Rewrite ethread library

Large parts of the ethread library have been rewritten. The
ethread library is an Erlang runtime system internal, portable
thread library used by the runtime system itself.

Most notable improvement is a reader optimized rwlock
implementation which dramatically improve the performance of
read-lock/read-unlock operations on multi processor systems by
avoiding ping-ponging of the rwlock cache lines. The reader
optimized rwlock implementation is used by miscellaneous
rwlocks in the runtime system that are known to be read-locked
frequently, and can be enabled on ETS tables by passing the
`{read_concurrency, true}' option upon table creation. See the
documentation of `ets:new/2' for more information.

The ethread library can now also use the libatomic_ops library
for atomic memory accesses. This makes it possible for the
Erlang runtime system to utilize optimized atomic operations
on more platforms than before. Use the
`--with-libatomic_ops=PATH' configure command line argument
when specifying where the libatomic_ops installation is
located. The libatomic_ops library can be downloaded from:
http://www.hpl.hp.com/research/linux/atomic_ops/

The changed API of the ethread library has also caused
modifications in the Erlang runtime system. Preparations for
the to come "delayed deallocation" feature has also been done
since it depends on the ethread library.

Note: When building for x86, the ethread library will now use
instructions that first appeared on the pentium 4 processor. If
you want the runtime system to be compatible with older
processors (back to 486) you need to pass the
`--enable-ethread-pre-pentium4-compatibility' configure command
line argument when configuring the system.

1 files changed, 234 insertions, 85 deletions

diff --git a/erts/emulator/beam/erl_db.c b/erts/emulator/beam/erl_db.c
index cbdaa459de..b0369a402b 100644
--- a/erts/emulator/beam/erl_db.c
+++ b/erts/emulator/beam/erl_db.c
@@ -78,11 +78,19 @@ enum DbIterSafety {
 ** The main meta table, containing all ets tables.
 */
 #ifdef ERTS_SMP
-#  define META_MAIN_TAB_LOCK_CNT 16
-static union {
-    erts_smp_spinlock_t lck;
-    byte _cache_line_alignment[64];
-}meta_main_tab_locks[META_MAIN_TAB_LOCK_CNT];
+
+#define ERTS_META_MAIN_TAB_LOCK_TAB_BITS 8
+#define ERTS_META_MAIN_TAB_LOCK_TAB_SIZE (1 << ERTS_META_MAIN_TAB_LOCK_TAB_BITS)
+#define ERTS_META_MAIN_TAB_LOCK_TAB_MASK (ERTS_META_MAIN_TAB_LOCK_TAB_SIZE - 1)
+
+typedef union {
+    erts_smp_rwmtx_t rwmtx;
+    byte cache_line_align__[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(
+				sizeof(erts_smp_rwmtx_t))];
+} erts_meta_main_tab_lock_t;
+
+static erts_meta_main_tab_lock_t *meta_main_tab_locks;
+
 #endif
 static struct {
     union {
@@ -104,17 +112,13 @@ static struct {
 #define MARK_SLOT_DEAD(i) (meta_main_tab[(i)].u.next_free |= 2)
 #define GET_ANY_SLOT_TAB(i) ((DbTable*)(meta_main_tab[(i)].u.next_free & ~(1|2))) /* dead or alive */
 
-static ERTS_INLINE void meta_main_tab_lock(unsigned slot)
+static ERTS_INLINE erts_smp_rwmtx_t *
+get_meta_main_tab_lock(unsigned slot)
 {
 #ifdef ERTS_SMP
-    erts_smp_spin_lock(&meta_main_tab_locks[slot % META_MAIN_TAB_LOCK_CNT].lck);
-#endif
-}
-
-static ERTS_INLINE void meta_main_tab_unlock(unsigned slot)
-{
-#ifdef ERTS_SMP
-    erts_smp_spin_unlock(&meta_main_tab_locks[slot % META_MAIN_TAB_LOCK_CNT].lck);
+    return &meta_main_tab_locks[slot & ERTS_META_MAIN_TAB_LOCK_TAB_MASK].rwmtx;
+#else
+    return NULL;
 #endif
 }
 
@@ -166,7 +170,8 @@ struct meta_name_tab_entry* meta_name_tab_bucket(Eterm name,
 typedef enum {
     LCK_READ=1,     /* read only access */
     LCK_WRITE=2,    /* exclusive table write access */
-    LCK_WRITE_REC=3 /* record write access */
+    LCK_WRITE_REC=3, /* record write access */
+    LCK_NONE=4
 } db_lock_kind_t;
 
 extern DbTableMethod db_hash;
@@ -214,17 +219,17 @@ Export ets_select_continue_exp;
  */
 static Export ets_delete_continue_exp;
 
-static ERTS_INLINE DbTable* db_ref(DbTable* tb)
+static ERTS_INLINE DbTable* db_ref(DbTable* tb, db_lock_kind_t kind)
 {
-    if (tb != NULL) {
+    if (tb != NULL && kind != LCK_READ) {
 	erts_refc_inc(&tb->common.ref, 2);
     }
     return tb;
 }
 
-static ERTS_INLINE DbTable* db_unref(DbTable* tb)
+static ERTS_INLINE DbTable* db_unref(DbTable* tb, db_lock_kind_t kind)
 {
-    if (!erts_refc_dectest(&tb->common.ref, 0)) {
+    if (kind != LCK_READ && !erts_refc_dectest(&tb->common.ref, 0)) {
 #ifdef HARDDEBUG
 	if (erts_smp_atomic_read(&tb->common.memory_size) != sizeof(DbTable)) {
 	    erts_fprintf(stderr, "ets: db_unref memory remain=%ld fix=%x\n",
@@ -256,12 +261,19 @@ static ERTS_INLINE DbTable* db_unref(DbTable* tb)
     return tb;
 }
 
-static ERTS_INLINE void db_init_lock(DbTable* tb, char *rwname, char* fixname)
+static ERTS_INLINE void db_init_lock(DbTable* tb, int use_frequent_read_lock,
+				     char *rwname, char* fixname)
 {
+#ifdef ERTS_SMP
+    erts_smp_rwmtx_opt_t rwmtx_opt = ERTS_SMP_THR_OPTS_DEFAULT_INITER;
+    if (use_frequent_read_lock)
+	rwmtx_opt.type = ERTS_SMP_RWMTX_TYPE_FREQUENT_READ;
+#endif
     erts_refc_init(&tb->common.ref, 1);
     erts_refc_init(&tb->common.fixref, 0);
 #ifdef ERTS_SMP
-    erts_smp_rwmtx_init_x(&tb->common.rwlock, rwname, tb->common.the_name);
+    erts_smp_rwmtx_init_opt_x(&tb->common.rwlock, &rwmtx_opt,
+			      rwname, tb->common.the_name);
     erts_smp_mtx_init_x(&tb->common.fixlock, fixname, tb->common.the_name);
     tb->common.is_thread_safe = !(tb->common.status & DB_FINE_LOCKED);
 #endif
@@ -297,7 +309,7 @@ static ERTS_INLINE void db_lock_take_over_ref(DbTable* tb, db_lock_kind_t kind)
 
 static ERTS_INLINE void db_lock(DbTable* tb, db_lock_kind_t kind)
 {
-    (void) db_ref(tb);
+    (void) db_ref(tb, kind);
 #ifdef ERTS_SMP
     db_lock_take_over_ref(tb, kind);
 #endif
@@ -331,7 +343,7 @@ static ERTS_INLINE void db_unlock(DbTable* tb, db_lock_kind_t kind)
 	}
     }
 #endif
-    (void) db_unref(tb); /* May delete table... */
+    (void) db_unref(tb, kind); /* May delete table... */
 }
 
 
@@ -349,32 +361,49 @@ static ERTS_INLINE void db_meta_unlock(DbTable* tb, db_lock_kind_t kind)
 }
 
 static ERTS_INLINE
-DbTable* db_get_table(Process *p,
-		      Eterm id,
-		      int what,
-		      db_lock_kind_t kind)
+DbTable* db_get_table_aux(Process *p,
+			  Eterm id,
+			  int what,
+			  db_lock_kind_t kind,
+			  int meta_already_locked)
 {
     DbTable *tb = NULL;
+    erts_smp_rwmtx_t *mtl = NULL;
 
     if (is_small(id)) {
 	Uint slot = unsigned_val(id) & meta_main_tab_slot_mask;
-	meta_main_tab_lock(slot);
+	if (!meta_already_locked) {
+	    mtl = get_meta_main_tab_lock(slot);
+	    erts_smp_rwmtx_rlock(mtl);
+	}
+#if defined(ERTS_SMP) && defined(ERTS_ENABLE_LOCK_CHECK)
+	else {
+	    erts_smp_rwmtx_t *test_mtl = get_meta_main_tab_lock(slot);
+	    ERTS_SMP_LC_ASSERT(erts_lc_rwmtx_is_rlocked(test_mtl)
+			       || erts_lc_rwmtx_is_rwlocked(test_mtl));
+	}
+#endif
 	if (slot < db_max_tabs && IS_SLOT_ALIVE(slot)) {
 	    /* SMP: inc to prevent race, between unlock of meta_main_tab_lock
 	     * and the table locking outside the meta_main_tab_lock
 	     */
-	    tb = db_ref(meta_main_tab[slot].u.tb);
+	    tb = db_ref(meta_main_tab[slot].u.tb, kind);
 	}
-	meta_main_tab_unlock(slot);
     }
     else if (is_atom(id)) {
-	erts_smp_rwmtx_t* rwlock;
-	struct meta_name_tab_entry* bucket = meta_name_tab_bucket(id,&rwlock);
-	erts_smp_rwmtx_rlock(rwlock);
+	struct meta_name_tab_entry* bucket = meta_name_tab_bucket(id,&mtl);
+	if (!meta_already_locked)
+	    erts_smp_rwmtx_rlock(mtl);
+	else{
+	    ERTS_SMP_LC_ASSERT(erts_lc_rwmtx_is_rlocked(mtl)
+			       || erts_lc_rwmtx_is_rwlocked(mtl));
+	    mtl = NULL;
+	}
+
 	if (bucket->pu.tb != NULL) {
 	    if (is_atom(bucket->u.name_atom)) { /* single */
 		if (bucket->u.name_atom == id) {
-		    tb = db_ref(bucket->pu.tb);
+		    tb = db_ref(bucket->pu.tb, kind);
 		}
 	    }
 	    else { /* multi */
@@ -382,23 +411,33 @@ DbTable* db_get_table(Process *p,
 		Uint i;
 		for (i=0; i<cnt; i++) {
 		    if (bucket->pu.mvec[i].u.name_atom == id) {
-			tb = db_ref(bucket->pu.mvec[i].pu.tb);
+			tb = db_ref(bucket->pu.mvec[i].pu.tb, kind);
 			break;
 		    }
 		}
 	    }
 	}
-	erts_smp_rwmtx_runlock(rwlock);
     }
     if (tb) {
 	db_lock_take_over_ref(tb, kind);
-	if (tb->common.id == id && ((tb->common.status & what) != 0 || 
-				    p->id == tb->common.owner)) {
-	    return tb;
+	if (tb->common.id != id
+	    || ((tb->common.status & what) == 0 && p->id != tb->common.owner)) {
+	    db_unlock(tb, kind);
+	    tb = NULL;
 	}
-	db_unlock(tb, kind);
     }
-    return NULL;
+    if (mtl)
+	erts_smp_rwmtx_runlock(mtl);
+    return tb;
+}
+
+static ERTS_INLINE
+DbTable* db_get_table(Process *p,
+		      Eterm id,
+		      int what,
+		      db_lock_kind_t kind)
+{
+    return db_get_table_aux(p, id, what, kind, 0);
 }
 
 /* Requires meta_main_tab_locks[slot] locked.
@@ -413,15 +452,15 @@ static ERTS_INLINE void free_slot(int slot)
     erts_smp_spin_unlock(&meta_main_tab_main_lock);
 }
 
-static int insert_named_tab(Eterm name_atom, DbTable* tb)
+static int insert_named_tab(Eterm name_atom, DbTable* tb, int have_lock)
 {
     int ret = 0;
     erts_smp_rwmtx_t* rwlock;
     struct meta_name_tab_entry* new_entry;
     struct meta_name_tab_entry* bucket = meta_name_tab_bucket(name_atom,
 							      &rwlock);
-
-    erts_smp_rwmtx_rwlock(rwlock);
+    if (!have_lock)
+	erts_smp_rwmtx_rwlock(rwlock);
 
     if (bucket->pu.tb == NULL) { /* empty */
 	new_entry = bucket;
@@ -468,17 +507,32 @@ static int insert_named_tab(Eterm name_atom, DbTable* tb)
     ret = 1; /* Ok */
 
 done:
-    erts_smp_rwmtx_rwunlock(rwlock);
+    if (!have_lock)
+	erts_smp_rwmtx_rwunlock(rwlock);
     return ret;
 }
 
-static int remove_named_tab(Eterm name_atom)
+static int remove_named_tab(DbTable *tb, int have_lock)
 {
     int ret = 0;
     erts_smp_rwmtx_t* rwlock;
+    Eterm name_atom = tb->common.id;
     struct meta_name_tab_entry* bucket = meta_name_tab_bucket(name_atom,
 							      &rwlock);
-    erts_smp_rwmtx_rwlock(rwlock);
+#ifdef ERTS_SMP
+    if (!have_lock && erts_smp_rwmtx_tryrwlock(rwlock) == EBUSY) {
+	/*
+	 * We keep our increased refc over this op in order to
+	 * prevent the table from disapearing.
+	 */
+	erts_smp_rwmtx_rwunlock(&tb->common.rwlock);
+	erts_smp_rwmtx_rwlock(rwlock);
+	erts_smp_rwmtx_rwlock(&tb->common.rwlock);
+    }
+#endif
+
+    ERTS_SMP_LC_ASSERT(erts_lc_rwmtx_is_rwlocked(rwlock));
+
     if (bucket->pu.tb == NULL) {
 	goto done;
     }
@@ -529,7 +583,8 @@ static int remove_named_tab(Eterm name_atom)
     ret = 1; /* Ok */
 
 done:
-    erts_smp_rwmtx_rwunlock(rwlock);
+    if (!have_lock)
+	erts_smp_rwmtx_rwunlock(rwlock);
     return ret;
 }
 
@@ -1133,6 +1188,7 @@ BIF_RETTYPE ets_rename_2(BIF_ALIST_2)
 {
     DbTable* tb;
     Eterm ret;
+    erts_smp_rwmtx_t *lck1, *lck2;
 
 #ifdef HARDDEBUG
     erts_fprintf(stderr,
@@ -1141,34 +1197,65 @@ BIF_RETTYPE ets_rename_2(BIF_ALIST_2)
 		BIF_P->initial[0], BIF_P->initial[1], BIF_P->initial[2]);
 #endif
 
-    if ((tb = db_get_table(BIF_P, BIF_ARG_1, DB_WRITE, LCK_WRITE)) == NULL) {
+
+    if (is_not_atom(BIF_ARG_2)) {
 	BIF_ERROR(BIF_P, BADARG);
     }
 
-    if (is_not_atom(BIF_ARG_2)) {
-	goto badarg;
+    (void) meta_name_tab_bucket(BIF_ARG_2, &lck1);
+
+    if (is_small(BIF_ARG_1)) {
+	Uint slot = unsigned_val(BIF_ARG_1) & meta_main_tab_slot_mask;
+	lck2 = get_meta_main_tab_lock(slot);
+    }
+    else if (is_atom(BIF_ARG_1)) {
+	(void) meta_name_tab_bucket(BIF_ARG_1, &lck2);
+	if (lck1 == lck2)
+	    lck2 = NULL;
+	else if (lck1 > lck2) {
+	    erts_smp_rwmtx_t *tmp = lck1;
+	    lck1 = lck2;
+	    lck2 = tmp;
+	}
+    }
+    else {
+	BIF_ERROR(BIF_P, BADARG);
     }
 
+    erts_smp_rwmtx_rwlock(lck1);
+    if (lck2)
+	erts_smp_rwmtx_rwlock(lck2);
+
+    tb = db_get_table_aux(BIF_P, BIF_ARG_1, DB_WRITE, LCK_WRITE, 1);
+    if (!tb)
+	goto badarg;
+
     if (is_not_atom(tb->common.id)) { /* Not a named table */
 	tb->common.the_name = BIF_ARG_2;
 	goto done;
     }
 
-    if (!insert_named_tab(BIF_ARG_2,tb)) {
+    if (!insert_named_tab(BIF_ARG_2, tb, 1))
 	goto badarg;
-    }
-    if (!remove_named_tab(tb->common.id)) {
+
+    if (!remove_named_tab(tb, 1))
 	erl_exit(1,"Could not find named tab %s", tb->common.id);
-    }
 
     tb->common.id = tb->common.the_name = BIF_ARG_2;
 
  done:
     ret = tb->common.id;
     db_unlock(tb, LCK_WRITE);
+    erts_smp_rwmtx_rwunlock(lck1);
+    if (lck2)
+	erts_smp_rwmtx_rwunlock(lck2);
     BIF_RET(ret);
  badarg:
-    db_unlock(tb, LCK_WRITE);
+    if (tb)
+	db_unlock(tb, LCK_WRITE);
+    erts_smp_rwmtx_rwunlock(lck1);
+    if (lck2)
+	erts_smp_rwmtx_rwunlock(lck2);
     BIF_ERROR(BIF_P, BADARG);    
 }
 
@@ -1189,10 +1276,11 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2)
     UWord heir_data;
     Uint32 status;
     Sint keypos;
-    int is_named, is_fine_locked;
+    int is_named, is_fine_locked, frequent_read;
     int cret;
     DeclareTmpHeap(meta_tuple,3,BIF_P);
     DbTableMethod* meth;
+    erts_smp_rwmtx_t *mmtl;
 
     if (is_not_atom(BIF_ARG_1)) {
 	BIF_ERROR(BIF_P, BADARG);
@@ -1205,6 +1293,7 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2)
     keypos = 1;
     is_named = 0;
     is_fine_locked = 0;
+    frequent_read = 0;
     heir = am_none;
     heir_data = (UWord) am_undefined;
 
@@ -1238,6 +1327,13 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2)
 			is_fine_locked = 0;
 		    } else break;
 		}
+		else if (tp[1] == am_read_concurrency) {
+		    if (tp[2] == am_true) {
+			frequent_read = 1;
+		    } else if (tp[2] == am_false) {
+			frequent_read = 0;
+		    } else break;
+		}
 		else if (tp[1] == am_heir && tp[2] == am_none) {
 		    heir = am_none;
 		    heir_data = am_undefined;
@@ -1286,6 +1382,11 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2)
 	BIF_ERROR(BIF_P, BADARG);
     }
 
+#ifdef ERTS_SMP
+    if (frequent_read && !(status & DB_PRIVATE))
+	status |= DB_FREQ_READ;
+#endif
+
     /* we create table outside any table lock
      * and take the unusal cost of destroy table if it
      * fails to find a slot 
@@ -1308,7 +1409,8 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2)
     tb->common.type = status & ERTS_ETS_TABLE_TYPES;
     /* Note, 'type' is *read only* from now on... */
 #endif
-    db_init_lock(tb, "db_tab", "db_tab_fix");
+    db_init_lock(tb, status & (DB_FINE_LOCKED|DB_FREQ_READ),
+		 "db_tab", "db_tab_fix");
     tb->common.keypos = keypos;
     tb->common.owner = BIF_P->id;
     set_heir(BIF_P, tb, heir, heir_data);
@@ -1351,15 +1453,17 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2)
     tb->common.id = ret;
     tb->common.slot = slot;           /* store slot for erase */
 
-    meta_main_tab_lock(slot);
+    mmtl = get_meta_main_tab_lock(slot);
+    erts_smp_rwmtx_rwlock(mmtl);
     meta_main_tab[slot].u.tb = tb;
     ASSERT(IS_SLOT_ALIVE(slot));
-    meta_main_tab_unlock(slot);
+    erts_smp_rwmtx_rwunlock(mmtl);
 
-    if (is_named && !insert_named_tab(BIF_ARG_1, tb)) {
-	meta_main_tab_lock(slot);
+    if (is_named && !insert_named_tab(BIF_ARG_1, tb, 0)) {
+	mmtl = get_meta_main_tab_lock(slot);
+	erts_smp_rwmtx_rwlock(mmtl);
 	free_slot(slot);
-	meta_main_tab_unlock(slot);
+	erts_smp_rwmtx_rwunlock(mmtl);
 
 	db_lock_take_over_ref(tb,LCK_WRITE);
 	free_heir_data(tb);
@@ -1499,6 +1603,7 @@ BIF_RETTYPE ets_delete_1(BIF_ALIST_1)
 {
     int trap;
     DbTable* tb;
+    erts_smp_rwmtx_t *mmtl;
 
 #ifdef HARDDEBUG
     erts_fprintf(stderr,
@@ -1520,13 +1625,23 @@ BIF_RETTYPE ets_delete_1(BIF_ALIST_1)
     tb->common.status &= ~(DB_PROTECTED|DB_PUBLIC|DB_PRIVATE);
     tb->common.status |= DB_DELETE;
 
-    meta_main_tab_lock(tb->common.slot);
+    mmtl = get_meta_main_tab_lock(tb->common.slot);
+#ifdef ERTS_SMP
+    if (erts_smp_rwmtx_tryrwlock(mmtl) == EBUSY) {
+	/*
+	 * We keep our increased refc over this op in order to
+	 * prevent the table from disapearing.
+	 */
+	erts_smp_rwmtx_rwunlock(&tb->common.rwlock);
+	erts_smp_rwmtx_rwlock(mmtl);
+	erts_smp_rwmtx_rwlock(&tb->common.rwlock);
+    }
+#endif
     /* We must keep the slot, to be found by db_proc_dead() if process dies */
     MARK_SLOT_DEAD(tb->common.slot);
-    meta_main_tab_unlock(tb->common.slot);
-    if (is_atom(tb->common.id)) {
-	remove_named_tab(tb->common.id);
-    }
+    erts_smp_rwmtx_rwunlock(mmtl);
+    if (is_atom(tb->common.id))
+	remove_named_tab(tb, 0);
     
     if (tb->common.owner != BIF_P->id) {
 	DeclareTmpHeap(meta_tuple,3,BIF_P);
@@ -1919,14 +2034,15 @@ BIF_RETTYPE ets_all_0(BIF_ALIST_0)
     previous = NIL;
     j = 0;
     for(i = 0; (i < t_max_tabs && j < t_tabs_cnt); i++) {
-	meta_main_tab_lock(i);
+	erts_smp_rwmtx_t *mmtl = get_meta_main_tab_lock(i);
+	erts_smp_rwmtx_rlock(mmtl);
 	if (IS_SLOT_ALIVE(i)) {
 	    j++;
 	    tb = meta_main_tab[i].u.tb;
 	    previous = CONS(hp, tb->common.id, previous);
 	    hp += 2;
 	}
-	meta_main_tab_unlock(i);
+	erts_smp_rwmtx_runlock(mmtl);
     }
     HRelease(BIF_P, hendp, hp);
     BIF_RET(previous);
@@ -2630,12 +2746,30 @@ void init_db(void)
     size_t size;
 
 #ifdef ERTS_SMP
-    for (i=0; i<META_MAIN_TAB_LOCK_CNT; i++) {
-	erts_smp_spinlock_init_x(&meta_main_tab_locks[i].lck, "meta_main_tab_slot", make_small(i));
+    erts_smp_rwmtx_opt_t rwmtx_opt = ERTS_SMP_THR_OPTS_DEFAULT_INITER;
+    rwmtx_opt.type = ERTS_SMP_RWMTX_TYPE_FREQUENT_READ;
+    rwmtx_opt.lived = ERTS_SMP_RWMTX_LONG_LIVED;
+
+    meta_main_tab_locks = erts_alloc(ERTS_ALC_T_DB_TABLES,
+				     (sizeof(erts_meta_main_tab_lock_t)
+				      * (ERTS_META_MAIN_TAB_LOCK_TAB_SIZE+1)));
+
+    if ((((Uint) meta_main_tab_locks) & ERTS_CACHE_LINE_MASK) != 0)
+	meta_main_tab_locks = ((erts_meta_main_tab_lock_t *)
+			       ((((Uint) meta_main_tab_locks)
+				 & ~ERTS_CACHE_LINE_MASK)
+				+ ERTS_CACHE_LINE_SIZE));
+
+    ASSERT((((Uint) meta_main_tab_locks) & ERTS_CACHE_LINE_MASK) == 0);
+
+    for (i = 0; i < ERTS_META_MAIN_TAB_LOCK_TAB_SIZE; i++) {
+	erts_smp_rwmtx_init_opt_x(&meta_main_tab_locks[i].rwmtx, &rwmtx_opt,
+				  "meta_main_tab_slot", make_small(i));
     }
     erts_smp_spinlock_init(&meta_main_tab_main_lock, "meta_main_tab_main");
     for (i=0; i<META_NAME_TAB_LOCK_CNT; i++) {
-	erts_smp_rwmtx_init_x(&meta_name_tab_rwlocks[i].lck, "meta_name_tab", make_small(i));
+	erts_smp_rwmtx_init_opt_x(&meta_name_tab_rwlocks[i].lck, &rwmtx_opt,
+				  "meta_name_tab", make_small(i));
     }
 #endif
 
@@ -2895,12 +3029,12 @@ retry:
 				to_pid, to_locks,
 				ERTS_P2P_FLG_TRY_LOCK);
     if (to_proc == ERTS_PROC_LOCK_BUSY) {
-	db_ref(tb); /* while unlocked */
+	db_ref(tb, LCK_NONE); /* while unlocked */
 	db_unlock(tb,LCK_WRITE);    
 	to_proc = erts_pid2proc(p, ERTS_PROC_LOCK_MAIN,
 				to_pid, to_locks);    
 	db_lock(tb,LCK_WRITE);
-	tb = db_unref(tb);
+	tb = db_unref(tb, LCK_NONE);
 	ASSERT(tb != NULL);
     
 	if (tb->common.owner != p->id) {
@@ -3008,12 +3142,13 @@ erts_db_process_exiting(Process *c_p, ErtsProcLocks c_p_locks)
 	    while (state->slots.ix < state->slots.size) {
 		DbTable *tb = NULL;
 		Sint ix = unsigned_val(state->slots.arr[state->slots.ix]);
-		meta_main_tab_lock(ix);
+		erts_smp_rwmtx_t *mmtl = get_meta_main_tab_lock(ix);
+		erts_smp_rwmtx_rlock(mmtl);
 		if (!IS_SLOT_FREE(ix)) {
-		    tb = db_ref(GET_ANY_SLOT_TAB(ix));
+		    tb = db_ref(GET_ANY_SLOT_TAB(ix), LCK_WRITE);
 		    ASSERT(tb);
 		}
-		meta_main_tab_unlock(ix);
+		erts_smp_rwmtx_runlock(mmtl);
 		if (tb) {
 		    int do_yield;
 		    db_lock_take_over_ref(tb, LCK_WRITE);
@@ -3045,7 +3180,7 @@ erts_db_process_exiting(Process *c_p, ErtsProcLocks c_p_locks)
 			    tb->common.status |= DB_DELETE;
 
 			    if (is_atom(tb->common.id))
-				remove_named_tab(tb->common.id);
+				remove_named_tab(tb, 0);
 
 			    free_heir_data(tb);
 			    free_fixations_locked(tb);
@@ -3095,12 +3230,13 @@ erts_db_process_exiting(Process *c_p, ErtsProcLocks c_p_locks)
 	    while (state->slots.ix < state->slots.size) {
 		DbTable *tb = NULL;
 		Sint ix = unsigned_val(state->slots.arr[state->slots.ix]);
-		meta_main_tab_lock(ix);
+		erts_smp_rwmtx_t *mmtl = get_meta_main_tab_lock(ix);
+		erts_smp_rwmtx_rlock(mmtl);
 		if (IS_SLOT_ALIVE(ix)) {
-		    tb = db_ref(meta_main_tab[ix].u.tb);
+		    tb = db_ref(meta_main_tab[ix].u.tb, LCK_WRITE_REC);
 		    ASSERT(tb);
 		}
-		meta_main_tab_unlock(ix);
+		erts_smp_rwmtx_runlock(mmtl);
 		if (tb) {
 		    int reds;
 		    DbFixation** pp;
@@ -3274,6 +3410,7 @@ unlocked:
 #ifdef ERTS_SMP
 	if (*kind_p == LCK_READ && tb->common.is_thread_safe) {
 	    /* Must have write lock while purging pseudo-deleted (OTP-8166) */
+	    db_ref(tb, LCK_WRITE); /* LCK_WRITE need it, but not LCK_READ */
 	    erts_smp_rwmtx_runlock(&tb->common.rwlock);
 	    erts_smp_rwmtx_rwlock(&tb->common.rwlock);
 	    *kind_p = LCK_WRITE;
@@ -3386,6 +3523,7 @@ static int free_table_cont(Process *p,
 			   int clean_meta_tab)
 {
     Eterm result;
+    erts_smp_rwmtx_t *mmtl;
 
 #ifdef HARDDEBUG
     if (!first) {
@@ -3411,9 +3549,20 @@ static int free_table_cont(Process *p,
 		     tb->common.id);
 #endif
 	/* Completely done - we will not get called again. */
-	meta_main_tab_lock(tb->common.slot);
+	mmtl = get_meta_main_tab_lock(tb->common.slot);
+#ifdef ERTS_SMP
+	if (erts_smp_rwmtx_tryrwlock(mmtl) == EBUSY) {
+	    /*
+	     * We keep our increased refc over this op in order to
+	     * prevent the table from disapearing.
+	     */
+	    erts_smp_rwmtx_rwunlock(&tb->common.rwlock);
+	    erts_smp_rwmtx_rwlock(mmtl);
+	    erts_smp_rwmtx_rwlock(&tb->common.rwlock);
+	}
+#endif
 	free_slot(tb->common.slot);
-	meta_main_tab_unlock(tb->common.slot);
+	erts_smp_rwmtx_rwunlock(mmtl);
 
 	if (clean_meta_tab) {
 	    db_meta_lock(meta_pid_to_tab, LCK_WRITE_REC);
@@ -3421,7 +3570,7 @@ static int free_table_cont(Process *p,
 				make_small(tb->common.slot));
 	    db_meta_unlock(meta_pid_to_tab, LCK_WRITE_REC);
 	}
-	db_unref(tb);
+	db_unref(tb, LCK_NONE);
 	BUMP_REDS(p, 100);
 	return 0;
     }