 
    256 /*
    257  * Allocate and initialize a pool.
    258  */
    259 struct pool *
    260 pool_create(size, align, ioff, nitems, wchan, pagesz, alloc, release, mtype)
    261 	size_t	size;
    262 	u_int	align;
    263 	u_int	ioff;
    264 	int	nitems;
    265 	char	*wchan;
    266 	size_t	pagesz;
    267 	void	*(*alloc) __P((unsigned long, int, int));
    268 	void	(*release) __P((void *, unsigned long, int));
    269 	int	mtype;
    270 {
    271 	struct pool *pp;
    272 	int flags;
    273 
    274 	pp = (struct pool *)malloc(sizeof(*pp), M_POOL, M_NOWAIT);
    275 	if (pp == NULL)
    276 		return (NULL);
    277 
    278 	flags = PR_FREEHEADER;
    279 #ifdef POOL_DIAGNOSTIC
    280 	if (pool_logsize != 0)
    281 		flags |= PR_LOGGING;
    282 #endif
    283 
    284 	pool_init(pp, size, align, ioff, flags, wchan, pagesz,
    285 		  alloc, release, mtype);
    286 
    287 	if (nitems != 0) {
    288 		if (pool_prime(pp, nitems, NULL) != 0) {
    289 			pool_destroy(pp);
    290 			return (NULL);
    291 		}
    292 	}
    293 
    294 	return (pp);
    295 }
    296 
    297 /*
    298  * Initialize the given pool resource structure.
    299  *
    300  * We export this routine to allow other kernel parts to declare
    301  * static pools that must be initialized before malloc() is available.
    302  */
    303 void
    304 pool_init(pp, size, align, ioff, flags, wchan, pagesz, alloc, release, mtype)
    305 	struct pool	*pp;
    306 	size_t		size;
    307 	u_int		align;
    308 	u_int		ioff;
    309 	int		flags;
    310 	char		*wchan;
    311 	size_t		pagesz;
    312 	void		*(*alloc) __P((unsigned long, int, int));
    313 	void		(*release) __P((void *, unsigned long, int));
    314 	int		mtype;
    315 {
    316 	int off, slack;
    317 
    318 	/*
    319 	 * Check arguments and construct default values.
    320 	 */
    321 	if (!powerof2(pagesz) || pagesz > PAGE_SIZE)
    322 		panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);
    323 
    324 	if (alloc == NULL && release == NULL) {
    325 		alloc = pool_page_alloc;
    326 		release = pool_page_free;
    327 		pagesz = PAGE_SIZE;	/* Rounds to PAGE_SIZE anyhow. */
    328 	} else if ((alloc != NULL && release != NULL) == 0) {
    329 		/* If you specifiy one, must specify both. */
    330 		panic("pool_init: must specify alloc and release together");
    331 	}
    332 
    333 	if (pagesz == 0)
    334 		pagesz = PAGE_SIZE;
    335 
    336 	if (align == 0)
    337 		align = ALIGN(1);
    338 
    339 	/*
    340 	 * Initialize the pool structure.
    341 	 */
    342 	TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
    343 	TAILQ_INIT(&pp->pr_pagelist);
    344 	pp->pr_curpage = NULL;
    345 	pp->pr_npages = 0;
    346 	pp->pr_minitems = 0;
    347 	pp->pr_minpages = 0;
    348 	pp->pr_maxpages = UINT_MAX;
    349 	pp->pr_flags = flags;
    350 	pp->pr_size = ALIGN(size);
    351 	pp->pr_align = align;
    352 	pp->pr_wchan = wchan;
    353 	pp->pr_mtype = mtype;
    354 	pp->pr_alloc = alloc;
    355 	pp->pr_free = release;
    356 	pp->pr_pagesz = pagesz;
    357 	pp->pr_pagemask = ~(pagesz - 1);
    358 	pp->pr_pageshift = ffs(pagesz) - 1;
    359 
    360 	/*
    361 	 * Decide whether to put the page header off page to avoid
    362 	 * wasting too large a part of the page. Off-page page headers
    363 	 * go on a hash table, so we can match a returned item
    364 	 * with its header based on the page address.
    365 	 * We use 1/16 of the page size as the threshold (XXX: tune)
    366 	 */
    367 	if (pp->pr_size < pagesz/16) {
    368 		/* Use the end of the page for the page header */
    369 		pp->pr_flags |= PR_PHINPAGE;
    370 		pp->pr_phoffset = off =
    371 			pagesz - ALIGN(sizeof(struct pool_item_header));
    372 	} else {
    373 		/* The page header will be taken from our page header pool */
    374 		pp->pr_phoffset = 0;
    375 		off = pagesz;
    376 		bzero(pp->pr_hashtab, sizeof(pp->pr_hashtab));
    377 	}
    378 
    379 	/*
    380 	 * Alignment is to take place at `ioff' within the item. This means
    381 	 * we must reserve up to `align - 1' bytes on the page to allow
    382 	 * appropriate positioning of each item.
    383 	 *
    384 	 * Silently enforce `0 <= ioff < align'.
    385 	 */
    386 	pp->pr_itemoffset = ioff = ioff % align;
    387 	pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
    388 
    389 	/*
    390 	 * Use the slack between the chunks and the page header
    391 	 * for "cache coloring".
    392 	 */
    393 	slack = off - pp->pr_itemsperpage * pp->pr_size;
    394 	pp->pr_maxcolor = (slack / align) * align;
    395 	pp->pr_curcolor = 0;
    396 
    397 	pp->pr_nget = 0;
    398 	pp->pr_nfail = 0;
    399 	pp->pr_nput = 0;
    400 	pp->pr_npagealloc = 0;
    401 	pp->pr_npagefree = 0;
    402 	pp->pr_hiwat = 0;
    403 
    404 #ifdef POOL_DIAGNOSTIC
    405 	if ((flags & PR_LOGGING) != 0) {
    406 		pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
    407 				    M_TEMP, M_NOWAIT);
    408 		if (pp->pr_log == NULL)
    409 			pp->pr_flags &= ~PR_LOGGING;
    410 		pp->pr_curlogentry = 0;
    411 		pp->pr_logsize = pool_logsize;
    412 	}
    413 #endif
    414 
    415 	simple_lock_init(&pp->pr_lock);
    416 
    417 	/*
    418 	 * Initialize private page header pool if we haven't done so yet.
    419 	 */
    420 	if (phpool.pr_size == 0) {
    421 		pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
    422 			  0, "phpool", 0, 0, 0, 0);
    423 	}
    424 
    425 	return;
    426 }
    427 
    428 /*
    429  * De-commision a pool resource.
    430  */
    431 void
    432 pool_destroy(pp)
    433 	struct pool *pp;
    434 {
    435 	struct pool_item_header *ph;
    436 
    437 #ifdef DIAGNOSTIC
    438 	if (pp->pr_nget - pp->pr_nput != 0) {
    439 		pr_printlog(pp);
    440 		panic("pool_destroy: pool busy: still out: %lu\n",
    441 		      pp->pr_nget - pp->pr_nput);
    442 	}
    443 #endif
    444 
    445 	/* Remove all pages */
    446 	if ((pp->pr_flags & PR_STATIC) == 0)
    447 		while ((ph = pp->pr_pagelist.tqh_first) != NULL)
    448 			pr_rmpage(pp, ph);
    449 
    450 	/* Remove from global pool list */
    451 	TAILQ_REMOVE(&pool_head, pp, pr_poollist);
    452 	drainpp = NULL;
    453 
    454 #ifdef POOL_DIAGNOSTIC
    455 	if ((pp->pr_flags & PR_LOGGING) != 0)
    456 		free(pp->pr_log, M_TEMP);
    457 #endif
    458 
    459 	if (pp->pr_flags & PR_FREEHEADER)
    460 		free(pp, M_POOL);
    461 }
    462 
    463 
    464 /*
    465  * Grab an item from the pool; must be called at appropriate spl level
    466  */
    467 #ifdef POOL_DIAGNOSTIC
    468 void *
    469 _pool_get(pp, flags, file, line)
    470 	struct pool *pp;
    471 	int flags;
    472 	const char *file;
    473 	long line;
    474 #else
    475 void *
    476 pool_get(pp, flags)
    477 	struct pool *pp;
    478 	int flags;
    479 #endif
    480 {
    481 	void *v;
    482 	struct pool_item *pi;
    483 	struct pool_item_header *ph;
    484 
    485 #ifdef DIAGNOSTIC
    486 	if ((pp->pr_flags & PR_STATIC) && (flags & PR_MALLOCOK)) {
    487 		pr_printlog(pp);
    488 		panic("pool_get: static");
    489 	}
    490 #endif
    491 
    492 	simple_lock(&pp->pr_lock);
    493 	if (curproc == NULL && (flags & PR_WAITOK) != 0)
    494 		panic("pool_get: must have NOWAIT");
    495 
    496 	/*
    497 	 * The convention we use is that if `curpage' is not NULL, then
    498 	 * it points at a non-empty bucket. In particular, `curpage'
    499 	 * never points at a page header which has PR_PHINPAGE set and
    500 	 * has no items in its bucket.
    501 	 */
    502 again:
    503 	if ((ph = pp->pr_curpage) == NULL) {
    504 		void *v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
    505 		if (v == NULL) {
    506 			if (flags & PR_URGENT)
    507 				panic("pool_get: urgent");
    508 			if ((flags & PR_WAITOK) == 0) {
    509 				pp->pr_nfail++;
    510 				simple_unlock(&pp->pr_lock);
    511 				return (NULL);
    512 			}
    513 
    514 			pp->pr_flags |= PR_WANTED;
    515 			simple_unlock(&pp->pr_lock);
    516 			tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0);
    517 			simple_lock(&pp->pr_lock);
    518 		} else {
    519 			pp->pr_npagealloc++;
    520 			pool_prime_page(pp, v);
    521 		}
    522 
    523 		goto again;
    524 	}
    525 
    526 	if ((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)
    527 		panic("pool_get: %s: page empty", pp->pr_wchan);
    528 
    529 	pr_log(pp, v, PRLOG_GET, file, line);
    530 
    531 #ifdef DIAGNOSTIC
    532 	if (pi->pi_magic != PI_MAGIC) {
    533 		pr_printlog(pp);
    534 		panic("pool_get(%s): free list modified: magic=%x; page %p;"
    535 		       " item addr %p\n",
    536 			pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
    537 	}
    538 #endif
    539 
    540 	/*
    541 	 * Remove from item list.
    542 	 */
    543 	TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
    544 	ph->ph_nmissing++;
    545 	if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
    546 		/*
    547 		 * Find a new non-empty page header, if any.
    548 		 * Start search from the page head, to increase
    549 		 * the chance for "high water" pages to be freed.
    550 		 *
    551 		 * First, move the now empty page to the head of
    552 		 * the page list.
    553 		 */
    554 		TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
    555 		TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
    556 		while ((ph = TAILQ_NEXT(ph, ph_pagelist)) != NULL)
    557 			if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
    558 				break;
    559 
    560 		pp->pr_curpage = ph;
    561 	}
    562 
    563 	pp->pr_nget++;
    564 	simple_unlock(&pp->pr_lock);
    565 	return (v);
    566 }
    567 
    568 /*
    569  * Return resource to the pool; must be called at appropriate spl level
    570  */
    571 #ifdef POOL_DIAGNOSTIC
    572 void
    573 _pool_put(pp, v, file, line)
    574 	struct pool *pp;
    575 	void *v;
    576 	const char *file;
    577 	long line;
    578 #else
    579 void
    580 pool_put(pp, v)
    581 	struct pool *pp;
    582 	void *v;
    583 #endif
    584 {
    585 	struct pool_item *pi = v;
    586 	struct pool_item_header *ph;
    587 	caddr_t page;
    588 
    589 	page = (caddr_t)((u_long)v & pp->pr_pagemask);
    590 
    591 	simple_lock(&pp->pr_lock);
    592 
    593 	pr_log(pp, v, PRLOG_PUT, file, line);
    594 
    595 	if ((ph = pr_find_pagehead(pp, page)) == NULL) {
    596 		pr_printlog(pp);
    597 		panic("pool_put: %s: page header missing", pp->pr_wchan);
    598 	}
    599 
    600 	/*
    601 	 * Return to item list.
    602 	 */
    603 #ifdef DIAGNOSTIC
    604 	pi->pi_magic = PI_MAGIC;
    605 #endif
    606 	TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
    607 	ph->ph_nmissing--;
    608 	pp->pr_nput++;
    609 
    610 	/* Cancel "pool empty" condition if it exists */
    611 	if (pp->pr_curpage == NULL)
    612 		pp->pr_curpage = ph;
    613 
    614 	if (pp->pr_flags & PR_WANTED) {
    615 		pp->pr_flags &= ~PR_WANTED;
    616 		wakeup((caddr_t)pp);
    617 		simple_unlock(&pp->pr_lock);
    618 		return;
    619 	}
    620 
    621 	/*
    622 	 * If this page is now complete, move it to the end of the pagelist.
    623 	 * If this page has just become un-empty, move it the head.
    624 	 */
    625 	if (ph->ph_nmissing == 0) {
    626 		if (pp->pr_npages > pp->pr_maxpages) {
    627 #if 0
    628 			timeout(pool_drain, 0, pool_inactive_time*hz);
    629 #else
    630 			pr_rmpage(pp, ph);
    631 #endif
    632 		} else {
    633 			TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
    634 			TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
    635 			ph->ph_time = time;
    636 
    637 			/* XXX - update curpage */
    638 			for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
    639 			     ph = TAILQ_NEXT(ph, ph_pagelist))
    640 				if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
    641 					break;
    642 
    643 			pp->pr_curpage = ph;
    644 		}
    645 	}
    646 
    647 	simple_unlock(&pp->pr_lock);
    648 }
    649 
    650 /*
    651  * Add N items to the pool.
    652  */
    653 int
    654 pool_prime(pp, n, storage)
    655 	struct pool *pp;
    656 	int n;
    657 	caddr_t storage;
    658 {
    659 	caddr_t cp;
    660 	int newnitems, newpages;
    661 
    662 #ifdef DIAGNOSTIC
    663 	if (storage && !(pp->pr_flags & PR_STATIC))
    664 		panic("pool_prime: static");
    665 	/* !storage && static caught below */
    666 #endif
    667 
    668 	newnitems = pp->pr_minitems + n;
    669 	newpages =
    670 		roundup(pp->pr_itemsperpage,newnitems) / pp->pr_itemsperpage
    671 		- pp->pr_minpages;
    672 
    673 	simple_lock(&pp->pr_lock);
    674 	while (newpages-- > 0) {
    675 
    676 		if (pp->pr_flags & PR_STATIC) {
    677 			cp = storage;
    678 			storage += pp->pr_pagesz;
    679 		} else {
    680 			cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
    681 		}
    682 
    683 		if (cp == NULL) {
    684 			simple_unlock(&pp->pr_lock);
    685 			return (ENOMEM);
    686 		}
    687 
    688 		pool_prime_page(pp, cp);
    689 		pp->pr_minpages++;
    690 	}
    691 
    692 	pp->pr_minitems = newnitems;
    693 
    694 	if (pp->pr_minpages >= pp->pr_maxpages)
    695 		pp->pr_maxpages = pp->pr_minpages + 1;	/* XXX */
    696 
    697 	simple_unlock(&pp->pr_lock);
    698 	return (0);
    699 }
    700 
    701 /*
    702  * Add a page worth of items to the pool.
    703  */
    704 int
    705 pool_prime_page(pp, storage)
    706 	struct pool *pp;
    707 	caddr_t storage;
    708 {
    709 	struct pool_item *pi;
    710 	struct pool_item_header *ph;
    711 	caddr_t cp = storage;
    712 	unsigned int align = pp->pr_align;
    713 	unsigned int ioff = pp->pr_itemoffset;
    714 	int n;
    715 
    716 	if ((pp->pr_flags & PR_PHINPAGE) != 0) {
    717 		ph = (struct pool_item_header *)(cp + pp->pr_phoffset);
    718 	} else {
    719 		ph = pool_get(&phpool, PR_URGENT);
    720 		LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
    721 				 ph, ph_hashlist);
    722 	}
    723 
    724 	/*
    725 	 * Insert page header.
    726 	 */
    727 	TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
    728 	TAILQ_INIT(&ph->ph_itemlist);
    729 	ph->ph_page = storage;
    730 	ph->ph_nmissing = 0;
    731 	ph->ph_time.tv_sec = ph->ph_time.tv_usec = 0;
    732 
    733 	/*
    734 	 * Color this page.
    735 	 */
    736 	cp = (caddr_t)(cp + pp->pr_curcolor);
    737 	if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
    738 		pp->pr_curcolor = 0;
    739 
    740 	/*
    741 	 * Adjust storage to apply aligment to `pr_itemoffset' in each item.
    742 	 */
    743 	if (ioff != 0)
    744 		cp = (caddr_t)(cp + (align - ioff));
    745 
    746 	/*
    747 	 * Insert remaining chunks on the bucket list.
    748 	 */
    749 	n = pp->pr_itemsperpage;
    750 
    751 	while (n--) {
    752 		pi = (struct pool_item *)cp;
    753 
    754 		/* Insert on page list */
    755 		TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
    756 #ifdef DIAGNOSTIC
    757 		pi->pi_magic = PI_MAGIC;
    758 #endif
    759 		cp = (caddr_t)(cp + pp->pr_size);
    760 	}
    761 
    762 	/*
    763 	 * If the pool was depleted, point at the new page.
    764 	 */
    765 	if (pp->pr_curpage == NULL)
    766 		pp->pr_curpage = ph;
    767 
    768 	if (++pp->pr_npages > pp->pr_hiwat)
    769 		pp->pr_hiwat = pp->pr_npages;
    770 
    771 	return (0);
    772 }
    773 
    774 void
    775 pool_setlowat(pp, n)
    776 	pool_handle_t	pp;
    777 	int n;
    778 {
    779 	pp->pr_minitems = n;
    780 	if (n == 0) {
    781 		pp->pr_minpages = 0;
    782 		return;
    783 	}
    784 	pp->pr_minpages =
    785 		roundup(pp->pr_itemsperpage,n) / pp->pr_itemsperpage;
    786 }
    787 
    788 void
    789 pool_sethiwat(pp, n)
    790 	pool_handle_t	pp;
    791 	int n;
    792 {
    793 	if (n == 0) {
    794 		pp->pr_maxpages = 0;
    795 		return;
    796 	}
    797 	pp->pr_maxpages =
    798 		roundup(pp->pr_itemsperpage,n) / pp->pr_itemsperpage;
    799 }
    800 
    801 
    802 /*
    803  * Default page allocator.
    804  */
    805 static void *
    806 pool_page_alloc(sz, flags, mtype)
    807 	unsigned long sz;
    808 	int flags;
    809 	int mtype;
    810 {
    811 
    812 #if defined(UVM)
    813 	return ((void *)uvm_km_alloc_poolpage());
    814 #else
    815 	return ((void *)kmem_alloc_poolpage());
    816 #endif
    817 }
    818 
    819 static void
    820 pool_page_free(v, sz, mtype)
    821 	void *v;
    822 	unsigned long sz;
    823 	int mtype;
    824 {
    825 
    826 #if defined(UVM)
    827 	uvm_km_free_poolpage((vaddr_t)v);
    828 #else
    829 	kmem_free_poolpage((vaddr_t)v);
    830 #endif
    831 }
    832 
    833 /*
    834  * Release all complete pages that have not been used recently.
    835  */
    836 void
    837 pool_reclaim (pp)
    838 	pool_handle_t pp;
    839 {
    840 	struct pool_item_header *ph, *phnext;
    841 	struct timeval curtime = time;
    842 
    843 	if (pp->pr_flags & PR_STATIC)
    844 		return;
    845 
    846 	if (simple_lock_try(&pp->pr_lock) == 0)
    847 		return;
    848 
    849 	for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
    850 		phnext = TAILQ_NEXT(ph, ph_pagelist);
    851 
    852 		/* Check our minimum page claim */
    853 		if (pp->pr_npages <= pp->pr_minpages)
    854 			break;
    855 
    856 		if (ph->ph_nmissing == 0) {
    857 			struct timeval diff;
    858 			timersub(&curtime, &ph->ph_time, &diff);
    859 			if (diff.tv_sec < pool_inactive_time)
    860 				continue;
    861 			pr_rmpage(pp, ph);
    862 		}
    863 	}
    864 
    865 	simple_unlock(&pp->pr_lock);
    866 }
    867 
    868 
    869 /*
    870  * Drain pools, one at a time.
    871  */
    872 void
    873 pool_drain(arg)
    874 	void *arg;
    875 {
    876 	struct pool *pp;
    877 	int s = splimp();
    878 
    879 	/* XXX:lock pool head */
    880 	if (drainpp == NULL && (drainpp = TAILQ_FIRST(&pool_head)) == NULL) {
    881 		splx(s);
    882 		return;
    883 	}
    884 
    885 	pp = drainpp;
    886 	drainpp = TAILQ_NEXT(pp, pr_poollist);
    887 	/* XXX:unlock pool head */
    888 
    889 	pool_reclaim(pp);
    890 	splx(s);
    891 }
    892 
    893 
    894 #ifdef DEBUG
    895 /*
    896  * Diagnostic helpers.
    897  */
    898 void
    899 pool_print(pp, label)
    900 	struct pool *pp;
    901 	char *label;
    902 {
    903 
    904 	if (label != NULL)
    905 		printf("%s: ", label);
    906 
    907 	printf("pool %s: nalloc %lu nfree %lu npagealloc %lu npagefree %lu\n"
    908 	       "         npages %u minitems %u itemsperpage %u itemoffset %u\n",
    909 		pp->pr_wchan,
    910 		pp->pr_nget,
    911 		pp->pr_nput,
    912 		pp->pr_npagealloc,
    913 		pp->pr_npagefree,
    914 		pp->pr_npages,
    915 		pp->pr_minitems,
    916 		pp->pr_itemsperpage,
    917 		pp->pr_itemoffset);
    918 }
    919 
    920 int
    921 pool_chk(pp, label)
    922 	struct pool *pp;
    923 	char *label;
    924 {
    925 	struct pool_item_header *ph;
    926 	int r = 0;
    927 
    928 	simple_lock(&pp->pr_lock);
    929 
    930 	for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
    931 	     ph = TAILQ_NEXT(ph, ph_pagelist)) {
    932 
    933 		struct pool_item *pi;
    934 		int n;
    935 		caddr_t page;
    936 
    937 		page = (caddr_t)((u_long)ph & pp->pr_pagemask);
    938 		if (page != ph->ph_page) {
    939 			if (label != NULL)
    940 				printf("%s: ", label);
    941 			printf("pool(%s): page inconsistency: page %p;"
    942 			       " at page head addr %p (p %p)\n",
    943 				pp->pr_wchan, ph->ph_page,
    944 				ph, page);
    945 			r++;
    946 			goto out;
    947 		}
    948 
    949 		for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
    950 		     pi != NULL;
    951 		     pi = TAILQ_NEXT(pi,pi_list), n++) {
    952 
    953 #ifdef DIAGNOSTIC
    954 			if (pi->pi_magic != PI_MAGIC) {
    955 				if (label != NULL)
    956 					printf("%s: ", label);
    957 				printf("pool(%s): free list modified: magic=%x;"
    958 				       " page %p; item ordinal %d;"
    959 				       " addr %p (p %p)\n",
    960 					pp->pr_wchan, pi->pi_magic, ph->ph_page,
    961 					n, pi, page);
    962 				panic("pool");
    963 			}
    964 #endif
    965 			page = (caddr_t)((u_long)pi & pp->pr_pagemask);
    966 			if (page == ph->ph_page)
    967 				continue;
    968 
    969 			if (label != NULL)
    970 				printf("%s: ", label);
    971 			printf("pool(%s): page inconsistency: page %p;"
    972 			       " item ordinal %d; addr %p (p %p)\n",
    973 				pp->pr_wchan, ph->ph_page,
    974 				n, pi, page);
    975 			r++;
    976 			goto out;
    977 		}
    978 	}
    979 out:
    980 	simple_unlock(&pp->pr_lock);
    981 	return (r);
    982 }
    983 #endif
    984