SDRAM row page size wierdness

[Eric W. Biederman]

Richard Smith <rsmith at bitworks.com> writes:

> Richard Smith wrote:
> 
> Ok...  I got the page size detection worked out.  I'm still confused but at
> least it works.
> 
> Going back over the original  440BX code I noticed that it used the number of
> _column_ address rather than the number of row adresses.  At first I thought
> this was a typo (spd loc 4 rather than 3).  This was further reenforced by the
> 440gx code and Eriks general code which use the number of row adress bits.

My code was wrong.
 
> Sounds logical... The number of row adress bits indicating row page size but...

The number of column address bits determine the size of a row.
the number of row address bits determine the number of rows you have.
> 
> After testing several DIMMs using the row address bits failed in about 50% of
> the cases so I started looking back at the column address bits. I don't
> understand why but if you use the number of column address bits as a index for
> what page size to use it works great.
> 
> Basically just subtract 8 from the value to shift it down to zero.
> 8 column bits = 2k page
> 9 = 4k
> 10 = 8k
> 
> The 440bx dosen't do beyond 8k pages.
> 
> I tested as many different DIMMs as I could find here and they all work.
> 
> Any ideas on why this would be the case?

See above.  I think I spotted this a long time ago, but forget to do
anything about it as I don't have a 440 board.

Eric