New Serial Problems (was Re: [Soekris] gotcha with the net4501 clock)

David Kelly dkelly at WarningSystems.com
Fri Feb 14 15:57:46 UTC 2003


On Friday, February 14, 2003, at 02:00 AM, phk at phk.freebsd.dk wrote:

> In message <4D297825-3F6B-11D7-9565-000393BB56F2 at WarningSystems.com>, 
> David Kel
>
>> Very interesting because "me too, the older 4501's work better." While
>> I "knew" the main crystal wasn't part of the problem, it was the only
>> visible difference between the working units and randomly failing
>> units. All the failing units had 33.3333 MHz main crystals. We tried
>> swapping a 32.768 MHz crystal from a working unit and the problem
>> stayed with the PCB. So its not some oddball thing happening in the
>> scheduler.
>
> The serial port is clocked from the 32.768 kHz "cmos" crystal.  If
> the PLL which runs the frequency up is unstable, the serial port
> would suffer, but I doubt that is the problem.

Yup. As I said, knew the MHz crystal was not supposed to apply to the 
serial port. But at the time we swapped crystals around I don't 
remember if we had caught it in the act of sending the previous 
character twice in place of a correct character.

Its "on my list" to make another foray into FreeBSD kernel-land. Have 
looked at code before but never got a big enough picture to know where 
I was. Once Upon A Time I was trying to understand a Crystal 
Semiconductor soundcard.

Looking at the Linux patch, appears a polling loop is implemented as a 
fix. Again, not having the big picture I couldn't tell if it 
busy-waited or rescheduled itself to check again in the near future.

--
David Kelly N4HHE, dkelly at interactplus.com, (256) 704-8774 x105
========================================================================
Whom computers would destroy, they must first drive mad.




More information about the Soekris-tech mailing list