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 1.190.2.2 retrieving revision 1.204 diff -u -p -r1.190.2.2 -r1.204 --- src/sys/kern/subr_pool.c 2012/05/23 10:08:11 1.190.2.2 +++ src/sys/kern/subr_pool.c 2015/07/28 12:32:44 1.204 @@ -1,13 +1,14 @@ -/* $NetBSD: subr_pool.c,v 1.190.2.2 2012/05/23 10:08:11 yamt Exp $ */ +/* $NetBSD: subr_pool.c,v 1.204 2015/07/28 12:32:44 maxv Exp $ */ /*- - * Copyright (c) 1997, 1999, 2000, 2002, 2007, 2008, 2010 + * Copyright (c) 1997, 1999, 2000, 2002, 2007, 2008, 2010, 2014, 2015 * The NetBSD Foundation, Inc. * All rights reserved. * * This code is derived from software contributed to The NetBSD Foundation * by Paul Kranenburg; by Jason R. Thorpe of the Numerical Aerospace - * Simulation Facility, NASA Ames Research Center, and by Andrew Doran. + * Simulation Facility, NASA Ames Research Center; by Andrew Doran, and by + * Maxime Villard. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions @@ -32,13 +33,14 @@ */ #include -__KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.190.2.2 2012/05/23 10:08:11 yamt Exp $"); +__KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.204 2015/07/28 12:32:44 maxv Exp $"); #include "opt_ddb.h" #include "opt_lockdebug.h" #include #include +#include #include #include #include @@ -67,8 +69,8 @@ __KERNEL_RCSID(0, "$NetBSD: subr_pool.c, * an internal pool of page headers (`phpool'). */ -/* List of all pools */ -static TAILQ_HEAD(, pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head); +/* List of all pools. Non static as needed by 'vmstat -i' */ +TAILQ_HEAD(, pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head); /* Private pool for page header structures */ #define PHPOOL_MAX 8 @@ -81,6 +83,17 @@ static struct pool phpool[PHPOOL_MAX]; static struct pool psppool; #endif +#ifdef POOL_REDZONE +# define POOL_REDZONE_SIZE 2 +static void pool_redzone_init(struct pool *, size_t); +static void pool_redzone_fill(struct pool *, void *); +static void pool_redzone_check(struct pool *, void *); +#else +# define pool_redzone_init(pp, sz) /* NOTHING */ +# define pool_redzone_fill(pp, ptr) /* NOTHING */ +# define pool_redzone_check(pp, ptr) /* NOTHING */ +#endif + static void *pool_page_alloc_meta(struct pool *, int); static void pool_page_free_meta(struct pool *, void *); @@ -191,7 +204,7 @@ 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 *, @@ -203,9 +216,9 @@ 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 *); @@ -458,10 +471,12 @@ pool_init(struct pool *pp, size_t size, const char *wchan, struct pool_allocator *palloc, int ipl) { struct pool *pp1; - size_t trysize, phsize; + size_t trysize, phsize, prsize; 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. @@ -471,6 +486,8 @@ pool_init(struct pool *pp, size_t size, panic("pool_init: pool %s already initialised", wchan); } + if (__predict_true(!cold)) + mutex_exit(&pool_head_lock); #endif if (palloc == NULL) @@ -501,13 +518,14 @@ pool_init(struct pool *pp, size_t size, if (align == 0) align = ALIGN(1); - if ((flags & PR_NOTOUCH) == 0 && size < sizeof(struct pool_item)) - size = sizeof(struct pool_item); + prsize = size; + if ((flags & PR_NOTOUCH) == 0 && prsize < sizeof(struct pool_item)) + prsize = sizeof(struct pool_item); - size = roundup(size, align); + prsize = roundup(prsize, align); #ifdef DIAGNOSTIC - if (size > palloc->pa_pagesz) - panic("pool_init: pool item size (%zu) too large", size); + if (prsize > palloc->pa_pagesz) + panic("pool_init: pool item size (%zu) too large", prsize); #endif /* @@ -524,7 +542,7 @@ pool_init(struct pool *pp, size_t size, pp->pr_maxpages = UINT_MAX; pp->pr_roflags = flags; pp->pr_flags = 0; - pp->pr_size = size; + pp->pr_size = prsize; pp->pr_align = align; pp->pr_wchan = wchan; pp->pr_alloc = palloc; @@ -539,6 +557,7 @@ pool_init(struct pool *pp, size_t size, pp->pr_drain_hook = NULL; pp->pr_drain_hook_arg = NULL; pp->pr_freecheck = NULL; + pool_redzone_init(pp, size); /* * Decide whether to put the page header off page to avoid @@ -557,9 +576,10 @@ pool_init(struct pool *pp, size_t size, /* See the comment below about reserved bytes. */ trysize = palloc->pa_pagesz - ((align - ioff) % align); phsize = ALIGN(sizeof(struct pool_item_header)); - if ((pp->pr_roflags & (PR_NOTOUCH | PR_NOALIGN)) == 0 && + if (pp->pr_roflags & PR_PHINPAGE || + ((pp->pr_roflags & (PR_NOTOUCH | PR_NOALIGN)) == 0 && (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) || - trysize / pp->pr_size == (trysize - phsize) / pp->pr_size)) { + trysize / pp->pr_size == (trysize - phsize) / pp->pr_size))) { /* Use the end of the page for the page header */ pp->pr_roflags |= PR_PHINPAGE; pp->pr_phoffset = off = palloc->pa_pagesz - phsize; @@ -929,6 +949,7 @@ pool_get(struct pool *pp, int flags) mutex_exit(&pp->pr_lock); KASSERT((((vaddr_t)v + pp->pr_itemoffset) & (pp->pr_align - 1)) == 0); FREECHECK_OUT(&pp->pr_freecheck, v); + pool_redzone_fill(pp, v); return (v); } @@ -942,6 +963,7 @@ pool_do_put(struct pool *pp, void *v, st struct pool_item_header *ph; KASSERT(mutex_owned(&pp->pr_lock)); + pool_redzone_check(pp, v); FREECHECK_IN(&pp->pr_freecheck, v); LOCKDEBUG_MEM_CHECK(v, pp->pr_size); @@ -1300,7 +1322,7 @@ 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 pool_reclaim(struct pool *pp) @@ -1311,9 +1333,7 @@ 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) { /* @@ -1387,17 +1407,14 @@ 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)); @@ -1422,28 +1439,6 @@ 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); @@ -1453,6 +1448,9 @@ pool_drain_end(struct pool *pp, uint64_t cv_broadcast(&pool_busy); mutex_exit(&pool_head_lock); + if (ppp != NULL) + *ppp = pp; + return reclaimed; } @@ -2007,31 +2005,39 @@ 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; @@ -2198,6 +2204,7 @@ pool_cache_get_slow(pool_cache_cpu_t *cc } FREECHECK_OUT(&pc->pc_freecheck, object); + pool_redzone_fill(&pc->pc_pool, object); return false; } @@ -2243,6 +2250,7 @@ pool_cache_get_paddr(pool_cache_t pc, in cc->cc_hits++; splx(s); FREECHECK_OUT(&pc->pc_freecheck, object); + pool_redzone_fill(&pc->pc_pool, object); return object; } @@ -2273,6 +2281,7 @@ 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; @@ -2283,6 +2292,7 @@ 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 @@ -2292,6 +2302,16 @@ 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; @@ -2300,7 +2320,6 @@ 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++; @@ -2308,7 +2327,7 @@ 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); @@ -2375,6 +2394,7 @@ pool_cache_put_paddr(pool_cache_t pc, vo int s; KASSERT(object != NULL); + pool_redzone_check(&pc->pc_pool, object); FREECHECK_IN(&pc->pc_freecheck, object); /* Lock out interrupts and disable preemption. */ @@ -2415,13 +2435,13 @@ 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; @@ -2596,6 +2616,120 @@ pool_page_free_meta(struct pool *pp, voi vmem_free(kmem_meta_arena, (vmem_addr_t)v, pp->pr_alloc->pa_pagesz); } +#ifdef POOL_REDZONE +#if defined(_LP64) +# define PRIME 0x9e37fffffffc0000UL +#else /* defined(_LP64) */ +# define PRIME 0x9e3779b1 +#endif /* defined(_LP64) */ +#define STATIC_BYTE 0xFE +CTASSERT(POOL_REDZONE_SIZE > 1); + +static inline uint8_t +pool_pattern_generate(const void *p) +{ + return (uint8_t)(((uintptr_t)p) * PRIME + >> ((sizeof(uintptr_t) - sizeof(uint8_t))) * CHAR_BIT); +} + +static void +pool_redzone_init(struct pool *pp, size_t requested_size) +{ + size_t nsz; + + if (pp->pr_roflags & PR_NOTOUCH) { + pp->pr_reqsize = 0; + pp->pr_redzone = false; + return; + } + + /* + * We may have extended the requested size earlier; check if + * there's naturally space in the padding for a red zone. + */ + if (pp->pr_size - requested_size >= POOL_REDZONE_SIZE) { + pp->pr_reqsize = requested_size; + pp->pr_redzone = true; + return; + } + + /* + * No space in the natural padding; check if we can extend a + * bit the size of the pool. + */ + nsz = roundup(pp->pr_size + POOL_REDZONE_SIZE, pp->pr_align); + if (nsz <= pp->pr_alloc->pa_pagesz) { + /* Ok, we can */ + pp->pr_size = nsz; + pp->pr_reqsize = requested_size; + pp->pr_redzone = true; + } else { + /* No space for a red zone... snif :'( */ + pp->pr_reqsize = 0; + pp->pr_redzone = false; + printf("pool redzone disabled for '%s'\n", pp->pr_wchan); + } +} + +static void +pool_redzone_fill(struct pool *pp, void *p) +{ + uint8_t *cp, pat; + const uint8_t *ep; + + if (!pp->pr_redzone) + return; + + cp = (uint8_t *)p + pp->pr_reqsize; + ep = cp + POOL_REDZONE_SIZE; + + /* + * We really don't want the first byte of the red zone to be '\0'; + * an off-by-one in a string may not be properly detected. + */ + pat = pool_pattern_generate(cp); + *cp = (pat == '\0') ? STATIC_BYTE: pat; + cp++; + + while (cp < ep) { + *cp = pool_pattern_generate(cp); + cp++; + } +} + +static void +pool_redzone_check(struct pool *pp, void *p) +{ + uint8_t *cp, pat, expected; + const uint8_t *ep; + + if (!pp->pr_redzone) + return; + + cp = (uint8_t *)p + pp->pr_reqsize; + ep = cp + POOL_REDZONE_SIZE; + + pat = pool_pattern_generate(cp); + expected = (pat == '\0') ? STATIC_BYTE: pat; + if (expected != *cp) { + panic("%s: %p: 0x%02x != 0x%02x\n", + __func__, cp, *cp, expected); + } + cp++; + + while (cp < ep) { + expected = pool_pattern_generate(cp); + if (*cp != expected) { + panic("%s: %p: 0x%02x != 0x%02x\n", + __func__, cp, *cp, expected); + } + cp++; + } +} + +#endif /* POOL_REDZONE */ + + #ifdef POOL_SUBPAGE /* Sub-page allocator, for machines with large hardware pages. */ void * @@ -2751,3 +2885,100 @@ print: } } #endif /* defined(DDB) */ + +static int +pool_sysctl(SYSCTLFN_ARGS) +{ + struct pool_sysctl data; + struct pool *pp; + struct pool_cache *pc; + pool_cache_cpu_t *cc; + int error; + size_t i, written; + + if (oldp == NULL) { + *oldlenp = 0; + TAILQ_FOREACH(pp, &pool_head, pr_poollist) + *oldlenp += sizeof(data); + return 0; + } + + memset(&data, 0, sizeof(data)); + error = 0; + written = 0; + TAILQ_FOREACH(pp, &pool_head, pr_poollist) { + if (written + sizeof(data) > *oldlenp) + break; + strlcpy(data.pr_wchan, pp->pr_wchan, sizeof(data.pr_wchan)); + data.pr_pagesize = pp->pr_alloc->pa_pagesz; + data.pr_flags = pp->pr_roflags | pp->pr_flags; +#define COPY(field) data.field = pp->field + COPY(pr_size); + + COPY(pr_itemsperpage); + COPY(pr_nitems); + COPY(pr_nout); + COPY(pr_hardlimit); + COPY(pr_npages); + COPY(pr_minpages); + COPY(pr_maxpages); + + COPY(pr_nget); + COPY(pr_nfail); + COPY(pr_nput); + COPY(pr_npagealloc); + COPY(pr_npagefree); + COPY(pr_hiwat); + COPY(pr_nidle); +#undef COPY + + data.pr_cache_nmiss_pcpu = 0; + data.pr_cache_nhit_pcpu = 0; + if (pp->pr_cache) { + pc = pp->pr_cache; + data.pr_cache_meta_size = pc->pc_pcgsize; + data.pr_cache_nfull = pc->pc_nfull; + data.pr_cache_npartial = pc->pc_npart; + data.pr_cache_nempty = pc->pc_nempty; + data.pr_cache_ncontended = pc->pc_contended; + data.pr_cache_nmiss_global = pc->pc_misses; + data.pr_cache_nhit_global = pc->pc_hits; + for (i = 0; i < pc->pc_ncpu; ++i) { + cc = pc->pc_cpus[i]; + if (cc == NULL) + continue; + data.pr_cache_nmiss_pcpu = cc->cc_misses; + data.pr_cache_nhit_pcpu = cc->cc_hits; + } + } else { + data.pr_cache_meta_size = 0; + data.pr_cache_nfull = 0; + data.pr_cache_npartial = 0; + data.pr_cache_nempty = 0; + data.pr_cache_ncontended = 0; + data.pr_cache_nmiss_global = 0; + data.pr_cache_nhit_global = 0; + } + + error = sysctl_copyout(l, &data, oldp, sizeof(data)); + if (error) + break; + written += sizeof(data); + oldp = (char *)oldp + sizeof(data); + } + + *oldlenp = written; + return error; +} + +SYSCTL_SETUP(sysctl_pool_setup, "sysctl kern.pool setup") +{ + const struct sysctlnode *rnode = NULL; + + sysctl_createv(clog, 0, NULL, &rnode, + CTLFLAG_PERMANENT, + CTLTYPE_STRUCT, "pool", + SYSCTL_DESCR("Get pool statistics"), + pool_sysctl, 0, NULL, 0, + CTL_KERN, CTL_CREATE, CTL_EOL); +}