[#169] STM32: make UART thread safe #27
Conversation
|
|
Viatorus
changed the title
Viatorus
changed the title
| if(value == -1) | ||
| { | ||
| Logger.LogAs(this, LogLevel.Debug, "Client disconnected, stream closed."); | ||
| writerCancellationToken.Cancel(); | ||
| break; | ||
| } | ||
|
|
||
| var dataReceived = DataReceived; |
There was a problem hiding this comment.
FYI:
I changed the order, otherwise the value -1 would be used as dataReceived!
| receiveFifo.Enqueue(value); | ||
| readFifoNotEmpty.Value = true; | ||
| Update(); | ||
| lock (receiveFifo) |
There was a problem hiding this comment.
FYI: The queue was used in two threads but never locked. This led to crashes.
| (transmitDataRegisterEmptyInterruptEnabled.Value) || // TXE is assumed to be true | ||
| (transmissionCompleteInterruptEnabled.Value && transmissionComplete.Value) | ||
| ); | ||
| machine.ClockSource.ExecuteInLock(() => |
There was a problem hiding this comment.
FYI:
Here is a possible dead lock. I don't like my fix but it worked for me.
machine thread is inside BaseClockSource (locked) and tries to lock the NVIC.
_Reader_Thread updates the NVIC (locked) and then tries to lock the BaseClockSource which is locked by machine thread.
There was a problem hiding this comment.
@Viatorus Sorry, How do you generate this figure? Could you please tell me the name of the tool?
There was a problem hiding this comment.
It's a debug tool from Rider: https://www.jetbrains.com/help/rider/Debugging_Multithreaded_Applications.html#parallel-stacks
| .WithFlag(5, out readFifoNotEmpty, FieldMode.Read | FieldMode.WriteZeroToClear, name: "RXNE") // as these two flags are WZTC, we cannot just calculate their results | ||
| .WithFlag(5, FieldMode.Read | FieldMode.WriteZeroToClear, name: "RXNE", valueProviderCallback: _ => | ||
| { | ||
| return readFifoNotEmpty; |
There was a problem hiding this comment.
FYI:
This workaround I don't understand my self but otherwise, it is not working.
if readFifoNotEmpty is a IFlagRegisterField, it happens that the machine does not see the change and the application stucks.
If we put inside the callback function, this is not happening anymore.
|
|
||
| machine.ClockSource.ExecuteInLock(() => | ||
| { | ||
| if (receiverNotEmptyInterruptEnabled.Value) |
There was a problem hiding this comment.
FYI:
So there is the case, that the machine is not yet done with the last interrupt handling:
receiverNotEmptyInterruptEnabled is not yet set to true but transmissionCompleteInterruptEnabled and transmissionComplete are so the Update() method would wrongly set IRQ to true again.
If this happens, the IRQ hangs somehow.
Overall I am not a fan of updating the IRQ from a different thread than the machine thread.
renode/renode#169