James Hendricks <[EMAIL PROTECTED]> writes:
> > I'll try to take a look as well. I occasionally retest on that
> > board to make certain everything hasn't regressed badly.
>
> Shoot.. Looks like we may be reviving our L440GX+ cluster for LinuxBIOS
> and SCYLD... This means I'll have to look at it in the next few weeks.
> As I recall:
>
> * We have SMP working and implemented
> * The framebuffer working and implemented (I should do a proper scan for
> it, which takes another 10 lines of code)
> * The L1 and L2 cache on the second processor is working, but not
> implemented (last I knew) -- sometimes the box will go very slowly because
> of this
> * The flash update from Linux is working, but only partially implemented
> and completely untested
>
> It looks like I need to do 3 changes, 1 trivial, 1 moderately difficult,
> and 1 difficult change (respectively). I hope to start work on this the
> 30th.. It looks like the ATI video driver will get delayed to Aug 14th..
>
> Eric, any progress you get on this would be greatly appreciated! Also,
> more importantly, where should second processor init code be located? In
> particular, where should I put the second processor's L1/L2 cache init
> (did you have to do this or does the AMD not suck and just start with L1 &
> L@ on?). If nobody cares, I'll choose.
The Athlons come up with caches ``enabled''. But until you set some
registers you can't access ram. Also both processors start out
running instead of the intel case where all but one processor is
initially stopped. So I have to handle the second processor in
linuxBIOS.
Most of the code is already in. Please take a look at my latest diff
on ftp://download.linuxnetorx.com/pub/src/linuxBIOS. Everything that
must be there is there, but what I'm doing in C code is a little
lacking.
So I would say we should put the code in linuxBIOS as it is already
there, plus this makes using prexisting kernels easier. We should
make it a build option so you don't always have the code compiled in.
The biggest reason for putting in linuxBIOS though is you can get some
really weird things happening when you don't have both processors
initialized. My favorite is DMA transactions lockup up the entire
machine because the processor busses aren't initialized correctly.
Eric
