5.1beta2 on Alpha ES40 w/ 32GB pys ram

[Bernd Walter]

On Wed, Jun 18, 2003 at 12:59:16PM -0400, Andrew Gallatin wrote:
> 
> Bernd Walter writes:
> 
>  > > How do you tell?  Is there someting in config space, or do you just
>  > > check the device's docs?
>  > 
>  > I've checked the memory base registers.
>  > A memory base register return bit set to zero 0 and bit2/1 tell
>  > the width (00 = 32bit, 10 = 64bit).
>  > I don't know if this is a shure indicator.
> 
> Doesn't this just mean that it can be the target for a DAC
> transaction?  That's not the same thing as being a master.  

Yes, but that's the best that I could find in the specs.
The exact information seems to be chip specific so it's left to the
driver and the bridges between.
PCI-PCI bridges are well defined.
The 2105x specs don't mention DAC - maybe the newer 2115x can.

> Our older Myrinet cards have a 32-bit memory base, but can be a 64-bit
> busmaster.  (ie, are what I consider to be DAC capable).  But maybe
> they are not typical..

I have no clue.

>  > But there are cards that definitive can't without replacement.
>  > E.g. USB OHCI controllers are defined with 32bit DMA registers.
> 
> I don't use USB on my alpha ;)  

Me too, but I do tests on alpha.

> My concern is that at least one cheap NIC, and one cheap SCSI
> controller is DAC capable.  I don't think anything else is really
> critical.  If we had that, and we could made busdma DAC aware, and we
> got the chipset goo right, we'd have an alpha which could be used with > 2GB.

Mmm...
sym defines nc_dnad64 in a header but it's not used anywhere.

-- 
B.Walter                   BWCT                http://www.bwct.de
ticso at bwct.de                                  info at bwct.de