On May 4, 2005, at 7:28 AM, Gavan Fantom wrote:

> For those folks with non-booting EPIA boards, could you please try 
> this patch?
>
> http://frog.coolfactor.org/~gavan/epia_hack.diff
>
> It works on my ML5000, and I'd like to get some data on whether it 
> works for other people, too.
>
Is an EPIA with a C3 or an EDEN ?  The Eden problem described in list 
before
was exactly the same the problem I had with my C3 / Ezra.. Video going 
out of range
etc,

So, I tried the patch, tested via building and using dosboot, no luck 
though.

> Also, here's a challenge. If you can come up with a satisfactory 
> explanation of *why* this patch works, and some experiments to 
> demonstrate the correctness of the explanation, you get a cookie.
>
> My assessment is that this is a CPU bug, and that this patch is 
> flushing internal buffers that the ljmp should have done. Again, why 
> the 1.6 bootloader didn't hit this bug, I don't know.
>
> Any theories or opinions welcome!
Well, I noticed that from 1.6->2.x eax is preserved, but in some places 
in 2.x the
caller still saves it.

gcc 2.95.3 , as 2.11.2 in 1.6 vs gcc 3.3.3 as 2.14 in 2.x
producing different code, patch causes gcc to produce code that avoids 
bug

>
> (please CC me on replies, as I'm not currently subscribed to port-i386)
>
> -- 
> Gilette - the best a man can forget
>