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 1.186.2.1 retrieving revision 1.198.2.2 diff -u -p -r1.186.2.1 -r1.198.2.2 --- src/sys/kern/subr_pool.c 2011/06/06 09:09:35 1.186.2.1 +++ src/sys/kern/subr_pool.c 2013/06/23 06:18:58 1.198.2.2 @@ -1,4 +1,4 @@ -/* $NetBSD: subr_pool.c,v 1.186.2.1 2011/06/06 09:09:35 jruoho Exp $ */ +/* $NetBSD: subr_pool.c,v 1.198.2.2 2013/06/23 06:18:58 tls Exp $ */ /*- * Copyright (c) 1997, 1999, 2000, 2002, 2007, 2008, 2010 @@ -32,11 +32,9 @@ */ #include -__KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.186.2.1 2011/06/06 09:09:35 jruoho Exp $"); +__KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.198.2.2 2013/06/23 06:18:58 tls Exp $"); #include "opt_ddb.h" -#include "opt_pool.h" -#include "opt_poollog.h" #include "opt_lockdebug.h" #include @@ -45,7 +43,7 @@ __KERNEL_RCSID(0, "$NetBSD: subr_pool.c, #include #include #include -#include +#include #include #include #include @@ -55,9 +53,6 @@ __KERNEL_RCSID(0, "$NetBSD: subr_pool.c, #include #include -#ifdef DIAGNOSTIC -#include /* uvm_km_va_drain */ -#endif /* * Pool resource management utility. @@ -86,16 +81,14 @@ static struct pool phpool[PHPOOL_MAX]; static struct pool psppool; #endif -static SLIST_HEAD(, pool_allocator) pa_deferinitq = - SLIST_HEAD_INITIALIZER(pa_deferinitq); - static void *pool_page_alloc_meta(struct pool *, int); static void pool_page_free_meta(struct pool *, void *); /* allocator for pool metadata */ struct pool_allocator pool_allocator_meta = { - pool_page_alloc_meta, pool_page_free_meta, - .pa_backingmapptr = &kmem_map, + .pa_alloc = pool_page_alloc_meta, + .pa_free = pool_page_free_meta, + .pa_pagesz = 0 }; /* # of seconds to retain page after last use */ @@ -198,7 +191,7 @@ static bool pool_cache_get_slow(pool_cac static void pool_cache_cpu_init1(struct cpu_info *, pool_cache_t); static void pool_cache_invalidate_groups(pool_cache_t, pcg_t *); static void pool_cache_invalidate_cpu(pool_cache_t, u_int); -static void pool_cache_xcall(pool_cache_t); +static void pool_cache_transfer(pool_cache_t); static int pool_catchup(struct pool *); static void pool_prime_page(struct pool *, void *, @@ -210,143 +203,13 @@ static void *pool_allocator_alloc(struct static void pool_allocator_free(struct pool *, void *); static void pool_print_pagelist(struct pool *, struct pool_pagelist *, - void (*)(const char *, ...)); + void (*)(const char *, ...) __printflike(1, 2)); static void pool_print1(struct pool *, const char *, - void (*)(const char *, ...)); + void (*)(const char *, ...) __printflike(1, 2)); 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(). - */ -struct pool_log { - const char *pl_file; - long pl_line; - int pl_action; -#define PRLOG_GET 1 -#define PRLOG_PUT 2 - void *pl_addr; -}; - -#ifdef POOL_DIAGNOSTIC -/* Number of entries in pool log buffers */ -#ifndef POOL_LOGSIZE -#define POOL_LOGSIZE 10 -#endif - -int pool_logsize = POOL_LOGSIZE; - -static inline void -pr_log(struct pool *pp, void *v, int action, const char *file, long line) -{ - int n; - struct pool_log *pl; - - if ((pp->pr_roflags & PR_LOGGING) == 0) - return; - - if (pp->pr_log == NULL) { - if (kmem_map != NULL) - pp->pr_log = malloc( - pool_logsize * sizeof(struct pool_log), - M_TEMP, M_NOWAIT | M_ZERO); - if (pp->pr_log == NULL) - return; - pp->pr_curlogentry = 0; - pp->pr_logsize = pool_logsize; - } - - /* - * Fill in the current entry. Wrap around and overwrite - * the oldest entry if necessary. - */ - n = pp->pr_curlogentry; - pl = &pp->pr_log[n]; - pl->pl_file = file; - pl->pl_line = line; - pl->pl_action = action; - pl->pl_addr = v; - if (++n >= pp->pr_logsize) - n = 0; - pp->pr_curlogentry = n; -} - -static void -pr_printlog(struct pool *pp, struct pool_item *pi, - void (*pr)(const char *, ...)) -{ - int i = pp->pr_logsize; - int n = pp->pr_curlogentry; - - if (pp->pr_log == NULL) - return; - - /* - * Print all entries in this pool's log. - */ - while (i-- > 0) { - struct pool_log *pl = &pp->pr_log[n]; - if (pl->pl_action != 0) { - if (pi == NULL || pi == pl->pl_addr) { - (*pr)("\tlog entry %d:\n", i); - (*pr)("\t\taction = %s, addr = %p\n", - pl->pl_action == PRLOG_GET ? "get" : "put", - pl->pl_addr); - (*pr)("\t\tfile: %s at line %lu\n", - pl->pl_file, pl->pl_line); - } - } - if (++n >= pp->pr_logsize) - n = 0; - } -} - -static inline void -pr_enter(struct pool *pp, const char *file, long line) -{ - - if (__predict_false(pp->pr_entered_file != NULL)) { - printf("pool %s: reentrancy at file %s line %ld\n", - pp->pr_wchan, file, line); - printf(" previous entry at file %s line %ld\n", - pp->pr_entered_file, pp->pr_entered_line); - panic("pr_enter"); - } - - pp->pr_entered_file = file; - pp->pr_entered_line = line; -} - -static inline void -pr_leave(struct pool *pp) -{ - - if (__predict_false(pp->pr_entered_file == NULL)) { - printf("pool %s not entered?\n", pp->pr_wchan); - panic("pr_leave"); - } - - pp->pr_entered_file = NULL; - pp->pr_entered_line = 0; -} - -static inline void -pr_enter_check(struct pool *pp, void (*pr)(const char *, ...)) -{ - - if (pp->pr_entered_file != NULL) - (*pr)("\n\tcurrently entered from file %s line %ld\n", - pp->pr_entered_file, pp->pr_entered_line); -} -#else -#define pr_log(pp, v, action, file, line) -#define pr_printlog(pp, pi, pr) -#define pr_enter(pp, file, line) -#define pr_leave(pp) -#define pr_enter_check(pp, pr) -#endif /* POOL_DIAGNOSTIC */ - static inline unsigned int pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph, const void *v) @@ -529,115 +392,66 @@ pr_rmpage(struct pool *pp, struct pool_i pool_update_curpage(pp); } -static bool -pa_starved_p(struct pool_allocator *pa) -{ - - if (pa->pa_backingmap != NULL) { - return vm_map_starved_p(pa->pa_backingmap); - } - return false; -} - -static int -pool_reclaim_callback(struct callback_entry *ce, void *obj, void *arg) -{ - struct pool *pp = obj; - struct pool_allocator *pa = pp->pr_alloc; - - KASSERT(&pp->pr_reclaimerentry == ce); - pool_reclaim(pp); - if (!pa_starved_p(pa)) { - return CALLBACK_CHAIN_ABORT; - } - return CALLBACK_CHAIN_CONTINUE; -} - -static void -pool_reclaim_register(struct pool *pp) -{ - struct vm_map *map = pp->pr_alloc->pa_backingmap; - int s; - - if (map == NULL) { - return; - } - - s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */ - callback_register(&vm_map_to_kernel(map)->vmk_reclaim_callback, - &pp->pr_reclaimerentry, pp, pool_reclaim_callback); - splx(s); - -#ifdef DIAGNOSTIC - /* Diagnostic drain attempt. */ - uvm_km_va_drain(map, 0); -#endif -} - -static void -pool_reclaim_unregister(struct pool *pp) -{ - struct vm_map *map = pp->pr_alloc->pa_backingmap; - int s; - - if (map == NULL) { - return; - } - - s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */ - callback_unregister(&vm_map_to_kernel(map)->vmk_reclaim_callback, - &pp->pr_reclaimerentry); - splx(s); -} - -static void -pa_reclaim_register(struct pool_allocator *pa) -{ - struct vm_map *map = *pa->pa_backingmapptr; - struct pool *pp; - - KASSERT(pa->pa_backingmap == NULL); - if (map == NULL) { - SLIST_INSERT_HEAD(&pa_deferinitq, pa, pa_q); - return; - } - pa->pa_backingmap = map; - TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) { - pool_reclaim_register(pp); - } -} - /* * Initialize all the pools listed in the "pools" link set. */ void pool_subsystem_init(void) { - struct pool_allocator *pa; + size_t size; + int idx; mutex_init(&pool_head_lock, MUTEX_DEFAULT, IPL_NONE); mutex_init(&pool_allocator_lock, MUTEX_DEFAULT, IPL_NONE); cv_init(&pool_busy, "poolbusy"); - while ((pa = SLIST_FIRST(&pa_deferinitq)) != NULL) { - KASSERT(pa->pa_backingmapptr != NULL); - KASSERT(*pa->pa_backingmapptr != NULL); - SLIST_REMOVE_HEAD(&pa_deferinitq, pa_q); - pa_reclaim_register(pa); + /* + * Initialize private page header pool and cache magazine pool if we + * haven't done so yet. + */ + for (idx = 0; idx < PHPOOL_MAX; idx++) { + static char phpool_names[PHPOOL_MAX][6+1+6+1]; + int nelem; + size_t sz; + + nelem = PHPOOL_FREELIST_NELEM(idx); + snprintf(phpool_names[idx], sizeof(phpool_names[idx]), + "phpool-%d", nelem); + sz = sizeof(struct pool_item_header); + if (nelem) { + sz = offsetof(struct pool_item_header, + ph_bitmap[howmany(nelem, BITMAP_SIZE)]); + } + pool_init(&phpool[idx], sz, 0, 0, 0, + phpool_names[idx], &pool_allocator_meta, IPL_VM); } +#ifdef POOL_SUBPAGE + pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0, + PR_RECURSIVE, "psppool", &pool_allocator_meta, IPL_VM); +#endif + + size = sizeof(pcg_t) + + (PCG_NOBJECTS_NORMAL - 1) * sizeof(pcgpair_t); + pool_init(&pcg_normal_pool, size, coherency_unit, 0, 0, + "pcgnormal", &pool_allocator_meta, IPL_VM); + + size = sizeof(pcg_t) + + (PCG_NOBJECTS_LARGE - 1) * sizeof(pcgpair_t); + pool_init(&pcg_large_pool, size, coherency_unit, 0, 0, + "pcglarge", &pool_allocator_meta, IPL_VM); pool_init(&cache_pool, sizeof(struct pool_cache), coherency_unit, - 0, 0, "pcache", &pool_allocator_nointr, IPL_NONE); + 0, 0, "pcache", &pool_allocator_meta, IPL_NONE); pool_init(&cache_cpu_pool, sizeof(pool_cache_cpu_t), coherency_unit, - 0, 0, "pcachecpu", &pool_allocator_nointr, IPL_NONE); + 0, 0, "pcachecpu", &pool_allocator_meta, IPL_NONE); } /* * Initialize the given pool resource structure. * * We export this routine to allow other kernel parts to declare - * static pools that must be initialized before malloc() is available. + * static pools that must be initialized before kmem(9) is available. */ void pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags, @@ -648,6 +462,8 @@ pool_init(struct pool *pp, size_t size, int off, slack; #ifdef DEBUG + if (__predict_true(!cold)) + mutex_enter(&pool_head_lock); /* * Check that the pool hasn't already been initialised and * added to the list of all pools. @@ -657,14 +473,8 @@ pool_init(struct pool *pp, size_t size, panic("pool_init: pool %s already initialised", wchan); } -#endif - -#ifdef POOL_DIAGNOSTIC - /* - * Always log if POOL_DIAGNOSTIC is defined. - */ - if (pool_logsize != 0) - flags |= PR_LOGGING; + if (__predict_true(!cold)) + mutex_exit(&pool_head_lock); #endif if (palloc == NULL) @@ -688,10 +498,6 @@ pool_init(struct pool *pp, size_t size, mutex_init(&palloc->pa_lock, MUTEX_DEFAULT, IPL_VM); palloc->pa_pagemask = ~(palloc->pa_pagesz - 1); palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1; - - if (palloc->pa_backingmapptr != NULL) { - pa_reclaim_register(palloc); - } } if (!cold) mutex_exit(&pool_allocator_lock); @@ -817,54 +623,10 @@ pool_init(struct pool *pp, size_t size, pp->pr_nidle = 0; pp->pr_refcnt = 0; - pp->pr_log = NULL; - - pp->pr_entered_file = NULL; - pp->pr_entered_line = 0; - mutex_init(&pp->pr_lock, MUTEX_DEFAULT, ipl); cv_init(&pp->pr_cv, wchan); pp->pr_ipl = ipl; - /* - * Initialize private page header pool and cache magazine pool if we - * haven't done so yet. - * XXX LOCKING. - */ - if (phpool[0].pr_size == 0) { - int idx; - for (idx = 0; idx < PHPOOL_MAX; idx++) { - static char phpool_names[PHPOOL_MAX][6+1+6+1]; - int nelem; - size_t sz; - - nelem = PHPOOL_FREELIST_NELEM(idx); - snprintf(phpool_names[idx], sizeof(phpool_names[idx]), - "phpool-%d", nelem); - sz = sizeof(struct pool_item_header); - if (nelem) { - sz = offsetof(struct pool_item_header, - ph_bitmap[howmany(nelem, BITMAP_SIZE)]); - } - pool_init(&phpool[idx], sz, 0, 0, 0, - phpool_names[idx], &pool_allocator_meta, IPL_VM); - } -#ifdef POOL_SUBPAGE - pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0, - PR_RECURSIVE, "psppool", &pool_allocator_meta, IPL_VM); -#endif - - size = sizeof(pcg_t) + - (PCG_NOBJECTS_NORMAL - 1) * sizeof(pcgpair_t); - pool_init(&pcg_normal_pool, size, coherency_unit, 0, 0, - "pcgnormal", &pool_allocator_meta, IPL_VM); - - size = sizeof(pcg_t) + - (PCG_NOBJECTS_LARGE - 1) * sizeof(pcgpair_t); - pool_init(&pcg_large_pool, size, coherency_unit, 0, 0, - "pcglarge", &pool_allocator_meta, IPL_VM); - } - /* Insert into the list of all pools. */ if (!cold) mutex_enter(&pool_head_lock); @@ -885,8 +647,6 @@ pool_init(struct pool *pp, size_t size, TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list); if (!cold) mutex_exit(&palloc->pa_lock); - - pool_reclaim_register(pp); } /* @@ -908,7 +668,6 @@ pool_destroy(struct pool *pp) mutex_exit(&pool_head_lock); /* Remove this pool from its allocator's list of pools. */ - pool_reclaim_unregister(pp); mutex_enter(&pp->pr_alloc->pa_lock); TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list); mutex_exit(&pp->pr_alloc->pa_lock); @@ -924,7 +683,6 @@ 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", pp->pr_nout); } @@ -941,14 +699,6 @@ pool_destroy(struct pool *pp) mutex_exit(&pp->pr_lock); pr_pagelist_free(pp, &pq); - -#ifdef POOL_DIAGNOSTIC - if (pp->pr_log != NULL) { - free(pp->pr_log, M_TEMP); - pp->pr_log = NULL; - } -#endif - cv_destroy(&pp->pr_cv); mutex_destroy(&pp->pr_lock); } @@ -983,11 +733,7 @@ pool_alloc_item_header(struct pool *pp, * Grab an item from the pool. */ void * -#ifdef POOL_DIAGNOSTIC -_pool_get(struct pool *pp, int flags, const char *file, long line) -#else pool_get(struct pool *pp, int flags) -#endif { struct pool_item *pi; struct pool_item_header *ph; @@ -1007,8 +753,6 @@ pool_get(struct pool *pp, int flags) } mutex_enter(&pp->pr_lock); - pr_enter(pp, file, line); - startover: /* * Check to see if we've reached the hard limit. If we have, @@ -1017,7 +761,6 @@ pool_get(struct pool *pp, int flags) */ #ifdef DIAGNOSTIC if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) { - pr_leave(pp); mutex_exit(&pp->pr_lock); panic("pool_get: %s: crossed hard limit", pp->pr_wchan); } @@ -1029,11 +772,9 @@ pool_get(struct pool *pp, int flags) * back to the pool, unlock, call the hook, re-lock, * and check the hardlimit condition again. */ - pr_leave(pp); mutex_exit(&pp->pr_lock); (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags); mutex_enter(&pp->pr_lock); - pr_enter(pp, file, line); if (pp->pr_nout < pp->pr_hardlimit) goto startover; } @@ -1044,9 +785,7 @@ pool_get(struct pool *pp, int flags) * it be? */ pp->pr_flags |= PR_WANTED; - pr_leave(pp); cv_wait(&pp->pr_cv, &pp->pr_lock); - pr_enter(pp, file, line); goto startover; } @@ -1060,7 +799,6 @@ pool_get(struct pool *pp, int flags) pp->pr_nfail++; - pr_leave(pp); mutex_exit(&pp->pr_lock); return (NULL); } @@ -1088,9 +826,7 @@ pool_get(struct pool *pp, int flags) * Release the pool lock, as the back-end page allocator * may block. */ - pr_leave(pp); error = pool_grow(pp, flags); - pr_enter(pp, file, line); if (error != 0) { /* * We were unable to allocate a page or item @@ -1102,7 +838,6 @@ pool_get(struct pool *pp, int flags) goto startover; pp->pr_nfail++; - pr_leave(pp); mutex_exit(&pp->pr_lock); return (NULL); } @@ -1113,25 +848,19 @@ pool_get(struct pool *pp, int flags) if (pp->pr_roflags & PR_NOTOUCH) { #ifdef DIAGNOSTIC if (__predict_false(ph->ph_nmissing == pp->pr_itemsperpage)) { - pr_leave(pp); mutex_exit(&pp->pr_lock); panic("pool_get: %s: page empty", pp->pr_wchan); } #endif v = pr_item_notouch_get(pp, ph); -#ifdef POOL_DIAGNOSTIC - pr_log(pp, v, PRLOG_GET, file, line); -#endif } else { v = pi = LIST_FIRST(&ph->ph_itemlist); if (__predict_false(v == NULL)) { - pr_leave(pp); mutex_exit(&pp->pr_lock); panic("pool_get: %s: page empty", pp->pr_wchan); } #ifdef DIAGNOSTIC if (__predict_false(pp->pr_nitems == 0)) { - pr_leave(pp); mutex_exit(&pp->pr_lock); printf("pool_get: %s: items on itemlist, nitems %u\n", pp->pr_wchan, pp->pr_nitems); @@ -1139,13 +868,8 @@ pool_get(struct pool *pp, int flags) } #endif -#ifdef POOL_DIAGNOSTIC - pr_log(pp, v, PRLOG_GET, file, line); -#endif - #ifdef DIAGNOSTIC if (__predict_false(pi->pi_magic != PI_MAGIC)) { - pr_printlog(pp, pi, printf); panic("pool_get(%s): free list modified: " "magic=%x; page %p; item addr %p\n", pp->pr_wchan, pi->pi_magic, ph->ph_page, pi); @@ -1178,7 +902,6 @@ pool_get(struct pool *pp, int flags) #ifdef DIAGNOSTIC if (__predict_false((pp->pr_roflags & PR_NOTOUCH) == 0 && !LIST_EMPTY(&ph->ph_itemlist))) { - pr_leave(pp); mutex_exit(&pp->pr_lock); panic("pool_get: %s: nmissing inconsistent", pp->pr_wchan); @@ -1194,7 +917,6 @@ pool_get(struct pool *pp, int flags) } pp->pr_nget++; - pr_leave(pp); /* * If we have a low water mark and we are now below that low @@ -1236,7 +958,6 @@ pool_do_put(struct pool *pp, void *v, st #endif if (__predict_false((ph = pr_find_pagehead(pp, v)) == NULL)) { - pr_printlog(pp, NULL, printf); panic("pool_put: %s: page header missing", pp->pr_wchan); } @@ -1325,32 +1046,6 @@ pool_do_put(struct pool *pp, void *v, st } } -/* - * Return resource to the pool. - */ -#ifdef POOL_DIAGNOSTIC -void -_pool_put(struct pool *pp, void *v, const char *file, long line) -{ - struct pool_pagelist pq; - - LIST_INIT(&pq); - - mutex_enter(&pp->pr_lock); - pr_enter(pp, file, line); - - pr_log(pp, v, PRLOG_PUT, file, line); - - pool_do_put(pp, v, &pq); - - pr_leave(pp); - mutex_exit(&pp->pr_lock); - - pr_pagelist_free(pp, &pq); -} -#undef pool_put -#endif /* POOL_DIAGNOSTIC */ - void pool_put(struct pool *pp, void *v) { @@ -1365,10 +1060,6 @@ pool_put(struct pool *pp, void *v) pr_pagelist_free(pp, &pq); } -#ifdef POOL_DIAGNOSTIC -#define pool_put(h, v) _pool_put((h), (v), __FILE__, __LINE__) -#endif - /* * pool_grow: grow a pool by a page. * @@ -1613,14 +1304,10 @@ pool_sethardlimit(struct pool *pp, int n /* * Release all complete pages that have not been used recently. * - * Might be called from interrupt context. + * Must not be called from interrupt context. */ int -#ifdef POOL_DIAGNOSTIC -_pool_reclaim(struct pool *pp, const char *file, long line) -#else pool_reclaim(struct pool *pp) -#endif { struct pool_item_header *ph, *phnext; struct pool_pagelist pq; @@ -1628,9 +1315,7 @@ pool_reclaim(struct pool *pp) bool klock; int rv; - if (cpu_intr_p() || cpu_softintr_p()) { - KASSERT(pp->pr_ipl != IPL_NONE); - } + KASSERT(!cpu_intr_p() && !cpu_softintr_p()); if (pp->pr_drain_hook != NULL) { /* @@ -1660,7 +1345,6 @@ pool_reclaim(struct pool *pp) } return (0); } - pr_enter(pp, file, line); LIST_INIT(&pq); @@ -1674,8 +1358,7 @@ pool_reclaim(struct pool *pp) break; KASSERT(ph->ph_nmissing == 0); - if (curtime - ph->ph_time < pool_inactive_time - && !pa_starved_p(pp->pr_alloc)) + if (curtime - ph->ph_time < pool_inactive_time) continue; /* @@ -1689,7 +1372,6 @@ pool_reclaim(struct pool *pp) pr_rmpage(pp, ph, &pq); } - pr_leave(pp); mutex_exit(&pp->pr_lock); if (LIST_EMPTY(&pq)) @@ -1707,17 +1389,14 @@ pool_reclaim(struct pool *pp) } /* - * Drain pools, one at a time. This is a two stage process; - * drain_start kicks off a cross call to drain CPU-level caches - * if the pool has an associated pool_cache. drain_end waits - * for those cross calls to finish, and then drains the cache - * (if any) and pool. + * Drain pools, one at a time. The drained pool is returned within ppp. * * Note, must never be called from interrupt context. */ -void -pool_drain_start(struct pool **ppp, uint64_t *wp) +bool +pool_drain(struct pool **ppp) { + bool reclaimed; struct pool *pp; KASSERT(!TAILQ_EMPTY(&pool_head)); @@ -1742,28 +1421,6 @@ pool_drain_start(struct pool **ppp, uint pp->pr_refcnt++; mutex_exit(&pool_head_lock); - /* If there is a pool_cache, drain CPU level caches. */ - *ppp = pp; - if (pp->pr_cache != NULL) { - *wp = xc_broadcast(0, (xcfunc_t)pool_cache_xcall, - pp->pr_cache, NULL); - } -} - -bool -pool_drain_end(struct pool *pp, uint64_t where) -{ - bool reclaimed; - - if (pp == NULL) - return false; - - KASSERT(pp->pr_refcnt > 0); - - /* Wait for remote draining to complete. */ - if (pp->pr_cache != NULL) - xc_wait(where); - /* Drain the cache (if any) and pool.. */ reclaimed = pool_reclaim(pp); @@ -1773,18 +1430,15 @@ pool_drain_end(struct pool *pp, uint64_t cv_broadcast(&pool_busy); mutex_exit(&pool_head_lock); + if (ppp != NULL) + *ppp = pp; + return reclaimed; } /* * Diagnostic helpers. */ -void -pool_print(struct pool *pp, const char *modif) -{ - - pool_print1(pp, modif, printf); -} void pool_printall(const char *modif, void (*pr)(const char *, ...)) @@ -1896,11 +1550,6 @@ pool_print1(struct pool *pp, const char goto skip_log; (*pr)("\n"); - if ((pp->pr_roflags & PR_LOGGING) == 0) - (*pr)("\tno log\n"); - else { - pr_printlog(pp, NULL, pr); - } skip_log: @@ -1950,8 +1599,6 @@ pool_print1(struct pool *pp, const char } } #undef PR_GROUPLIST - - pr_enter_check(pp, pr); } static int @@ -2157,6 +1804,19 @@ pool_cache_bootstrap(pool_cache_t pc, si void pool_cache_destroy(pool_cache_t pc) { + + pool_cache_bootstrap_destroy(pc); + pool_put(&cache_pool, pc); +} + +/* + * pool_cache_bootstrap_destroy: + * + * Destroy a pool cache. + */ +void +pool_cache_bootstrap_destroy(pool_cache_t pc) +{ struct pool *pp = &pc->pc_pool; u_int i; @@ -2182,7 +1842,6 @@ pool_cache_destroy(pool_cache_t pc) /* Finally, destroy it. */ mutex_destroy(&pc->pc_lock); pool_destroy(pp); - pool_put(&cache_pool, pc); } /* @@ -2328,31 +1987,39 @@ pool_cache_invalidate_groups(pool_cache_ * Note: For pool caches that provide constructed objects, there * is an assumption that another level of synchronization is occurring * between the input to the constructor and the cache invalidation. + * + * Invalidation is a costly process and should not be called from + * interrupt context. */ void pool_cache_invalidate(pool_cache_t pc) { - pcg_t *full, *empty, *part; -#if 0 uint64_t where; + pcg_t *full, *empty, *part; + + KASSERT(!cpu_intr_p() && !cpu_softintr_p()); if (ncpu < 2 || !mp_online) { /* * We might be called early enough in the boot process * for the CPU data structures to not be fully initialized. - * In this case, simply gather the local CPU's cache now - * since it will be the only one running. + * In this case, transfer the content of the local CPU's + * cache back into global cache as only this CPU is currently + * running. */ - pool_cache_xcall(pc); + pool_cache_transfer(pc); } else { /* - * Gather all of the CPU-specific caches into the - * global cache. + * Signal all CPUs that they must transfer their local + * cache back to the global pool then wait for the xcall to + * complete. */ - where = xc_broadcast(0, (xcfunc_t)pool_cache_xcall, pc, NULL); + where = xc_broadcast(0, (xcfunc_t)pool_cache_transfer, + pc, NULL); xc_wait(where); } -#endif + + /* Empty pool caches, then invalidate objects */ mutex_enter(&pc->pc_lock); full = pc->pc_fullgroups; empty = pc->pc_emptygroups; @@ -2384,7 +2051,6 @@ pool_cache_invalidate(pool_cache_t pc) static void pool_cache_invalidate_cpu(pool_cache_t pc, u_int index) { - pool_cache_cpu_t *cc; pcg_t *pcg; @@ -2539,8 +2205,8 @@ pool_cache_get_paddr(pool_cache_t pc, in KASSERTMSG((!cpu_intr_p() && !cpu_softintr_p()) || (pc->pc_pool.pr_ipl != IPL_NONE || cold || panicstr != NULL), - ("pool '%s' is IPL_NONE, but called from interrupt context\n", - pc->pc_pool.pr_wchan)); + "pool '%s' is IPL_NONE, but called from interrupt context\n", + pc->pc_pool.pr_wchan); if (flags & PR_WAITOK) { ASSERT_SLEEPABLE(); @@ -2595,6 +2261,7 @@ pool_cache_get_paddr(pool_cache_t pc, in static bool __noinline pool_cache_put_slow(pool_cache_cpu_t *cc, int s, void *object) { + struct lwp *l = curlwp; pcg_t *pcg, *cur; uint64_t ncsw; pool_cache_t pc; @@ -2605,6 +2272,7 @@ pool_cache_put_slow(pool_cache_cpu_t *cc pc = cc->cc_cache; pcg = NULL; cc->cc_misses++; + ncsw = l->l_ncsw; /* * If there are no empty groups in the cache then allocate one @@ -2614,6 +2282,16 @@ pool_cache_put_slow(pool_cache_cpu_t *cc if (__predict_true(!pool_cache_disable)) { pcg = pool_get(pc->pc_pcgpool, PR_NOWAIT); } + /* + * If pool_get() blocked, then our view of + * the per-CPU data is invalid: retry. + */ + if (__predict_false(l->l_ncsw != ncsw)) { + if (pcg != NULL) { + pool_put(pc->pc_pcgpool, pcg); + } + return true; + } if (__predict_true(pcg != NULL)) { pcg->pcg_avail = 0; pcg->pcg_size = pc->pc_pcgsize; @@ -2622,7 +2300,6 @@ pool_cache_put_slow(pool_cache_cpu_t *cc /* Lock the cache. */ if (__predict_false(!mutex_tryenter(&pc->pc_lock))) { - ncsw = curlwp->l_ncsw; mutex_enter(&pc->pc_lock); pc->pc_contended++; @@ -2630,7 +2307,7 @@ pool_cache_put_slow(pool_cache_cpu_t *cc * If we context switched while locking, then our view of * the per-CPU data is invalid: retry. */ - if (__predict_false(curlwp->l_ncsw != ncsw)) { + if (__predict_false(l->l_ncsw != ncsw)) { mutex_exit(&pc->pc_lock); if (pcg != NULL) { pool_put(pc->pc_pcgpool, pcg); @@ -2737,13 +2414,13 @@ pool_cache_put_paddr(pool_cache_t pc, vo } /* - * pool_cache_xcall: + * pool_cache_transfer: * * Transfer objects from the per-CPU cache to the global cache. * Run within a cross-call thread. */ static void -pool_cache_xcall(pool_cache_t pc) +pool_cache_transfer(pool_cache_t pc) { pool_cache_cpu_t *cc; pcg_t *prev, *cur, **list; @@ -2806,28 +2483,29 @@ void pool_page_free(struct pool *, void #ifdef POOL_SUBPAGE struct pool_allocator pool_allocator_kmem_fullpage = { - pool_page_alloc, pool_page_free, 0, - .pa_backingmapptr = &kmem_map, + .pa_alloc = pool_page_alloc, + .pa_free = pool_page_free, + .pa_pagesz = 0 }; #else struct pool_allocator pool_allocator_kmem = { - pool_page_alloc, pool_page_free, 0, - .pa_backingmapptr = &kmem_map, + .pa_alloc = pool_page_alloc, + .pa_free = pool_page_free, + .pa_pagesz = 0 }; #endif -void *pool_page_alloc_nointr(struct pool *, int); -void pool_page_free_nointr(struct pool *, void *); - #ifdef POOL_SUBPAGE struct pool_allocator pool_allocator_nointr_fullpage = { - pool_page_alloc_nointr, pool_page_free_nointr, 0, - .pa_backingmapptr = &kernel_map, + .pa_alloc = pool_page_alloc, + .pa_free = pool_page_free, + .pa_pagesz = 0 }; #else struct pool_allocator pool_allocator_nointr = { - pool_page_alloc_nointr, pool_page_free_nointr, 0, - .pa_backingmapptr = &kernel_map, + .pa_alloc = pool_page_alloc, + .pa_free = pool_page_free, + .pa_pagesz = 0 }; #endif @@ -2836,16 +2514,15 @@ void *pool_subpage_alloc(struct pool *, void pool_subpage_free(struct pool *, void *); struct pool_allocator pool_allocator_kmem = { - pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE, - .pa_backingmapptr = &kmem_map, + .pa_alloc = pool_subpage_alloc, + .pa_free = pool_subpage_free, + .pa_pagesz = POOL_SUBPAGE }; -void *pool_subpage_alloc_nointr(struct pool *, int); -void pool_subpage_free_nointr(struct pool *, void *); - struct pool_allocator pool_allocator_nointr = { - pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE, - .pa_backingmapptr = &kmem_map, + .pa_alloc = pool_subpage_alloc, + .pa_free = pool_subpage_free, + .pa_pagesz = POOL_SUBPAGE }; #endif /* POOL_SUBPAGE */ @@ -2881,31 +2558,41 @@ pool_allocator_free(struct pool *pp, voi void * pool_page_alloc(struct pool *pp, int flags) { - bool waitok = (flags & PR_WAITOK) ? true : false; + const vm_flag_t vflags = (flags & PR_WAITOK) ? VM_SLEEP: VM_NOSLEEP; + vmem_addr_t va; + int ret; - return ((void *) uvm_km_alloc_poolpage_cache(kmem_map, waitok)); + ret = uvm_km_kmem_alloc(kmem_va_arena, pp->pr_alloc->pa_pagesz, + vflags | VM_INSTANTFIT, &va); + + return ret ? NULL : (void *)va; } void pool_page_free(struct pool *pp, void *v) { - uvm_km_free_poolpage_cache(kmem_map, (vaddr_t) v); + uvm_km_kmem_free(kmem_va_arena, (vaddr_t)v, pp->pr_alloc->pa_pagesz); } static void * pool_page_alloc_meta(struct pool *pp, int flags) { - bool waitok = (flags & PR_WAITOK) ? true : false; + const vm_flag_t vflags = (flags & PR_WAITOK) ? VM_SLEEP: VM_NOSLEEP; + vmem_addr_t va; + int ret; + + ret = vmem_alloc(kmem_meta_arena, pp->pr_alloc->pa_pagesz, + vflags | VM_INSTANTFIT, &va); - return ((void *) uvm_km_alloc_poolpage(kmem_map, waitok)); + return ret ? NULL : (void *)va; } static void pool_page_free_meta(struct pool *pp, void *v) { - uvm_km_free_poolpage(kmem_map, (vaddr_t) v); + vmem_free(kmem_meta_arena, (vmem_addr_t)v, pp->pr_alloc->pa_pagesz); } #ifdef POOL_SUBPAGE @@ -2922,35 +2609,7 @@ pool_subpage_free(struct pool *pp, void pool_put(&psppool, v); } -/* We don't provide a real nointr allocator. Maybe later. */ -void * -pool_subpage_alloc_nointr(struct pool *pp, int flags) -{ - - return (pool_subpage_alloc(pp, flags)); -} - -void -pool_subpage_free_nointr(struct pool *pp, void *v) -{ - - pool_subpage_free(pp, v); -} #endif /* POOL_SUBPAGE */ -void * -pool_page_alloc_nointr(struct pool *pp, int flags) -{ - bool waitok = (flags & PR_WAITOK) ? true : false; - - return ((void *) uvm_km_alloc_poolpage_cache(kernel_map, waitok)); -} - -void -pool_page_free_nointr(struct pool *pp, void *v) -{ - - uvm_km_free_poolpage_cache(kernel_map, (vaddr_t) v); -} #if defined(DDB) static bool