Home | History | Annotate | Line # | Download | only in kern
subr_pool.c revision 1.9
      1 /*	$NetBSD: subr_pool.c,v 1.9 1998/08/04 04:03:15 perry 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 	 * If the page was idle, decrement the idle page count.
    230 	 */
    231 	if (ph->ph_nmissing == 0) {
    232 #ifdef DIAGNOSTIC
    233 		if (pp->pr_nidle == 0)
    234 			panic("pr_rmpage: nidle inconsistent");
    235 #endif
    236 		pp->pr_nidle--;
    237 	}
    238 
    239 	/*
    240 	 * Unlink a page from the pool and release it.
    241 	 */
    242 	TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
    243 	(*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
    244 	pp->pr_npages--;
    245 	pp->pr_npagefree++;
    246 
    247 	if ((pp->pr_flags & PR_PHINPAGE) == 0) {
    248 		LIST_REMOVE(ph, ph_hashlist);
    249 		pool_put(&phpool, ph);
    250 	}
    251 
    252 	if (pp->pr_curpage == ph) {
    253 		/*
    254 		 * Find a new non-empty page header, if any.
    255 		 * Start search from the page head, to increase the
    256 		 * chance for "high water" pages to be freed.
    257 		 */
    258 		for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
    259 		     ph = TAILQ_NEXT(ph, ph_pagelist))
    260 			if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
    261 				break;
    262 
    263 		pp->pr_curpage = ph;
    264 	}
    265 }
    266 
    267 /*
    268  * Allocate and initialize a pool.
    269  */
    270 struct pool *
    271 pool_create(size, align, ioff, nitems, wchan, pagesz, alloc, release, mtype)
    272 	size_t	size;
    273 	u_int	align;
    274 	u_int	ioff;
    275 	int	nitems;
    276 	char	*wchan;
    277 	size_t	pagesz;
    278 	void	*(*alloc) __P((unsigned long, int, int));
    279 	void	(*release) __P((void *, unsigned long, int));
    280 	int	mtype;
    281 {
    282 	struct pool *pp;
    283 	int flags;
    284 
    285 	pp = (struct pool *)malloc(sizeof(*pp), M_POOL, M_NOWAIT);
    286 	if (pp == NULL)
    287 		return (NULL);
    288 
    289 	flags = PR_FREEHEADER;
    290 #ifdef POOL_DIAGNOSTIC
    291 	if (pool_logsize != 0)
    292 		flags |= PR_LOGGING;
    293 #endif
    294 
    295 	pool_init(pp, size, align, ioff, flags, wchan, pagesz,
    296 		  alloc, release, mtype);
    297 
    298 	if (nitems != 0) {
    299 		if (pool_prime(pp, nitems, NULL) != 0) {
    300 			pool_destroy(pp);
    301 			return (NULL);
    302 		}
    303 	}
    304 
    305 	return (pp);
    306 }
    307 
    308 /*
    309  * Initialize the given pool resource structure.
    310  *
    311  * We export this routine to allow other kernel parts to declare
    312  * static pools that must be initialized before malloc() is available.
    313  */
    314 void
    315 pool_init(pp, size, align, ioff, flags, wchan, pagesz, alloc, release, mtype)
    316 	struct pool	*pp;
    317 	size_t		size;
    318 	u_int		align;
    319 	u_int		ioff;
    320 	int		flags;
    321 	char		*wchan;
    322 	size_t		pagesz;
    323 	void		*(*alloc) __P((unsigned long, int, int));
    324 	void		(*release) __P((void *, unsigned long, int));
    325 	int		mtype;
    326 {
    327 	int off, slack;
    328 
    329 	/*
    330 	 * Check arguments and construct default values.
    331 	 */
    332 	if (!powerof2(pagesz) || pagesz > PAGE_SIZE)
    333 		panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);
    334 
    335 	if (alloc == NULL && release == NULL) {
    336 		alloc = pool_page_alloc;
    337 		release = pool_page_free;
    338 		pagesz = PAGE_SIZE;	/* Rounds to PAGE_SIZE anyhow. */
    339 	} else if ((alloc != NULL && release != NULL) == 0) {
    340 		/* If you specifiy one, must specify both. */
    341 		panic("pool_init: must specify alloc and release together");
    342 	}
    343 
    344 	if (pagesz == 0)
    345 		pagesz = PAGE_SIZE;
    346 
    347 	if (align == 0)
    348 		align = ALIGN(1);
    349 
    350 	/*
    351 	 * Initialize the pool structure.
    352 	 */
    353 	TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
    354 	TAILQ_INIT(&pp->pr_pagelist);
    355 	pp->pr_curpage = NULL;
    356 	pp->pr_npages = 0;
    357 	pp->pr_minitems = 0;
    358 	pp->pr_minpages = 0;
    359 	pp->pr_maxpages = UINT_MAX;
    360 	pp->pr_flags = flags;
    361 	pp->pr_size = ALIGN(size);
    362 	pp->pr_align = align;
    363 	pp->pr_wchan = wchan;
    364 	pp->pr_mtype = mtype;
    365 	pp->pr_alloc = alloc;
    366 	pp->pr_free = release;
    367 	pp->pr_pagesz = pagesz;
    368 	pp->pr_pagemask = ~(pagesz - 1);
    369 	pp->pr_pageshift = ffs(pagesz) - 1;
    370 
    371 	/*
    372 	 * Decide whether to put the page header off page to avoid
    373 	 * wasting too large a part of the page. Off-page page headers
    374 	 * go on a hash table, so we can match a returned item
    375 	 * with its header based on the page address.
    376 	 * We use 1/16 of the page size as the threshold (XXX: tune)
    377 	 */
    378 	if (pp->pr_size < pagesz/16) {
    379 		/* Use the end of the page for the page header */
    380 		pp->pr_flags |= PR_PHINPAGE;
    381 		pp->pr_phoffset = off =
    382 			pagesz - ALIGN(sizeof(struct pool_item_header));
    383 	} else {
    384 		/* The page header will be taken from our page header pool */
    385 		pp->pr_phoffset = 0;
    386 		off = pagesz;
    387 		memset(pp->pr_hashtab, 0, sizeof(pp->pr_hashtab));
    388 	}
    389 
    390 	/*
    391 	 * Alignment is to take place at `ioff' within the item. This means
    392 	 * we must reserve up to `align - 1' bytes on the page to allow
    393 	 * appropriate positioning of each item.
    394 	 *
    395 	 * Silently enforce `0 <= ioff < align'.
    396 	 */
    397 	pp->pr_itemoffset = ioff = ioff % align;
    398 	pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
    399 
    400 	/*
    401 	 * Use the slack between the chunks and the page header
    402 	 * for "cache coloring".
    403 	 */
    404 	slack = off - pp->pr_itemsperpage * pp->pr_size;
    405 	pp->pr_maxcolor = (slack / align) * align;
    406 	pp->pr_curcolor = 0;
    407 
    408 	pp->pr_nget = 0;
    409 	pp->pr_nfail = 0;
    410 	pp->pr_nput = 0;
    411 	pp->pr_npagealloc = 0;
    412 	pp->pr_npagefree = 0;
    413 	pp->pr_hiwat = 0;
    414 	pp->pr_nidle = 0;
    415 
    416 #ifdef POOL_DIAGNOSTIC
    417 	if ((flags & PR_LOGGING) != 0) {
    418 		pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
    419 				    M_TEMP, M_NOWAIT);
    420 		if (pp->pr_log == NULL)
    421 			pp->pr_flags &= ~PR_LOGGING;
    422 		pp->pr_curlogentry = 0;
    423 		pp->pr_logsize = pool_logsize;
    424 	}
    425 #endif
    426 
    427 	simple_lock_init(&pp->pr_lock);
    428 
    429 	/*
    430 	 * Initialize private page header pool if we haven't done so yet.
    431 	 */
    432 	if (phpool.pr_size == 0) {
    433 		pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
    434 			  0, "phpool", 0, 0, 0, 0);
    435 	}
    436 
    437 	return;
    438 }
    439 
    440 /*
    441  * De-commision a pool resource.
    442  */
    443 void
    444 pool_destroy(pp)
    445 	struct pool *pp;
    446 {
    447 	struct pool_item_header *ph;
    448 
    449 #ifdef DIAGNOSTIC
    450 	if (pp->pr_nget - pp->pr_nput != 0) {
    451 		pr_printlog(pp);
    452 		panic("pool_destroy: pool busy: still out: %lu\n",
    453 		      pp->pr_nget - pp->pr_nput);
    454 	}
    455 #endif
    456 
    457 	/* Remove all pages */
    458 	if ((pp->pr_flags & PR_STATIC) == 0)
    459 		while ((ph = pp->pr_pagelist.tqh_first) != NULL)
    460 			pr_rmpage(pp, ph);
    461 
    462 	/* Remove from global pool list */
    463 	TAILQ_REMOVE(&pool_head, pp, pr_poollist);
    464 	drainpp = NULL;
    465 
    466 #ifdef POOL_DIAGNOSTIC
    467 	if ((pp->pr_flags & PR_LOGGING) != 0)
    468 		free(pp->pr_log, M_TEMP);
    469 #endif
    470 
    471 	if (pp->pr_flags & PR_FREEHEADER)
    472 		free(pp, M_POOL);
    473 }
    474 
    475 
    476 /*
    477  * Grab an item from the pool; must be called at appropriate spl level
    478  */
    479 #ifdef POOL_DIAGNOSTIC
    480 void *
    481 _pool_get(pp, flags, file, line)
    482 	struct pool *pp;
    483 	int flags;
    484 	const char *file;
    485 	long line;
    486 #else
    487 void *
    488 pool_get(pp, flags)
    489 	struct pool *pp;
    490 	int flags;
    491 #endif
    492 {
    493 	void *v;
    494 	struct pool_item *pi;
    495 	struct pool_item_header *ph;
    496 
    497 #ifdef DIAGNOSTIC
    498 	if ((pp->pr_flags & PR_STATIC) && (flags & PR_MALLOCOK)) {
    499 		pr_printlog(pp);
    500 		panic("pool_get: static");
    501 	}
    502 #endif
    503 
    504 	simple_lock(&pp->pr_lock);
    505 	if (curproc == NULL && (flags & PR_WAITOK) != 0)
    506 		panic("pool_get: must have NOWAIT");
    507 
    508 	/*
    509 	 * The convention we use is that if `curpage' is not NULL, then
    510 	 * it points at a non-empty bucket. In particular, `curpage'
    511 	 * never points at a page header which has PR_PHINPAGE set and
    512 	 * has no items in its bucket.
    513 	 */
    514 again:
    515 	if ((ph = pp->pr_curpage) == NULL) {
    516 		void *v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
    517 		if (v == NULL) {
    518 			if (flags & PR_URGENT)
    519 				panic("pool_get: urgent");
    520 			if ((flags & PR_WAITOK) == 0) {
    521 				pp->pr_nfail++;
    522 				simple_unlock(&pp->pr_lock);
    523 				return (NULL);
    524 			}
    525 
    526 			pp->pr_flags |= PR_WANTED;
    527 			simple_unlock(&pp->pr_lock);
    528 			tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0);
    529 			simple_lock(&pp->pr_lock);
    530 		} else {
    531 			pp->pr_npagealloc++;
    532 			pool_prime_page(pp, v);
    533 		}
    534 
    535 		goto again;
    536 	}
    537 
    538 	if ((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)
    539 		panic("pool_get: %s: page empty", pp->pr_wchan);
    540 
    541 	pr_log(pp, v, PRLOG_GET, file, line);
    542 
    543 #ifdef DIAGNOSTIC
    544 	if (pi->pi_magic != PI_MAGIC) {
    545 		pr_printlog(pp);
    546 		panic("pool_get(%s): free list modified: magic=%x; page %p;"
    547 		       " item addr %p\n",
    548 			pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
    549 	}
    550 #endif
    551 
    552 	/*
    553 	 * Remove from item list.
    554 	 */
    555 	TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
    556 	if (ph->ph_nmissing == 0) {
    557 #ifdef DIAGNOSTIC
    558 		if (pp->pr_nidle == 0)
    559 			panic("pool_get: nidle inconsistent");
    560 #endif
    561 		pp->pr_nidle--;
    562 	}
    563 	ph->ph_nmissing++;
    564 	if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
    565 		/*
    566 		 * Find a new non-empty page header, if any.
    567 		 * Start search from the page head, to increase
    568 		 * the chance for "high water" pages to be freed.
    569 		 *
    570 		 * First, move the now empty page to the head of
    571 		 * the page list.
    572 		 */
    573 		TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
    574 		TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
    575 		while ((ph = TAILQ_NEXT(ph, ph_pagelist)) != NULL)
    576 			if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
    577 				break;
    578 
    579 		pp->pr_curpage = ph;
    580 	}
    581 
    582 	pp->pr_nget++;
    583 	simple_unlock(&pp->pr_lock);
    584 	return (v);
    585 }
    586 
    587 /*
    588  * Return resource to the pool; must be called at appropriate spl level
    589  */
    590 #ifdef POOL_DIAGNOSTIC
    591 void
    592 _pool_put(pp, v, file, line)
    593 	struct pool *pp;
    594 	void *v;
    595 	const char *file;
    596 	long line;
    597 #else
    598 void
    599 pool_put(pp, v)
    600 	struct pool *pp;
    601 	void *v;
    602 #endif
    603 {
    604 	struct pool_item *pi = v;
    605 	struct pool_item_header *ph;
    606 	caddr_t page;
    607 
    608 	page = (caddr_t)((u_long)v & pp->pr_pagemask);
    609 
    610 	simple_lock(&pp->pr_lock);
    611 
    612 	pr_log(pp, v, PRLOG_PUT, file, line);
    613 
    614 	if ((ph = pr_find_pagehead(pp, page)) == NULL) {
    615 		pr_printlog(pp);
    616 		panic("pool_put: %s: page header missing", pp->pr_wchan);
    617 	}
    618 
    619 	/*
    620 	 * Return to item list.
    621 	 */
    622 #ifdef DIAGNOSTIC
    623 	pi->pi_magic = PI_MAGIC;
    624 #endif
    625 	TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
    626 	ph->ph_nmissing--;
    627 	pp->pr_nput++;
    628 
    629 	/* Cancel "pool empty" condition if it exists */
    630 	if (pp->pr_curpage == NULL)
    631 		pp->pr_curpage = ph;
    632 
    633 	if (pp->pr_flags & PR_WANTED) {
    634 		pp->pr_flags &= ~PR_WANTED;
    635 		wakeup((caddr_t)pp);
    636 		simple_unlock(&pp->pr_lock);
    637 		return;
    638 	}
    639 
    640 	/*
    641 	 * If this page is now complete, move it to the end of the pagelist.
    642 	 * If this page has just become un-empty, move it the head.
    643 	 */
    644 	if (ph->ph_nmissing == 0) {
    645 		pp->pr_nidle++;
    646 		if (pp->pr_npages > pp->pr_maxpages) {
    647 #if 0
    648 			timeout(pool_drain, 0, pool_inactive_time*hz);
    649 #else
    650 			pr_rmpage(pp, ph);
    651 #endif
    652 		} else {
    653 			TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
    654 			TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
    655 			ph->ph_time = time;
    656 
    657 			/* XXX - update curpage */
    658 			for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
    659 			     ph = TAILQ_NEXT(ph, ph_pagelist))
    660 				if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
    661 					break;
    662 
    663 			pp->pr_curpage = ph;
    664 		}
    665 	}
    666 
    667 	simple_unlock(&pp->pr_lock);
    668 }
    669 
    670 /*
    671  * Add N items to the pool.
    672  */
    673 int
    674 pool_prime(pp, n, storage)
    675 	struct pool *pp;
    676 	int n;
    677 	caddr_t storage;
    678 {
    679 	caddr_t cp;
    680 	int newnitems, newpages;
    681 
    682 #ifdef DIAGNOSTIC
    683 	if (storage && !(pp->pr_flags & PR_STATIC))
    684 		panic("pool_prime: static");
    685 	/* !storage && static caught below */
    686 #endif
    687 
    688 	newnitems = pp->pr_minitems + n;
    689 	newpages =
    690 		roundup(pp->pr_itemsperpage,newnitems) / pp->pr_itemsperpage
    691 		- pp->pr_minpages;
    692 
    693 	simple_lock(&pp->pr_lock);
    694 	while (newpages-- > 0) {
    695 
    696 		if (pp->pr_flags & PR_STATIC) {
    697 			cp = storage;
    698 			storage += pp->pr_pagesz;
    699 		} else {
    700 			cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
    701 		}
    702 
    703 		if (cp == NULL) {
    704 			simple_unlock(&pp->pr_lock);
    705 			return (ENOMEM);
    706 		}
    707 
    708 		pool_prime_page(pp, cp);
    709 		pp->pr_minpages++;
    710 	}
    711 
    712 	pp->pr_minitems = newnitems;
    713 
    714 	if (pp->pr_minpages >= pp->pr_maxpages)
    715 		pp->pr_maxpages = pp->pr_minpages + 1;	/* XXX */
    716 
    717 	simple_unlock(&pp->pr_lock);
    718 	return (0);
    719 }
    720 
    721 /*
    722  * Add a page worth of items to the pool.
    723  */
    724 int
    725 pool_prime_page(pp, storage)
    726 	struct pool *pp;
    727 	caddr_t storage;
    728 {
    729 	struct pool_item *pi;
    730 	struct pool_item_header *ph;
    731 	caddr_t cp = storage;
    732 	unsigned int align = pp->pr_align;
    733 	unsigned int ioff = pp->pr_itemoffset;
    734 	int n;
    735 
    736 	if ((pp->pr_flags & PR_PHINPAGE) != 0) {
    737 		ph = (struct pool_item_header *)(cp + pp->pr_phoffset);
    738 	} else {
    739 		ph = pool_get(&phpool, PR_URGENT);
    740 		LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
    741 				 ph, ph_hashlist);
    742 	}
    743 
    744 	/*
    745 	 * Insert page header.
    746 	 */
    747 	TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
    748 	TAILQ_INIT(&ph->ph_itemlist);
    749 	ph->ph_page = storage;
    750 	ph->ph_nmissing = 0;
    751 	ph->ph_time.tv_sec = ph->ph_time.tv_usec = 0;
    752 
    753 	pp->pr_nidle++;
    754 
    755 	/*
    756 	 * Color this page.
    757 	 */
    758 	cp = (caddr_t)(cp + pp->pr_curcolor);
    759 	if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
    760 		pp->pr_curcolor = 0;
    761 
    762 	/*
    763 	 * Adjust storage to apply aligment to `pr_itemoffset' in each item.
    764 	 */
    765 	if (ioff != 0)
    766 		cp = (caddr_t)(cp + (align - ioff));
    767 
    768 	/*
    769 	 * Insert remaining chunks on the bucket list.
    770 	 */
    771 	n = pp->pr_itemsperpage;
    772 
    773 	while (n--) {
    774 		pi = (struct pool_item *)cp;
    775 
    776 		/* Insert on page list */
    777 		TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
    778 #ifdef DIAGNOSTIC
    779 		pi->pi_magic = PI_MAGIC;
    780 #endif
    781 		cp = (caddr_t)(cp + pp->pr_size);
    782 	}
    783 
    784 	/*
    785 	 * If the pool was depleted, point at the new page.
    786 	 */
    787 	if (pp->pr_curpage == NULL)
    788 		pp->pr_curpage = ph;
    789 
    790 	if (++pp->pr_npages > pp->pr_hiwat)
    791 		pp->pr_hiwat = pp->pr_npages;
    792 
    793 	return (0);
    794 }
    795 
    796 void
    797 pool_setlowat(pp, n)
    798 	pool_handle_t	pp;
    799 	int n;
    800 {
    801 	pp->pr_minitems = n;
    802 	if (n == 0) {
    803 		pp->pr_minpages = 0;
    804 		return;
    805 	}
    806 	pp->pr_minpages =
    807 		roundup(pp->pr_itemsperpage,n) / pp->pr_itemsperpage;
    808 }
    809 
    810 void
    811 pool_sethiwat(pp, n)
    812 	pool_handle_t	pp;
    813 	int n;
    814 {
    815 	if (n == 0) {
    816 		pp->pr_maxpages = 0;
    817 		return;
    818 	}
    819 	pp->pr_maxpages =
    820 		roundup(pp->pr_itemsperpage,n) / pp->pr_itemsperpage;
    821 }
    822 
    823 
    824 /*
    825  * Default page allocator.
    826  */
    827 static void *
    828 pool_page_alloc(sz, flags, mtype)
    829 	unsigned long sz;
    830 	int flags;
    831 	int mtype;
    832 {
    833 
    834 #if defined(UVM)
    835 	return ((void *)uvm_km_alloc_poolpage());
    836 #else
    837 	return ((void *)kmem_alloc_poolpage());
    838 #endif
    839 }
    840 
    841 static void
    842 pool_page_free(v, sz, mtype)
    843 	void *v;
    844 	unsigned long sz;
    845 	int mtype;
    846 {
    847 
    848 #if defined(UVM)
    849 	uvm_km_free_poolpage((vm_offset_t)v);
    850 #else
    851 	kmem_free_poolpage((vm_offset_t)v);
    852 #endif
    853 }
    854 
    855 /*
    856  * Release all complete pages that have not been used recently.
    857  */
    858 void
    859 pool_reclaim (pp)
    860 	pool_handle_t pp;
    861 {
    862 	struct pool_item_header *ph, *phnext;
    863 	struct timeval curtime = time;
    864 
    865 	if (pp->pr_flags & PR_STATIC)
    866 		return;
    867 
    868 	if (simple_lock_try(&pp->pr_lock) == 0)
    869 		return;
    870 
    871 	for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
    872 		phnext = TAILQ_NEXT(ph, ph_pagelist);
    873 
    874 		/* Check our minimum page claim */
    875 		if (pp->pr_npages <= pp->pr_minpages)
    876 			break;
    877 
    878 		if (ph->ph_nmissing == 0) {
    879 			struct timeval diff;
    880 			timersub(&curtime, &ph->ph_time, &diff);
    881 			if (diff.tv_sec < pool_inactive_time)
    882 				continue;
    883 			pr_rmpage(pp, ph);
    884 		}
    885 	}
    886 
    887 	simple_unlock(&pp->pr_lock);
    888 }
    889 
    890 
    891 /*
    892  * Drain pools, one at a time.
    893  */
    894 void
    895 pool_drain(arg)
    896 	void *arg;
    897 {
    898 	struct pool *pp;
    899 	int s = splimp();
    900 
    901 	/* XXX:lock pool head */
    902 	if (drainpp == NULL && (drainpp = TAILQ_FIRST(&pool_head)) == NULL) {
    903 		splx(s);
    904 		return;
    905 	}
    906 
    907 	pp = drainpp;
    908 	drainpp = TAILQ_NEXT(pp, pr_poollist);
    909 	/* XXX:unlock pool head */
    910 
    911 	pool_reclaim(pp);
    912 	splx(s);
    913 }
    914 
    915 
    916 #ifdef DEBUG
    917 /*
    918  * Diagnostic helpers.
    919  */
    920 void
    921 pool_print(pp, label)
    922 	struct pool *pp;
    923 	char *label;
    924 {
    925 
    926 	if (label != NULL)
    927 		printf("%s: ", label);
    928 
    929 	printf("pool %s: nalloc %lu nfree %lu npagealloc %lu npagefree %lu\n"
    930 	       "         npages %u minitems %u itemsperpage %u itemoffset %u\n"
    931 	       "         nidle %lu\n",
    932 		pp->pr_wchan,
    933 		pp->pr_nget,
    934 		pp->pr_nput,
    935 		pp->pr_npagealloc,
    936 		pp->pr_npagefree,
    937 		pp->pr_npages,
    938 		pp->pr_minitems,
    939 		pp->pr_itemsperpage,
    940 		pp->pr_itemoffset,
    941 		pp->pr_nidle);
    942 }
    943 
    944 int
    945 pool_chk(pp, label)
    946 	struct pool *pp;
    947 	char *label;
    948 {
    949 	struct pool_item_header *ph;
    950 	int r = 0;
    951 
    952 	simple_lock(&pp->pr_lock);
    953 
    954 	for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
    955 	     ph = TAILQ_NEXT(ph, ph_pagelist)) {
    956 
    957 		struct pool_item *pi;
    958 		int n;
    959 		caddr_t page;
    960 
    961 		page = (caddr_t)((u_long)ph & pp->pr_pagemask);
    962 		if (page != ph->ph_page) {
    963 			if (label != NULL)
    964 				printf("%s: ", label);
    965 			printf("pool(%s): page inconsistency: page %p;"
    966 			       " at page head addr %p (p %p)\n",
    967 				pp->pr_wchan, ph->ph_page,
    968 				ph, page);
    969 			r++;
    970 			goto out;
    971 		}
    972 
    973 		for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
    974 		     pi != NULL;
    975 		     pi = TAILQ_NEXT(pi,pi_list), n++) {
    976 
    977 #ifdef DIAGNOSTIC
    978 			if (pi->pi_magic != PI_MAGIC) {
    979 				if (label != NULL)
    980 					printf("%s: ", label);
    981 				printf("pool(%s): free list modified: magic=%x;"
    982 				       " page %p; item ordinal %d;"
    983 				       " addr %p (p %p)\n",
    984 					pp->pr_wchan, pi->pi_magic, ph->ph_page,
    985 					n, pi, page);
    986 				panic("pool");
    987 			}
    988 #endif
    989 			page = (caddr_t)((u_long)pi & pp->pr_pagemask);
    990 			if (page == ph->ph_page)
    991 				continue;
    992 
    993 			if (label != NULL)
    994 				printf("%s: ", label);
    995 			printf("pool(%s): page inconsistency: page %p;"
    996 			       " item ordinal %d; addr %p (p %p)\n",
    997 				pp->pr_wchan, ph->ph_page,
    998 				n, pi, page);
    999 			r++;
   1000 			goto out;
   1001 		}
   1002 	}
   1003 out:
   1004 	simple_unlock(&pp->pr_lock);
   1005 	return (r);
   1006 }
   1007 #endif
   1008