Add runtime option to enable/disable lcnt stats

Add erts_debug:lock_counters({copy_save, bool()}). This option
enables or disables statistics saving for destroyed processes and
ets-tables. Enabling this might consume a lot of memory.

Add id-numbering for lock classes which is otherwise undefined.

7 files changed, 110 insertions, 82 deletions

diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c
index 5ff1f794df..a34d400ed8 100644
--- a/erts/emulator/beam/erl_bif_info.c
+++ b/erts/emulator/beam/erl_bif_info.c
@@ -3704,12 +3704,11 @@ static Eterm lcnt_build_lock_term(Eterm **hpp, Uint *szp, erts_lcnt_lock_t *lock
     ASSERT(ltype);
     
     type  = am_atom_put(ltype, strlen(ltype));           
-
     name  = am_atom_put(lock->name, strlen(lock->name)); 
 
     if (lock->flag & ERTS_LCNT_LT_ALLOC) {
 	/* use allocator types names as id's for allocator locks */
-	ltype = ERTS_ALC_A2AD(signed_val(lock->id));
+	ltype = (char *) ERTS_ALC_A2AD(signed_val(lock->id));
 	id    = am_atom_put(ltype, strlen(ltype));
     } else if (lock->flag & ERTS_LCNT_LT_PROCLOCK) {
 	/* use registered names as id's for process locks if available */
@@ -3778,17 +3777,28 @@ BIF_RETTYPE erts_debug_lock_counters_1(BIF_ALIST_1)
 {
 #ifdef ERTS_ENABLE_LOCK_COUNT
     Eterm res = NIL;
-    erts_smp_proc_unlock(BIF_P, ERTS_PROC_LOCK_MAIN);
-    erts_smp_block_system(0);
+#endif
+
+
+    if (BIF_ARG_1 == am_enabled) {
+#ifdef ERTS_ENABLE_LOCK_COUNT
+	BIF_RET(am_true);
+#else
+	BIF_RET(am_false);
+#endif
+    }
+#ifdef ERTS_ENABLE_LOCK_COUNT
 
-    if (BIF_ARG_1 == am_info) {
+    else if (BIF_ARG_1 == am_info) {
 	erts_lcnt_data_t *data; 
 	Uint hsize = 0;
 	Uint *szp;
     	Eterm* hp;
 
-	erts_lcnt_set_rt_opt(ERTS_LCNT_OPT_SUSPEND);
+	erts_smp_proc_unlock(BIF_P, ERTS_PROC_LOCK_MAIN);
+	erts_smp_block_system(0);
 
+	erts_lcnt_set_rt_opt(ERTS_LCNT_OPT_SUSPEND);
 	data = erts_lcnt_get_data();
 
 	/* calculate size */
@@ -3803,29 +3813,65 @@ BIF_RETTYPE erts_debug_lock_counters_1(BIF_ALIST_1)
 	res = lcnt_build_result_term(&hp, NULL, data, res);
 	
 	erts_lcnt_clear_rt_opt(ERTS_LCNT_OPT_SUSPEND);
+
+	erts_smp_release_system();
+	erts_smp_proc_lock(BIF_P, ERTS_PROC_LOCK_MAIN);
 	
-	goto done;
+	BIF_RET(res);
     } else if (BIF_ARG_1 == am_clear) {
+	erts_smp_proc_unlock(BIF_P, ERTS_PROC_LOCK_MAIN);
+	erts_smp_block_system(0);
+
 	erts_lcnt_clear_counters();
-	res = am_ok;
-	goto done;
+
+	erts_smp_release_system();
+	erts_smp_proc_lock(BIF_P, ERTS_PROC_LOCK_MAIN);
+
+	BIF_RET(am_ok);
     } else if (is_tuple(BIF_ARG_1)) {
-	Uint prev = 0;
 	Eterm* tp = tuple_val(BIF_ARG_1);
+
 	switch (arityval(tp[0])) {
 	    case 2:
-	    	if (ERTS_IS_ATOM_STR("process_locks", tp[1])) {
+		if (ERTS_IS_ATOM_STR("copy_save", tp[1])) {
+		    erts_smp_proc_unlock(BIF_P, ERTS_PROC_LOCK_MAIN);
+		    erts_smp_block_system(0);
 		    if (tp[2] == am_true) {
-			prev = erts_lcnt_set_rt_opt(ERTS_LCNT_OPT_PROCLOCK);
-			if (prev) res = am_true;
-			else res = am_false;
-			goto done;
+
+			res = erts_lcnt_set_rt_opt(ERTS_LCNT_OPT_COPYSAVE) ? am_true : am_false;
+
 		    } else if (tp[2] == am_false) {
-			prev = erts_lcnt_clear_rt_opt(ERTS_LCNT_OPT_PROCLOCK);
-			if (prev) res = am_true;
-			else res = am_false;
-			goto done;
+
+			res = erts_lcnt_clear_rt_opt(ERTS_LCNT_OPT_COPYSAVE) ? am_true : am_false;
+
+		    } else {
+			erts_smp_release_system();
+			erts_smp_proc_lock(BIF_P, ERTS_PROC_LOCK_MAIN);
+			BIF_ERROR(BIF_P, BADARG);
 		    }
+		    erts_smp_release_system();
+		    erts_smp_proc_lock(BIF_P, ERTS_PROC_LOCK_MAIN);
+		    BIF_RET(res);
+
+		} else if (ERTS_IS_ATOM_STR("process_locks", tp[1])) {
+		    erts_smp_proc_unlock(BIF_P, ERTS_PROC_LOCK_MAIN);
+		    erts_smp_block_system(0);
+		    if (tp[2] == am_true) {
+
+			res = erts_lcnt_set_rt_opt(ERTS_LCNT_OPT_PROCLOCK) ? am_true : am_false;
+
+		    } else if (tp[2] == am_false) {
+
+			res = erts_lcnt_set_rt_opt(ERTS_LCNT_OPT_PROCLOCK) ? am_true : am_false;
+
+		    } else {
+			erts_smp_release_system();
+			erts_smp_proc_lock(BIF_P, ERTS_PROC_LOCK_MAIN);
+			BIF_ERROR(BIF_P, BADARG);
+		    }
+		    erts_smp_release_system();
+		    erts_smp_proc_lock(BIF_P, ERTS_PROC_LOCK_MAIN);
+		    BIF_RET(res);
 		 }
 	    break;
      
@@ -3834,16 +3880,8 @@ BIF_RETTYPE erts_debug_lock_counters_1(BIF_ALIST_1)
 	}
     } 
 
-    erts_smp_release_system();
-    erts_smp_proc_lock(BIF_P, ERTS_PROC_LOCK_MAIN);
 #endif 
     BIF_ERROR(BIF_P, BADARG);
-#ifdef ERTS_ENABLE_LOCK_COUNT
-done:    
-    erts_smp_release_system();
-    erts_smp_proc_lock(BIF_P, ERTS_PROC_LOCK_MAIN);
-    BIF_RET(res);
-#endif 
 }
 
 void
diff --git a/erts/emulator/beam/erl_db.c b/erts/emulator/beam/erl_db.c
index b02150008f..e07a76835b 100644
--- a/erts/emulator/beam/erl_db.c
+++ b/erts/emulator/beam/erl_db.c
@@ -261,13 +261,8 @@ static ERTS_INLINE void db_init_lock(DbTable* tb, char *rwname, char* fixname)
     erts_refc_init(&tb->common.ref, 1);
     erts_refc_init(&tb->common.fixref, 0);
 #ifdef ERTS_SMP
-# ifdef ERTS_ENABLE_LOCK_COUNT
     erts_smp_rwmtx_init_x(&tb->common.rwlock, rwname, tb->common.the_name);
     erts_smp_mtx_init_x(&tb->common.fixlock, fixname, tb->common.the_name);
-# else
-    erts_smp_rwmtx_init(&tb->common.rwlock, rwname);
-    erts_smp_mtx_init(&tb->common.fixlock, fixname);
-# endif
     tb->common.is_thread_safe = !(tb->common.status & DB_FINE_LOCKED);
 #endif
 }
@@ -2597,19 +2592,11 @@ void init_db(void)
 
 #ifdef ERTS_SMP
     for (i=0; i<META_MAIN_TAB_LOCK_CNT; i++) {
-#ifdef ERTS_ENABLE_LOCK_COUNT
 	erts_smp_spinlock_init_x(&meta_main_tab_locks[i].lck, "meta_main_tab_slot", make_small(i));
-#else
-	erts_smp_spinlock_init(&meta_main_tab_locks[i].lck, "meta_main_tab_slot");
-#endif
     }
     erts_smp_spinlock_init(&meta_main_tab_main_lock, "meta_main_tab_main");
     for (i=0; i<META_NAME_TAB_LOCK_CNT; i++) {
-#ifdef ERTS_ENABLE_LOCK_COUNT
 	erts_smp_rwmtx_init_x(&meta_name_tab_rwlocks[i].lck, "meta_name_tab", make_small(i));
-#else
-	erts_smp_rwmtx_init(&meta_name_tab_rwlocks[i].lck, "meta_name_tab");
-#endif
     }
 #endif
 
diff --git a/erts/emulator/beam/erl_db_hash.c b/erts/emulator/beam/erl_db_hash.c
index dea45053df..dbe13204d9 100644
--- a/erts/emulator/beam/erl_db_hash.c
+++ b/erts/emulator/beam/erl_db_hash.c
@@ -628,11 +628,7 @@ int db_create_hash(Process *p, DbTable *tbl)
 							      (DbTable *) tb,
 							      sizeof(DbTableHashFineLocks));	    	    
 	for (i=0; i<DB_HASH_LOCK_CNT; ++i) {
-	    #ifdef ERTS_ENABLE_LOCK_COUNT	
-	    erts_rwmtx_init_x(&tb->locks->lck_vec[i].lck, "db_hash_slot", tb->common.the_name); 
-	    #else		
-	    erts_rwmtx_init(&tb->locks->lck_vec[i].lck, "db_hash_slot");
-	    #endif	
+	    erts_rwmtx_init_x(&tb->locks->lck_vec[i].lck, "db_hash_slot", make_small(i));
 	}
 	/* This important property is needed to guarantee that the buckets
     	 * involved in a grow/shrink operation it protected by the same lock:
diff --git a/erts/emulator/beam/erl_lock_count.c b/erts/emulator/beam/erl_lock_count.c
index 6211983f4b..32fddce38b 100644
--- a/erts/emulator/beam/erl_lock_count.c
+++ b/erts/emulator/beam/erl_lock_count.c
@@ -162,7 +162,6 @@ static void print_lock_x(erts_lcnt_lock_t *lock, Uint16 flag, char *action, char
     long int colls, tries, w_state, r_state;
     erts_lcnt_lock_stats_t *stats = NULL;
     
-    float rate;
     char *type;
     int i;
     
@@ -170,8 +169,6 @@ static void print_lock_x(erts_lcnt_lock_t *lock, Uint16 flag, char *action, char
     ethr_atomic_read(&lock->r_state, &r_state);
     ethr_atomic_read(&lock->w_state, &w_state);
 
-    if (tries > 0) rate = (float)(colls/(float)tries)*100;
-    else rate = 0.0f;
     
     if (lock->flag & flag) {
         erts_printf("%20s [%30s] [r/w state %4ld/%4ld] id %T %s\r\n", 
@@ -181,26 +178,13 @@ static void print_lock_x(erts_lcnt_lock_t *lock, Uint16 flag, char *action, char
 		w_state, 
 		lock->id, 
 		extra);
-	
-	for(i = 0; i < lock->n_stats; i++) {
-	    stats = &(lock->stats[i]);
-    	    ethr_atomic_read(&stats->tries, &tries);
-    	    ethr_atomic_read(&stats->colls, &colls);
-	    fprintf(stderr, "%69s:%5d [tries %9ld] [colls %9ld] [timer_n %8ld] [timer %4ld s %6ld us]\r\n",
-		    stats->file,
-		    stats->line,
-		    tries,
-		    colls,
-		    stats->timer_n,
-		    stats->timer.s,
-		    (unsigned long)stats->timer.ns/1000);
-	}
-	fprintf(stderr, "\r\n");
     }
 }
 
 static void print_lock(erts_lcnt_lock_t *lock, char *action) {
-    print_lock_x(lock, ERTS_LCNT_LT_ALL, action, "");
+    if (strcmp(lock->name, "proc_main") == 0) {
+        print_lock_x(lock, ERTS_LCNT_LT_ALL, action, "");
+    }
 }
 
 #endif
@@ -230,8 +214,8 @@ static void lcnt_update_stats(erts_lcnt_lock_stats_t *stats, int lock_in_conflic
     
     ethr_atomic_inc(&stats->tries);
 
-    /* beware of trylock */
-    if (lock_in_conflict) ethr_atomic_inc(&stats->colls);
+    if (lock_in_conflict)
+	ethr_atomic_inc(&stats->colls);
 
     if (time_wait) {
 	lcnt_time_add(&(stats->timer), time_wait);
@@ -366,6 +350,7 @@ void erts_lcnt_init_lock_x(erts_lcnt_lock_t *lock, char *name, Uint16 flag, Eter
     for (i = 0; i < ERTS_LCNT_MAX_LOCK_LOCATIONS; i++) {
 	lcnt_clear_stats(&lock->stats[i]);
     }
+
     erts_lcnt_list_insert(erts_lcnt_data->current_locks, lock);
     
     lcnt_unlock();
@@ -373,18 +358,20 @@ void erts_lcnt_init_lock_x(erts_lcnt_lock_t *lock, char *name, Uint16 flag, Eter
 
 void erts_lcnt_destroy_lock(erts_lcnt_lock_t *lock) {
     erts_lcnt_lock_t *deleted_lock;
-    
-    /* copy structure and insert the copy */
-    deleted_lock = (erts_lcnt_lock_t*)malloc(sizeof(erts_lcnt_lock_t));
 
     lcnt_lock();
 
-    memcpy(deleted_lock, lock, sizeof(erts_lcnt_lock_t));
-    deleted_lock->next = NULL;
-    deleted_lock->prev = NULL;
+    if (erts_lcnt_rt_options & ERTS_LCNT_OPT_COPYSAVE) {
+	/* copy structure and insert the copy */
+
+	deleted_lock = (erts_lcnt_lock_t*)malloc(sizeof(erts_lcnt_lock_t));
+	memcpy(deleted_lock, lock, sizeof(erts_lcnt_lock_t));
 
-    erts_lcnt_list_insert(erts_lcnt_data->deleted_locks, deleted_lock);
+	deleted_lock->next = NULL;
+	deleted_lock->prev = NULL;
 
+	erts_lcnt_list_insert(erts_lcnt_data->deleted_locks, deleted_lock);
+    }
     /* delete original */
     erts_lcnt_list_delete(erts_lcnt_data->current_locks, lock);
     
@@ -416,9 +403,10 @@ void erts_lcnt_lock_opt(erts_lcnt_lock_t *lock, Uint16 option) {
     /* we cannot acquire w_lock if either w or r are taken */
     /* we cannot acquire r_lock if w_lock is taken */	
     
-    if ((w_state > 0) || (r_state > 0)){
+    if ((w_state > 0) || (r_state > 0)) {
 	eltd->lock_in_conflict = 1;
-	if (eltd->timer_set == 0) lcnt_time(&eltd->timer);
+	if (eltd->timer_set == 0)
+	    lcnt_time(&eltd->timer);
 	eltd->timer_set++;
     } else {
 	eltd->lock_in_conflict = 0;
@@ -445,7 +433,8 @@ void erts_lcnt_lock(erts_lcnt_lock_t *lock) {
 	 * 'atomicly'. All other locks will block the thread if w_state > 0
 	 * i.e. locked.
 	 */
-	if (eltd->timer_set == 0) lcnt_time(&eltd->timer);
+	if (eltd->timer_set == 0)
+	    lcnt_time(&eltd->timer);
 	eltd->timer_set++;
 
     } else {
@@ -494,24 +483,23 @@ void erts_lcnt_lock_post_x(erts_lcnt_lock_t *lock, char *file, unsigned int line
     eltd = lcnt_get_thread_data();
     
     ASSERT(eltd);
-    
+
     /* if lock was in conflict, time it */
 	
     stats = lcnt_get_lock_stats(lock, file, line);
     
     if (eltd->timer_set) {
 	lcnt_time(&timer);
-
-	eltd->timer_set--;
 	
 	lcnt_time_diff(&time_wait, &timer, &(eltd->timer));
 	lcnt_update_stats(stats, eltd->lock_in_conflict, &time_wait);
 	
+	eltd->timer_set--;
 	ASSERT(eltd->timer_set >= 0);
     } else {
 	lcnt_update_stats(stats, eltd->lock_in_conflict, NULL);
     }
-	
+
 }
 
 /* unlock */
diff --git a/erts/emulator/beam/erl_lock_count.h b/erts/emulator/beam/erl_lock_count.h
index 8564c36203..86c85226e4 100644
--- a/erts/emulator/beam/erl_lock_count.h
+++ b/erts/emulator/beam/erl_lock_count.h
@@ -39,6 +39,20 @@
  * 	Each instance of a lock is the unique lock, i.e. set and id in that set.
  * 	For each lock there is a set of statistics with where and what impact
  * 	the lock aqusition had.
+ *
+ * 	Runtime options
+ * 	- suspend, used when internal lock-counting can't be applied. For instance
+ * 	  when allocating a term for the outside and halloc needs to be used.
+ * 	  Default: off.
+ * 	- location, reserved and not used.
+ * 	- proclock, disable proclock counting. Used when performance might be an
+ * 	  issue. Accessible from erts_debug:lock_counters({process_locks, bool()}).
+ * 	  Default: off.
+ * 	- copysave, enable saving of destroyed locks (and thereby its statistics).
+ * 	  If memory constraints is an issue this need to be disabled.
+ * 	  Accessible from erts_debug:lock_counters({copy_save, bool()}).
+ * 	  Default: off.
+ *
  */
 
 #include "sys.h"
@@ -74,6 +88,7 @@
 #define ERTS_LCNT_OPT_SUSPEND   (((Uint16) 1) << 0)
 #define ERTS_LCNT_OPT_LOCATION  (((Uint16) 1) << 1)
 #define ERTS_LCNT_OPT_PROCLOCK  (((Uint16) 1) << 2)
+#define ERTS_LCNT_OPT_COPYSAVE  (((Uint16) 1) << 3)
 
 typedef struct {
     unsigned long s;
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index 2789691c55..bf08bc7a86 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -2095,7 +2095,11 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online)
 	rq->ix = ix;
 	erts_smp_atomic_init(&rq->info_flags, ERTS_RUNQ_IFLG_NONEMPTY);
 
-	erts_smp_mtx_init(&rq->mtx, "run_queue");
+	/* make sure that the "extra" id correponds to the schedulers
+	 * id if the esdp->no <-> ix+1 mapping change.
+	 */
+
+	erts_smp_mtx_init_x(&rq->mtx, "run_queue", make_small(ix + 1));
 	erts_smp_cnd_init(&rq->cnd);
 
 	erts_smp_atomic_init(&rq->spin_waiter, 0);
diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c
index 61985271e6..34afe5656c 100644
--- a/erts/emulator/beam/io.c
+++ b/erts/emulator/beam/io.c
@@ -1227,7 +1227,7 @@ void init_io(void)
 	erts_smp_atomic_init(&erts_port[i].refc, 0);
 	erts_port[i].lock = NULL;
 	erts_port[i].xports = NULL;
-	erts_smp_spinlock_init(&erts_port[i].state_lck, "port_state");
+	erts_smp_spinlock_init_x(&erts_port[i].state_lck, "port_state", make_small(i));
 #endif
 	erts_port[i].tracer_proc = NIL;
 	erts_port[i].trace_flags = 0;