[#420] In SimpleThreadAPI mode, forward signal to ydb engine mutex lo…

…ck holder thread even in TP (or else we call to ydb_ci_t that does a HANG command while inside TP could nhang eternally because SIGALRM gets delivered to MAIN worker thread which has no way of interrupting the ydb engine lock holder thread which could be stuck in a sigsuspend() call waiting for SIGALRM)

sr_unix/sig_init.h

@@ -245,7 +245,7 @@ GBLREF	sig_info_context_t	stapi_signal_handler_oscontext[sig_hndlr_num_entries];
 	GBLREF	boolean_t	gtm_jvm_process;										\
 																\
 	pthread_t		mutex_holder_thread_id, this_thread_id;								\
-	int			i;												\
+	int			i, tLevel;											\
 	boolean_t		signalForwarded;										\
 																\
 	if (DUMMY_SIG_NUM != SIG)												\
@@ -272,21 +272,19 @@ GBLREF	sig_info_context_t	stapi_signal_handler_oscontext[sig_hndlr_num_entries];
 			 * current lock holder thread if it is non-zero). But if we are in TP, then lock index could be		\
 			 * "dollar_tlevel"     : e.g. if a "ydb_get_st" call occurs inside of the "ydb_tp_st" call OR		\
 			 * "dollar_tlevel - 1" : if control is in the TP callback function inside "ydb_tp_st" but not a		\
-			 *	SimpleThreadAPI call like "ydb_get_st" etc. In this latter case, it is possible any		\
-			 *	number of threads could get the dollar_tlevel index lock concurrently so we cannot		\
-			 *	be sure about signal forwarding. Therefore forward only in non-TP. In case of TP,		\
-			 *	"timer_handler" will be invoked just "ydb_tp_st" is done finishing one TP level			\
-			 *	when it does the THREADED_API_YDB_ENGINE_UNLOCK call. Therefore it is okay to skip		\
-			 *	signal forwarding in the TP case.								\
+			 *	SimpleThreadAPI call like "ydb_get_st" etc.							\
 			 */													\
-			if (dollar_tlevel)											\
-			{													\
-				STAPI_SET_SIGNAL_HANDLER_DEFERRED(SIGHNDLRTYPE, SIG, INFO, CONTEXT);				\
-				return;												\
-			}													\
 			this_thread_id = pthread_self();									\
 			assert(this_thread_id);											\
-			mutex_holder_thread_id = ydb_engine_threadsafe_mutex_holder[0];						\
+			tLevel = dollar_tlevel;	/* take a local copy of global variable as it could be concurrently changing */	\
+			if (!tLevel)												\
+				mutex_holder_thread_id = ydb_engine_threadsafe_mutex_holder[0];					\
+			else													\
+			{													\
+				mutex_holder_thread_id = ydb_engine_threadsafe_mutex_holder[tLevel];				\
+				if (!mutex_holder_thread_id)									\
+					mutex_holder_thread_id = ydb_engine_threadsafe_mutex_holder[tLevel - 1];		\
+			}													\
 			if (!pthread_equal(mutex_holder_thread_id, this_thread_id))						\
 			{	/* It is possible we are the MAIN worker thread that gets the SIGALRM signal but		\
 				 * another thread that is holding the YottaDB engine lock is waiting for the signal		\