Re: [avrdude-dev] Problem with avr910

[Jan-Hinnerk Reichert]

On Monday 16 February 2004 22:45, Charles Steinkuehler wrote:
> Charles Steinkuehler wrote:

> Sorry about replying to my own e-mail...I forgot to propose a fix.
>
> I'm not sure what kind of sync problems TRoth was having (I'm not
> seeing a similar problem), but the correct solution to a loss of
> sync should be sending a few 0x27 escape characters (which are
> ignored by the programmer if it's waiting for a command) and
> flushing anything the programmer sends back.

> If it is actually necessary to send the 'y' command for some
> reason, perhaps it could be followed by an escape, so programmers
> that expect a two byte LED command would still work?

Sorry, for this. I have disussed this issue with Ted _before_ the 
4.3.0 release, IIRC. He has suggested the ESC-solution as well.
However, I simply forgot about it...

Rethinking the problem, I might have found the cause for the trouble 
Ted was having. IMHO it is *not* sync-related. The original 
AVR910-code does only delay 9ms after chip erase. Even if we consider 
the transmission time of the answer and the system latency, it could 
be too short for chips requiring 20ms. The additional command _seems_ 
to fix it, but it just adds additional delay.

So, it would be cleaner (but not perfect) to add an usleep() to 
avr_chip_erase() and drop the whole pgm-write_setup(). Actually, we 
should need the usleep() to get the cycle-counter working on broken 
programmer-firmware ;-(

@Ted: What do you think?

Cheers
  Jan-Hinnerk