Re: flexfax: Large number of errors with Hylafax and ZyXEL Omni modem

[David Woolley <david@djwhome.demon.co.uk>]

You've probably already said this, but what operating system, what type
of serial interrface and what driver are you using with that interface?
If this is a PC type machine and a dumb serial interface, exactly which
chipset (not 16550, but the ASIC identity)?

The reason for asking is that I have fairly good circumstancial evidence
that the Linux standard serial driver is flawed, at least for some chipsets.
I've tried to raise this before, but I think most people are using error
corrected protocols and don't notice, whereas fax is less tolerant.  It
seems likely that many tpc.int Hylafax installations are Linux based.

The problem, I suspect, is that it treats 8250 type UARTs (emulation, not
chipset, now) as though they were any other type, and uses interrupt enable
to control the starting and stopping of output.  This is not the way 8250s were
designed to be used, and I have empirical evidence that they can lose or
corrupt characters under these conditions.  (On the other hand, I can think
of no internal design reason why they should and don't understand why the 
problem would apply to emulations as well as the real thing - unless the
emulations were based on the original logic design.)

I keep meaning to try changing the code.  It's fairly self contained, but
not so trivial that one can try it in an hour.

(If you also run UUCP over error correcting modems and get any errors at all,
but no overruns logged, this may be the same problem.  In the UUCP context,
Comtrol Rocket cards are clean.)

(8250s have edge triggered interrupts, so once the interrupt is acknowledged,
it goes away, whereas other designs will jam the interrupt line until a new
character is sent, or transmit interrupts are turned off.)