Home | History | Annotate | Line # | Download | only in kern
subr_pool.c revision 1.6
      1 /*	$NetBSD: subr_pool.c,v 1.6 1998/08/01 23:44:21 thorpej 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 (ph->ph_nmissing == 0) {
    237 #ifdef DIAGNOSTIC
    238 		if (pp->pr_nidle == 0)
    239 			panic("pr_rmpage: nidle inconsistent");
    240 #endif
    241 		pp->pr_nidle--;
    242 	}
    243 
    244 	if ((pp->pr_flags & PR_PHINPAGE) == 0) {
    245 		LIST_REMOVE(ph, ph_hashlist);
    246 		pool_put(&phpool, ph);
    247 	}
    248 
    249 	if (pp->pr_curpage == ph) {
    250 		/*
    251 		 * Find a new non-empty page header, if any.
    252 		 * Start search from the page head, to increase the
    253 		 * chance for "high water" pages to be freed.
    254 		 */
    255 		for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
    256 		     ph = TAILQ_NEXT(ph, ph_pagelist))
    257 			if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
    258 				break;
    259 
    260 		pp->pr_curpage = ph;
    261 	}
    262 }
    263 
    264 /*
    265  * Allocate and initialize a pool.
    266  */
    267 struct pool *
    268 pool_create(size, align, ioff, nitems, wchan, pagesz, alloc, release, mtype)
    269 	size_t	size;
    270 	u_int	align;
    271 	u_int	ioff;
    272 	int	nitems;
    273 	char	*wchan;
    274 	size_t	pagesz;
    275 	void	*(*alloc) __P((unsigned long, int, int));
    276 	void	(*release) __P((void *, unsigned long, int));
    277 	int	mtype;
    278 {
    279 	struct pool *pp;
    280 	int flags;
    281 
    282 	pp = (struct pool *)malloc(sizeof(*pp), M_POOL, M_NOWAIT);
    283 	if (pp == NULL)
    284 		return (NULL);
    285 
    286 	flags = PR_FREEHEADER;
    287 #ifdef POOL_DIAGNOSTIC
    288 	if (pool_logsize != 0)
    289 		flags |= PR_LOGGING;
    290 #endif
    291 
    292 	pool_init(pp, size, align, ioff, flags, wchan, pagesz,
    293 		  alloc, release, mtype);
    294 
    295 	if (nitems != 0) {
    296 		if (pool_prime(pp, nitems, NULL) != 0) {
    297 			pool_destroy(pp);
    298 			return (NULL);
    299 		}
    300 	}
    301 
    302 	return (pp);
    303 }
    304 
    305 /*
    306  * Initialize the given pool resource structure.
    307  *
    308  * We export this routine to allow other kernel parts to declare
    309  * static pools that must be initialized before malloc() is available.
    310  */
    311 void
    312 pool_init(pp, size, align, ioff, flags, wchan, pagesz, alloc, release, mtype)
    313 	struct pool	*pp;
    314 	size_t		size;
    315 	u_int		align;
    316 	u_int		ioff;
    317 	int		flags;
    318 	char		*wchan;
    319 	size_t		pagesz;
    320 	void		*(*alloc) __P((unsigned long, int, int));
    321 	void		(*release) __P((void *, unsigned long, int));
    322 	int		mtype;
    323 {
    324 	int off, slack;
    325 
    326 	/*
    327 	 * Check arguments and construct default values.
    328 	 */
    329 	if (!powerof2(pagesz) || pagesz > PAGE_SIZE)
    330 		panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);
    331 
    332 	if (alloc == NULL && release == NULL) {
    333 		alloc = pool_page_alloc;
    334 		release = pool_page_free;
    335 		pagesz = PAGE_SIZE;	/* Rounds to PAGE_SIZE anyhow. */
    336 	} else if ((alloc != NULL && release != NULL) == 0) {
    337 		/* If you specifiy one, must specify both. */
    338 		panic("pool_init: must specify alloc and release together");
    339 	}
    340 
    341 	if (pagesz == 0)
    342 		pagesz = PAGE_SIZE;
    343 
    344 	if (align == 0)
    345 		align = ALIGN(1);
    346 
    347 	/*
    348 	 * Initialize the pool structure.
    349 	 */
    350 	TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
    351 	TAILQ_INIT(&pp->pr_pagelist);
    352 	pp->pr_curpage = NULL;
    353 	pp->pr_npages = 0;
    354 	pp->pr_minitems = 0;
    355 	pp->pr_minpages = 0;
    356 	pp->pr_maxpages = UINT_MAX;
    357 	pp->pr_flags = flags;
    358 	pp->pr_size = ALIGN(size);
    359 	pp->pr_align = align;
    360 	pp->pr_wchan = wchan;
    361 	pp->pr_mtype = mtype;
    362 	pp->pr_alloc = alloc;
    363 	pp->pr_free = release;
    364 	pp->pr_pagesz = pagesz;
    365 	pp->pr_pagemask = ~(pagesz - 1);
    366 	pp->pr_pageshift = ffs(pagesz) - 1;
    367 
    368 	/*
    369 	 * Decide whether to put the page header off page to avoid
    370 	 * wasting too large a part of the page. Off-page page headers
    371 	 * go on a hash table, so we can match a returned item
    372 	 * with its header based on the page address.
    373 	 * We use 1/16 of the page size as the threshold (XXX: tune)
    374 	 */
    375 	if (pp->pr_size < pagesz/16) {
    376 		/* Use the end of the page for the page header */
    377 		pp->pr_flags |= PR_PHINPAGE;
    378 		pp->pr_phoffset = off =
    379 			pagesz - ALIGN(sizeof(struct pool_item_header));
    380 	} else {
    381 		/* The page header will be taken from our page header pool */
    382 		pp->pr_phoffset = 0;
    383 		off = pagesz;
    384 		bzero(pp->pr_hashtab, sizeof(pp->pr_hashtab));
    385 	}
    386 
    387 	/*
    388 	 * Alignment is to take place at `ioff' within the item. This means
    389 	 * we must reserve up to `align - 1' bytes on the page to allow
    390 	 * appropriate positioning of each item.
    391 	 *
    392 	 * Silently enforce `0 <= ioff < align'.
    393 	 */
    394 	pp->pr_itemoffset = ioff = ioff % align;
    395 	pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
    396 
    397 	/*
    398 	 * Use the slack between the chunks and the page header
    399 	 * for "cache coloring".
    400 	 */
    401 	slack = off - pp->pr_itemsperpage * pp->pr_size;
    402 	pp->pr_maxcolor = (slack / align) * align;
    403 	pp->pr_curcolor = 0;
    404 
    405 	pp->pr_nget = 0;
    406 	pp->pr_nfail = 0;
    407 	pp->pr_nput = 0;
    408 	pp->pr_npagealloc = 0;
    409 	pp->pr_npagefree = 0;
    410 	pp->pr_hiwat = 0;
    411 
    412 #ifdef POOL_DIAGNOSTIC
    413 	if ((flags & PR_LOGGING) != 0) {
    414 		pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
    415 				    M_TEMP, M_NOWAIT);
    416 		if (pp->pr_log == NULL)
    417 			pp->pr_flags &= ~PR_LOGGING;
    418 		pp->pr_curlogentry = 0;
    419 		pp->pr_logsize = pool_logsize;
    420 	}
    421 #endif
    422 
    423 	simple_lock_init(&pp->pr_lock);
    424 
    425 	/*
    426 	 * Initialize private page header pool if we haven't done so yet.
    427 	 */
    428 	if (phpool.pr_size == 0) {
    429 		pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
    430 			  0, "phpool", 0, 0, 0, 0);
    431 	}
    432 
    433 	return;
    434 }
    435 
    436 /*
    437  * De-commision a pool resource.
    438  */
    439 void
    440 pool_destroy(pp)
    441 	struct pool *pp;
    442 {
    443 	struct pool_item_header *ph;
    444 
    445 #ifdef DIAGNOSTIC
    446 	if (pp->pr_nget - pp->pr_nput != 0) {
    447 		pr_printlog(pp);
    448 		panic("pool_destroy: pool busy: still out: %lu\n",
    449 		      pp->pr_nget - pp->pr_nput);
    450 	}
    451 #endif
    452 
    453 	/* Remove all pages */
    454 	if ((pp->pr_flags & PR_STATIC) == 0)
    455 		while ((ph = pp->pr_pagelist.tqh_first) != NULL)
    456 			pr_rmpage(pp, ph);
    457 
    458 	/* Remove from global pool list */
    459 	TAILQ_REMOVE(&pool_head, pp, pr_poollist);
    460 	drainpp = NULL;
    461 
    462 #ifdef POOL_DIAGNOSTIC
    463 	if ((pp->pr_flags & PR_LOGGING) != 0)
    464 		free(pp->pr_log, M_TEMP);
    465 #endif
    466 
    467 	if (pp->pr_flags & PR_FREEHEADER)
    468 		free(pp, M_POOL);
    469 }
    470 
    471 
    472 /*
    473  * Grab an item from the pool; must be called at appropriate spl level
    474  */
    475 #ifdef POOL_DIAGNOSTIC
    476 void *
    477 _pool_get(pp, flags, file, line)
    478 	struct pool *pp;
    479 	int flags;
    480 	const char *file;
    481 	long line;
    482 #else
    483 void *
    484 pool_get(pp, flags)
    485 	struct pool *pp;
    486 	int flags;
    487 #endif
    488 {
    489 	void *v;
    490 	struct pool_item *pi;
    491 	struct pool_item_header *ph;
    492 
    493 #ifdef DIAGNOSTIC
    494 	if ((pp->pr_flags & PR_STATIC) && (flags & PR_MALLOCOK)) {
    495 		pr_printlog(pp);
    496 		panic("pool_get: static");
    497 	}
    498 #endif
    499 
    500 	simple_lock(&pp->pr_lock);
    501 	if (curproc == NULL && (flags & PR_WAITOK) != 0)
    502 		panic("pool_get: must have NOWAIT");
    503 
    504 	/*
    505 	 * The convention we use is that if `curpage' is not NULL, then
    506 	 * it points at a non-empty bucket. In particular, `curpage'
    507 	 * never points at a page header which has PR_PHINPAGE set and
    508 	 * has no items in its bucket.
    509 	 */
    510 again:
    511 	if ((ph = pp->pr_curpage) == NULL) {
    512 		void *v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
    513 		if (v == NULL) {
    514 			if (flags & PR_URGENT)
    515 				panic("pool_get: urgent");
    516 			if ((flags & PR_WAITOK) == 0) {
    517 				pp->pr_nfail++;
    518 				simple_unlock(&pp->pr_lock);
    519 				return (NULL);
    520 			}
    521 
    522 			pp->pr_flags |= PR_WANTED;
    523 			simple_unlock(&pp->pr_lock);
    524 			tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0);
    525 			simple_lock(&pp->pr_lock);
    526 		} else {
    527 			pp->pr_npagealloc++;
    528 			pool_prime_page(pp, v);
    529 		}
    530 
    531 		goto again;
    532 	}
    533 
    534 	if ((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)
    535 		panic("pool_get: %s: page empty", pp->pr_wchan);
    536 
    537 	pr_log(pp, v, PRLOG_GET, file, line);
    538 
    539 #ifdef DIAGNOSTIC
    540 	if (pi->pi_magic != PI_MAGIC) {
    541 		pr_printlog(pp);
    542 		panic("pool_get(%s): free list modified: magic=%x; page %p;"
    543 		       " item addr %p\n",
    544 			pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
    545 	}
    546 #endif
    547 
    548 	/*
    549 	 * Remove from item list.
    550 	 */
    551 	TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
    552 	if (ph->ph_nmissing == 0) {
    553 #ifdef DIAGNOSTIC
    554 		if (pp->pr_nidle == 0)
    555 			panic("pool_get: nidle inconsistent");
    556 #endif
    557 		pp->pr_nidle--;
    558 	}
    559 	ph->ph_nmissing++;
    560 	if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
    561 		/*
    562 		 * Find a new non-empty page header, if any.
    563 		 * Start search from the page head, to increase
    564 		 * the chance for "high water" pages to be freed.
    565 		 *
    566 		 * First, move the now empty page to the head of
    567 		 * the page list.
    568 		 */
    569 		TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
    570 		TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
    571 		while ((ph = TAILQ_NEXT(ph, ph_pagelist)) != NULL)
    572 			if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
    573 				break;
    574 
    575 		pp->pr_curpage = ph;
    576 	}
    577 
    578 	pp->pr_nget++;
    579 	simple_unlock(&pp->pr_lock);
    580 	return (v);
    581 }
    582 
    583 /*
    584  * Return resource to the pool; must be called at appropriate spl level
    585  */
    586 #ifdef POOL_DIAGNOSTIC
    587 void
    588 _pool_put(pp, v, file, line)
    589 	struct pool *pp;
    590 	void *v;
    591 	const char *file;
    592 	long line;
    593 #else
    594 void
    595 pool_put(pp, v)
    596 	struct pool *pp;
    597 	void *v;
    598 #endif
    599 {
    600 	struct pool_item *pi = v;
    601 	struct pool_item_header *ph;
    602 	caddr_t page;
    603 
    604 	page = (caddr_t)((u_long)v & pp->pr_pagemask);
    605 
    606 	simple_lock(&pp->pr_lock);
    607 
    608 	pr_log(pp, v, PRLOG_PUT, file, line);
    609 
    610 	if ((ph = pr_find_pagehead(pp, page)) == NULL) {
    611 		pr_printlog(pp);
    612 		panic("pool_put: %s: page header missing", pp->pr_wchan);
    613 	}
    614 
    615 	/*
    616 	 * Return to item list.
    617 	 */
    618 #ifdef DIAGNOSTIC
    619 	pi->pi_magic = PI_MAGIC;
    620 #endif
    621 	TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
    622 	ph->ph_nmissing--;
    623 	pp->pr_nput++;
    624 
    625 	/* Cancel "pool empty" condition if it exists */
    626 	if (pp->pr_curpage == NULL)
    627 		pp->pr_curpage = ph;
    628 
    629 	if (pp->pr_flags & PR_WANTED) {
    630 		pp->pr_flags &= ~PR_WANTED;
    631 		wakeup((caddr_t)pp);
    632 		simple_unlock(&pp->pr_lock);
    633 		return;
    634 	}
    635 
    636 	/*
    637 	 * If this page is now complete, move it to the end of the pagelist.
    638 	 * If this page has just become un-empty, move it the head.
    639 	 */
    640 	if (ph->ph_nmissing == 0) {
    641 		pp->pr_nidle++;
    642 		if (pp->pr_npages > pp->pr_maxpages) {
    643 #if 0
    644 			timeout(pool_drain, 0, pool_inactive_time*hz);
    645 #else
    646 			pr_rmpage(pp, ph);
    647 #endif
    648 		} else {
    649 			TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
    650 			TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
    651 			ph->ph_time = time;
    652 
    653 			/* XXX - update curpage */
    654 			for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
    655 			     ph = TAILQ_NEXT(ph, ph_pagelist))
    656 				if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
    657 					break;
    658 
    659 			pp->pr_curpage = ph;
    660 		}
    661 	}
    662 
    663 	simple_unlock(&pp->pr_lock);
    664 }
    665 
    666 /*
    667  * Add N items to the pool.
    668  */
    669 int
    670 pool_prime(pp, n, storage)
    671 	struct pool *pp;
    672 	int n;
    673 	caddr_t storage;
    674 {
    675 	caddr_t cp;
    676 	int newnitems, newpages;
    677 
    678 #ifdef DIAGNOSTIC
    679 	if (storage && !(pp->pr_flags & PR_STATIC))
    680 		panic("pool_prime: static");
    681 	/* !storage && static caught below */
    682 #endif
    683 
    684 	newnitems = pp->pr_minitems + n;
    685 	newpages =
    686 		roundup(pp->pr_itemsperpage,newnitems) / pp->pr_itemsperpage
    687 		- pp->pr_minpages;
    688 
    689 	simple_lock(&pp->pr_lock);
    690 	while (newpages-- > 0) {
    691 
    692 		if (pp->pr_flags & PR_STATIC) {
    693 			cp = storage;
    694 			storage += pp->pr_pagesz;
    695 		} else {
    696 			cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
    697 		}
    698 
    699 		if (cp == NULL) {
    700 			simple_unlock(&pp->pr_lock);
    701 			return (ENOMEM);
    702 		}
    703 
    704 		pool_prime_page(pp, cp);
    705 		pp->pr_minpages++;
    706 	}
    707 
    708 	pp->pr_minitems = newnitems;
    709 
    710 	if (pp->pr_minpages >= pp->pr_maxpages)
    711 		pp->pr_maxpages = pp->pr_minpages + 1;	/* XXX */
    712 
    713 	simple_unlock(&pp->pr_lock);
    714 	return (0);
    715 }
    716 
    717 /*
    718  * Add a page worth of items to the pool.
    719  */
    720 int
    721 pool_prime_page(pp, storage)
    722 	struct pool *pp;
    723 	caddr_t storage;
    724 {
    725 	struct pool_item *pi;
    726 	struct pool_item_header *ph;
    727 	caddr_t cp = storage;
    728 	unsigned int align = pp->pr_align;
    729 	unsigned int ioff = pp->pr_itemoffset;
    730 	int n;
    731 
    732 	if ((pp->pr_flags & PR_PHINPAGE) != 0) {
    733 		ph = (struct pool_item_header *)(cp + pp->pr_phoffset);
    734 	} else {
    735 		ph = pool_get(&phpool, PR_URGENT);
    736 		LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
    737 				 ph, ph_hashlist);
    738 	}
    739 
    740 	/*
    741 	 * Insert page header.
    742 	 */
    743 	TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
    744 	TAILQ_INIT(&ph->ph_itemlist);
    745 	ph->ph_page = storage;
    746 	ph->ph_nmissing = 0;
    747 	ph->ph_time.tv_sec = ph->ph_time.tv_usec = 0;
    748 
    749 	pp->pr_nidle++;
    750 
    751 	/*
    752 	 * Color this page.
    753 	 */
    754 	cp = (caddr_t)(cp + pp->pr_curcolor);
    755 	if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
    756 		pp->pr_curcolor = 0;
    757 
    758 	/*
    759 	 * Adjust storage to apply aligment to `pr_itemoffset' in each item.
    760 	 */
    761 	if (ioff != 0)
    762 		cp = (caddr_t)(cp + (align - ioff));
    763 
    764 	/*
    765 	 * Insert remaining chunks on the bucket list.
    766 	 */
    767 	n = pp->pr_itemsperpage;
    768 
    769 	while (n--) {
    770 		pi = (struct pool_item *)cp;
    771 
    772 		/* Insert on page list */
    773 		TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
    774 #ifdef DIAGNOSTIC
    775 		pi->pi_magic = PI_MAGIC;
    776 #endif
    777 		cp = (caddr_t)(cp + pp->pr_size);
    778 	}
    779 
    780 	/*
    781 	 * If the pool was depleted, point at the new page.
    782 	 */
    783 	if (pp->pr_curpage == NULL)
    784 		pp->pr_curpage = ph;
    785 
    786 	if (++pp->pr_npages > pp->pr_hiwat)
    787 		pp->pr_hiwat = pp->pr_npages;
    788 
    789 	return (0);
    790 }
    791 
    792 void
    793 pool_setlowat(pp, n)
    794 	pool_handle_t	pp;
    795 	int n;
    796 {
    797 	pp->pr_minitems = n;
    798 	if (n == 0) {
    799 		pp->pr_minpages = 0;
    800 		return;
    801 	}
    802 	pp->pr_minpages =
    803 		roundup(pp->pr_itemsperpage,n) / pp->pr_itemsperpage;
    804 }
    805 
    806 void
    807 pool_sethiwat(pp, n)
    808 	pool_handle_t	pp;
    809 	int n;
    810 {
    811 	if (n == 0) {
    812 		pp->pr_maxpages = 0;
    813 		return;
    814 	}
    815 	pp->pr_maxpages =
    816 		roundup(pp->pr_itemsperpage,n) / pp->pr_itemsperpage;
    817 }
    818 
    819 
    820 /*
    821  * Default page allocator.
    822  */
    823 static void *
    824 pool_page_alloc(sz, flags, mtype)
    825 	unsigned long sz;
    826 	int flags;
    827 	int mtype;
    828 {
    829 
    830 #if defined(UVM)
    831 	return ((void *)uvm_km_alloc_poolpage());
    832 #else
    833 	return ((void *)kmem_alloc_poolpage());
    834 #endif
    835 }
    836 
    837 static void
    838 pool_page_free(v, sz, mtype)
    839 	void *v;
    840 	unsigned long sz;
    841 	int mtype;
    842 {
    843 
    844 #if defined(UVM)
    845 	uvm_km_free_poolpage((vm_offset_t)v);
    846 #else
    847 	kmem_free_poolpage((vm_offset_t)v);
    848 #endif
    849 }
    850 
    851 /*
    852  * Release all complete pages that have not been used recently.
    853  */
    854 void
    855 pool_reclaim (pp)
    856 	pool_handle_t pp;
    857 {
    858 	struct pool_item_header *ph, *phnext;
    859 	struct timeval curtime = time;
    860 
    861 	if (pp->pr_flags & PR_STATIC)
    862 		return;
    863 
    864 	if (simple_lock_try(&pp->pr_lock) == 0)
    865 		return;
    866 
    867 	for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
    868 		phnext = TAILQ_NEXT(ph, ph_pagelist);
    869 
    870 		/* Check our minimum page claim */
    871 		if (pp->pr_npages <= pp->pr_minpages)
    872 			break;
    873 
    874 		if (ph->ph_nmissing == 0) {
    875 			struct timeval diff;
    876 			timersub(&curtime, &ph->ph_time, &diff);
    877 			if (diff.tv_sec < pool_inactive_time)
    878 				continue;
    879 			pr_rmpage(pp, ph);
    880 		}
    881 	}
    882 
    883 	simple_unlock(&pp->pr_lock);
    884 }
    885 
    886 
    887 /*
    888  * Drain pools, one at a time.
    889  */
    890 void
    891 pool_drain(arg)
    892 	void *arg;
    893 {
    894 	struct pool *pp;
    895 	int s = splimp();
    896 
    897 	/* XXX:lock pool head */
    898 	if (drainpp == NULL && (drainpp = TAILQ_FIRST(&pool_head)) == NULL) {
    899 		splx(s);
    900 		return;
    901 	}
    902 
    903 	pp = drainpp;
    904 	drainpp = TAILQ_NEXT(pp, pr_poollist);
    905 	/* XXX:unlock pool head */
    906 
    907 	pool_reclaim(pp);
    908 	splx(s);
    909 }
    910 
    911 
    912 #ifdef DEBUG
    913 /*
    914  * Diagnostic helpers.
    915  */
    916 void
    917 pool_print(pp, label)
    918 	struct pool *pp;
    919 	char *label;
    920 {
    921 
    922 	if (label != NULL)
    923 		printf("%s: ", label);
    924 
    925 	printf("pool %s: nalloc %lu nfree %lu npagealloc %lu npagefree %lu\n"
    926 	       "         npages %u minitems %u itemsperpage %u itemoffset %u\n"
    927 	       "         nidle %lu\n",
    928 		pp->pr_wchan,
    929 		pp->pr_nget,
    930 		pp->pr_nput,
    931 		pp->pr_npagealloc,
    932 		pp->pr_npagefree,
    933 		pp->pr_npages,
    934 		pp->pr_minitems,
    935 		pp->pr_itemsperpage,
    936 		pp->pr_itemoffset,
    937 		pp->pr_nidle);
    938 }
    939 
    940 int
    941 pool_chk(pp, label)
    942 	struct pool *pp;
    943 	char *label;
    944 {
    945 	struct pool_item_header *ph;
    946 	int r = 0;
    947 
    948 	simple_lock(&pp->pr_lock);
    949 
    950 	for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
    951 	     ph = TAILQ_NEXT(ph, ph_pagelist)) {
    952 
    953 		struct pool_item *pi;
    954 		int n;
    955 		caddr_t page;
    956 
    957 		page = (caddr_t)((u_long)ph & pp->pr_pagemask);
    958 		if (page != ph->ph_page) {
    959 			if (label != NULL)
    960 				printf("%s: ", label);
    961 			printf("pool(%s): page inconsistency: page %p;"
    962 			       " at page head addr %p (p %p)\n",
    963 				pp->pr_wchan, ph->ph_page,
    964 				ph, page);
    965 			r++;
    966 			goto out;
    967 		}
    968 
    969 		for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
    970 		     pi != NULL;
    971 		     pi = TAILQ_NEXT(pi,pi_list), n++) {
    972 
    973 #ifdef DIAGNOSTIC
    974 			if (pi->pi_magic != PI_MAGIC) {
    975 				if (label != NULL)
    976 					printf("%s: ", label);
    977 				printf("pool(%s): free list modified: magic=%x;"
    978 				       " page %p; item ordinal %d;"
    979 				       " addr %p (p %p)\n",
    980 					pp->pr_wchan, pi->pi_magic, ph->ph_page,
    981 					n, pi, page);
    982 				panic("pool");
    983 			}
    984 #endif
    985 			page = (caddr_t)((u_long)pi & pp->pr_pagemask);
    986 			if (page == ph->ph_page)
    987 				continue;
    988 
    989 			if (label != NULL)
    990 				printf("%s: ", label);
    991 			printf("pool(%s): page inconsistency: page %p;"
    992 			       " item ordinal %d; addr %p (p %p)\n",
    993 				pp->pr_wchan, ph->ph_page,
    994 				n, pi, page);
    995 			r++;
    996 			goto out;
    997 		}
    998 	}
    999 out:
   1000 	simple_unlock(&pp->pr_lock);
   1001 	return (r);
   1002 }
   1003 #endif
   1004