Issue #096 [New/open] - Agree on delivered values when reading an GPIO output channel

Posted by andreas.riegg

Some chips have an electronic architecture for their digital ports that can lead to different values when reading the value of an output port depending on whats going on behind the pin. Mostly this is the case with flip-flop driven open drain ports that invert the signal. Examples for such chips are the DS2408 and the DS2413, others may also exist.

To change the output, some value is written direct to the output latches. This value gets inverted by the latch flip-flop and then drives the port FET. The sensed input level is not measured by taking the current value of the latches, it is mesured at the open drain position. This leads to the situation that the values of the latch and those at the open drain sensed level may be seldom but still sometime different depending on whether weak pullups or pulldowns are used or othr kinds of driving components.

Aside those electronic details, the simple question at the bottom line now is:

If a digital port is configured as an output and reading methods are called (e.g. digitalRead or portRead), which values have then to be sent back? The values of the output latches or the sensed levels of the pins? And, will there be a difference between reading the whole port at once with portRead and reading only a single channel with digitalRead. Plus, keep in mind the situation where you have a mixture of input and output ports (e.g. 3 inputs and 5 outputs), what does readPort deliver then?

Last point, many drivers read the status of all output latches (as byte value) when they handle digitalWrite, mask the channel output value bit to be updated into the byte and write this byte back to the output latches. However, if the read values are not taken from the latches but form the sensed level and if for some reason the sensed level is different to the current latch level, this may lead to an unwanted accidential update of an other port. To avaid this there is a clear reason to be able to read the output latches and many chips give direct access to those latches by special registers.

I don't have an answer to this right now, but when looking at drivers code I have the impression that this is not handeled unique in all drivers. So best would be to agree on what way to follow and check all digital port drivers if they implement this way. My proposal would be to deliver the values of the output latches back when reading an output port and not the sensed levels. But, if there is a mixed input/output configuration, this will mean that portRead will have to mix the values form sensing for inputs and latches for outputs to be consistent.

Hope you understand my question, maybe a more detailed discussion is necessary in this topic.


Comment 1

Posted by andreas.riegg

In the meanwhile I finished intensive work and testing with the DS2408/DS2413 driver as well with the upcoming PCA9698 and PCA9530 drivers.

I came to the conclusion that the following strategy is best for getting most functionality and stability for digital I/O drivers:

I have implemented this concept within the drivers above, you find all generic code for the input/output masking e.g. in the PCA9698 class. Maybe that code should make its way up to __init__ to avoid copied code, I did this not for the moment.

In order to have this concept used on all digital I/O drivers it will be necessary to update the MCPxxxx driver and use the output latch registers of those chips as well. As I have a MCP23008 and a MCP23017 at hand, I may be able to do and test this, the SPI chips testing would have to be provided by someone else.

The PCF8574 seems to need no update as it has quasi-bidirectional ports and has no output latch registers that could be used.