Interrupt Driven PSP Port Communication Problems

[bfemmel]

I am having problems with communications over the PSP port. We use it to communicate between a Windows CE system and a PIC18F6520. I recently rewrote the PSP routines to be interrupt driven since we were having problems with the previous communications routines after we revised the CE code from VB to C++. The C++ code was much faster than the VB code and showed a bug in the code prompting the rewrite.

The first problem was that we started missing bytes and having the communications hang. It seems that the CE code was so fast that if the PIC received or sent a byte and then set the external ready bit to signal the CE computer that the CE computer may have serviced the PSP port again before the PIC had left the interrupt routine. That at first did not sound like a problem to me, I just figured that the PIC would exit the interrupt and reenter right away to service the port again. But this was not happening. When I checked the Assembly listing I noticed that the code clears the PSP interrupt bit at the end of the routine right before it jumps back to the interrupt handler:

BCF F9E.7
GOTO 0084

This seems a bit strange to me. Is there a simple way around this? I would think that the interrupt should be cleared on entering the routine not on exiting. Right now I am setting a flag in the INT _PSP instead of setting the external PIC_Ready bit and then setting that external signal in my main loop outside of the interrupt. Of course this makes the PSP interrupt kind of a mute thing. I would do just as well with a polled routine because of that problem.

The second problem, and one that I have not fixed yet, is that even after we fixed the first problem, we still get a lot of errors in the communications, usually bad data in the command or status response bytes. We can lessen the errors by sprinkling the C++ code with 1 millisecond sleep statements but why do we need to do that?

Here is an example of the C++ Code that reads from and writes to the PIC without any sleep statements. The 0x188 port address is the external PIC Ready bit which is cleared in hardware whenever a read or write is done to port 0x180 which is the PSP port of the PIC. When we insert sleep(1) statements in between all the reads and writes we end up with communications that is a lot more reliable.

[PCM programmer]

[bfemmel]

PCM programmer;

Thanks for the tip on instructing the compiler to not clear the int flag. I added the NOCLEAR option to the interrupt declaration and then added a bit clear to the beginning of the routine like this;