From owner-svn-src-all@freebsd.org Sun Dec 8 01:15:07 2019 Return-Path: Delivered-To: svn-src-all@mailman.nyi.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2610:1c1:1:606c::19:1]) by mailman.nyi.freebsd.org (Postfix) with ESMTP id 80E0E1D267B; Sun, 8 Dec 2019 01:15:07 +0000 (UTC) (envelope-from jeff@FreeBSD.org) Received: from mxrelay.nyi.freebsd.org (mxrelay.nyi.freebsd.org [IPv6:2610:1c1:1:606c::19:3]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) server-signature RSA-PSS (4096 bits) client-signature RSA-PSS (4096 bits) client-digest SHA256) (Client CN "mxrelay.nyi.freebsd.org", Issuer "Let's Encrypt Authority X3" (verified OK)) by mx1.freebsd.org (Postfix) with ESMTPS id 47VpK72tmTz4RY1; Sun, 8 Dec 2019 01:15:07 +0000 (UTC) (envelope-from jeff@FreeBSD.org) Received: from repo.freebsd.org (repo.freebsd.org [IPv6:2610:1c1:1:6068::e6a:0]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (Client did not present a certificate) by mxrelay.nyi.freebsd.org (Postfix) with ESMTPS id 476E822730; Sun, 8 Dec 2019 01:15:07 +0000 (UTC) (envelope-from jeff@FreeBSD.org) Received: from repo.freebsd.org ([127.0.1.37]) by repo.freebsd.org (8.15.2/8.15.2) with ESMTP id xB81F7lk004417; Sun, 8 Dec 2019 01:15:07 GMT (envelope-from jeff@FreeBSD.org) Received: (from jeff@localhost) by repo.freebsd.org (8.15.2/8.15.2/Submit) id xB81F6LT004415; Sun, 8 Dec 2019 01:15:06 GMT (envelope-from jeff@FreeBSD.org) Message-Id: <201912080115.xB81F6LT004415@repo.freebsd.org> X-Authentication-Warning: repo.freebsd.org: jeff set sender to jeff@FreeBSD.org using -f From: Jeff Roberson Date: Sun, 8 Dec 2019 01:15:06 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r355510 - head/sys/vm X-SVN-Group: head X-SVN-Commit-Author: jeff X-SVN-Commit-Paths: head/sys/vm X-SVN-Commit-Revision: 355510 X-SVN-Commit-Repository: base MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-BeenThere: svn-src-all@freebsd.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: "SVN commit messages for the entire src tree \(except for " user" and " projects" \)" List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sun, 08 Dec 2019 01:15:07 -0000 Author: jeff Date: Sun Dec 8 01:15:06 2019 New Revision: 355510 URL: https://svnweb.freebsd.org/changeset/base/355510 Log: Use a variant slab structure for offpage zones. This saves space in embedded slabs but also is an opportunity to tidy up code and add accessor inlines. Reviewed by: markj, rlibby Differential Revision: https://reviews.freebsd.org/D22609 Modified: head/sys/vm/uma_core.c head/sys/vm/uma_int.h Modified: head/sys/vm/uma_core.c ============================================================================== --- head/sys/vm/uma_core.c Sun Dec 8 00:06:00 2019 (r355509) +++ head/sys/vm/uma_core.c Sun Dec 8 01:15:06 2019 (r355510) @@ -664,8 +664,7 @@ hash_alloc(struct uma_hash *hash, u_int size) if (size > UMA_HASH_SIZE_INIT) { hash->uh_hashsize = size; alloc = sizeof(hash->uh_slab_hash[0]) * hash->uh_hashsize; - hash->uh_slab_hash = (struct slabhead *)malloc(alloc, - M_UMAHASH, M_NOWAIT); + hash->uh_slab_hash = malloc(alloc, M_UMAHASH, M_NOWAIT); } else { alloc = sizeof(hash->uh_slab_hash[0]) * UMA_HASH_SIZE_INIT; hash->uh_slab_hash = zone_alloc_item(hashzone, NULL, @@ -698,7 +697,7 @@ hash_alloc(struct uma_hash *hash, u_int size) static int hash_expand(struct uma_hash *oldhash, struct uma_hash *newhash) { - uma_slab_t slab; + uma_hash_slab_t slab; u_int hval; u_int idx; @@ -714,12 +713,12 @@ hash_expand(struct uma_hash *oldhash, struct uma_hash */ for (idx = 0; idx < oldhash->uh_hashsize; idx++) - while (!SLIST_EMPTY(&oldhash->uh_slab_hash[idx])) { - slab = SLIST_FIRST(&oldhash->uh_slab_hash[idx]); - SLIST_REMOVE_HEAD(&oldhash->uh_slab_hash[idx], us_hlink); - hval = UMA_HASH(newhash, slab->us_data); - SLIST_INSERT_HEAD(&newhash->uh_slab_hash[hval], - slab, us_hlink); + while (!LIST_EMPTY(&oldhash->uh_slab_hash[idx])) { + slab = LIST_FIRST(&oldhash->uh_slab_hash[idx]); + LIST_REMOVE(slab, uhs_hlink); + hval = UMA_HASH(newhash, slab->uhs_data); + LIST_INSERT_HEAD(&newhash->uh_slab_hash[hval], + slab, uhs_hlink); } return (1); @@ -992,7 +991,7 @@ keg_free_slab(uma_keg_t keg, uma_slab_t slab, int star CTR4(KTR_UMA, "keg_free_slab keg %s(%p) slab %p, returning %d bytes", keg->uk_name, keg, slab, PAGE_SIZE * keg->uk_ppera); - mem = slab->us_data; + mem = slab_data(slab, keg); flags = slab->us_flags; i = start; if (keg->uk_fini != NULL) { @@ -1006,11 +1005,10 @@ keg_free_slab(uma_keg_t keg, uma_slab_t slab, int star * albeit we don't make skip check for other init/fini * invocations. */ - if (!uma_dbg_kskip(keg, slab->us_data + (keg->uk_rsize * i)) || + if (!uma_dbg_kskip(keg, slab_item(slab, keg, i)) || keg->uk_fini != trash_fini) #endif - keg->uk_fini(slab->us_data + (keg->uk_rsize * i), - keg->uk_size); + keg->uk_fini(slab_item(slab, keg, i), keg->uk_size); } if (keg->uk_flags & UMA_ZONE_OFFPAGE) zone_free_item(keg->uk_slabzone, slab, NULL, SKIP_NONE); @@ -1057,18 +1055,17 @@ keg_drain(uma_keg_t keg) keg->uk_free -= keg->uk_ipers; if (keg->uk_flags & UMA_ZONE_HASH) - UMA_HASH_REMOVE(&keg->uk_hash, slab, - slab->us_data); + UMA_HASH_REMOVE(&keg->uk_hash, slab); - SLIST_INSERT_HEAD(&freeslabs, slab, us_hlink); + LIST_INSERT_HEAD(&freeslabs, slab, us_link); } } finished: KEG_UNLOCK(keg); - while ((slab = SLIST_FIRST(&freeslabs)) != NULL) { - SLIST_REMOVE(&freeslabs, slab, uma_slab, us_hlink); + while ((slab = LIST_FIRST(&freeslabs)) != NULL) { + LIST_REMOVE(slab, us_link); keg_free_slab(keg, slab, keg->uk_ipers); } } @@ -1190,13 +1187,14 @@ keg_alloc_slab(uma_keg_t keg, uma_zone_t zone, int dom /* Point the slab into the allocated memory */ if (!(keg->uk_flags & UMA_ZONE_OFFPAGE)) slab = (uma_slab_t )(mem + keg->uk_pgoff); + else + ((uma_hash_slab_t)slab)->uhs_data = mem; if (keg->uk_flags & UMA_ZONE_VTOSLAB) for (i = 0; i < keg->uk_ppera; i++) vsetzoneslab((vm_offset_t)mem + (i * PAGE_SIZE), zone, slab); - slab->us_data = mem; slab->us_freecount = keg->uk_ipers; slab->us_flags = sflags; slab->us_domain = domain; @@ -1207,7 +1205,7 @@ keg_alloc_slab(uma_keg_t keg, uma_zone_t zone, int dom if (keg->uk_init != NULL) { for (i = 0; i < keg->uk_ipers; i++) - if (keg->uk_init(slab->us_data + (keg->uk_rsize * i), + if (keg->uk_init(slab_item(slab, keg, i), keg->uk_size, flags) != 0) break; if (i != keg->uk_ipers) { @@ -2393,15 +2391,12 @@ uma_startup(void *mem, int npages) zone_ctor(zones, zsize, &args, M_WAITOK); /* Now make a zone for slab headers */ - slabzone = uma_zcreate("UMA Slabs", - slab_sizeof(SLAB_MAX_SETSIZE), - NULL, NULL, NULL, NULL, - UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL); + slabzone = uma_zcreate("UMA Slabs", sizeof(struct uma_hash_slab), + NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL); hashzone = uma_zcreate("UMA Hash", sizeof(struct slabhead *) * UMA_HASH_SIZE_INIT, - NULL, NULL, NULL, NULL, - UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL); + NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL); bucket_init(); @@ -3097,7 +3092,7 @@ slab_alloc_item(uma_keg_t keg, uma_slab_t slab) freei = BIT_FFS(keg->uk_ipers, &slab->us_free) - 1; BIT_CLR(keg->uk_ipers, freei, &slab->us_free); - item = slab->us_data + (keg->uk_rsize * freei); + item = slab_item(slab, keg, freei); slab->us_freecount--; keg->uk_free--; @@ -3609,7 +3604,7 @@ slab_free_item(uma_zone_t zone, uma_slab_t slab, void } /* Slab management. */ - freei = ((uintptr_t)item - (uintptr_t)slab->us_data) / keg->uk_rsize; + freei = slab_item_index(slab, keg, item); BIT_SET(keg->uk_ipers, freei, &slab->us_free); slab->us_freecount++; @@ -4491,7 +4486,7 @@ uma_dbg_alloc(uma_zone_t zone, uma_slab_t slab, void * item, zone->uz_name); } keg = zone->uz_keg; - freei = ((uintptr_t)item - (uintptr_t)slab->us_data) / keg->uk_rsize; + freei = slab_item_index(slab, keg, item); if (BIT_ISSET(SLAB_MAX_SETSIZE, freei, &slab->us_debugfree)) panic("Duplicate alloc of %p from zone %p(%s) slab %p(%d)\n", @@ -4519,13 +4514,13 @@ uma_dbg_free(uma_zone_t zone, uma_slab_t slab, void *i item, zone->uz_name); } keg = zone->uz_keg; - freei = ((uintptr_t)item - (uintptr_t)slab->us_data) / keg->uk_rsize; + freei = slab_item_index(slab, keg, item); if (freei >= keg->uk_ipers) panic("Invalid free of %p from zone %p(%s) slab %p(%d)\n", item, zone, zone->uz_name, slab, freei); - if (((freei * keg->uk_rsize) + slab->us_data) != item) + if (slab_item(slab, keg, freei) != item) panic("Unaligned free of %p from zone %p(%s) slab %p(%d)\n", item, zone, zone->uz_name, slab, freei); Modified: head/sys/vm/uma_int.h ============================================================================== --- head/sys/vm/uma_int.h Sun Dec 8 00:06:00 2019 (r355509) +++ head/sys/vm/uma_int.h Sun Dec 8 01:15:06 2019 (r355510) @@ -139,33 +139,28 @@ /* Max waste percentage before going to off page slab management */ #define UMA_MAX_WASTE 10 + /* - * I doubt there will be many cases where this is exceeded. This is the initial - * size of the hash table for uma_slabs that are managed off page. This hash - * does expand by powers of two. Currently it doesn't get smaller. + * Hash table for freed address -> slab translation. + * + * Only zones with memory not touchable by the allocator use the + * hash table. Otherwise slabs are found with vtoslab(). */ #define UMA_HASH_SIZE_INIT 32 -/* - * I should investigate other hashing algorithms. This should yield a low - * number of collisions if the pages are relatively contiguous. - */ - #define UMA_HASH(h, s) ((((uintptr_t)s) >> UMA_SLAB_SHIFT) & (h)->uh_hashmask) #define UMA_HASH_INSERT(h, s, mem) \ - SLIST_INSERT_HEAD(&(h)->uh_slab_hash[UMA_HASH((h), \ - (mem))], (s), us_hlink) -#define UMA_HASH_REMOVE(h, s, mem) \ - SLIST_REMOVE(&(h)->uh_slab_hash[UMA_HASH((h), \ - (mem))], (s), uma_slab, us_hlink) + LIST_INSERT_HEAD(&(h)->uh_slab_hash[UMA_HASH((h), \ + (mem))], (uma_hash_slab_t)(s), uhs_hlink) -/* Hash table for freed address -> slab translation */ +#define UMA_HASH_REMOVE(h, s) \ + LIST_REMOVE((uma_hash_slab_t)(s), uhs_hlink) -SLIST_HEAD(slabhead, uma_slab); +LIST_HEAD(slabhashhead, uma_hash_slab); struct uma_hash { - struct slabhead *uh_slab_hash; /* Hash table for slabs */ + struct slabhashhead *uh_slab_hash; /* Hash table for slabs */ u_int uh_hashsize; /* Current size of the hash table */ u_int uh_hashmask; /* Mask used during hashing */ }; @@ -202,13 +197,15 @@ struct uma_cache { typedef struct uma_cache * uma_cache_t; +LIST_HEAD(slabhead, uma_slab); + /* * Per-domain memory list. Embedded in the kegs. */ struct uma_domain { - LIST_HEAD(,uma_slab) ud_part_slab; /* partially allocated slabs */ - LIST_HEAD(,uma_slab) ud_free_slab; /* empty slab list */ - LIST_HEAD(,uma_slab) ud_full_slab; /* full slabs */ + struct slabhead ud_part_slab; /* partially allocated slabs */ + struct slabhead ud_free_slab; /* completely unallocated slabs */ + struct slabhead ud_full_slab; /* fully allocated slabs */ }; typedef struct uma_domain * uma_domain_t; @@ -271,8 +268,6 @@ BITSET_DEFINE(noslabbits, 0); */ struct uma_slab { LIST_ENTRY(uma_slab) us_link; /* slabs in zone */ - SLIST_ENTRY(uma_slab) us_hlink; /* Link for hash table */ - uint8_t *us_data; /* First item */ uint16_t us_freecount; /* How many are free? */ uint8_t us_flags; /* Page flags see uma.h */ uint8_t us_domain; /* Backing NUMA domain. */ @@ -281,7 +276,6 @@ struct uma_slab { #endif struct noslabbits us_free; /* Free bitmask. */ }; - #if MAXMEMDOM >= 255 #error "Slab domain type insufficient" #endif @@ -293,6 +287,47 @@ size_t slab_sizeof(int nitems); size_t slab_space(int nitems); int slab_ipers(size_t size, int align); +/* + * Slab structure with a full sized bitset and hash link for both + * HASH and OFFPAGE zones. + */ +struct uma_hash_slab { + struct uma_slab uhs_slab; /* Must be first. */ + struct slabbits uhs_bits; /* Must be second. */ + LIST_ENTRY(uma_hash_slab) uhs_hlink; /* Link for hash table */ + uint8_t *uhs_data; /* First item */ +}; + +typedef struct uma_hash_slab * uma_hash_slab_t; + +static inline void * +slab_data(uma_slab_t slab, uma_keg_t keg) +{ + + if ((keg->uk_flags & UMA_ZONE_OFFPAGE) == 0) + return ((void *)((uintptr_t)slab - keg->uk_pgoff)); + else + return (((uma_hash_slab_t)slab)->uhs_data); +} + +static inline void * +slab_item(uma_slab_t slab, uma_keg_t keg, int index) +{ + uintptr_t data; + + data = (uintptr_t)slab_data(slab, keg); + return ((void *)(data + keg->uk_rsize * index)); +} + +static inline int +slab_item_index(uma_slab_t slab, uma_keg_t keg, void *item) +{ + uintptr_t data; + + data = (uintptr_t)slab_data(slab, keg); + return (((uintptr_t)item - data) / keg->uk_rsize); +} + TAILQ_HEAD(uma_bucketlist, uma_bucket); struct uma_zone_domain { @@ -444,14 +479,14 @@ static __inline uma_slab_t hash_sfind(struct uma_hash static __inline uma_slab_t hash_sfind(struct uma_hash *hash, uint8_t *data) { - uma_slab_t slab; + uma_hash_slab_t slab; u_int hval; hval = UMA_HASH(hash, data); - SLIST_FOREACH(slab, &hash->uh_slab_hash[hval], us_hlink) { - if ((uint8_t *)slab->us_data == data) - return (slab); + LIST_FOREACH(slab, &hash->uh_slab_hash[hval], uhs_hlink) { + if ((uint8_t *)slab->uhs_data == data) + return (&slab->uhs_slab); } return (NULL); }