Please note that diffs are not public domain; they are subject to the copyright notices on the relevant files. =================================================================== RCS file: /ftp/cvs/cvsroot/src/sys/kern/subr_pool.c,v rcsdiff: /ftp/cvs/cvsroot/src/sys/kern/subr_pool.c,v: warning: Unknown phrases like `commitid ...;' are present. retrieving revision 1.77 retrieving revision 1.91 diff -u -p -r1.77 -r1.91 --- src/sys/kern/subr_pool.c 2002/07/11 17:18:48 1.77 +++ src/sys/kern/subr_pool.c 2004/01/16 12:47:37 1.91 @@ -1,4 +1,4 @@ -/* $NetBSD: subr_pool.c,v 1.77 2002/07/11 17:18:48 matt Exp $ */ +/* $NetBSD: subr_pool.c,v 1.91 2004/01/16 12:47:37 yamt Exp $ */ /*- * Copyright (c) 1997, 1999, 2000 The NetBSD Foundation, Inc. @@ -38,7 +38,7 @@ */ #include -__KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.77 2002/07/11 17:18:48 matt Exp $"); +__KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.91 2004/01/16 12:47:37 yamt Exp $"); #include "opt_pool.h" #include "opt_poollog.h" @@ -59,12 +59,14 @@ __KERNEL_RCSID(0, "$NetBSD: subr_pool.c, /* * Pool resource management utility. * - * Memory is allocated in pages which are split into pieces according - * to the pool item size. Each page is kept on a list headed by `pr_pagelist' - * in the pool structure and the individual pool items are on a linked list - * headed by `ph_itemlist' in each page header. The memory for building - * the page list is either taken from the allocated pages themselves (for - * small pool items) or taken from an internal pool of page headers (`phpool'). + * Memory is allocated in pages which are split into pieces according to + * the pool item size. Each page is kept on one of three lists in the + * pool structure: `pr_emptypages', `pr_fullpages' and `pr_partpages', + * for empty, full and partially-full pages respectively. The individual + * pool items are on a linked list headed by `ph_itemlist' in each page + * header. The memory for building the page list is either taken from + * the allocated pages themselves (for small pool items) or taken from + * an internal pool of page headers (`phpool'). */ /* List of all pools */ @@ -89,30 +91,25 @@ struct simplelock pool_head_slock = SIMP struct pool_item_header { /* Page headers */ - TAILQ_ENTRY(pool_item_header) + LIST_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 */ + SPLAY_ENTRY(pool_item_header) + ph_node; /* Off-page page headers */ + unsigned int ph_nmissing; /* # of chunks in use */ caddr_t ph_page; /* this page's address */ struct timeval ph_time; /* last referenced */ }; -TAILQ_HEAD(pool_pagelist,pool_item_header); struct pool_item { #ifdef DIAGNOSTIC - int pi_magic; + u_int pi_magic; #endif -#define PI_MAGIC 0xdeadbeef +#define PI_MAGIC 0xdeadbeefU /* Other entries use only this list entry */ TAILQ_ENTRY(pool_item) pi_list; }; -#define PR_HASH_INDEX(pp,addr) \ - (((u_long)(addr) >> (pp)->pr_alloc->pa_pageshift) & \ - (PR_HASHTABSIZE - 1)) - #define POOL_NEEDS_CATCHUP(pp) \ ((pp)->pr_nitems < (pp)->pr_minitems) @@ -150,13 +147,19 @@ static void pool_cache_reclaim(struct po static int pool_catchup(struct pool *); static void pool_prime_page(struct pool *, caddr_t, struct pool_item_header *); +static void pool_update_curpage(struct pool *); void *pool_allocator_alloc(struct pool *, int); void pool_allocator_free(struct pool *, void *); +static void pool_print_pagelist(struct pool_pagelist *, + void (*)(const char *, ...)); static void pool_print1(struct pool *, const char *, void (*)(const char *, ...)); +static int pool_chk_page(struct pool *, const char *, + struct pool_item_header *); + /* * Pool log entry. An array of these is allocated in pool_init(). */ @@ -169,6 +172,7 @@ struct pool_log { void *pl_addr; }; +#ifdef POOL_DIAGNOSTIC /* Number of entries in pool log buffers */ #ifndef POOL_LOGSIZE #define POOL_LOGSIZE 10 @@ -176,7 +180,6 @@ struct pool_log { int pool_logsize = POOL_LOGSIZE; -#ifdef POOL_DIAGNOSTIC static __inline void pr_log(struct pool *pp, void *v, int action, const char *file, long line) { @@ -275,24 +278,34 @@ pr_enter_check(struct pool *pp, void (*p #define pr_enter_check(pp, pr) #endif /* POOL_DIAGNOSTIC */ +static __inline int +phtree_compare(struct pool_item_header *a, struct pool_item_header *b) +{ + if (a->ph_page < b->ph_page) + return (-1); + else if (a->ph_page > b->ph_page) + return (1); + else + return (0); +} + +SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare); +SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare); + /* * Return the pool page header based on page address. */ static __inline struct pool_item_header * pr_find_pagehead(struct pool *pp, caddr_t page) { - struct pool_item_header *ph; + struct pool_item_header *ph, tmp; if ((pp->pr_roflags & PR_PHINPAGE) != 0) return ((struct pool_item_header *)(page + pp->pr_phoffset)); - for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]); - ph != NULL; - ph = LIST_NEXT(ph, ph_hashlist)) { - if (ph->ph_page == page) - return (ph); - } - return (NULL); + tmp.ph_page = page; + ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp); + return ph; } /* @@ -304,6 +317,8 @@ pr_rmpage(struct pool *pp, struct pool_i { int s; + LOCK_ASSERT(!simple_lock_held(&pp->pr_slock) || pq != NULL); + /* * If the page was idle, decrement the idle page count. */ @@ -322,14 +337,15 @@ pr_rmpage(struct pool *pp, struct pool_i /* * Unlink a page from the pool and release it (or queue it for release). */ - TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist); + LIST_REMOVE(ph, ph_pagelist); + if ((pp->pr_roflags & PR_PHINPAGE) == 0) + SPLAY_REMOVE(phtree, &pp->pr_phtree, ph); if (pq) { - TAILQ_INSERT_HEAD(pq, ph, ph_pagelist); + LIST_INSERT_HEAD(pq, ph, ph_pagelist); } else { pool_allocator_free(pp, ph->ph_page); if ((pp->pr_roflags & PR_PHINPAGE) == 0) { - LIST_REMOVE(ph, ph_hashlist); - s = splhigh(); + s = splvm(); pool_put(&phpool, ph); splx(s); } @@ -337,18 +353,7 @@ pr_rmpage(struct pool *pp, struct pool_i pp->pr_npages--; pp->pr_npagefree++; - 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. - */ - TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist) - if (TAILQ_FIRST(&ph->ph_itemlist) != NULL) - break; - - pp->pr_curpage = ph; - } + pool_update_curpage(pp); } /* @@ -361,7 +366,7 @@ void pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags, const char *wchan, struct pool_allocator *palloc) { - int off, slack, i; + int off, slack; #ifdef POOL_DIAGNOSTIC /* @@ -415,7 +420,7 @@ pool_init(struct pool *pp, size_t size, if (size < sizeof(struct pool_item)) size = sizeof(struct pool_item); - size = ALIGN(size); + size = roundup(size, align); #ifdef DIAGNOSTIC if (size > palloc->pa_pagesz) panic("pool_init: pool item size (%lu) too large", @@ -425,7 +430,9 @@ pool_init(struct pool *pp, size_t size, /* * Initialize the pool structure. */ - TAILQ_INIT(&pp->pr_pagelist); + LIST_INIT(&pp->pr_emptypages); + LIST_INIT(&pp->pr_fullpages); + LIST_INIT(&pp->pr_partpages); TAILQ_INIT(&pp->pr_cachelist); pp->pr_curpage = NULL; pp->pr_npages = 0; @@ -465,9 +472,7 @@ pool_init(struct pool *pp, size_t size, /* The page header will be taken from our page header pool */ pp->pr_phoffset = 0; off = palloc->pa_pagesz; - for (i = 0; i < PR_HASHTABSIZE; i++) { - LIST_INIT(&pp->pr_hashtab[i]); - } + SPLAY_INIT(&pp->pr_phtree); } /* @@ -564,14 +569,16 @@ pool_destroy(struct pool *pp) #ifdef DIAGNOSTIC if (pp->pr_nout != 0) { pr_printlog(pp, NULL, printf); - panic("pool_destroy: pool busy: still out: %u\n", + panic("pool_destroy: pool busy: still out: %u", pp->pr_nout); } #endif /* Remove all pages */ - while ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL) + while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL) pr_rmpage(pp, ph, NULL); + KASSERT(LIST_EMPTY(&pp->pr_fullpages)); + KASSERT(LIST_EMPTY(&pp->pr_partpages)); /* Remove from global pool list */ simple_lock(&pool_head_slock); @@ -600,7 +607,7 @@ pool_set_drain_hook(struct pool *pp, voi pp->pr_drain_hook_arg = arg; } -static __inline struct pool_item_header * +static struct pool_item_header * pool_alloc_item_header(struct pool *pp, caddr_t storage, int flags) { struct pool_item_header *ph; @@ -611,7 +618,7 @@ pool_alloc_item_header(struct pool *pp, if ((pp->pr_roflags & PR_PHINPAGE) != 0) ph = (struct pool_item_header *) (storage + pp->pr_phoffset); else { - s = splhigh(); + s = splvm(); ph = pool_get(&phpool, flags); splx(s); } @@ -634,7 +641,7 @@ pool_get(struct pool *pp, int flags) void *v; #ifdef DIAGNOSTIC - if (__predict_false(curproc == NULL && doing_shutdown == 0 && + if (__predict_false(curlwp == NULL && doing_shutdown == 0 && (flags & PR_WAITOK) != 0)) panic("pool_get: %s: must have NOWAIT", pp->pr_wchan); @@ -715,7 +722,7 @@ pool_get(struct pool *pp, int flags) simple_unlock(&pp->pr_slock); printf("pool_get: %s: curpage NULL, nitems %u\n", pp->pr_wchan, pp->pr_nitems); - panic("pool_get: nitems inconsistent\n"); + panic("pool_get: nitems inconsistent"); } #endif @@ -729,13 +736,14 @@ pool_get(struct pool *pp, int flags) v = pool_allocator_alloc(pp, flags); if (__predict_true(v != NULL)) ph = pool_alloc_item_header(pp, v, flags); - simple_lock(&pp->pr_slock); - pr_enter(pp, file, line); if (__predict_false(v == NULL || ph == NULL)) { if (v != NULL) pool_allocator_free(pp, v); + simple_lock(&pp->pr_slock); + pr_enter(pp, file, line); + /* * We were unable to allocate a page or item * header, but we released the lock during @@ -767,13 +775,14 @@ pool_get(struct pool *pp, int flags) } /* We have more memory; add it to the pool */ + simple_lock(&pp->pr_slock); + pr_enter(pp, file, line); pool_prime_page(pp, v, ph); pp->pr_npagealloc++; /* Start the allocation process over. */ goto startover; } - if (__predict_false((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)) { pr_leave(pp); simple_unlock(&pp->pr_slock); @@ -785,7 +794,7 @@ pool_get(struct pool *pp, int flags) simple_unlock(&pp->pr_slock); printf("pool_get: %s: items on itemlist, nitems %u\n", pp->pr_wchan, pp->pr_nitems); - panic("pool_get: nitems inconsistent\n"); + panic("pool_get: nitems inconsistent"); } #endif @@ -814,9 +823,16 @@ pool_get(struct pool *pp, int flags) panic("pool_get: nidle inconsistent"); #endif pp->pr_nidle--; + + /* + * This page was previously empty. Move it to the list of + * partially-full pages. This page is already curpage. + */ + LIST_REMOVE(ph, ph_pagelist); + LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist); } ph->ph_nmissing++; - if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) { + if (TAILQ_EMPTY(&ph->ph_itemlist)) { #ifdef DIAGNOSTIC if (__predict_false(ph->ph_nmissing != pp->pr_itemsperpage)) { pr_leave(pp); @@ -826,23 +842,12 @@ pool_get(struct pool *pp, int flags) } #endif /* - * Find a new non-empty page header, if any. - * Start search from the page head, to increase - * the chance for "high water" pages to be freed. - * - * Migrate empty pages to the end of the list. This - * will speed the update of curpage as pages become - * idle. Empty pages intermingled with idle pages - * is no big deal. As soon as a page becomes un-empty, - * it will move back to the head of the list. + * This page is now full. Move it to the full list + * and select a new current page. */ - TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist); - TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist); - TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist) - if (TAILQ_FIRST(&ph->ph_itemlist) != NULL) - break; - - pp->pr_curpage = ph; + LIST_REMOVE(ph, ph_pagelist); + LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist); + pool_update_curpage(pp); } pp->pr_nget++; @@ -916,6 +921,7 @@ pool_do_put(struct pool *pp, void *v) #endif TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list); + KDASSERT(ph->ph_nmissing != 0); ph->ph_nmissing--; pp->pr_nput++; pp->pr_nitems++; @@ -934,27 +940,31 @@ pool_do_put(struct pool *pp, void *v) } /* - * If this page is now complete, do one of two things: + * If this page is now empty, do one of two things: * - * (1) If we have more pages than the page high water - * mark, free the page back to the system. + * (1) If we have more pages than the page high water mark, + * or if we are flagged as immediately freeing back idle + * pages, free the page back to the system. ONLY CONSIDER + * FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE + * CLAIM. * - * (2) Move it to the end of the page list, so that - * we minimize our chances of fragmenting the - * pool. Idle pages migrate to the end (along with - * completely empty pages, so that we find un-empty - * pages more quickly when we update curpage) of the - * list so they can be more easily swept up by - * the pagedaemon when pages are scarce. + * (2) Otherwise, move the page to the empty page list. + * + * Either way, select a new current page (so we use a partially-full + * page if one is available). */ if (ph->ph_nmissing == 0) { pp->pr_nidle++; - if (pp->pr_npages > pp->pr_maxpages || - (pp->pr_alloc->pa_flags & PA_WANT) != 0) { + if (pp->pr_npages > pp->pr_minpages && + (pp->pr_npages > pp->pr_maxpages || + (pp->pr_roflags & PR_IMMEDRELEASE) != 0 || + (pp->pr_alloc->pa_flags & PA_WANT) != 0)) { + simple_unlock(&pp->pr_slock); pr_rmpage(pp, ph, NULL); + simple_lock(&pp->pr_slock); } else { - TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist); - TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist); + LIST_REMOVE(ph, ph_pagelist); + LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist); /* * Update the timestamp on the page. A page must @@ -965,31 +975,19 @@ pool_do_put(struct pool *pp, void *v) s = splclock(); ph->ph_time = mono_time; splx(s); - - /* - * Update the current page pointer. Just look for - * the first page with any free items. - * - * XXX: Maybe we want an option to look for the - * page with the fewest available items, to minimize - * fragmentation? - */ - TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist) - if (TAILQ_FIRST(&ph->ph_itemlist) != NULL) - break; - - pp->pr_curpage = ph; } + pool_update_curpage(pp); } + /* - * If the page has just become un-empty, move it to the head of - * the list, and make it the current page. The next allocation - * will get the item from this page, instead of further fragmenting - * the pool. + * If the page was previously completely full, move it to the + * partially-full list and make it the current page. The next + * allocation will get the item from this page, instead of + * further fragmenting the pool. */ else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) { - TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist); - TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist); + LIST_REMOVE(ph, ph_pagelist); + LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist); pp->pr_curpage = ph; } } @@ -1036,7 +1034,7 @@ pool_put(struct pool *pp, void *v) int pool_prime(struct pool *pp, int n) { - struct pool_item_header *ph; + struct pool_item_header *ph = NULL; caddr_t cp; int newpages; @@ -1049,14 +1047,15 @@ pool_prime(struct pool *pp, int n) cp = pool_allocator_alloc(pp, PR_NOWAIT); if (__predict_true(cp != NULL)) ph = pool_alloc_item_header(pp, cp, PR_NOWAIT); - simple_lock(&pp->pr_slock); if (__predict_false(cp == NULL || ph == NULL)) { if (cp != NULL) pool_allocator_free(pp, cp); + simple_lock(&pp->pr_slock); break; } + simple_lock(&pp->pr_slock); pool_prime_page(pp, cp, ph); pp->pr_npagealloc++; pp->pr_minpages++; @@ -1082,24 +1081,27 @@ pool_prime_page(struct pool *pp, caddr_t unsigned int align = pp->pr_align; unsigned int ioff = pp->pr_itemoffset; int n; + int s; + + LOCK_ASSERT(simple_lock_held(&pp->pr_slock)); #ifdef DIAGNOSTIC if (((u_long)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0) panic("pool_prime_page: %s: unaligned page", pp->pr_wchan); #endif - if ((pp->pr_roflags & PR_PHINPAGE) == 0) - LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)], - ph, ph_hashlist); - /* * Insert page header. */ - TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist); + LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist); TAILQ_INIT(&ph->ph_itemlist); ph->ph_page = storage; ph->ph_nmissing = 0; - memset(&ph->ph_time, 0, sizeof(ph->ph_time)); + s = splclock(); + ph->ph_time = mono_time; + splx(s); + if ((pp->pr_roflags & PR_PHINPAGE) == 0) + SPLAY_INSERT(phtree, &pp->pr_phtree, ph); pp->pr_nidle++; @@ -1125,6 +1127,8 @@ pool_prime_page(struct pool *pp, caddr_t while (n--) { pi = (struct pool_item *)cp; + KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0); + /* Insert on page list */ TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list); #ifdef DIAGNOSTIC @@ -1145,7 +1149,7 @@ pool_prime_page(struct pool *pp, caddr_t /* * Used by pool_get() when nitems drops below the low water mark. This - * is used to catch up nitmes with the low water mark. + * is used to catch up pr_nitems with the low water mark. * * Note 1, we never wait for memory here, we let the caller decide what to do. * @@ -1155,7 +1159,7 @@ pool_prime_page(struct pool *pp, caddr_t static int pool_catchup(struct pool *pp) { - struct pool_item_header *ph; + struct pool_item_header *ph = NULL; caddr_t cp; int error = 0; @@ -1170,13 +1174,14 @@ pool_catchup(struct pool *pp) cp = pool_allocator_alloc(pp, PR_NOWAIT); if (__predict_true(cp != NULL)) ph = pool_alloc_item_header(pp, cp, PR_NOWAIT); - simple_lock(&pp->pr_slock); if (__predict_false(cp == NULL || ph == NULL)) { if (cp != NULL) pool_allocator_free(pp, cp); error = ENOMEM; + simple_lock(&pp->pr_slock); break; } + simple_lock(&pp->pr_slock); pool_prime_page(pp, cp, ph); pp->pr_npagealloc++; } @@ -1184,6 +1189,16 @@ pool_catchup(struct pool *pp) return (error); } +static void +pool_update_curpage(struct pool *pp) +{ + + pp->pr_curpage = LIST_FIRST(&pp->pr_partpages); + if (pp->pr_curpage == NULL) { + pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages); + } +} + void pool_setlowat(struct pool *pp, int n) { @@ -1257,6 +1272,7 @@ pool_reclaim(struct pool *pp) struct pool_cache *pc; struct timeval curtime; struct pool_pagelist pq; + struct timeval diff; int s; if (pp->pr_drain_hook != NULL) { @@ -1270,7 +1286,7 @@ pool_reclaim(struct pool *pp) return (0); pr_enter(pp, file, line); - TAILQ_INIT(&pq); + LIST_INIT(&pq); /* * Reclaim items from the pool's caches. @@ -1282,44 +1298,41 @@ pool_reclaim(struct pool *pp) curtime = mono_time; splx(s); - for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) { - phnext = TAILQ_NEXT(ph, ph_pagelist); + for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) { + phnext = LIST_NEXT(ph, ph_pagelist); /* Check our minimum page claim */ if (pp->pr_npages <= pp->pr_minpages) break; - if (ph->ph_nmissing == 0) { - struct timeval diff; - timersub(&curtime, &ph->ph_time, &diff); - if (diff.tv_sec < pool_inactive_time) - continue; + KASSERT(ph->ph_nmissing == 0); + timersub(&curtime, &ph->ph_time, &diff); + if (diff.tv_sec < pool_inactive_time) + continue; - /* - * If freeing this page would put us below - * the low water mark, stop now. - */ - if ((pp->pr_nitems - pp->pr_itemsperpage) < - pp->pr_minitems) - break; + /* + * If freeing this page would put us below + * the low water mark, stop now. + */ + if ((pp->pr_nitems - pp->pr_itemsperpage) < + pp->pr_minitems) + break; - pr_rmpage(pp, ph, &pq); - } + pr_rmpage(pp, ph, &pq); } pr_leave(pp); simple_unlock(&pp->pr_slock); - if (TAILQ_EMPTY(&pq)) + if (LIST_EMPTY(&pq)) return (0); - while ((ph = TAILQ_FIRST(&pq)) != NULL) { - TAILQ_REMOVE(&pq, ph, ph_pagelist); + while ((ph = LIST_FIRST(&pq)) != NULL) { + LIST_REMOVE(ph, ph_pagelist); pool_allocator_free(pp, ph->ph_page); if (pp->pr_roflags & PR_PHINPAGE) { continue; } - LIST_REMOVE(ph, ph_hashlist); - s = splhigh(); + s = splvm(); pool_put(&phpool, ph); splx(s); } @@ -1404,14 +1417,35 @@ pool_printit(struct pool *pp, const char } static void -pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...)) +pool_print_pagelist(struct pool_pagelist *pl, void (*pr)(const char *, ...)) { struct pool_item_header *ph; - struct pool_cache *pc; - struct pool_cache_group *pcg; #ifdef DIAGNOSTIC struct pool_item *pi; #endif + + LIST_FOREACH(ph, pl, ph_pagelist) { + (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n", + ph->ph_page, ph->ph_nmissing, + (u_long)ph->ph_time.tv_sec, + (u_long)ph->ph_time.tv_usec); +#ifdef DIAGNOSTIC + TAILQ_FOREACH(pi, &ph->ph_itemlist, pi_list) { + if (pi->pi_magic != PI_MAGIC) { + (*pr)("\t\t\titem %p, magic 0x%x\n", + pi, pi->pi_magic); + } + } +#endif + } +} + +static void +pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...)) +{ + struct pool_item_header *ph; + struct pool_cache *pc; + struct pool_cache_group *pcg; int i, print_log = 0, print_pagelist = 0, print_cache = 0; char c; @@ -1422,7 +1456,6 @@ pool_print1(struct pool *pp, const char print_pagelist = 1; if (c == 'c') print_cache = 1; - modif++; } (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n", @@ -1442,29 +1475,22 @@ pool_print1(struct pool *pp, const char if (print_pagelist == 0) goto skip_pagelist; - if ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL) - (*pr)("\n\tpage list:\n"); - for (; ph != NULL; ph = TAILQ_NEXT(ph, ph_pagelist)) { - (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n", - ph->ph_page, ph->ph_nmissing, - (u_long)ph->ph_time.tv_sec, - (u_long)ph->ph_time.tv_usec); -#ifdef DIAGNOSTIC - TAILQ_FOREACH(pi, &ph->ph_itemlist, pi_list) { - if (pi->pi_magic != PI_MAGIC) { - (*pr)("\t\t\titem %p, magic 0x%x\n", - pi, pi->pi_magic); - } - } -#endif - } + if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL) + (*pr)("\n\tempty page list:\n"); + pool_print_pagelist(&pp->pr_emptypages, pr); + if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL) + (*pr)("\n\tfull page list:\n"); + pool_print_pagelist(&pp->pr_fullpages, pr); + if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL) + (*pr)("\n\tpartial-page list:\n"); + pool_print_pagelist(&pp->pr_partpages, pr); + if (pp->pr_curpage == NULL) (*pr)("\tno current page\n"); else (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page); skip_pagelist: - if (print_log == 0) goto skip_log; @@ -1475,7 +1501,6 @@ pool_print1(struct pool *pp, const char pr_printlog(pp, NULL, pr); skip_log: - if (print_cache == 0) goto skip_cache; @@ -1486,73 +1511,103 @@ pool_print1(struct pool *pp, const char pc->pc_hits, pc->pc_misses, pc->pc_ngroups, pc->pc_nitems); TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) { (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail); - for (i = 0; i < PCG_NOBJECTS; i++) - (*pr)("\t\t\t%p\n", pcg->pcg_objects[i]); + for (i = 0; i < PCG_NOBJECTS; i++) { + if (pcg->pcg_objects[i].pcgo_pa != + POOL_PADDR_INVALID) { + (*pr)("\t\t\t%p, 0x%llx\n", + pcg->pcg_objects[i].pcgo_va, + (unsigned long long) + pcg->pcg_objects[i].pcgo_pa); + } else { + (*pr)("\t\t\t%p\n", + pcg->pcg_objects[i].pcgo_va); + } + } } } skip_cache: - pr_enter_check(pp, pr); } -int -pool_chk(struct pool *pp, const char *label) +static int +pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph) { - struct pool_item_header *ph; - int r = 0; + struct pool_item *pi; + caddr_t page; + int n; - simple_lock(&pp->pr_slock); + page = (caddr_t)((u_long)ph & pp->pr_alloc->pa_pagemask); + if (page != ph->ph_page && + (pp->pr_roflags & PR_PHINPAGE) != 0) { + if (label != NULL) + printf("%s: ", label); + printf("pool(%p:%s): page inconsistency: page %p;" + " at page head addr %p (p %p)\n", pp, + pp->pr_wchan, ph->ph_page, + ph, page); + return 1; + } + + for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0; + pi != NULL; + pi = TAILQ_NEXT(pi,pi_list), n++) { - TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist) { - struct pool_item *pi; - int n; - caddr_t page; - - page = (caddr_t)((u_long)ph & pp->pr_alloc->pa_pagemask); - if (page != ph->ph_page && - (pp->pr_roflags & PR_PHINPAGE) != 0) { +#ifdef DIAGNOSTIC + if (pi->pi_magic != PI_MAGIC) { if (label != NULL) printf("%s: ", label); - printf("pool(%p:%s): page inconsistency: page %p;" - " at page head addr %p (p %p)\n", pp, - pp->pr_wchan, ph->ph_page, - ph, page); - r++; - goto out; + printf("pool(%s): free list modified: magic=%x;" + " page %p; item ordinal %d;" + " addr %p (p %p)\n", + pp->pr_wchan, pi->pi_magic, ph->ph_page, + n, pi, page); + panic("pool"); } +#endif + page = + (caddr_t)((u_long)pi & pp->pr_alloc->pa_pagemask); + if (page == ph->ph_page) + continue; - for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0; - pi != NULL; - pi = TAILQ_NEXT(pi,pi_list), n++) { + if (label != NULL) + printf("%s: ", label); + printf("pool(%p:%s): page inconsistency: page %p;" + " item ordinal %d; addr %p (p %p)\n", pp, + pp->pr_wchan, ph->ph_page, + n, pi, page); + return 1; + } + return 0; +} -#ifdef DIAGNOSTIC - if (pi->pi_magic != PI_MAGIC) { - if (label != NULL) - printf("%s: ", label); - printf("pool(%s): free list modified: magic=%x;" - " page %p; item ordinal %d;" - " addr %p (p %p)\n", - pp->pr_wchan, pi->pi_magic, ph->ph_page, - n, pi, page); - panic("pool"); - } -#endif - page = - (caddr_t)((u_long)pi & pp->pr_alloc->pa_pagemask); - if (page == ph->ph_page) - continue; - if (label != NULL) - printf("%s: ", label); - printf("pool(%p:%s): page inconsistency: page %p;" - " item ordinal %d; addr %p (p %p)\n", pp, - pp->pr_wchan, ph->ph_page, - n, pi, page); - r++; +int +pool_chk(struct pool *pp, const char *label) +{ + struct pool_item_header *ph; + int r = 0; + + simple_lock(&pp->pr_slock); + LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) { + r = pool_chk_page(pp, label, ph); + if (r) { goto out; } } + LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) { + r = pool_chk_page(pp, label, ph); + if (r) { + goto out; + } + } + LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) { + r = pool_chk_page(pp, label, ph); + if (r) { + goto out; + } + } + out: simple_unlock(&pp->pr_slock); return (r); @@ -1616,7 +1671,7 @@ pool_cache_destroy(struct pool_cache *pc } static __inline void * -pcg_get(struct pool_cache_group *pcg) +pcg_get(struct pool_cache_group *pcg, paddr_t *pap) { void *object; u_int idx; @@ -1625,32 +1680,36 @@ pcg_get(struct pool_cache_group *pcg) KASSERT(pcg->pcg_avail != 0); idx = --pcg->pcg_avail; - KASSERT(pcg->pcg_objects[idx] != NULL); - object = pcg->pcg_objects[idx]; - pcg->pcg_objects[idx] = NULL; + KASSERT(pcg->pcg_objects[idx].pcgo_va != NULL); + object = pcg->pcg_objects[idx].pcgo_va; + if (pap != NULL) + *pap = pcg->pcg_objects[idx].pcgo_pa; + pcg->pcg_objects[idx].pcgo_va = NULL; return (object); } static __inline void -pcg_put(struct pool_cache_group *pcg, void *object) +pcg_put(struct pool_cache_group *pcg, void *object, paddr_t pa) { u_int idx; KASSERT(pcg->pcg_avail < PCG_NOBJECTS); idx = pcg->pcg_avail++; - KASSERT(pcg->pcg_objects[idx] == NULL); - pcg->pcg_objects[idx] = object; + KASSERT(pcg->pcg_objects[idx].pcgo_va == NULL); + pcg->pcg_objects[idx].pcgo_va = object; + pcg->pcg_objects[idx].pcgo_pa = pa; } /* - * pool_cache_get: + * pool_cache_get{,_paddr}: * - * Get an object from a pool cache. + * Get an object from a pool cache (optionally returning + * the physical address of the object). */ void * -pool_cache_get(struct pool_cache *pc, int flags) +pool_cache_get_paddr(struct pool_cache *pc, int flags, paddr_t *pap) { struct pool_cache_group *pcg; void *object; @@ -1685,13 +1744,20 @@ pool_cache_get(struct pool_cache *pc, in return (NULL); } } + if (object != NULL && pap != NULL) { +#ifdef POOL_VTOPHYS + *pap = POOL_VTOPHYS(object); +#else + *pap = POOL_PADDR_INVALID; +#endif + } return (object); } have_group: pc->pc_hits++; pc->pc_nitems--; - object = pcg_get(pcg); + object = pcg_get(pcg, pap); if (pcg->pcg_avail == 0) pc->pc_allocfrom = NULL; @@ -1702,12 +1768,13 @@ pool_cache_get(struct pool_cache *pc, in } /* - * pool_cache_put: + * pool_cache_put{,_paddr}: * - * Put an object back to the pool cache. + * Put an object back to the pool cache (optionally caching the + * physical address of the object). */ void -pool_cache_put(struct pool_cache *pc, void *object) +pool_cache_put_paddr(struct pool_cache *pc, void *object, paddr_t pa) { struct pool_cache_group *pcg; int s; @@ -1750,7 +1817,7 @@ pool_cache_put(struct pool_cache *pc, vo have_group: pc->pc_nitems++; - pcg_put(pcg, object); + pcg_put(pcg, object, pa); if (pcg->pcg_avail == PCG_NOBJECTS) pc->pc_freeto = NULL; @@ -1792,7 +1859,7 @@ pool_cache_do_invalidate(struct pool_cac npcg = TAILQ_NEXT(pcg, pcg_list); while (pcg->pcg_avail != 0) { pc->pc_nitems--; - object = pcg_get(pcg); + object = pcg_get(pcg, NULL); if (pcg->pcg_avail == 0 && pc->pc_allocfrom == pcg) pc->pc_allocfrom = NULL; if (pc->pc_dtor != NULL) @@ -1899,6 +1966,8 @@ pool_allocator_alloc(struct pool *org, i int s, freed; void *res; + LOCK_ASSERT(!simple_lock_held(&org->pr_slock)); + do { if ((res = (*pa->pa_alloc)(org, flags)) != NULL) return (res); @@ -1969,6 +2038,8 @@ pool_allocator_free(struct pool *pp, voi struct pool_allocator *pa = pp->pr_alloc; int s; + LOCK_ASSERT(!simple_lock_held(&pp->pr_slock)); + (*pa->pa_free)(pp, v); s = splvm();