suspend processors while waiting

Before checking the atomic variable, disable IRQs to avoid missing the wake-up event.

https://marc.info/?l=linux-arm-kernel&m=114684008423863&w=2
https://electronics.stackexchange.com/questions/12601/best-pattern-for-wfi-wait-for-interrupt-on-cortex-arm-microcontrolers

add dsb before wfi

torvalds/linux@8553cb6
https://github.com/torvalds/linux/blob/master/arch/arm/mm/proc-v7.S
https://github.com/torvalds/linux/blob/master/arch/arm64/kernel/idle.c
torvalds/linux@9cce7a4

src/minidexed.cpp

@@ -431,13 +431,16 @@ void CMiniDexed::Run (unsigned nCore)
 		m_CoreStatus[nCore] = CoreStatusIdle;			// core 1 ready
 
 		// wait for cores 2 and 3 to be ready
+		DisableIRQs ();
 		for (unsigned nCore = 2; nCore < CORES; nCore++)
 		{
 			while (m_CoreStatus[nCore] != CoreStatusIdle)
 			{
-				// just wait
+				DataSyncBarrier ();
+				WaitForInterrupt ();
 			}
 		}
+		EnableIRQs ();
 
 		while (m_CoreStatus[nCore] != CoreStatusExit)
 		{
@@ -449,10 +452,15 @@ void CMiniDexed::Run (unsigned nCore)
 		while (1)
 		{
 			m_CoreStatus[nCore] = CoreStatusIdle;		// ready to be kicked
+			SendIPI (1, IPI_USER);
+
+			DisableIRQs ();
 			while (m_CoreStatus[nCore] == CoreStatusIdle)
 			{
-				// just wait
+				DataSyncBarrier ();
+				WaitForInterrupt ();
 			}
+			EnableIRQs ();
 
 			// now kicked from core 1
 
@@ -1230,6 +1238,7 @@ void CMiniDexed::ProcessSound (void)
 		{
 			assert (m_CoreStatus[nCore] == CoreStatusIdle);
 			m_CoreStatus[nCore] = CoreStatusBusy;
+			SendIPI (nCore, IPI_USER);
 		}
 
 		// process the TGs assigned to core 1
@@ -1241,13 +1250,16 @@ void CMiniDexed::ProcessSound (void)
 		}
 
 		// wait for cores 2 and 3 to complete their work
+		DisableIRQs ();
 		for (unsigned nCore = 2; nCore < CORES; nCore++)
 		{
 			while (m_CoreStatus[nCore] != CoreStatusIdle)
 			{
-				// just wait
+				DataSyncBarrier ();
+				WaitForInterrupt ();
 			}
 		}
+		EnableIRQs ();
 
 		//
 		// Audio signal path after tone generators starts here