Jonathan Stone <jonathan@DSG.Stanford.EDU> writes:

> 
> Is anyone testing the new com driver on  FIFO-less chips (16450s, 8250s)?

Yes.  On a fairly slow 486SL, I can do 57.6k reliably, but get a few
silo overflows at 115.2k with no FIFO.  I'll probably improve this a
bit.

With FIFOs, I get no overflows, even during disk access.  What I do
get is tty buffer overflows, because kermit takes too long to read the
data while it's writing to disk.  You shouldn't notice this with PPP
(and of course enabling a flow control mode should prevent losing
data).

There are also a few things related to buffers filling up and
asserting backpressure that aren't quite finished yet, although this
shouldn't crop up in practice unless you have a lot of ports.


P.S.  Actually, I had to disassemble a DOS configuration program to
figure out how to change the UART emulation mode on my test machine,
in order to get FIFOs at all!  Just say `NO!' to bizzare multi-I/O
chips.