Re: [avrdude-dev] avr910 speed-testing/tuning

[Joerg Wunsch]

As Jan-Hinnerk Reichert wrote:

> > The designers of UUCP already knew that request-response protocols
> > suck.  The designers of AVR910 apparently didn't have this 20-year
> > old knowledge...

> Ancient knowledge often gets lost ;-)

Some Atmel engineers seem to have a tendency to repeat history.  See
their use of the COFF object file format...

> What do you mean by disable? Don't rely on it or disable it?

Disable.

There's also some magic to dynamically enable it, but I don't really
grok what it does exactly.

> Under Linux you can disable the FIFO with "setserial" by selecting a 
> "16450"-UART. However, this really disables the FIFO in hardware, so 
> you get lost bytes at higher baudrates ;-(

The FreeBSD serial driver could stay up with 38400 bps links without
flow control, and with 115200 bps links with RTS/CTS flow control even
on a 386/16 machine with a non-FIFO UART.  Given that our current
machines are more than an order of magnitude faster, and that AVR910
only runs at slow baud rates anyway, I don't think that disabling the
FIFO will hurt as long as nothing else than an AVR910-style programmer
is connected to that port.

Of course, instead of using AVR910 protocol at 115200 bps, it would
make more sense to use STK500 protocol then.  AFAIK, there's currently
no opensource firmware solution for that, alas.

Anyway, yes, I agree that the FIFO issue with AVR910 ought to be
mentioned in the documentation.
-- 
J"org Wunsch                                           Unix support engineer
address@hidden        http://www.interface-systems.de/~j/