ETS busy wait option

Conflicts:
	erts/emulator/beam/erl_init.c
	erts/etc/common/erlexec.c

4 files changed, 92 insertions, 7 deletions

diff --git a/erts/emulator/beam/erl_db.c b/erts/emulator/beam/erl_db.c
index 2e2cb98354..844272c699 100644
--- a/erts/emulator/beam/erl_db.c
+++ b/erts/emulator/beam/erl_db.c
@@ -279,6 +279,8 @@ static ERTS_INLINE void db_init_lock(DbTable* tb, int use_frequent_read_lock,
     erts_smp_rwmtx_opt_t rwmtx_opt = ERTS_SMP_RWMTX_OPT_DEFAULT_INITER;
     if (use_frequent_read_lock)
 	rwmtx_opt.type = ERTS_SMP_RWMTX_TYPE_FREQUENT_READ;
+    if (erts_ets_rwmtx_spin_count >= 0)
+	rwmtx_opt.main_spincount = erts_ets_rwmtx_spin_count;
 #endif
 #ifdef ERTS_SMP
     erts_smp_rwmtx_init_opt_x(&tb->common.rwlock, &rwmtx_opt,
@@ -2856,10 +2858,11 @@ BIF_RETTYPE ets_match_spec_run_r_3(BIF_ALIST_3)
 ** External interface (NOT BIF's)
 */
 
+int erts_ets_rwmtx_spin_count = -1;
 
 /* Init the db */
 
-void init_db(void)
+void init_db(ErtsDbSpinCount db_spin_count)
 {
     DbTable init_tb;
     int i;
@@ -2868,10 +2871,48 @@ void init_db(void)
     size_t size;
 
 #ifdef ERTS_SMP
+    int max_spin_count = (1 << 15) - 1; /* internal limit */
     erts_smp_rwmtx_opt_t rwmtx_opt = ERTS_SMP_RWMTX_OPT_DEFAULT_INITER;
     rwmtx_opt.type = ERTS_SMP_RWMTX_TYPE_FREQUENT_READ;
     rwmtx_opt.lived = ERTS_SMP_RWMTX_LONG_LIVED;
 
+    switch (db_spin_count) {
+    case ERTS_DB_SPNCNT_NONE:
+	erts_ets_rwmtx_spin_count = 0;
+	break;
+    case ERTS_DB_SPNCNT_VERY_LOW:
+	erts_ets_rwmtx_spin_count = 100;
+	break;
+    case ERTS_DB_SPNCNT_LOW:
+	erts_ets_rwmtx_spin_count = 200;
+	erts_ets_rwmtx_spin_count += erts_no_schedulers * 50;
+	if (erts_ets_rwmtx_spin_count > 1000)
+	    erts_ets_rwmtx_spin_count = 1000;
+	break;
+    case ERTS_DB_SPNCNT_HIGH:
+	erts_ets_rwmtx_spin_count = 2000;
+	erts_ets_rwmtx_spin_count += erts_no_schedulers * 100;
+	if (erts_ets_rwmtx_spin_count > 15000)
+	    erts_ets_rwmtx_spin_count = 15000;
+	break;
+    case ERTS_DB_SPNCNT_VERY_HIGH:
+	erts_ets_rwmtx_spin_count = 15000;
+	erts_ets_rwmtx_spin_count += erts_no_schedulers * 500;
+	if (erts_ets_rwmtx_spin_count > max_spin_count)
+	    erts_ets_rwmtx_spin_count = max_spin_count;
+	break;
+    case ERTS_DB_SPNCNT_EXTREMELY_HIGH:
+	erts_ets_rwmtx_spin_count = max_spin_count;
+	break;
+    case ERTS_DB_SPNCNT_NORMAL:
+    default:
+	erts_ets_rwmtx_spin_count = -1;
+	break;
+    }
+
+    if (erts_ets_rwmtx_spin_count >= 0)
+	rwmtx_opt.main_spincount = erts_ets_rwmtx_spin_count;
+
     meta_main_tab_locks =
 	erts_alloc_permanent_cache_aligned(ERTS_ALC_T_DB_TABLES,
 					   sizeof(erts_meta_main_tab_lock_t)
diff --git a/erts/emulator/beam/erl_db.h b/erts/emulator/beam/erl_db.h
index 5b4681fc90..5039b71108 100644
--- a/erts/emulator/beam/erl_db.h
+++ b/erts/emulator/beam/erl_db.h
@@ -61,7 +61,17 @@ union db_table {
 				"ERL_MAX_ETS_TABLES" */
 #define ERL_MAX_ETS_TABLES_ENV "ERL_MAX_ETS_TABLES"
 
-void init_db(void);
+typedef enum {
+    ERTS_DB_SPNCNT_NONE,
+    ERTS_DB_SPNCNT_VERY_LOW,
+    ERTS_DB_SPNCNT_LOW,
+    ERTS_DB_SPNCNT_NORMAL,
+    ERTS_DB_SPNCNT_HIGH,
+    ERTS_DB_SPNCNT_VERY_HIGH,
+    ERTS_DB_SPNCNT_EXTREMELY_HIGH
+} ErtsDbSpinCount;
+
+void init_db(ErtsDbSpinCount);
 int erts_db_process_exiting(Process *, ErtsProcLocks);
 void db_info(int, void *, int);
 void erts_db_foreach_table(void (*)(DbTable *, void *), void *);
@@ -69,6 +79,7 @@ void erts_db_foreach_offheap(DbTable *,
 			     void (*func)(ErlOffHeap *, void *),
 			     void *);
 
+extern int erts_ets_rwmtx_spin_count;
 extern int user_requested_db_max_tabs; /* set in erl_init */
 extern int erts_ets_realloc_always_moves;  /* set in erl_init */
 extern int erts_ets_always_compress;  /* set in erl_init */
diff --git a/erts/emulator/beam/erl_db_hash.c b/erts/emulator/beam/erl_db_hash.c
index 383ee7c430..f6ea57e513 100644
--- a/erts/emulator/beam/erl_db_hash.c
+++ b/erts/emulator/beam/erl_db_hash.c
@@ -670,6 +670,8 @@ int db_create_hash(Process *p, DbTable *tbl)
 	int i;
 	if (tb->common.type & DB_FREQ_READ)
 	    rwmtx_opt.type = ERTS_SMP_RWMTX_TYPE_FREQUENT_READ;
+	if (erts_ets_rwmtx_spin_count >= 0)
+	    rwmtx_opt.main_spincount = erts_ets_rwmtx_spin_count;
 	tb->locks = (DbTableHashFineLocks*) erts_db_alloc_fnf(ERTS_ALC_T_DB_SEG, /* Other type maybe? */ 
 							      (DbTable *) tb,
 							      sizeof(DbTableHashFineLocks));	    	    
diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c
index 574a49dc7a..c676682463 100644
--- a/erts/emulator/beam/erl_init.c
+++ b/erts/emulator/beam/erl_init.c
@@ -142,7 +142,8 @@ static void erl_init(int ncpu,
 		     int legacy_port_tab,
 		     int time_correction,
 		     ErtsTimeWarpMode time_warp_mode,
-		     int node_tab_delete_delay);
+		     int node_tab_delete_delay,
+		     ErtsDbSpinCount db_spin_count);
 
 static erts_atomic_t exiting;
 
@@ -316,7 +317,8 @@ erts_short_init(void)
 	     0,
 	     time_correction,
 	     time_warp_mode,
-	     ERTS_NODE_TAB_DELAY_GC_DEFAULT);
+	     ERTS_NODE_TAB_DELAY_GC_DEFAULT,
+	     ERTS_DB_SPNCNT_NORMAL);
     erts_initialized = 1;
 }
 
@@ -329,7 +331,8 @@ erl_init(int ncpu,
 	 int legacy_port_tab,
 	 int time_correction,
 	 ErtsTimeWarpMode time_warp_mode,
-	 int node_tab_delete_delay)
+	 int node_tab_delete_delay,
+	 ErtsDbSpinCount db_spin_count)
 {
     init_benchmarking();
 
@@ -369,7 +372,7 @@ erl_init(int ncpu,
     erts_ptab_init(); /* Must be after init_emulator() */
     erts_init_binary(); /* Must be after init_emulator() */
     erts_bp_init();
-    init_db(); /* Must be after init_emulator */
+    init_db(db_spin_count); /* Must be after init_emulator */
     erts_init_node_tables(node_tab_delete_delay);
     init_dist();
     erl_drv_thr_init();
@@ -635,6 +638,10 @@ void erts_usage(void)
     erts_fprintf(stderr, "-zdntgc time   set delayed node table gc in seconds\n");
     erts_fprintf(stderr, "               valid values are infinity or intergers in the range [0-%d]\n",
 		 ERTS_NODE_TAB_DELAY_GC_MAX);
+#if 0
+    erts_fprintf(stderr, "-zebwt  val    set ets busy wait threshold, valid values are:\n");
+    erts_fprintf(stderr, "               none|very_short|short|medium|long|very_long|extremely_long\n");
+#endif
     erts_fprintf(stderr, "\n");
     erts_fprintf(stderr, "Note that if the emulator is started with erlexec (typically\n");
     erts_fprintf(stderr, "from the erl script), these flags should be specified with +.\n");
@@ -1226,6 +1233,7 @@ erl_start(int argc, char **argv)
     int time_correction;
     ErtsTimeWarpMode time_warp_mode;
     int node_tab_delete_delay = ERTS_NODE_TAB_DELAY_GC_DEFAULT;
+    ErtsDbSpinCount db_spin_count = ERTS_DB_SPNCNT_NORMAL;
 
     set_default_time_adj(&time_correction,
 			 &time_warp_mode);
@@ -2040,6 +2048,28 @@ erl_start(int argc, char **argv)
 		    }
 		}
 		node_tab_delete_delay = (int) secs;
+	    }
+	    else if (has_prefix("ebwt", sub_param)) {
+		arg = get_arg(sub_param+4, argv[i+1], &i);
+		if (sys_strcmp(arg, "none") == 0)
+		    db_spin_count = ERTS_DB_SPNCNT_NONE;
+		else if (sys_strcmp(arg, "very_short") == 0)
+		    db_spin_count = ERTS_DB_SPNCNT_VERY_LOW;
+		else if (sys_strcmp(arg, "short") == 0)
+		    db_spin_count = ERTS_DB_SPNCNT_LOW;
+		else if (sys_strcmp(arg, "medium") == 0)
+		    db_spin_count = ERTS_DB_SPNCNT_NORMAL;
+		else if (sys_strcmp(arg, "long") == 0)
+		    db_spin_count = ERTS_DB_SPNCNT_HIGH;
+		else if (sys_strcmp(arg, "very_long") == 0)
+		    db_spin_count = ERTS_DB_SPNCNT_VERY_HIGH;
+		else if (sys_strcmp(arg, "extremely_long") == 0)
+		    db_spin_count = ERTS_DB_SPNCNT_EXTREMELY_HIGH;
+		else {
+		    erts_fprintf(stderr,
+				 "Invalid ets busy wait threshold: %s\n", arg);
+		    erts_usage();
+		}
 	    } else {
 		erts_fprintf(stderr, "bad -z option %s\n", argv[i]);
 		erts_usage();
@@ -2114,7 +2144,8 @@ erl_start(int argc, char **argv)
 	     legacy_port_tab,
 	     time_correction,
 	     time_warp_mode,
-	     node_tab_delete_delay);
+	     node_tab_delete_delay,
+	     db_spin_count);
 
     load_preloaded();
     erts_end_staging_code_ix();