magma@4800 => ttya@9600

CR: 0x0d -> 0xc7  0101100001  -> 0111000111
 0: 0x30 -> 0xc8  0000011001  -> 0000100111
 1: 0x31 -> 0xc9  0100011001  -> 0100100111
 2: 0x32 -> 0xc8  0010011001  -> 0000100111
 3: 0x33 -> 0xc9  0110011001  -> 0100100111
 4: 0x34 -> 0xca  0001011001  -> 0110100111
 5: 0x35 -> 0xcb  0101011001  -> 0001100111
 6: 0x36 -> 0xcc  0011011001  -> 0101100111
 7: 0x37 -> 0xcd  0111011001  -> 0011100111
 8: 0x38 -> 0xde  0000111001  -> 0011110111
 9: 0x39 -> 0xc6  0100111001  -> 0011000111

magma@2400 => ttya@9600
CR: 0x0d -> 0xf8  0101100001  -> 0000111111
<didn't bother going further, this seems to be too far>

Before a wife and kid I would rip this open -- kernel drivers were my
"thing" a couple years ago.  I'll have to see what I can do to get a
handle on this.

> > Ideas?
> 
> Yes, the clock constant used for the baud rate generator is wrong.
> 
> Given an 8-bit byte HGFEDCBA (i.e. H is bit 7, G is bit 6, etc.),
> a serial chip will transmit 0ABCDEFGH1 where 0 and 1 are start and stop
> bits (with value 0 and 1 respectivly). Normally the line is transmitting a
> continuoys 1.
> 
>    -1: 0x0d -> 0xfe  0101100001  -> 00111111111
> >   0: 0x30 -> 0xfa  0000011001  -> 00101111111
> >   1: 0x31 -> 0xfd  0100011001  -> 01011111111
> >   2: 0x32 -> 0xfa  0010011001  -> 00101111111
> >   3: 0x33 -> 0xff  0110011001  -> 01111111111
> >   4: 0x34 -> 0xfb  0001011001  -> 01101111111
> >   5: 0x35 -> 0xff  0101011001  -> 01111111111
> >   6: 0x36 -> 0xfb  0011011001  -> 01101111111
> >   7: 0x37 -> 0xff  0111011001  -> 01111111111
> >   8: 0x38 -> 0xfa  0000111001  -> 00101111111
> >   9: 0x39 -> 0xfd  0100111001  -> 01011111111
> 
> For all of them except for 0x33, the pattern is 00XAXBXCXDXEXFXGXH where X
> is don't care (0x33 would have needed to be 0xfd).
> 
> Looks like your serial port is running at twice the rate you expect. Try
> different baud rates and see what happens. Like ask it to run at 1/2 the
> rate you want. :-)
>